Sundhedsudvalget 2020-21
L 61 Bilag 1
Offentligt
2259709_0001.png
of the 2012 Packaging Changes. A more meaningful way to address this issue is by
focusing on the pooled agreement responses (“strongly agree” and “agree”) and pooled
disagreement responses (“strongly disagree” and “disagree”). Thus, while I will report
the distribution of the findings with respect to various statements regarding cigarettes, my
emphasis will be on the pooled set of responses who indicate agreement in terms of either
“strongly agree” or “agree.” Similarly, I will pool the disagreement responses for whether
the respondent indicates “strongly disagree” or “disagree” with the various statements.
40.
I divide the survey questions into three groups. The first set of questions explores
consumption-related questions pertaining to quit behavior and smoking. Whether the
2012 Packaging Changes are associated with actual smoking behaviors is the
fundamental policy issue. The next set of questions consists of belief and smoking
attitude questions. The overwhelming result is that there is no evidence that the 2012
Packaging Changes policy has succeeded in any of these dimensions. I then address the
questions pertaining to pack appearance. These are the questions that have received the
greatest emphasis in the published articles on the 2012 Packaging Changes, perhaps
because they are most consistent with some authors’ efforts to support the policy.
However, the absence of any impact of the 2012 Packaging Changes on the belief and
behavior responses suggests that even though the trademarks and the brands they
represent have been removed from cigarette packaging, that change has not advanced
more fundamental smoking policy objectives. Unlike Dunlop et al. (2014) that does not
consider the broad set of survey questions in the CITTS data that I analyze here, the
39
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0002.png
tables below and in Appendix B provide a more comprehensive and balanced perspective
on the implications of the data from the post-2012 Packaging Changes era.
36
B.
41.
Smoking and Quit Behavior in the CITTS Data
The first set of regression results reported in Appendix B estimates the effect of the 2012
Packaging Changes with the number of cigarettes smoked, controlling for the other
variables and the time trend variables listed in the tables. Tables B1 and B2 report similar
results for different specifications of the policy time period. These regressions pertain to
daily, weekly, and occasional smokers.
37
The table below reports only on the estimates of
the 2012 Packaging Changes variable in a large series of regressions for both the full
sample in the first column and excluding quitters in the second column. I exclude quitters
in some instances both because the questions may not be pertinent to quitters and because
it is also instructive to analyze the results for the current smoking population. Although
only the 2012 Packaging Changes coefficients are reported, the regressions included the
complete set of variables in the regressions reported in Appendix B. There is a
statistically significant increase in the number of cigarettes smoked both for the sample
overall and excluding quitters in the second column. Total cigarettes smoked rose after
the advent of the 2012 Packaging Changes policy by 0.9 cigarettes per day overall, and
by 1.4 cigarettes per day excluding quitters. That cigarettes per day increased even while
36
37
Multiple regression analysis makes it possible to estimate statistically the effect of the PP requirements
controlling for other factors, in particular, the shift in the sample recruitment to include mobile phone users
and the existing trends in smoking-related attitudes and behaviors. The regression coefficients have direct
interpretations. For 0-1 questions such as whether the respondent plans to quit in the next month, the
estimates are probit regressions where the coefficients have been transformed to correspond to marginal
effects. Probit regressions are appropriate when the dependent variable is dichotomous rather than
continuous. The other equations are estimated using ordinary least squares regressions.
Daily and weekly smokers are measured by self-reported number of cigarettes smoked per day. Monthly or
less frequent smokers are deemed to smoke zero cigarettes per day.
40
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0003.png
including recent quitters indicates that the rise in cigarettes per day is not because the
2012 Packaging Changes reduced the smoking rates of those who had less intensive
smoking histories, thus altering the mix of smokers. These results are consistent for each
possible marker for policy implementation, either full implementation from December 1,
2012 as shown here, partial implementation from October 1, 2012, or full implementation
excluding partial observations. The latter two results are included as Appendix Tables B1
and B2.
Regressions, Reporting Only 2012 Packaging Changes Coefficient
Variable Name
Number of cigarettes per day (OLS)
Plan to quit in the next month (yes/no)
Smoker: Daily
Smoker: At least weekly (not daily)
Smoker: Less often than weekly
Smoker: Not at all, but yes in last year
How difficult would it be to quit (OLS)
How difficult did you think (OLS)
How difficult last attempt (OLS)
How confident that you can quit (OLS)
How confident that you can stay (OLS)
GWL stop agree
GWL stop disagree
GWL worry agree
GWL worry disagree
GWL exaggerate agree
GWL exaggerate disagree
38
38
2012 Packaging
Changes
0.8904***
0.0095
0.0474***
–0.0284***
–0.0496***
0.0194*
0.6523***
0.7043**
0.2080
–0.5354***
–0.2906
0.0149
0.0015
0.0405**
–0.0296*
0.1608***
–0.1403***
2012 Packaging
Changes Quitters
Excluded
1.4234***
0.0095
0.0884***
–0.0324***
–0.0574***
.
0.6523***
.
.
–0.5354***
.
0.0099
0.0100
0.0405**
–0.0296*
0.1782***
–0.1554***
Results are probit regressions reporting marginal effects, unless identified with (OLS), in which case
ordinary least squares regressions.
41
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0004.png
Government pesters agree
Government pesters disagree
Health effects exaggerated agree
Health effects exaggerated disagree
Only harmful to heavy agree
Only harmful to heavy disagree
GWL hide pack agree
GWL hide pack disagree
GWL only thing notice agree
GWL only thing notice disagree
GWL notice of the warning yes/no
GWL don’t look at warning agree
GWL don’t look at warning disagree
0.0746***
–0.0319*
0.0467***
–0.0080
0.0308**
–0.0001
0.1078***
–0.1022***
0.0954***
–0.0934***
0.0268***
-0.0248
0.0258
0.0832***
–0.0441**
0.0591***
–0.0230
0.0346**
0.0012
0.1078***
–0.1022***
0.0954***
–0.0093***
0.0233***
–0.0248
0.0258
Notes:
Significance levels: *0.10, **0.05, ***0.01. GWL denotes graphic warning labels.
42.
The change in consumption is illustrated in the figure below. The number of cigarettes
smoked per day rises by a statistically significant amount of 0.9 cigarettes in the post-
implementation period for the full sample and by 1.4 cigarettes per day excluding
quitters. While these results are not pertinent to analyzing smoking prevalence due to the
fact that the sample is restricted to smokers and recent quitters (within the past twelve
months), they do indicate that within this population smoking intensity and consumption
has not declined but on the contrary has increased. To calculate the post-implementation
value, I added the incremental increase in the number of cigarettes smoked based on the
2012 Packaging Changes coefficient in the smoking probability regression. Thus, this and
all subsequent discussions of the CITTS data control for a detailed set of variables and
the impact of time trends.
39
39
To establish a parallel with studies of smoking prevalence rates, I include time trends in the analysis of the
attitudinal variables, and they show a significant quadratic effect consistent with the RMSS analysis.
42
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0005.png
Incremental Effect of Plain Packaging on
Cigarettes per Day, CITTS Data
16.0
12.1
13.0
Cigarettes per Day
12.0
8.0
4.0
0.0
Before Dec. 1, 2012
After Dec. 1, 2012 through June 2016
I note that this result is based on three and a half years of CITTS data after the
implementation of the 2012 Packaging Changes in Australia.
43.
The other smoking-related measures of interest pertain to smoking-related behaviors, and
another set pertains to various attitudinal variables such as whether the respondent
believes that graphic warning labels exaggerate the risk of smoking. All members of the
CITTS sample were smokers or recent quitters. However, there is a shift in the mix of
their smoking-related behaviors after the advent of the 2012 Packaging Changes.
Whether the respondent is a daily smoker has risen by a statistically significant value of
5%, while there has been a statistically significant decline in the “at least weekly (not
daily)” and “less often than weekly” smoking, reflecting an increase in the intensity of
smoking behavior. The figure below illustrates the shift in daily smoking after the advent
of the 2012 Packaging Changes policy. The baseline daily smoking rate in the period
before December 1, 2012 is the mean value of daily smoking in that sample period.
43
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0006.png
Incremental Effect of Plain Packaging on Daily
Smoking, CITTS Data
100%
80%
73.4%
78.1%
Daily Smoker
60%
40%
20%
0%
Before Dec. 1, 2012
After Dec. 1, 2012 through June 2016
44.
The next series of questions I examine pertains to quit-related behaviors. There is no
significant change in the proportion of respondents who plan to quit in the next month.
Based on analysis of these measures, the 2012 Packaging Changes policy has not
influenced these quit intentions.
45.
This overall quit intention question was followed by a series of questions in which
respondents rated the difficulty of quitting on a scale of 0 to 10. After the advent of the
2012 Packaging Changes policy, respondents rate it significantly more difficult to quit
both in terms of how difficult it would be to quit and how difficult they thought it would
be to quit. Respondents are significantly less confident that they can quit, as the average
ratings before and after December 1, 2012 policy implementation date in the figure below
indicate.
44
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0007.png
Incremental Effect of Plain Packaging on Belief in
Ability to Quit Smoking, CITTS Data
10.0
How Confident You cCan Quit (0-10)
9.0
8.0
7.0
6.0
5.0
4.0
3.0
2.0
1.0
0.0
Before Dec. 1, 2012
After Dec. 1, 2012 through June 2016
5.8
5.3
There were no statistically significant changes in the perceived difficulty of the last quit
attempt or confidence that they can stay as quitters after the advent of the 2012 Packaging
Changes. These outcomes indicate increased pessimism with regards to the possibility of
quitting smoking after the advent of the 2012 Packaging Changes.
46.
Dunlop et al. (2014) examine the CITTS data with respect to smoking emotions and pack
perceptions, but do not examine the effects on behavior. They conclude: “[f]urther
research should extend this study by considering any relationship between smokers’
responses to their plain packaging packs and changes in smoking behaviours….”
40
My
analysis of the CITTS data considered the behaviors that Dunlop et al. (2014) did not
explore, both with respect to cigarette consumption and quit behavior. All statistically
significant effects that I found are the opposite of what one would expect if the 2012
Packaging Changes policy was achieving its intended policy objective.
40
Dunlop et al. (2014),
supra,
at 6.
45
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0008.png
C.
47.
Risk Beliefs and Smoking Attitudes in the CITTS Data
The next series of questions pertains to a series of attitudinal variables of the type
considered by Dunlop et al. (2014). However, rather than focusing on the extreme
responses of those who “strongly agree” or “strongly disagree,” I pool the two “agree”
response categories and the two “disagree” response categories. The effect of the 2012
Packaging Changes on these attitudinal variables is mixed and on balance do not indicate
a shift in attitudes consistent with a beneficial effect of the 2012 Packaging Changes
policies. Consider the following effects, all of which are based on the regression analysis
controlling for demographic factors and other influential covariates. There is a
statistically significant 16% increase in whether respondents believe that the graphic
warning labels policy exaggerate the risk of smoking, a statistically significant 7%
increase in beliefs that the government pesters people too much about smoking risks, a
statistically significant 5% increase in beliefs that the health effects are exaggerated, and
a statistically significant 3% increase in the belief that smoking is only harmful to heavy
smokers. There is also a significant 4% increase in graphic warning labels making them
worry more when they get a cigarette out to smoke. The largest perceptional shift in the
post-implementation period is for whether respondents believe that the graphic warning
labels policy exaggerates the risk, and this shift is illustrated in the figure below.
46
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0009.png
Incremental Effect of Plain Packaging on Belief
that Labels Exaggerate Risk, CITTS Data
100%
Warning Labels Exaggerate Risk
80%
60%
40%
20%
0%
49.7%
33.6%
Before Dec. 1, 2012
After Dec. 1, 2012 through June 2016
48.
The overriding implication of these findings is that the 2012 Packaging Changes policy is
associated with no beneficial impact on risk beliefs, coupled with a substantially
increased degree of rejection of the graphic health warnings message. Other pertinent
policy performance metrics, such as the number of cigarettes smoked per day, daily
smoking behavior, and the perceived ability to quit smoking, show no net beneficial
effect and actually show modest impacts that are counterproductive to the intended
objective.
D.
49.
Perceptions of Pack Appearance in the CITTS Data
The final set of regression estimates focuses on the series of questions that pertain to pack
appearance, which clearly has not been enhanced by the advent of the 2012 Packaging
Changes. However, shifts in views toward pack appearance and associated changes in
warnings are not associated with increased risk beliefs or behavioral changes, as
documented above.
47
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0010.png
50.
The percentage of those who agree that they would hide the cigarette pack rose by a
statistically significant value of 11%. There consequently has been some increase in the
tendency to hide the pack, which will also affect whether the respondent or others can
view the graphic health warning.
51.
The percentage of those who viewed the warning as the only thing they saw on the
package rose by a statistically significant value of 10%. There was also a statistically
significant 3% increase in those who notice the warning. However, since there is very
little else on the pack that respondents can report seeing after the advent of 75% warnings
on the front of packs, these results are not surprising and could be a result of the increase
in the size of the warning to 75% rather than Plain Packaging. That Plain Packaging
removed trademarks and branding from cigarette packaging is apparent. However, the
responses to these questions largely characterize what the policy has done rather than
indicating any effect on risk beliefs, smoking cessation, or decreased cigarette
consumption.
E.
52.
Critique of Literature using CITTS Data
Dunlop et al. (2014) utilize the CITTS data comparing various pre-2012 Packaging
Changes periods from April 2006 to 2012 to the post-2012 Packaging Changes period
through to May 2013 (which is shorter than the period of the dataset that I present above
which goes through to June 2016). Although the responses considered gradations of
qualitative scores on a five point scale, the article by Dunlop et al. (2014) reports only the
results for the most extreme category of whether the respondent indicates “strongly
agree” with a favorable aspect of the 2012 Packaging Changes, such as increased
48
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0011.png
emotional response, or “strongly disagree” with an aspect that the 2012 Packaging
Changes policy is intended to diminish, such as pack attractiveness. As my analysis of the
CITTS data presented above indicates, that approach is an incomplete and misleading
characterization of the data. Focusing on the responses at one extreme end of the
spectrum ignores the countervailing movements at the other extreme for that question. In
addition, pooling the two “agree” responses and the two “disagree” responses eliminates
the apparent effects shown in Dunlop et al. (2014) by focusing only on the “strongly
agree” or only on the “strongly disagree” responses. Because there is no objective
reference point to distinguish degrees of agreement or disagreement, respondents may
differ as to what it means to only agree (disagree) or strongly agree (strongly disagree).
To establish a categorization that does not combine responses that may have quite
different meanings across respondents, I combine the two agree categories in one group
and the two disagree categories in a second group. I also examine many questions that
Dunlop et al. (2014) ignored. By not cherry picking the data to focus only on those
questions that appear to provide evidence favorable to the 2012 Packaging Changes, we
can obtain a more balanced policy perspective. My more comprehensive exploration of
the CITTS survey results above indicates quite different implications of the CITTS data.
The implication of the findings from my analysis is that the 2012 Packaging Changes
have been counterproductive in a number of respects, including that they are associated
with an increased belief that the warnings exaggerate the risk, a decreased belief in
respondents’ ability to stop smoking, and a slight increase in smoking behaviors.
49
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0012.png
V.
A.
53.
IMPLICATIONS OF THE NTPPTS DATA
Overview of the NTPPTS Data
The NTPPTS data is a national telephone adult smoker survey from which I analyze
10,308 observations from April 2012 to March 2014.
41
Further details of this survey are
provided in the paper by Wakefield et al. (2015).
42
Because the pre-implementation time
period covered was shorter than for the CITTS data and the sampling procedure did not
change over time as was the case with the CITTS data, I do not include a time trend
variable in the regression analysis. Similar to the CITTS, many of the NTPPTS questions
are in the form of qualitative Likert rating scales, although a different scale is used where
the survey posed statements and asked which of the following four levels of agreement
the respondent had to the statement: not at all; a little; somewhat; and much. The results
for the NTPPTS data are reported in Appendix C.
B.
54.
Smoking and Quit Behavior in the NTPPTS Data
The first set of regression results reported in Appendix C assesses the estimate of the
2012 Packaging Changes variable in analyses of the number of cigarettes smoked for the
NTPPTS sample. Controlling for other factors, the number of cigarettes smoked per day
has increased by 0.1 cigarettes, which falls short of statistical significance, providing no
support for the effectiveness of the policy. Tables in Appendix C report similar results for
other specifications of the starting date of the 2012 Packaging Changes time period.
41
42
The data collected in the National Tobacco Plain Packaging Monthly Tracking Survey is available on
request from the Australian Department of Health, see
http://www.health.gov.au/internet/main/publishing.nsf/Content/tobacco-plain-packaging-
evaluation#%5B%3Ch2%3E%5DNational%20Monthly%20Tobacco%20Pl
Melanie Wakefield, et al., "Australian Adult Smokers’ Responses to Plain Packaging after Larger Graphic
Health Warnings 1 Year after Implementation: Results from a National Cross-Sectional Tracking Survey",
Tobacco Control
2015; 24:ii17-ii25.
50
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0013.png
2012 Packaging Changes Estimates in Regression Predicting Number of
Cigarettes Per Day Smoked
43
Plain packaging policy (Dec. 1, 2012)
Coefficient
0.1198
Standard Error
0.2029
Notes:
The 2012 Packaging Changes variable is not statistically significant at the 0.10, 0.05, or 0.01
levels.
55.
I note that consistent with my analysis, Scollo et al. (2015),
44
which is the only published
study of the NTPPTS data that discusses the data on actual consumption behavior, also
found that the 2012 Packaging Changes had no impact on consumption. The authors state
the following regarding the impact of the implementation of the 2012 Packaging Changes
(which they called PP for “plain packs”) in Australia:
“Among daily cigarette smokers, there was no change in consumption between
pre-PP and the transition phase or PP year 1 period…Nor was any change
detected when mean daily consumption was analysed among regular
smokers…Mean daily consumption also did not change from the pre-PP to
subsequent two phases among current smokers…Furthermore consumption did
not change from pre-PP to the subsequent two phases among current smokers of
brands of any market segment…"
45
.
56.
The other questions analyzed in the table below regarding smoking cessation related
behaviors and thoughts about quitting also indicate almost a complete lack of any
statistically significant effects relating to advancing the policy objectives of the 2012
43
44
45
Regressions also include variables identifying whether data for demographic variables are missing. Missing
data coded as zero. See Appendix C for fuller regression results.
Michelle Scollo, Meghan Zacher, Kerri Coomber, Megan Bayly, and Melanie Wakefield, "Changes in Use
of Types of Tobacco Products by Pack Sizes and Price Segments, Prices Paid and Consumption Following
the Introduction of Plain Packaging in Australia,"
Tobacco Control
2015;24:ii66-ii75.
Ibid,
at 10, ii73; see also McNeill A, Gravely S , Hitchman SC, Bauld L, Hammond D, Hartmann-Boyce J.
"Tobacco packaging design for reducing tobacco use" Cochrane Database of Systematic Reviews 2017, at
p 20.
51
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0014.png
Packaging Changes. To the contrary, post-policy respondents were less likely to think
about quitting either once or once every few days over the previous week, less likely to
stub out many times after the policy, and similarly were less likely to stop many times
upon having the urge to smoke.
NTPPTS Data: Regressions for Consumption with December 1 Break
Point, Reporting Only 2012 Packaging Changes Coefficient
2012 Packaging
Variable Name
Changes
Cigarettes per day (OLS)
0.1198
How important quitting for good (0-10) (OLS)
–0.0798
How important quitting for good (0)
0.0048
How important quitting for good (10)
0.0063
Intend to quit in next month
–0.0195 *
Think quitting past week not at all
0.0173 *
Think quitting past week once
–0.0186 **
Think quitting past week once every few days
–0.0210 **
Think quitting past week once per day
0.0046
Think quitting past week several times per day
0.0172
Have you ever attempted quitting smoking
–0.0131
How long ago last quit attempt (days) (OLS)
3.4713
Stub out when thought harms never
0.0168
Stub out when thought harms once or twice
0.0112
Stub out when thought harms several times
0.0004
Stub out when thought harms many times
–0.0282 ***
Stop when had urge to smoke never
0.0145
Stop when had urge to smoke once or twice
–0.0036
Stop when had urge to smoke several times
0.0112
Stop when had urge to smoke many times
–0.0223 ***
Notes:
Probit regressions reporting marginal effects unless noted by (OLS), indicating
ordinary least squares; Significance levels: *0.10, **0.05, ***0.01. Regressions
include all other variables listed in Table C1.
57.
A series of questions focused on different aspects of smoking cessation behaviors. Rating
the importance of quitting for good on a 10-point scale, there is a statistically
insignificant decline of this score from 7.4 to 7.3.
52
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0015.png
58.
Whether the respondent intends to quit smoking in the next month exhibits a 2% decrease
in intentions to quit that is significant at the 0.10 level but not the 0.05 level, shown in the
figure below. This effect is the opposite of what one would expect if the 2012 Packaging
Changes policy fostered quit behavior.
59.
In terms of how much the respondent thought about quitting in the past week, there is
also a change that is opposite of presumed policy goals. The categories of “not at all”
reflects an increase, though one that does not meet the 0.05 statistical level. There are
also small but statistically significant decreases in two categories for thoughts of quitting
(“once” and “once every few days”). Overall, the responses to this question are consistent
with stable or even decreased quit intentions after the advent of the 2012 Packaging
Changes.
60.
The fraction of respondents who have ever attempted to quit smoking is unchanged
between the periods before and after the advent of the 2012 Packaging Changes policy.
61.
There is also no statistically significant difference in the number of days since the
respondent’s last quit attempt, with this value having risen from 142 days to 146 days.
53
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0016.png
62.
The next question reported in the table pertains to whether the smoker stubs out the
cigarette after thinking about the harms. The only statistically significant changes indicate
a decrease in this behavior. There is a 3% decrease in those who stub out their cigarette
many times for this reason. This result is the opposite of the policy goals of the 2012
Packaging Changes.
63.
Choosing to stop smoking when the respondent has the urge to smoke elicits similar
responses to stubbing out before and after the introduction of the 2012 Packaging
Changes requirements. The only statistically significant difference is for people who
would stop “many times,” with this value declining by 2% which is the opposite of the
policy goals of the 2012 Packaging Changes.
C.
64.
Risk Beliefs and Smoking Attitudes in the NTPPTS Data
The next table summarizes the series of risk belief and smoking attitude questions. A
broad overview of the statistical significance of the differences before and after the 2012
Packaging Changes indicates that, with a few exceptions, there are no statistically
significant differences in the responses to the risk belief and preference questions before
and after the 2012 Packaging Changes were enacted. Below I discuss each of the specific
questions and their implications, but this lack of significance is the general outcome from
these data.
54
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0017.png
Variable Name
Trouble believe harmful agree
Trouble believe harmful disagree
Trouble believe harmful strongly agree
Trouble believe harmful agree
Trouble believe harmful neither
Trouble believe harmful disagree
Trouble believe harmful strongly disagree
GWL motivate quit not at all
GWL motivate quit a little
GWL motivate quit somewhat
GWL motivate quit much
Harmfulness vs. year ago lower
Harmfulness vs. year ago same
Harmfulness vs. year ago higher
Smoking affect own health not at all
Smoking affect own health a little
Smoking affect own health somewhat
Smoking affect own health very
Smoking affect own health extremely
Lung cancer only old agree
Lung cancer only old disagree
Lung cancer old strongly agree
Lung cancer old agree
Lung cancer old neither
Lung cancer old disagree
Lung cancer old strongly disagree
Think about enjoy smoking never
Think about enjoy smoking once or twice
Think about enjoy smoking several
Think about enjoy smoking many
Think about money spent never
Think about money spent once or twice
Think about money spent several
Think about money spent many
Dangers exaggerated agree
Dangers exaggerated disagree
Dangers exaggerated strongly agree
Dangers exaggerated agree
Dangers exaggerated neither
Dangers exaggerated disagree
Dangers exaggerated strongly disagree
NTPPTS Data: Regressions for Risk Beliefs and Preferences with
December 1 Break Point, Reporting Only 2012 Packaging Changes
Coefficient
2012 Packaging Changes
0.0017
0.0049
0.0027
–0.0005
–0.0059
0.0236 **
–0.0191
–0.0437 ***
0.0019
0.0131 **
0.0282 ***
–0.0005
–0.0040
0.0045
0.0094
0.0031
–0.0091
–0.0033
0.0003
–0.0031
0.0044
–0.0038
0.0014
–0.0014
–0.0108
0.0163
0.0060
–0.0108
0.0025
0.0020
0.0067
–0.0038
–0.0085
0.0058
0.0203 **
–0.0254 **
0.0175 ***
0.0032
0.0053
–0.0171 *
–0.0069
Notes:
Probit regressions reporting marginal effects; significance levels: *0.10, **0.05, ***0.01.
Regressions include all other variables listed in Table C1.
55
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0018.png
65.
There is no statistically significant change in the percentage of respondents who agree
that they have trouble believing their current brand is harmful. There is also very little
change in the levels of agreement or disagreement responses, though one change
indicates a 2% increase among those who disagree.
66.
The question regarding the respondent’s motivation to quit is of a different form than
many of the CITTS questions in that it does not ask whether the respondent agrees or
disagrees with a particular statement. Instead, the quit intentions questions relate to how
motivated respondents are to quit. The survey asks if the respondent’s motivation to quit
is not at all, a little, somewhat, or much. There is a statistically significant 4% decrease in
those who indicate “not at all” to whether graphic health warnings have increased their
motivation to quit in the past month. While there is no statistically significant change to
those who indicate “a little” to whether graphic health warnings have increased their
motivation to quit in the past month, there is a statistically significant 1% and 2%
respective increases in responses indicating that the graphic health warnings did increase
their motivation to quit by “somewhat” and “much.” The results indicate some small
increases in stated motivations to quit. With respect to the respondents' belief regarding
the harmfulness of cigarettes compared to their beliefs a year ago, there is no statistically
significant effect for any of the response categories. The responses are consistent with a
zero association with the 2012 Packaging Changes.
67.
The next set of questions reported in this table pertains to whether the respondent
believes that smoking affects their own health, thus personalizing the risk assessment.
Here again, no response exhibits change due to policy implementation.
56
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0019.png
68.
The next two sets of questions, measuring perceptions of lung cancer risk, also show no
significant shifts due to policy change.
69.
There are also no evident effects with respect to all categories of responses regarding
thoughts about enjoying cigarettes. None of the differences are statistically significant.
70.
There are also zero significant changes with respect to all categories of responses
regarding whether the person thinks about the money spent on cigarettes. None of the
differences before and after the 2012 Packaging Changes policies are statistically
significant.
71.
Beliefs with respect to whether the dangers of cigarettes have been exaggerated do
exhibit statistically significant changes. There is a 2% increase in the agreement that the
dangers are exaggerated, as is indicated in the figure below. Consumers’ belief that the
risks are exaggerated is a survey response that indicates that the warnings policy is not
credible. If, as a result, consumers dismiss the warnings as being excessively alarmist,
they may be dismissed as being uninformative. Similarly, there is a 3% decrease in the
level of disagreement as fewer respondents disagree with the proposition that the risks of
smoking are exaggerated. These changes both reflect potentially counterproductive
results as there is evidence of increased opposition to the 2012 Packaging Changes
policy. The statistically significant shifts are restricted to the extreme responses of
“strongly agree” and the less pronounced “disagree” category. These various outcomes
suggest some resistance or even cynicism towards the new policy.
57
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0020.png
D.
72.
Perceptions of Pack Appearance in the NTPPTS Data
The results in the final NTPPTS table regarding changes in responses with respect to
cigarette pack appearance exhibit statistically significant shifts in almost every instance.
These shifts are consistent with what one would expect based on changes in the pack
after the introduction of 75% warnings on the front of packs in Australia which results in
the packs being dominated by graphic health warnings.
73.
The results indicate that the post-2012 Packaging Changes cigarettes are being perceived
as similar to generic and lower priced brands, but coupled with the findings in the
previous two tables, this shift has not translated into any discernible changes in on risk
beliefs or behavior (if anything, respondents are smoking more). Consideration of the
smoking risk belief questions tells quite a different story than considering only packaging
appearance effects.
58
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0021.png
Variable Name
Quality vs. year ago lower
Quality vs. year ago same
Quality vs. year ago higher
Satisfaction vs. year ago lower
Satisfaction vs. year ago same
Satisfaction vs. year ago higher
Value for money vs. year ago lower
Value for money vs. year ago same
Value for money vs. year ago higher
Pack appeal vs. year ago lower
Pack appeal vs. year ago same
Pack appeal vs. year ago higher
NTPPTS Data: Regressions for Pack Appearance with December 1 Break
Point, Reporting Only 2012 Packaging Changes Coefficient
2012 Packaging Changes
0.1160 ***
–0.0883 ***
–0.0266 ***
0.0743 ***
–0.0445 ***
–0.0284 ***
0.0830 ***
–0.0497 ***
–0.0321 ***
0.3835 ***
–0.3831 ***
–0.0017
Notes:
Probit regressions reporting marginal effects; significance levels: *0.10, **0.05, ***0.01.
Regressions include all other variables listed in Table C1.
74.
The implications of the NTPPTS data reinforce the findings based on the CITTS data and
the RMSS data. Survey evidence regarding smoking behavior, perceived risks of
smoking, and smoking attitudes are not consistent with the 2012 Packaging Changes
being significantly associated with fostering the avowed policy objectives. The only
consistent findings are with respect to package appearance, which could be a result of the
increase in the size of the graphic health warnings to 75% that result in the packs being
dominated by graphic health warnings, rather than Plain Packaging. Stripping away
everything from the pack other than the warning and the name of the product does not
translate into higher risk beliefs, reduced smoking, or quit-related behaviors. The quit
results are at best mixed, with no effect on quit attempts and a statistically insignificant
decrease in those who intend to quit in the next month. Statistically significant impacts
included a decrease in those who thought about quitting in the past week, a decrease in
those who stubbed out when they thought about the harms many times, and a decrease in
stopping smoking when they had the urge to smoke many times. At the very least, these
59
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0022.png
results indicate that there is no evidence of any beneficial policy effects on these
performance dimensions.
75.
The evidence of a lack of impact of the 2012 Packaging Changes in Australia and of a
number of potentially counterproductive effects is not unexpected. There is substantial
evidence in Australia that people are not only aware of the risks of smoking but also
personalize the risks to themselves.
46
In this environment, there is no beneficial role for
increased warnings where they are not providing any new information, as is the case with
the 2012 Packaging Changes. Bolder warnings do not convey unknown information and
telling people something that they already know in
bold
letters or
LARGE TYPE FACE
or with increased graphics does not change that. There is no empirical evidence that
“shouting” works in increasing behavioral compliance in this context, and it can have the
opposite effect.
76.
The evidence of the 2012 Packaging Changes having negative outcomes is also consistent
with research that demonstrates that fear-based warnings may in fact elicit responses that
are the opposite of their intended effect. For example, a meta-analysis of studies that
measure the impact of threatening communications on behavior found that: "… (1) there
are very few studies that could theoretically have supported the use of threatening
communications; and (2) those studies that do exist do not support the wide application of
threatening communications. Instead, they indicate that using threatening communication
is at best ineffective, and at worst causes health-defeating behaviour, unless the
intervention contains an element that effectively enhances response efficacy and
46
See e,g. P. Shanahan and D. Elliott,
Evaluation of the Effectiveness of the Graphic Health Warnings on
Tobacco Product Packaging 2008.
Australian Government Department of Health and Ageing, Canberra
(2009); Tobacco in Australia: Facts and Issues (3rd ed., M. M. Scollo and M. H. Winstanley, eds.,
Melbourne: Cancer Council Victoria, 2008), available at http://www.tobaccoinaustralia.org.au.
60
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0023.png
(especially) self-efficacy." The authors concluded that: "… warning labels on packs of
cigarettes seem ill-advised. They may in fact increase smoking among smokers who
derive self-esteem from their identity as a smoker."
47
A recent study by LaVoie et al.
(2015) also found results that suggest the using graphic health warnings on cigarette
packaging enhances freedom threat perceptions, reactance, and perceived source
domineeringness. They conclude: "These results indicate that utilizing graphic images on
tobacco packaging might not be as effective as some practitioners had originally hoped.
In fact, the messages designed to deter smoking behaviors ignite freedom threat appraisal,
which precedes reactance and, in turn, elevates source domineeringness. Each of these
outcomes is counterproductive to this antitobacco strategy."
48
E.
77.
Critique of Literature using NTPPTS Data
Wakefield et al. (2015),
49
focuses on components of the NTPPTS pertaining to matters
such as decreased pack appeal. Unlike my analysis of the NTPPTS data, the article does
not consider any of the cigarette consumption metrics in the NTPPTS data, such as the
number of cigarettes smoked per day. However, even with respect to the variables
considered, the data presented in Wakefield et al. (2015) do not provide consistent
evidence that is suggestive of a positive influence of the 2012 Packaging Changes. For
example, beliefs in the level of harmfulness of cigarettes are insignificantly lower after
47
48
49
Gjalt-Jorn Ygram Peters, Robert A.C. Ruiter & Gerjo Kok (2012): Threatening communication: a critical
re-analysis and a revised meta-analytic test of fear appeal theory, Health Psychology Review,
DOI:10.1080/17437199.2012.703527.
LaVoie et al (2015), Are Graphic Cigarette Warning Labels an Effective Message Strategy? A Test of
Psychological Reactance Theory and Source Appraisal, Communication Research, DOI:
10.1177/0093650215609669.
Melanie Wakefield, Kerri Coomber, Meghan Zacher, Sarah Durkin, Emily Brennan, and Michelle Scollo,
“Australian Adult Smokers’ Responses to Plain Packaging with Larger Graphic Health Warnings 1 Year
after Implementation: Results from a National Cross-Sectional Tracking Survey,” Tobacco Control
2015;24:ii17-ii25. doi:10.1136/tobaccocontrol- 2014-052050.
61
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0024.png
the 2012 Packaging Changes (24.2 pre-changes and 23.4 post-changes for the unadjusted
results, and 23.8 pre-changes adjusted and 23.6 post-changes for adjusted results). There
were also no statistically significant differences in perceived exaggeration of harms,
perceived differences in taste of different brands, or the belief that variants do not differ
in strength at one-year post-2012 Packaging Changes compared to pre-2012 Packaging
Changes. In addition, as my detailed review of the NTPPTS data presented above
indicates, there are numerous outcomes from the NTPPTS data that go against claims that
the 2012 Packaging Changes have been effective and which are not reported in
Wakefield et al. (2015). Unbiased assessments of the data require that one not select
isolated components of questions, but instead consider the implications of the full set of
responses. Failure to consider the full range of pertinent behavioral questions provides a
distorted perspective that makes it impossible to draw accurate conclusions about the
impact of the 2012 Packaging Changes. My analysis shows that the NTPPTS data do not
provide consistent support for the 2012 Packaging Changes being associated with even
these non-behavioral variables, leaving aside the questionable relevance of these
variables for examining the effect of the 2012 Packaging Changes on actual smoking
behavior. Wakefield et al. purport to find strong evidence that “the specific objectives of
plain packaging were achieved,” but this conclusion is based on mixed outcomes and an
incomplete analysis of the data. A proper analysis of the NTPPTS does not support the
2012 Packaging Changes as having been effective and indeed shows that the 2012
Packaging Changes have been counterproductive in many respects, not the least of which
is that there is no evidence that the 2012 Packaging Changes are associated with a
reduction in smoking behavior, which is the objective of the policy.
62
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0025.png
VI.
78.
ANALYSIS OF COHORT-BASED NTPPTS STUDIES
The NTPPTS survey also includes a cohort component in which a subsample of the
NTPPTS was re-interviewed at different times in the year following the implementation
of the 2012 Packaging Changes. By tracking individuals over time, these data could
potentially provide insight into how the 2012 Packaging Changes influenced behavior.
Thus, to take advantage of the cohort aspect of the data the researcher should examine
whether there was a change in behavior or smoking attitudes before and after the 2012
Packaging Changes. However, as I will demonstrate below, the published studies using
the cohort sample have not always utilized the cohort aspect of the data to examine
within person changes in behavior. The cohort aspect is sometimes ignored as there is no
exploration of changes. Rather the article simply inquires whether respondents who
happen to be members of the cohort subsample gave particular survey responses. Thus,
whether a person was in the cohort subsample or not is sometimes irrelevant to the
statistical analysis. And, more importantly, when they do analyze the changes in behavior
of members of the cohort subsample, they have ignored the biases in terms of which
sample members agreed to be re-interviewed. The differences in the baseline sample and
those who were re-interviewed are so stark that the cohort analysis is fundamentally
flawed.
79.
The Durkin et al. (2015)
50
study used the cohort subsample of the NTPPTS data to
examine both variables related to quitting and several smoking-related behaviors. I note
50
Sarah Durkin, Emily Brennan, Kerri Coomber, Meghan Zacher, Michelle Scollo, and Melanie Wakefield,
“Short-Term Changes in Quitting-Related Cognitions and Behaviours after the Implementation of Plain
Packaging with Larger Health Warnings: Findings from a National Cohort Study with Australian Adult
Smokers,” Tobacco Control 2015;24:ii26-ii32. doi:10.1136/tobaccocontrol-2014-052058
63
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0026.png
that the discussed data only concerned perceptions and intentions rather than actual
behavioral data. In addition, for many of the measures considered in Durkin et al. (2015)
there is not a statistically significant change between the pre-2012 Packaging Changes
period and the post-2012 Packaging Changes period. The response categories that
exhibited no statistically significant changes in the post-implementation period (i.e., zero
effects) include: daily thoughts about quitting in the past week, intend to quit next month,
firm date to quit next month, and stopped smoking several or many times in past month.
Accordingly, the NTPPTS data do not support the conclusion that Plain Packaging
contributed to increasing quitting cognitions and intentions. This is even clear from the
results reflected in the study by Durkin et al. (2015).
80.
The Durkin et al. (2015) article uses 4 cohorts of adult cigarette smokers sourced from
the NTPPTS sample surveyed before the 2012 Packaging Changes, followed up
approximately 1 month after their baseline interview. Logistic regression analyses
compared changes in selected quitting-related outcomes over this 1-month follow-up
period for the cohorts surveyed before the policy change, over the period of transition to
the policy change, and during the first year of the policy change. These periods are
labelled pre-Packaging Changes, early transition, late transition, and Packaging Changes
year 1. Given the short time frame that is involved as well as the failure to account for
any longer term trends, it is likely that any observed effects are spurious. If, however,
there are any meaningful statistically significant impacts, one would expect there to be a
consistently rising pattern of changes between the pre-Packaging Changes period and the
post-Packaging Changes period.
64
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0027.png
81.
The Durkin et al. (2015) article examined seven different variables. Several of the
response categories do not indicate any statistically significant differences, and those that
did typically exhibited mixed patterns. The “daily thoughts about quitting” variable
exhibited no changes based on either the unadjusted or adjusted (for baseline values and
covariates) model. The “intend to quit” measure is higher in the late transition but not in
either the early transition or the Packaging Changes year 1 results. The “firm date to quit
in next month” variable exhibits no significant differences from the baseline value. The
“concealed or covered pack several or many times in past month” variable exhibits a
rising pattern in the adjusted results, but rises and then declines in the unadjusted results.
This seems to be the strongest effect that was observed. The measure “stubbed out
several or many times in past month” is unchanged in the unadjusted results but has one
significant difference (for the Packaging Changes year 1) in the adjusted results. The
variable “stopped from smoking several or more times in past month” rises in the early
transition and declines back to the baseline level in the late transition and in the
Packaging Changes year 1 so that either the result is an aberration or the estimated effect
is ephemeral. Similarly, “attempt to quit in past month” exhibits no clear trend, with
significantly lower values in the unadjusted results in the late transition, and significantly
higher values in the unadjusted results in the early transition and Packaging Changes year
1. Accordingly, the NTPPTS data do not support the conclusion that the 2012 Packaging
Changes are associated with increased quitting cognitions and intentions. This is even
clear from the results reflected in the study by Durkin et al. (2015).
65
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0028.png
82.
Brennan et al. (2015)
51
use the NTPPTS cohort data to explore correlations of the quit
attempt variables with the attitudinal variables. However, the article focuses only on
those interviewed one month after the baseline period so that in that short time frame
there is unlikely to be any meaningful change in smoking behavior. The cohort aspect of
the study is largely irrelevant since the authors never analyze any changes in quit related
behaviors or changes in the various perception variables. Rather the focus of the article is
on correlations of levels of the variable. In particular, do people who express various anti-
smoking perceptions such as “dislikes pack” in the baseline period also express various
quit-related behaviors in the follow up survey? Even if there were such influences, they
have no bearing on whether there were any changes in behavior or perceptions since the
article does not address changes of any kind.
83.
In addition, the researchers also found very little in the way of statistically significant
correlations between baseline impressions and the follow-up answers. For example, the
following measures examined by Brennan et al. (2015) did not have a significant
association with quit attempts, much less actual quit behavior: dislikes pack, lower pack
appeal, lower quality, lower satisfaction, lower value for money, notices graphic health
warnings first when looking at pack, does not believe dangers of smoking are
exaggerated, and concealed pack in past month. This article focuses on subjective
impressions that have no validated link with actual smoking decisions and never analyzes
changes in the variables before and after the implementation of the 2012 Packaging
Changes. One cannot infer any causality from the correlations presented. Even the
51
Emily Brennan, Sarah Durkin, Kerri Coomber, Meghan Zacher, Michelle Scollo, and Melanie Wakefield,
“Are Quitting- Related Cognitions and Behaviours Predicted by Proximal Responses to Plain Packaging
with Larger Health Warnings? Findings from a National Cohort Study with Australian Adult Smokers,”
Tobacco Control
2015;24:ii33–ii41.doi:10.1136/tobaccocontrol-2014-052057., at 23.
66
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0029.png
number of insignificant correlations (i.e., those not distinguishable from zero)
outnumbered the number of significant correlations by two to one. The affirmative
conclusions of this article are misplaced. Given the design of the study, it has no bearing
whatsoever on whether the 2012 Packaging Changes have been effective.
84.
Both the Durkin et al. (2015) and the Brennan et al. (2015) articles also failed to account
for sample selection effects in terms of who agreed to be re-contacted for a follow-up
survey. The Durkin et al. (2015) study indicates that 95% agreed to be contacted and of
these 83% were successfully contacted, and the data reflect a follow-up participation rate
of 79%. There could be important sample selection biases if the characteristics of those
who participated in the follow-up survey differ significantly from those who did not. The
authors never address this issue, but using information from the NTPPTS regarding who
participated in the cohort and who did not we are able to test for such differences and
have identified some statistically significant gaps.
85.
The table below lists a series of sample characteristic variables and gives their value for
both the pre-2012 Packaging Changes period and the post implementation sample. As the
table indicates, there are 8 statistically significant differences in the sample characteristics
at the 0.05 level or better, as coefficients not statistically significant at the 0.10 level or
better are denoted by “ns.” Among the many notable demographic differences is that
those who participated in the follow-up survey are 4 years older and 8 percent more
likely to have a landline in the home.
67
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0030.png
NTPPTS Data: Personal Characteristics by Follow-up Participation
Variable Name
Age
Female
Years of education
Income, low
Income, medium
Income, high
Original peoples
English primary language
Phone: land line in home
Phone: mobile line in home
Hours of television per day
New South Wales
Victoria
Queensland
South Australia
Western Australia
Tasmania
Northern Territory
Australian Capital Territory
Capital city
No
Follow
39.61
13.17
0.2204
0.4932
0.2863
0.0433
0.8795
0.7033
0.9315
2.3513
0.3282
0.2494
0.2000
0.0587
0.1175
0.0186
0.0140
0.0135
0.6567
Yes
Follow
43.58
13.07
0.2385
0.5053
0.2562
0.0358
0.9391
0.7759
0.9260
2.6958
0.2948
0.2653
0.2089
0.0817
0.0960
0.0241
0.0116
0.0174
0.6370
t-statistic
–11.6014
–1.2809
1.5350
–1.5419
–0.8767
2.4846
1.6201
–9.4465
–6.9946
0.8783
–5.2806
2.9942
–1.4927
–0.9021
–3.5543
2.9398
–1.5084
0.8842
–1.2506
1.6760
Significance
***
ns
ns
ns
ns
**
ns
***
***
ns
***
***
ns
ns
***
***
ns
ns
ns
*
0.4471
0.4626
Notes:
Significance levels: *0.10, **0.05, ***0.01
86.
In addition to significant differences in respondent personal characteristics between the
entire sample of participants in the original survey and those who participated in the
follow-up survey, further problems are apparent between the two groups on the basis of
their answers to the questions related to cigarettes in the original survey. The tables
below show 8 statistically significant differences (at the 0.05 significance level) between
68
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0031.png
the two groups. Follow-up survey participants smoke 1.5 more cigarettes per day and are
6 percent more likely to have attempted to quit smoking.
Consumption by Follow-up Participation
Variable Name
Cigarettes per day
How important quitting for good (0-10)
How important quitting for good (0)
How important quitting for good (10)
Intend to quit in next month
Think quitting past week not at all
Think quitting past week once
Think quitting past week once every few days
Think quitting past week once per day
Think quitting past week several times per day
Have you ever attempted quitting smoking
How long ago last quit attempt (days)
Stub out when thought harms never
Stub out when thought harms once or twice
Stub out when thought harms several times
Stub out when thought harms many times
Stop when had urge to smoke never
Stop when had urge to smoke once or twice
Stop when had urge to smoke several times
Stop when had urge to smoke many times
Notes:
Significance levels: *0.10, **0.05, ***0.01.
No
Follow
13.09
Yes
Follow
14.65
t-statistic
–6.1623
–3.2386
2.5645
–1.3402
2.0158
–2.7967
–1.2956
–0.9463
–0.2198
4.4093
–6.0959
1.3049
–1.9687
1.9144
–0.5171
1.3664
1.5992
–1.0785
0.0782
–0.8694
Signifi-
cance
***
**
**
ns
**
***
ns
ns
ns
***
***
ns
**
*
ns
ns
ns
ns
ns
ns
7.0931
0.0766
0.3868
0.4149
0.2194
0.1323
0.1835
0.0822
0.3823
0.7477
150.4
0.5413
0.1872
0.1458
0.1257
0.3324
0.2566
0.2767
0.1343
7.3335
0.0612
0.4028
0.3881
0.2510
0.1442
0.1932
0.0838
0.3279
0.8116
143.7
0.5670
0.1682
0.1507
0.1142
0.3128
0.2691
0.2758
0.1422
Thus, consumption-related variables show many significant differences between the
original survey sample and the sample that participated in the follow-up survey.
Variables ranging from rate of consumption, attitudes or attempts of quitting, and stub-
out behavior are significantly different. These various differences in turn lead to
pronounced differences in risk beliefs and preferences. There are 24 variables pertaining
69
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0032.png
to risk beliefs and preferences that display statistically significant differences between the
sample that did not participate in the follow-up and the follow-up group.
Risk Beliefs and Preferences by Follow-up Participation
Variable Name
Trouble believe harmful agree
Trouble believe harmful disagree
Trouble believe harmful strongly agree
Trouble believe harmful agree
Trouble believe harmful neither
Trouble believe harmful disagree
Trouble believe harmful strongly disagree
GWL motivate quit not at all
GWL motivate quit a little
GWL motivate quit somewhat
GWL motivate quit much
Harmfulness vs. year ago higher
Harmfulness vs. year ago lower
Harmfulness vs. year ago same
Smoking affect own health not at all
Smoking affect own health a little
Smoking affect own health somewhat
Smoking affect own health very
Smoking affect own health extremely
Lung cancer only old agree
Lung cancer only old disagree
Lung cancer old strongly agree
Lung cancer old agree
Lung cancer old neither
Lung cancer old disagree
Lung cancer old strongly disagree
Think about enjoy smoking never
Think about enjoy smoking once or twice
Think about enjoy smoking several
Think about enjoy smoking many
Think about money spent never
Think about money spent once or twice
Think about money spent several
Think about money spent many
Dangers exaggerated agree
Dangers exaggerated disagree
Dangers exaggerated strongly agree
No
Follow
0.3305
0.6294
0.1183
0.2122
0.0401
0.2710
0.3584
0.5325
0.2010
0.1088
0.1577
0.2530
0.0568
0.6902
0.1536
0.2691
0.2691
0.1871
0.1211
0.1219
0.8661
0.0461
0.0759
0.0120
0.3932
0.4729
0.3555
0.2981
0.1955
0.1509
0.2289
0.1787
0.2057
0.3867
0.3321
0.6187
0.1190
Yes
Follow
0.2728
0.6907
0.0970
0.1758
0.0365
0.2534
0.4373
0.5770
0.1955
0.1084
0.1192
0.2192
0.0402
0.7406
0.1145
0.2563
0.2989
0.1976
0.1328
0.0682
0.9184
0.0268
0.0414
0.0134
0.3860
0.5324
0.3299
0.3088
0.2039
0.1574
0.1913
0.1759
0.2129
0.4199
0.2905
0.6691
0.1042
t-
statistic
4.2738
–4.3919
2.3586
3.1418
0.6503
1.3419
–5.3466
–3.6652
0.5595
0.0555
4.7325
2.9993
3.0178
–4.2184
4.9051
1.2016
–2.6863
–1.0883
–1.4246
8.1079
–7.3617
4.5336
6.5364
–0.5041
0.6004
–4.8478
2.2115
–0.9541
–0.8458
–0.7280
3.8592
0.2964
–0.7249
–2.7567
3.7035
–4.3340
1.9447
Significance
***
***
**
***
ns
ns
***
***
ns
ns
***
***
***
***
***
ns
***
ns
ns
***
***
***
***
ns
ns
***
**
ns
ns
ns
***
ns
ns
***
***
***
*
70
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0033.png
Dangers exaggerated agree
Dangers exaggerated neither
Dangers exaggerated disagree
Dangers exaggerated strongly disagree
Notes:
Significance levels: *0.10, **0.05, ***0.01.
0.2131
0.0492
0.3053
0.3134
0.1863
0.0404
0.3075
0.3616
2.7767
1.7831
–0.1950
–4.1147
***
*
ns
***
Risk belief variables show significant differences in literally every measured category
between the full sample and the follow-up participants. This is also true of the pack
appearance, shown below. There are 5 pack appearance variables that differ significantly
including the pack appeal variable as the follow-up group consists disproportionately of
those who answer negatively to whether the pack appeal is higher than a year ago.
Pack Appearance by Follow-up Participation
Variable Name
Quality vs. year ago lower
Quality vs. year ago same
Quality vs. year ago higher
Satisfaction vs. year ago lower
Satisfaction vs. year ago same
Satisfaction vs. year ago higher
Value for money vs. year ago lower
Value for money vs. year ago same
Value for money vs. year ago higher
Pack appeal vs. year ago lower
Pack appeal vs. year ago same
Pack appeal vs. year ago higher
Notes:
Significance levels: *0.10, **0.05, ***0.01
No
Follow
0.2165
0.7106
Yes
Follow
0.2217
0.7248
t-statistic
–0.4719
–1.1868
3.1239
–1.1253
–0.7969
3.4011
–4.3280
0.1834
4.9810
0.1906
–1.9161
3.4652
Significance
ns
ns
***
ns
ns
***
***
ns
***
ns
*
***
0.0729
0.1649
0.7663
0.0688
0.5184
0.2153
0.2663
0.4110
0.5029
0.0861
0.0535
0.1763
0.7752
0.0485
0.5758
0.2133
0.2109
0.4083
0.5303
0.0614
87.
The myriad of differences between those who participated in the follow-up survey and
those that did not create sample selection biases as the follow-up group is not a random
group of the original survey participants. These issues are not addressed in any way by
71
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0034.png
the Durkin et al. (2015) study, or the Brennan et al. (2015) study based on the cohort
data.
88.
In conclusion, the cohort component of the NTPPTS data has fundamental flaws and is
not representative of the underlying survey group. The Durkin et al. (2015) and Brennan
et al. (2015) studies ignore the sampling issues. However, even setting aside the sample
selection issues, the data provide no evidence of efficiency of the 2012 Packaging
Changes.
89.
I note that these published studies on the NTPPTS data are included in the Cochrane
Review,
52
which relies on these studies on their face without doing any critical review of
data analysis or any analysis of the original data on which the studies were based.
However, my examination of the outputs of each of the CITTS and NTPPTS datasets
indicates that the articles are disturbing from the standpoint of academic integrity and are
highly misleading. As such the conclusions drawn from the studies as presented in the
Cochrane Review are unjustified.
52
McNeill, A., Gravely, S., Hitchman, S.C., Bauld, L., Hammond, D., and Hartmann-Boyce, J., "Tobacco
Packaging Design for Reducing Tobacco Use," Cochrane Database of Systematic Reviews 2017, Issue 4,
Art. No.: CD011244.
72
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0035.png
VII.
REVIEW OF THE AUSTRALIAN GOVERNMENT'S AUSTRALIAN POST
IMPLEMENTATION REVIEW REPORT
90.
The Australian government’s Post-Implementation Review Tobacco Plain Packaging
2016 reached a favorable conclusion regarding the impact of Plain Packaging in
Australia, stating that: “[i]n light of the evidence, the PIR concludes that tobacco plain
packaging is achieving its aim of improving public health outcomes into the future.”
However, this conclusion is contradicted by my analysis presented in this report.
53
91.
The PIR’s only statistical evidence consistent with the 2012 Packaging Changes having a
positive effect on smoking behaviors is based on the highly flawed report by Dr. Chipty.
The PIR did not draw on any econometric analysis of other data, including the CITTS
and the NTPPTS data which also provide some information that can be used to evaluate
the association of the 2012 Packaging Changes with actual smoking behavior. Although
these datasets utilize different samples and provide different perspectives on smoking
behaviors, they provide a consistent theme that there is no evidence suggesting that the
2012 Packaging Changes are associated with a decrease in smoking. This failure to
examine these other available datasets produced excessive reliance on a single statistical
study of one dataset undertaken by Dr. Chipty. While the PIR also cites the existence of
downward trends in smoking behavior from other datasets, it reports no statistical
analysis linking these trends to the 2012 Packaging Changes.
92.
In addition, the PIR merely relies on published papers regarding the impact of Plain
Packaging without undertaking any critique or review of those papers. Based on my
53
I note that the PIR is also criticized in the paper by Sinclair Davidson and Ashton de Silva, “Stubbing out
the Evidence of Tobacco Plain Packaging Efficacy: An Analysis of the Australian National Tobacco Plain
Packaging Survey,” May 17, 2016, SSRN id=2780938.
73
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0036.png
review of the papers and the underlying NTPPTS and CITTS datasets, I conclude that the
published papers on these datasets cannot be relied upon.
93.
Accordingly, there is no sound basis for the PIR’s conclusion set out above. The PIR’s
reference to “all this evidence” is especially inappropriate because the cited studies did
not report all the evidence from the NTPPTS and CITTS datasets, but only the selected
results that provide the most favorable perspective on the performance of plain packs. In
addition, as noted above, my extended analysis of the RMSS data and the CITTS data,
which is the most extensive data analysis undertaken to date (and includes 15 months of
additional data to the analysis undertaken by Dr Chipty), confirms that there is no support
for the conclusion that Plain Packaging has been effective. The 2012 Packaging Changes
have a zero statistical association with smoking prevalence rates, which further
demonstrates that the conclusion reached in the PIR is unjustified.
74
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0037.png
VIII. CONCLUSION
94.
The availability of post-implementation data from Australia, including survey data
commissioned by the Australian Government to assess the impact of the 2012 Packaging
Changes and prevalence data relied on by the Australian Government, makes it possible
to assess whether the 2012 Packaging Changes policy in Australia has had the effects that
advocates of Plain Packaging claim.
95.
My analysis of 4 years of post-implementation RMSS data as well as 3 �½ years of post-
implementation CITTS data and the Australian Government commissioned NTPPTS data
indicates that the 2012 Packaging Changes in Australia are not associated with a
reduction in smoking or smoking consumption. My analysis of the RMSS data, which
includes 15 additional months of data in the post-2012 Packaging Changes period than
was considered in Dr. Chipty's report, found that the estimated effect of the 2012
Packaging Changes in Australia on smoking prevalence rates cannot be distinguished
statistically from zero. An evaluation of the CITTS and NTPPTS data relating to actual
cigarette consumption behavior in Australia also indicates that the 2012 Packaging
Changes are not associated with a decrease in smoking behaviors amongst current
smokers.
96.
There is also evidence consistent with either a mixed or an unfavorable effect of the 2012
Packaging Changes on a number of intermediary metrics
even setting aside issues
pertaining to the efficacy of these intermediate variables in predicting actual smoking
behaviors. Indeed the overriding implication of the findings from my analysis is that the
2012 Packaging Changes in Australia are associated with an increase degree of rejection
75
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0038.png
of the graphic warnings message.
There is also some evidence of mixed effects,
including some negative changes in risk beliefs and efforts to stop smoking.
97.
The favorable conclusions reached in other analyses of the RMSS data are unfounded.
The article by Diethelm and Farley (2015) had no original RMSS data but attempted to
impute monthly average values based on visual inspection of a figure from a working
paper by Kaul and Wolf. As a result, the statistical analysis included no demographic or
regional variables whatsoever, but only a 'bare-bones' group of four variables including a
linear trend. Cigarette prices and a potential nonlinear trend were completely ignored.
The report by Dr. Chipty likewise ignored these two sets of influences. While Dr.
Chipty's study did include a more extensive variable list than did Diethelm and Farley
(2015), and also utilized the actual RMSS data, the omission of the key factors driving
the temporal trend—cigarette prices and the nonlinear aspect of the trend—results in a
flawed analysis that cannot be relied on. I demonstrated the absence of a statistically
significant effect for the 2012 Packaging Changes variable using the time period used by
Dr. Chipty as well as a longer time period and a larger RMSS sample. These differences
with Dr. Chipty’s report are not differences of opinion but are matters of basic statistics
(i.e., a statistical test indicating a nonlinear trend) and fundamental economics (i.e., the
key role of prices in affecting consumer demand of all products).
98.
The Australian Government Post-Implementation Review Tobacco Plain Packaging 2016
report provided an inadequate and incorrect assessment of the effect of the 2012
Packaging Changes on smoking. The PIR’s only statistical evidence of an effect of the
2012 Packaging Changes on actual smoking behaviors is based on the flawed report by
Dr. Chipty.
76
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0039.png
99.
Moreover, unlike the presentations of the evidence provided in the PIR and in the
recently published Cochrane Review, which relies on the published papers, I also
undertake a full analysis of the underlying Australian NTPPTS data and the New South
Wales CITTS data. My analysis shows that the published articles analyzing these data
are disturbing from the standpoint of academic integrity and are highly misleading.
Rather than provide an unbiased assessment of the survey results, the studies present
selected findings that purport to demonstrate the efficacy of the 2012 Packaging Changes
policy, but after more thorough assessment do not. Unbiased assessments of the post-
implementation data require that one not select isolated components of questions but
instead consider the implications of the full set of responses. The published articles on
these data do not do this and cannot be relied on as being accurate or a complete
assessment of the data.
100.
My review of the Roy Morgan Research, CITTS, and NTPPTS data before and after the
imposition of the 2012 Packaging Changes in Australia, which I understand is the most
comprehensive analysis of these Australian datasets to date, compels the conclusion that
the 2012 Packaging Changes have not been effective in reducing smoking. In addition,
even with respect to the non-behavioral measures, which have no validated link to actual
smoking behaviors, there is no consistent evidence that the 2012 Packaging Changes are
achieving its stated aims.
Particularly with respect to the credibility of cigarette
warnings, the 2012 Packaging Changes appear to be having a counterproductive effect.
____________________
W. Kip Viscusi
2 January 2018
77
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0040.png
78
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0041.png
APPENDIX A
RMSS Data
1.
This section of the appendix provides details of my statistical analysis. Table A1 of
regression results analyzes the determinants of the probability that any particular
respondent in the RMSS data is a smoker. A respondent is defined as a smoker if they
smoke factory-made cigarettes, roll-your-own cigarettes, pipes, or cigars. To examine
whether a nonlinear relationship between smoking prevalence rates and time preceded the
2012 Packaging Changes, the first two equations consider only the pre-2012 Packaging
Changes period. Column 1 includes a linear time trend as well as a series of demographic
and policy variables, and column 2 adds a time squared variable to test for the
nonlinearity of the time trend.
While the linear time trend variable is statistically
significant in the first equation, the addition of the quadratic time trend variable in the
second equation is statistically significant whereas the linear trend no longer is
statistically significant. Note that the evidence of a nonlinear temporal trend in smoking
prevalence rates for the second column is for the pre-2012 Packaging Changes period.
Thus, even before the advent of the Packaging Changes policy in 2012, there is evidence
of a nonlinear trend.
54
The overall relationship is nonlinear in the pre-2012 Packaging
Changes era, controlling for excise taxes and a detailed set of sample characteristic
variables.
54
I note that this is the same conclusion reached by Diethelm P, McKee M. "Tobacco industry-funded
research on standardised packaging: there are none so blind as those who will not see!", Tob Control
2015;24:e113–e115 in respect of a sample of the RMSS data for 14-17 year olds.
79
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0042.png
Appendix Table A1
RMSS Data Regressions Predicting Smoking Behavior
Smoker
Smoker
(pre-policy) (pre-policy)
Smoker
2012 Packaging Changes, full
(Dec 1, 2012)
Graphic warning labels, 2006
Tax policy, 2010 (25%)
Tax policy, 2013 (12.5%)
Tax policy, 2014 (12.5%)
Tax policy, 2015 (12.5%)
Tax policy, 2016 (12.5%)
Time trend, months
Time trend, months (squared)
Female
Marital status, single
Marital status, divorced
Marital status, widowed
Marital status, separated
Student
Years of education
Age
Age (squared)
Non-adults (14-17)
Employed full time
–8.5E-5***
(3.2E-5)
–0.0396***
(0.0011)
0.0353***
(0.0016)
0.1000***
(0.0020)
0.0474***
(0.0022)
0.1278***
(0.0027)
–0.0748***
(0.0043)
–0.0204***
(0.0002)
0.0085***
(0.0002)
–1.4E-4***
(1.8E-6)
–0.2233***
(0.0030)
–0.0264***
4.8E-5
(6.2E-5)
–1.2E-6**
(4.8E-7)
–0.0396***
(0.0011)
0.0353***
(0.0016)
0.1000***
(0.0020)
0.0474***
(0.0022)
0.1278***
(0.0027)
–0.0744***
(0.0043)
–0.0204***
(0.0002)
0.0085***
(0.0002)
–1.4E-4***
(1.8E-6)
–0.2234***
(0.0030)
–0.0264***
0.0004
(0.0021)
–0.0108***
(0.0019)
0.0007
(0.0021)
–0.0059**
(0.0027)
–0.0061***
(0.0021)
0.0003
(0.0020)
–0.0105***
(0.0018)
–0.0052**
(0.0026)
–0.0009
(0.0026)
–0.0023
(0.0024)
–0.0073**
(0.0035)
–9.5E-5***
(3.1E-5)
–0.0396***
(0.0009)
0.0401***
(0.0013)
0.0985***
(0.0017)
0.0495***
(0.0019)
0.1278***
(0.0023)
–0.0783***
(0.0035)
–0.0199***
(0.0001)
0.0092***
(0.0002)
–1.4E-4***
(1.5E-6)
–0.2010***
(0.0027)
–0.0268***
Smoker
–0.0030
(0.0026)
0.0004
(0.0020)
–0.0069***
(0.0025)
–0.0033
(0.0028)
0.0011
(0.0028)
0.0003
(0.0027)
–0.0055
(0.0036)
7.8E-6
(5.9E-5)
–9.0E-7**
(4.4E-7)
–0.0396***
(0.0009)
0.0401***
(0.0013)
0.0985***
(0.0017)
0.0495***
(0.0019)
0.1278***
(0.0023)
–0.0781***
(0.0035)
–0.0199***
(0.0001)
0.0092***
(0.0002)
–1.4E-4***
(1.5E-6)
–0.2011***
(0.0027)
–0.0268***
80
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0043.png
Retired
Income
Income (squared)
Income, multiple
household members
Bread winner
Household size
Home owner
Victoria
Queensland
South Australia
Western Australia
Tasmania
Darwin-Alice Springs
Lives in capital city
Constant
R-squared
(0.0015)
–0.1022***
(0.0048)
–0.0005***
(3.8E-5)
1.2E-6***
(1.7E-7)
–0.0176***
(0.0012)
0.0099***
(0.0013)
–0.0010**
(0.0005)
–0.1150***
(0.0012)
0.0142***
(0.0013)
0.0067***
(0.0014)
0.0096***
(0.0020)
0.0064***
(0.0018)
0.0262***
(0.0023)
0.0364***
(0.0048)
–0.0251***
(0.0010)
0.5635***
(0.0054)
0.1133
(0.0015)
–0.1023***
(0.0048)
–0.0005***
(3.8E-5)
1.3E-6***
(1.7E-7)
–0.0176***
(0.0012)
0.0100***
(0.0013)
–0.0010**
(0.0005)
–0.1150***
(0.0012)
0.0143***
(0.0013)
0.0067***
(0.0014)
0.0097***
(0.0020)
0.0065***
(0.0018)
0.0262***
(0.0023)
0.0364***
(0.0048)
–0.0251***
(0.0010)
0.5609***
(0.0055)
0.1133
(0.0012)
–0.0966***
(0.0040)
–0.0005***
(3.1E-5)
1.2E-6***
(1.3E-7)
–0.0188***
(0.0011)
0.0084***
(0.0011)
–0.0009**
(0.0004)
–0.1115***
(0.0010)
0.0131***
(0.0011)
0.0057***
(0.0012)
0.0082***
(0.0016)
0.0055***
(0.0015)
0.0243***
(0.0020)
0.0348***
(0.0042)
–0.0259***
(0.0009)
0.5254***
(0.0046)
0.1139
(0.0012)
–0.0966***
(0.0040)
–0.0005***
(3.1E-5)
1.3E-6***
(1.3E-7)
–0.0188***
(0.0011)
0.0084***
(0.0011)
–0.0009**
(0.0004)
–0.1115***
(0.0010)
0.0131***
(0.0011)
0.0057***
(0.0012)
0.0082***
(0.0016)
0.0056***
(0.0015)
0.0243***
(0.0020)
0.0348***
(0.0042)
–0.0260***
(0.0009)
0.5234***
(0.0047)
0.1139
Notes:
Significance levels: *0.10, **0.05, ***0.01.
2.
The failure to consider the nonlinearity of the underlying relationship leads to Dr.
Chipty’s spurious claim that the post-2012 Packaging Changes departure of smoking
prevalence rates from the previous linear trend reflects the impact of the 2012 Packaging
Changes on smoking prevalence rates. Neither Dr. Chipty nor Diethelm and Farley
(2015) included a nonlinear time trend variable in their analyses.
81
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0044.png
3.
The final two equations in Table A1 are for the entire sample period. Column 3 included
a linear time trend variable, and column 4 also includes a squared time trend variable to
test for the potential nonlinearity of the results. In the first full sample period equation,
which is patterned generally after that of Dr. Chipty, the 2012 Packaging Changes
indicator variable has a statistically significant negative sign. However, addition of the
quadratic time trend variable in the final equation eliminates the statistical significance of
the 2012 Packaging Changes variable.
4.
The Table A2 regressions utilize the full sample as did columns 3 and 4 in Table A1, but
includes a set of variables to focus on the role of cigarette prices, setting aside the
potential influence of a quadratic time trend. Thus, I revert to including only the time
trend variable without the square of that variable (following the approach in column 3 of
Table A1), but instead of the excise tax indicator variables I include alternative measures
of prices. Column 1 replaces the excise taxes indicator variables used by Dr. Chipty with
a continuous measure of excise tax levels. Column 2 replaces excise taxes with an
overall price index for the economy. Column 3 substitutes the price index for Craven 20
cigarettes for that price variable. Because smoking prevalence rates and cigarette prices
may be mutually dependent, column 4 uses an instrumental variables version of Craven
20 prices in which these price levels are predicted by excise tax rates and the overall
consumer price index. The 2012 Packaging Changes variable remains negative and
significant in the first equation in which the excise tax level variable replaces the excise
tax indicator variable. However, the 2012 Packaging Changes variable loses its
significance once any measure of overall prices is included.
82
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0045.png
Appendix Table A2
RMSS Data Regressions with Linear Trend, Including Tax, CPI, or Pack Cost
55
Smoker
Smoker
Smoker
Smoker
(IV)
2012 Packaging Changes, full
–0.0058*** –0.0029
–0.0032
–0.0029
(Dec 1, 2012)
(0.0018)
(0.0021)
(0.0021)
(0.0021)
Graphic warning labels, 2006
0.0020
0.0013
0.0022
0.0020
(0.0020)
(0.0020)
(0.0019)
(0.0019)
Cigarette tax, per cigarette
–0.0658***
(0.0127)
Consumer price index (2012=100)
-0.0004***
(0.0001)
Cost per pack (Craven 20, real $)
–0.0016*** –0.0017***
(IV estimates using tax, CPI)
(0.0003)
(0.0003)
Time trend, months
–1.13E-4*** –5.5E-5
–9.0E-5**
–8.3E-5**
(3.2E-5)
(4.1E-5)
(3.5E-5)
(3.5E-5)
Constant
0.5401***
0.5443***
0.5389***
0.5393***
(0.0050)
(0.0053)
(0.0049)
(0.0049)
R-squared
0.1139
0.1139
0.1139
0.1139
Notes:
Significance levels: *0.10, **0.05, ***0.01. Other variables include all those in Table A1.
5.
The first price measure is the overall CPI for all products, not just cigarettes. The
Australian CPI variable alone eliminates the significance of the 2012 Packaging Changes
variable, indicating that the purported 2012 Packaging Changes effect simply tracks other
economic trends. The pertinent cigarette price variables are the Craven 20 price per pack
in inflation-adjusted terms in the third column and an instrumental variables ("IV")
version of the Craven price per pack variable in the fourth column. The instrumental
variables technique is a statistical procedure to account for any possible mutual
dependence of smoking rates and cigarette prices. The instrumental variables estimator
55
These regressions also include the demographic and geographic variables in Table A1. These variables
have coefficients and significance levels that are stable across these regressions.
83
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0046.png
uses the economy-wide CPI and excise tax rates, which are exogenous, to predict the
level of cigarette prices.
56
6.
Table A3 of regression results with the RMSS data adds the quadratic time variable to
this set of equations. Otherwise the columns in Table A3 directly parallel those in Table
A2 in terms of the set of variables included in the analysis.
Appendix Table A3
RMSS Data Regressions with Quadratic Trend, Including Tax, CPI, or Pack Cost
57
Smoker
Smoker
Smoker
Smoker
(IV)
2012 Packaging Changes, full
–0.0019
–0.0013
–0.0015
–0.0015
(Dec. 1, 2012)
(0.0023)
(0.0022)
(0.0022)
(0.0022)
Graphic warning labels, 2006
0.0018
0.0015
0.0019
0.0018
(0.0020)
(0.0020)
(0.0019)
(0.0019)
Cigarette tax, per cigarette
–0.0135
(0.0231)
Consumer price index (2012=100)
–0.0001
(0.0001)
Cost per pack (Craven 20, real $)
–0.0003
–0.0006
(IV estimates using tax, CPI)
(0.0006)
(0.0007)
Time trend, months
3.1E-6
1.5E-5
5.5E-6
–3.2E-6
(5.3E-5)
(4.8E-5)
(5.2E-5)
(5.3E-5)
Time trend, squared
–0.0118*** –0.0111*** –0.0117**
–0.0099*
(4.4E-7)
(4.2E-7)
(4.8E-7)
(5.1E-7)
Constant
0.5268***
0.5290***
0.5265***
0.5288***
(0.0070)
(0.0079)
(0.0071)
(0.0074)
R-squared
0.1139
0.1139
0.1139
0.1139
Notes:
Significance levels: *0.10, **0.05, ***0.01. Other demographic and geographic variables include all those
in Table A1.
7.
The 2012 Packaging Changes variable loses its statistical significance in every instance.
Because of the correlation of the time squared variable with the nonlinear trajectory of
56
57
Instrumental variables estimates using only excise taxes as instruments and not the overall CPI likewise do
not indicate significant effects of the 2012 Packaging Changes.
These regressions also include the demographic and geographic variables seen in Table A1. These variables
have coefficients and significance levels that are stable across these regressions.
84
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0047.png
cigarette prices, the price variables also drop out of significance given the dominant role
of the nonlinear temporal trend. This lack of a significant influence on prices is not
because prices are unimportant. Rather, as was illustrated with the Craven 20 price
trajectory, prices have been rising in a nonlinear fashion and this determinant of smoking
prevalence rates is being captured by the nonlinear trend variable.
8.
Finally, Table A4 presents representative regressions including a linear and quadratic
time trend variable for different specifications of the advent of the 2012 Packaging
Changes. Column 1 introduces the 2012 Packaging Changes starting on December 1 but
using the full sample, column 2 introduces the 2012 Packaging Changes starting on
October 1 and also using the full sample, while column 3 introduces the 2012 Packaging
Changes starting on December 1 but excluding the data from October and November,
2012, as does Dr. Chipty.
In every instance there is no evidence of a statistically
significant effect of the 2012 Packaging Changes.
Appendix Table A4
RMSS Data: Regressions Predicting Smoking Prevalence by Policy Date
Full Excluding
Full (Dec. 1)
Partial (Oct. 1)
Oct.–Nov.
2012 Packaging Changes
–0.0030
–0.0023
–0.0030
(0.0026)
(0.0025)
(0.0027)
Graphic warning labels, 2006
0.0004
0.0005
0.0004
(0.0020)
(0.0020)
(0.0020)
Tax policy, 2010 (25%)
–0.0069***
–0.0067***
–0.0069***
(0.0025)
(0.0025)
(0.0025)
Tax policy, 2013 (12.5%)
–0.0033
–0.0036
–0.0033
(0.0028)
(0.0027)
(0.0028)
Tax policy, 2014 (12.5%)
0.0011
0.0012
0.0012
(0.0028)
(0.0028)
(0.0028)
Tax policy, 2015 (12.5%)
0.0003
0.0004
0.0003
(0.0027)
(0.0027)
(0.0027)
Tax policy, 2016 (12.5%)
–0.0055
–0.0054
–0.0055
(0.0036)
(0.0036)
(0.0036)
85
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0048.png
Time trend, months
Time trend, months (squared)
Female
Marital status, single
Marital status, divorced
Marital status, widowed
Marital status, separated
Student
Years of education
Age
Age (squared)
Non-adults (14-17)
Employed full time
Retired
Income
Income (squared)
Income, multiple
household members
Bread winner
Household size
Home owner
Victoria
Queensland
South Australia
7.8E-6
(5.9E-5)
–9.0E-7**
(4.4E-7)
–0.0396***
(0.0009)
0.0401***
(0.0013)
0.0985***
(0.0017)
0.0495***
(0.0019)
0.1278***
(0.0023)
–0.0781***
(0.0035)
–0.0199***
(0.0001)
0.0092***
(0.0002)
–1.4E-4***
(1.5E-6)
–0.2011***
(0.0027)
–0.0268***
(0.0012)
–0.0966***
(0.0040)
–0.0005***
(3.1E-5)
1.3E-6***
(1.3E-7)
–0.0188***
(0.0011)
0.0084***
(0.0011)
–0.0009**
(0.0004)
–0.1115***
(0.0010)
0.0131***
(0.0011)
0.0057***
(0.0012)
0.0082***
(0.0016)
1.2E-5
(5.9E-5)
–9.5E-7**
(4.4E-7)
–0.0396***
(0.0009)
0.0401***
(0.0013)
0.0985***
(0.0017)
0.0495***
(0.0019)
0.1278***
(0.0023)
–0.0781***
(0.0035)
–0.0199***
(0.0001)
0.0092***
(0.0002)
–1.4E-4***
(1.5E-6)
–0.2011***
(0.0027)
–0.0268***
(0.0012)
–0.0966***
(0.0040)
–0.0005***
(3.1E-5)
1.3E-6***
(1.3E-7)
–0.0188***
(0.0011)
0.0084***
(0.0011)
–0.0009**
(0.0004)
–0.1115***
(0.0010)
0.0131***
(0.0011)
0.0057***
(0.0012)
0.0082***
(0.0016)
8.8E-6
(5.9E-5)
–9.1E-7**
(4.5E-7)
–0.0396***
(0.0009)
0.0399***
(0.0013)
0.0982***
(0.0017)
0.0493***
(0.0019)
0.1281***
(0.0023)
–0.0780***
(0.0035)
–0.0198***
(0.0001)
0.0092***
(0.0002)
–1.4E-4***
(1.5E-6)
–0.2016***
(0.0027)
–0.0267***
(0.0013)
–0.0963***
(0.0041)
–0.0005***
(3.1E-5)
1.3E-6***
(1.3E-7)
–0.0188***
(0.0011)
0.0085***
(0.0011)
–0.0009**
(0.0004)
–0.1114***
(0.0010)
0.0132***
(0.0011)
0.0059***
(0.0012)
0.0084***
(0.0017)
86
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0049.png
Western Australia
Tasmania
Darwin-Alice Springs
Lives in capital city
Constant
R-squared
0.0056***
(0.0015)
0.0243***
(0.0020)
0.0348***
(0.0042)
–0.0260***
(0.0009)
0.5234***
(0.0047)
0.1139
0.0056***
(0.0015)
0.0243***
(0.0020)
0.0348***
(0.0042)
–0.0260***
(0.0009)
0.5233***
(0.0047)
0. 1139
0.0057***
(0.0015)
0.0242***
(0.0020)
0.0350***
(0.0042)
–0.0260***
(0.0009)
0.5243***
(0.0048)
0. 1139
Notes:
Significance levels: *0.10, **0.05, ***0.01.
87
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0050.png
APPENDIX B
CITTS Data
Appendix Table B1
CITTS Data: Regressions Predicting Cigarettes Smoked Per Day by Policy Date
Full Exclude
Full (Dec. 1)
Partial (Oct. 1)
Partial
2012 Packaging Changes
0.8904***
0.6038*
0.8346**
(0.3438)
(0.3330)
(0.3571)
Year trend
0.3940*
0.3651*
0.3985*
(0.2112)
(0.2143)
(0.2152)
Year trend, squared
-0.0561**
-0.0474**
-0.0551**
(0.0238)
(0.0233)
(0.0238)
Mobile phone survey participation
-0.8646***
-0.7866***
-0.8782***
(0.2540)
(0.2506)
(0.2538)
Female
-1.7319***
-1.7372***
-1.7148***
(0.1639)
(0.1639)
(0.1656)
Age
0.5766***
0.5767***
0.5665***
(0.0296)
(0.0296)
(0.0299)
Age, squared
-0.0048***
-0.0048***
-0.0047***
(0.0003)
(0.0003)
(0.0003)
Income, low
-0.1570
-0.1467
-0.0706
(0.2070)
(0.2070)
(0.2095)
Income, high
-0.1510
-0.0836
-0.0682
(0.4268)
(0.4254)
(0.4269)
Education, low
2.3043***
2.3117***
2.3398***
(0.3267)
(0.3267)
(0.3301)
Education, high
-1.5282***
-1.5278***
-1.5522***
(0.1722)
(0.1722)
(0.1739)
Number of children
0.0583
0.0606
0.0435
(0.0789)
(0.0789)
(0.0796)
English primary language
1.6657***
1.6644***
1.6716***
(0.2247)
(0.2248)
(0.2267)
Original peoples
1.9389***
1.9354***
1.8136***
(0.4183)
(0.4184)
(0.4246)
Constant
-3.6559***
-3.6635***
-3.4774***
(0.8228)
(0.8262)
(0.8317)
R-squared
0.07
0.07
0.07
Notes:
Significance levels: *0.10, **0.05, ***0.01.
88
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0051.png
Notes:
Significance levels: *0.10, **0.05, ***0.01.
Appendix Table B2
CITTS Data: Regressions Predicting Cigarettes Smoked Per Day by Policy Date, Quitters
Excluded
Full Exclude
Full (Dec. 1)
Partial (Oct. 1)
Partial
2012 Packaging Changes
1.4234***
1.0850***
1.4021***
(0.3544)
(0.3430)
(0.3674)
Year trend
0.5636***
0.5040**
0.5520**
(0.2168)
(0.2199)
(0.2206)
Year trend, squared
-0.0787***
-0.0659***
-0.0767***
(0.0244)
(0.0240)
(0.0245)
Mobile phone survey participation
-1.2610***
-1.1551***
-1.2598***
(0.2631)
(0.2596)
(0.2625)
Female
-1.9620***
-1.9691***
-1.9341***
(0.1691)
(0.1692)
(0.1707)
Age
0.5960***
0.5961***
0.5852***
(0.0307)
(0.0307)
(0.0309)
Age, squared
-0.0050***
-0.0049***
-0.0048***
(0.0003)
(0.0003)
(0.0003)
Income, low
-0.0541
-0.0388
0.0241
(0.2126)
(0.2126)
(0.2151)
Income, high
-0.0398
0.0496
0.0358
(0.4426)
(0.4412)
(0.4422)
Education, low
2.6286***
2.6376***
2.6625***
(0.3334)
(0.3334)
(0.3364)
Education, high
-1.5553***
-1.5565***
-1.5678***
(0.1775)
(0.1775)
(0.1791)
Number of children
0.1514*
0.1535*
0.1515*
(0.0812)
(0.0813)
(0.0819)
English primary language
2.2907***
2.2877***
2.2985***
(0.2309)
(0.2309)
(0.2326)
Original peoples
1.4122***
1.4069***
1.2227***
(0.4218)
(0.4219)
(0.4272)
Constant
-3.0505***
-3.0155***
-2.8705***
(0.8551)
(0.8589)
(0.8634)
R-squared
0.09
0.09
0.09
89
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0052.png
APPENDIX C
NTPPTS Data
58
Appendix Table C1
NTPPTS Data: Regressions Predicting Cigarettes Smoked Per Day by Policy Date
Full Exclude
Full (Dec. 1)
Partial (Oct. 1)
Partial
2012 Packaging Changes
0.1198
0.3733*
0.3380
(0.2029)
(0.2206)
(0.2231)
Age
0.4916***
0.4915***
0.4965***
(0.0453)
(0.0453)
(0.0471)
Age, squared
–0.0036***
–0.0036***
–0.0037***
(0.0005)
(0.0005)
(0.0006)
Female
–2.6352***
–2.6431***
–2.6330***
(0.1941)
(0.1942)
(0.2019)
Years of education
–0.5142***
–0.5135***
–0.5583***
(0.0384)
(0.0384)
(0.0401)
Income, high (> $100k)
0.3168
0.3119
0.2199
(0.2589)
(0.2589)
(0.2696)
Income, low (< $30k)
1.1297***
1.1286***
1.0538***
(0.2712)
(0.2712)
(0.2813)
Original peoples
1.6490***
1.6553***
1.5516***
(0.5042)
(0.5041)
(0.5246)
English primary language
3.4684***
3.4641***
3.4061***
(0.3817)
(0.3816)
(0.3971)
Phone: land line in home
0.0552
0.0608
0.1234
(0.2374)
(0.2373)
(0.2463)
Phone: mobile line in home
–0.5414
–0.5441
–0.1341
(0.3812)
(0.3811)
(0.3951)
Hours of television per day
0.1944***
0.1956***
0.1910***
(0.0436)
(0.0436)
(0.0452)
Victoria
–0.0918
–0.0955
–0.1094
(0.2538)
(0.2538)
(0.2639)
Queensland
1.2380***
1.2375***
1.1982***
(0.2720)
(0.2720)
(0.2831)
South Australia
0.3751
0.3709
0.1701
(0.3886)
(0.3885)
(0.4012)
Western Australia
1.4274***
1.4254***
1.4012***
(0.3449)
(0.3448)
(0.3590)
Tasmania
0.6510
0.6541
0.7958
(0.6520)
(0.6519)
(0.6756)
Northern Territory
0.8806
0.8728
1.0230
58
Regressions also include variables identifying whether data for demographic variables are missing. Missing
data coded as zero.
90
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0053.png
Capital city
Constant
R-squared
Notes:
Significance levels: *0.10, **0.05, ***0.01.
(0.8681)
-0.8438***
(0.2119)
4.8122***
(1.1703)
0.15
(0.8680)
-0.8410***
(0.2119)
4.6228***
(1.1747)
0.15
(0.9040)
-0.8660***
(0.2209)
4.8703***
(1.2211)
0.15
91
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0054.png
APPENDIX D
EDUCATIONAL BACKGROUND AND PROFESSIONAL EXPERIENCE
1.
I am the University Distinguished Professor of Law, Economics, and Management at
Vanderbilt University, where I hold tenured appointments in the Vanderbilt University
Law School, the Department of Economics, and the Owen Graduate School of
Management. I have previously held tenured full professor positions at Harvard Law
School, Duke University, and Northwestern University. I hold a Bachelor’s degree in
Economics, two master’s degrees, and a Ph.D. in economics, all from Harvard
University. I graduated summa cum laude, Phi Beta Kappa, and won the awards at
Harvard University for the best undergraduate thesis in economics and the best doctoral
dissertation in economics.
2.
My research focuses on the economics of risk and uncertainty, with particular emphasis
on risks to health and safety. I have published more than 350 articles and 20 books
dealing primarily with health and safety risks. Most of these articles and books have been
peer reviewed. I have been ranked among the top 25 economists in the world based on
citations in economics journals and have been ranked as the leading contributor to the
health economics literature by
Health Economics
and the leading contributor to the risk
and insurance literature by the
Journal of Risk and Insurance.
My research has won
numerous article of the year and book of the year awards from organizations such as the
Royal Economic Society and the American Risk and Insurance Association. I am the
founding Editor of the
Journal of Risk and Uncertainty,
which is the leading international
journal in its field and which I continue to edit.
92
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0055.png
3.
My research currently focuses on how consumers make decisions involving products
such as cigarettes and drinking water that may pose precisely understood risks and more
uncertain risks. Much of my research has analyzed hazard warnings and how they affect
consumer behavior. I worked extensively with the U.S. Environmental Protection Agency
(“EPA”) on a continuous basis from 1983 to 2012, serving in several different roles.
Much of my work for EPA has focused on the development of guidelines for hazard
warnings for dangerous pesticides and chemicals. These studies involved an experimental
structure in which consumers reviewed different warnings, assessed the implied risks,
and indicated the precautions that they would take in using the product. This work has
appeared in numerous articles, and much of it is summarized in two books with Wesley
Magat:
Learning about Risk: Consumer and Worker Responses to Hazard Information
(Cambridge: Harvard University Press, 1987), and
Informational Approaches to
Regulation
(Cambridge: MIT Press, 1992). I have also written many articles and two
peer-reviewed books devoted to consumer decisions pertaining to smoking,
Smoking:
Making the Risky Decision
(Oxford University Press, 1992) and
Smoke-Filled Rooms: A
Postmortem on the Tobacco Deal
(University of Chicago Press, 2002). None of this
research has been funded by the tobacco industry or law firms representing the industry.
4.
In addition to my extensive work for EPA, I have consulted for several other
governmental and private entities on a variety of issues. I have taught courses about risk,
uncertainty, risk analysis, and hazard warnings to hundreds of U.S. Food and Drug
Administration officials, congressional staff, and federal and state judges. I served as the
Associate Reporter on The American Law Institute Study on Enterprise Responsibility
for Personal Injury and co-wrote the chapter on Product Defects and Warnings. I have
93
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0056.png
testified before the U.S. Congress on nine occasions as an expert in economics and risk
analysis. This testimony addressed such topics as, for example, alcoholic beverage
warnings.
5.
Apart from my academic and governmental work, I have consulted on matters such as
risk perception, hazard warnings design, and safety devices for large companies,
including Bic, Dupont, Becton Dickinson, Bristol-Meyers Squibb, Anheuser-Busch,
Black & Decker, R.J. Reynolds, and Medline Industries. During this period, I also
directed studies for the U.S. Environmental Protection Agency on the design and policy
role of hazard warnings for chemicals and pesticides. I testified on behalf of the Province
of Québec in the Loto Québec class action, Jean Brochu v. Loto Québec, regarding
warnings for video lottery terminals. I also have testified in Québec on behalf of JTI-
Macdonald Corp. and Rothmans, Benson & Hedges Inc. in the Blais and Létourneau
cigarette class actions. I have submitted several expert reports on behalf of British
American Tobacco group companies in relation to proposed tobacco regulation, including
the introduction of graphic health warning requirements and legal challenges to such
regulation. I have also testified on tobacco-related issues and have submitted expert
reports in various U.S. proceedings on behalf of cigarette companies. I have also served
as an expert witness on other matters, such as economic damages in wrongful death and
personal injury cases and hazardous waste site remediation efforts. My discussion below
draws on my professional expertise and knowledge of the literature on risk and warnings.
6.
I have extensive professional experience evaluating regulatory impact analysis and the
economic methodology used in benefit-cost analysis. From 1979-1980, I was the Deputy
Director of the President’s Council on Wage and Price Stability, which was responsible
94
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0057.png
for White House oversight of all new federal regulations during that period as well as
executive branch review of all regulatory impact analyses. I served as the President of the
Society for Benefit-Cost Analysis in 2015.
7.
A full copy of my Curriculum Vitae is available at
http://www.vanderbilt.edu/econ/faculty/cv/ViscusiCV.pdf
95
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0058.png
The effects of standardised
packaging: an empirical analysis
10 October 2017
Neil Dryden
1
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0059.png
Contents
Section 1
Introduction
Instructions
Assistance
Structure of the report
Summary of conclusions
The contribution of this report to the debate on standardised packaging
Empirical analysis
Introduction
Methodology
New Zealand is a good comparator for Australia
The impact of standardised packaging on the Australian tobacco market
Consumption
Prices
Down-trading
Alternative consumption analyses
Introduction
The need to account for the declining trend in cigarette consumption
The before-during approach
First method: linear trend and monthly indicator variables
Second method: including consumption in a different country as an
explanatory variable
The prediction approach
First method: linear trend and monthly indicator variables
Second method: including consumption in a different country as an
explanatory variable
Reasons for preferring the DID approach
The way my DID approach captures the declining trend in cigarette
consumption is superior to the linear trend method
The DID approach is superior to including New Zealand cigarette
consumption per capita as an explanatory variable
Conclusions
Curriculum vitae of Neil Dryden
Consumption analysis
Benchmark analysis
Quarterly regressions
Results with the model used in previous submissions
1
1
2
2
4
6
12
12
12
15
18
18
22
25
31
31
32
33
34
35
37
37
39
39
40
40
41
43
44
44
50
53
Section 2
Section 3
Section 4
Section 5
Annex A
Annex B
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0060.png
Annex C
Price analysis
Average price
Quarterly regressions
Brand level prices
Results with the model used in previous submissions
Alternative consumption analyses
Before-during approach
Prediction approach
55
55
59
62
70
72
72
74
Annex D
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0061.png
Section 1
Introduction
1.1
I am Neil Dryden, Executive Vice President at Compass Lexecon, an economic consulting
firm. Compass Lexecon is part of FTI Consulting Inc., a global business advisory firm. My
experience and expertise is as a micro-economist, specialised in the economics of
competition policy, regulation, public policy and market analysis. I have an M.Phil. in
Economics from Oxford University and a postgraduate diploma in EC competition law (with
distinction) from King’s College, London. I have worked as a professional economist for over
20 years, including advising on numerous mergers, agreement cases, dominance cases,
damages and market investigations. I was awarded the 2016 Economist of the Year by the
Global Competition Review.
My CV is included at Annex A.
1.2
Instructions
1.3
1.4
I have been commissioned to prepare this report for British American Tobacco (“BAT”).
I have been asked to:
a. set out my views on the expected impact that standardised packaging regulations will
have on the market for the supply of cigarettes from a theoretical point of view;
b. examine empirical data from Australia, where standardised packaging has been
2
implemented, to see what effect, if any, standardised packaging has had on:
i.
ii.
the consumption of cigarettes in Australia;
the prices of cigarettes in Australia; and
1
1
Standardised or plain packaging generally refers to the use of the same uniform colour on all tobacco
packs, with no brand imagery, and the brand name written in a specified font, colour and size.
Standardised packaging was introduced in Australia under the Tobacco Plain Packaging Act 2011, No.
148, 2011 with all tobacco products sold in Australia required to comply with the requirements from 1
December 2012.
2
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0062.png
iii.
consumers shifting from premium to non-premium brands (i.e., down-trading); and
c. where appropriate, use New Zealand as a comparator.
1.5
For this purpose, I have been asked to analyse the following datasets:
a. Nielsen and IRI–Aztec retail data regarding cigarettes and roll your own tobacco from
January 2009 to December 2016 for Australia; and
b. Nielsen scanner data regarding cigarettes and roll your own tobacco from January 2008
to December 2016 for New Zealand.
3
Assistance
1.6
I was assisted in preparing this report by Nadine Watson and Stefano Trento, Senior Vice
President and Vice President, respectively, at Compass Lexecon. However, the opinions
contained in this report are mine alone.
Structure of the report
1.7
1.8
The next section sets out a summary of my conclusions.
The remainder of the report is then structured in three parts as follows.
a. In Section 3, I explain why standardised packaging may in theory increase cigarette
consumption, contrary to the health objectives of such regulation. I also review the
current state of the published research on the effects of standardised packaging on
tobacco use, and explain how the data I have been provided with allows me to overcome
some of the limitations of this research and undertake a significantly more robust
analysis – on the questions that I have been asked – than any that has been published
to date.
b. In Section 4, I set out the main empirical analyses I have carried out to analyse the
effects of standardised packaging on cigarette consumption, prices and down-trading in
Australia using four years' worth of post-implementation data.
3
I have previously written a number of economic expert reports for BAT analysing the impact of various
regulations on competition in markets for tobacco products. In particular, I have submitted a number of
reports considering the impact of standardised packaging, and providing an empirical analysis of the
effects of standardised packaging on consumption, prices and down-trading in Australia, including
expert reports submitted in legal proceedings in the UK and France. In this report I examine a longer
data series than I had previously looked at, using data through to December 2016.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0063.png
c. In Section 5, I set out alternative empirical analyses on the effect of standardised
packaging on cigarette consumption, which I have carried out in addition to my main
analysis, and on the basis that similar approaches been adopted by other authors
analysing the effect of standardised packaging in Australia.
1.9
After Annex A, which sets out my CV, Annexes B, C and D set out the detailed results and
robustness checks of my empirical analysis.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0064.png
Section 2
Summary of conclusions
2.1
In this report, I first explain that, as a matter of economic theory, standardised packaging
may result in an increase of cigarette consumption, contrary to the heath objectives of the
regulation. I then briefly review the published academic research on the effects of
standardised packaging on cigarette consumption, set out its limitations, and explain how the
data I have been provided with allows me to overcome some of these limitations and
undertake a significantly more robust analysis – on the questions that I have been asked –
than any that has been published to date.
I secondly proceed to carry out my own empirical analysis on the effects of standardised
packaging in Australia, using scanner sales data (i.e., data collected using scanner systems
at retail store checkouts) up to and including December 2016, i.e., four years after the full
implementation of standardised packaging in Australia.
In particular, I analyse the effects of standardised packaging on cigarette consumption,
cigarette prices, and down-trading (i.e., consumers switching from premium to lower-quality
brands). For this empirical analysis, I proceed in four steps.
First,
I describe the methodology used to analyse consumption and pricing in my main
analysis. In particular I explain how I analyse the impact of standardised packaging in
Australia using New Zealand as a benchmark comparator (but taking account of differences
between the two countries, such as different tax policies or a different evolution of income
per capita).
I explain that, provided that New Zealand is a good benchmark comparator, such analysis,
known as difference-in-differences (“DID”), is more reliable than other statistical
methodologies (such as a before-during analysis of cigarette consumption in Australia) for
analysing the effects of standardised packaging on cigarette consumption in the presence of
confounding factors that also affect cigarette consumption but whose effects are difficult to
measure.
Secondly,
I carry out an econometric analysis to test whether New Zealand is a good
comparator for Australia in relation to tobacco consumption. I find that this is the case.
Thirdly,
I present the results of my main analysis on the effects of standardised packaging
on consumption, prices and down-trading. That analysis shows the following:
2.2
2.3
2.4
2.5
2.6
2.7
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0065.png
a. First, standardised packaging is associated with an increase in the per capita
consumption of cigarettes in Australia relative to the consumption that would have
prevailed had standardised packaging not been implemented, i.e., relative to the
counterfactual. In particular, I find that standardised packaging is associated with an
increase in per capita cigarette consumption (relative to the counterfactual and up to
December 2016) of 3.1%-3.5%, when control variables such as prices, excise taxes and
income per capita are expressed in local currencies; and an increase of 2.2%-3.0%
when these variables are expressed in purchasing power parity ("PPP").
b. Secondly, standardised packaging is associated with a reduction in the average retail
price paid by consumers for cigarettes in Australia relative to the counterfactual. While
average retail prices are also influenced by other factors, such as excise taxes and
consumers’ income, the methodology I use allows me to isolate the effect of
standardised packaging from the effect of these other factors on those prices. In
particular, I find that standardised packaging is associated with a decrease in the
average price paid by consumers (relative to the counterfactual and up to December
2016) of 2.0-2.6%. I also analyse the effect of standardised packaging on each brand
within the subset of brands (comprising 21 brands) that are sold both in Australia and in
New Zealand. I find that the weighted average price of these brands decreased by 6.8-
7.3% (relative to the counterfactual and up to December 2016).
c. Thirdly, while consumers have been shifting from premium to non-premium brands
(‘down-trading’) in Australia since at least 2009, the adoption of standardised packaging
is associated with a significant acceleration of this down-trading trend.
2.8
2.9
These results are reliable across a large set of robustness checks.
Fourthly,
in addition to my preferred DID analysis, I also carry out alternative analyses of the
effect of standardised packaging on cigarette consumption. Although they are not based on
my preferred method, I have included these alternative analyses on the basis that similar
approaches have been adopted by other authors for estimating the effect of standardised
packaging in Australia.
Although, in my opinion, these analyses are carried out in a less rigorous framework, the
results are consistent with those of my DID analysis, and indicate that standardised
packaging is associated with an increase in cigarette consumption per capita relative to the
counterfactual.
4
2.10
4
For this analysis, I refer to brands in a broad sense. For example, I refer to Dunhill as a brand, although
this brand comprises brand variants such as Dunhill, Dunhill Essence, Dunhill International and others.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0066.png
Section 3
The contribution of this report to the
debate on standardised packaging
3.1
Standardised packaging of tobacco products generally involves the use of the same uniform
colour on all tobacco packs, with no brand imagery, and the brand name written in a
5
specified font, colour and size.
The primary objective of standardised packaging is to reduce smoking by, among other
6
things, reducing the appeal of tobacco products to consumers. The key question is thus
whether standardised packaging reduces smoking. I note that this outcome is not
necessarily the case from the point of view of economic theory, as explained immediately
below.
Standardised packaging may lead to an increase in cigarette consumption
3.2
3.3
Economic theory shows that when products are similar to each other (or, in economic terms,
homogenous) the main way for suppliers to compete is by offering competitive prices. If they
were to attempt to charge a premium, they would experience an ‘exodus’ of consumers away
from their product towards cheaper and similar products. By contrast, when consumers
perceive the products as less similar (or, in economic terms, differentiated), price competition
is reduced because consumers are more ‘loyal’ to their preferred product (meaning that they
would not necessarily switch product as a result of a price increase) and thus suppliers can
charge higher prices.
I understand that, but for branding, tobacco products are highly substitutable in the eyes of
consumers, with research studies indicating that in the absence of brand information, many
3.4
5
McNeill, A., Gravely, S., Hitchman, S. C., Bauld, L., Hammond, D., & Hartmann Boyce, J. (2017).
Tobacco packaging design for reducing tobacco use. The Cochrane Library. Issue 4. Art. No.:
CD011244, page 3.
6
See for example UK Department of Health (2015) Standardised packaging of tobacco products: Impact
Assessment, IA No:3080, page 1.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0067.png
smokers are unable to identify different cigarette brands correctly. I also understand that
packaging helps cigarette producers build consumer loyalty and distinguish their products
8
from those of their competitors, thus allowing them to charge price premia for their products.
3.5
Standardised packaging eliminates by its very nature the ability of tobacco companies to
compete through packaging differentiation. Since – due to limitations in advertising and other
restrictions – packaging is the last means by which cigarette producers can communicate to
consumers, standardised packaging can be expected to intensify price competition among
tobacco companies and thus to reduce the price of tobacco products (relative to the
9
counterfactual of no standardised packaging).
To the extent that cigarette consumption responds to cigarette prices, of which there is
10
abundant evidence in the economic literature, standardised packaging may result in
increased cigarette consumption.
Even if standardised packaging also had the effect of reducing the appeal of tobacco
11
products in the eyes of consumers, as argued by proponents of such regulation, the overall
effect of standardised packaging on cigarette consumption would depend on the relative
strength of the ‘price’ effect and of the ‘appeal’ effect. A variety of theoretical models could
be used to shed light on the relative strength of these two effects. However, given that we
now have four years’ worth of post-implementation data for Australia, which allows for a
robust identification of the effects of standardised packaging, in my opinion these effects
(including those on consumption) are better analysed using an empirical approach.
In the next section I present the results of my empirical analysis, which uses cigarette
consumption data (as proxied by scanner sales data) in Australia from January 2009 to
7
3.6
3.7
3.8
7
Ramond, C. K.; Rachal, L. H.; Marks, M. R., (1950) Brand discrimination among cigarette smokers.
Journal of Applied Psychology, Vol 34(4); and Jaffe A.J., Glaros A.G. (1986) Taste dimensions in
cigarette discrimination: a multidimensional scaling approach. Addict Behav 11.
UK Department of Health (2015) Standardised packaging of tobacco products: Impact Assessment, IA
No:3080, paragraph 63.
Davidson, S., and de Silva, A. (2014). The plain truth about plain packaging: An econometric analysis
of the Australian 2011 tobacco plain packaging Act. Agenda: A Journal of Policy Analysis and Reform,
page 29.
See HMRC (2010) “Econometric Analysis of Cigarette Consumption in the UK”, Working Paper
Number 9; HMRC (2015) “Econometric Analysis of Cigarette Consumption in the UK”, Update to
Working Paper Number 9. Chaloupka F.J. (1991) “Rational addictive behavior and cigarette smoking”,
Journal of Political Economy 1991; 99(4), page 735 (“current cigarette consumption is found to be
significantly negatively related to the current price of cigarettes”). Chaloupka, F.J. (1992) “Clean indoor
air laws, addiction, and cigarette smoking”, Applied Economics 24(2), pages 202-203 (“increased
cigarette prices […] are also found to have a negative significant impact on cigarette consumption”).
See paragraph 3.2.
8
9
10
11
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0068.png
December 2016, i.e., data for four full years of post-implementation period. Before doing
so, I review below the academic research on the effect of standardised packaging on
tobacco use, and I explain how the data I have been provided with allows me to overcome
some of the limitations of this research and undertake a significantly more robust analysis –
on the questions that I have been asked – than any that has been published to date.
Studies on the effect of standardised packaging on tobacco use
3.9
A recent review of the academic literature on the effects of standardised packaging on
13
cigarette consumption can be found in the 2017 Cochrane Review. This review only
includes published, peer-reviewed articles and identifies five studies that analyse the effect
14
of standardised packaging on tobacco use. It grades these studies as either:
a. "low
quality",
meaning that the authors of the review had limited confidence in the study
and that "[t]he
true effect may be substantially different from the estimate of the effect";
or
b. "very
low quality",
meaning that the authors of the review had very little confidence in the
study conclusions and "the
true effect is likely to be substantially different from the
estimate of effect".
3.10
One of these five studies (Diethelm and Farley 2015 (‘[1]’)) focuses on the effects of
standardised packaging on smoking prevalence in Australia and finds that standardised
12
12
To date, only Australia (in December 2012) and France (in January 2017) have implemented
standardised tobacco packaging, although the French experience is too recent to allow for an empirical
analysis.
The Cochrane Database of Systematic Reviews is a leading resource for systematic reviews in health
care. The Cochrane Review on the effects of standardised packaging is McNeill, A., Gravely, S.,
Hitchman, S. C., Bauld, L., Hammond, D., & Hartmann Boyce, J. (2017). Tobacco packaging design for
reducing tobacco use. The Cochrane Library. Issue 4. Art. No.: CD011244
13
14
McNeill, A., Gravely, S., Hitchman, S. C., Bauld, L., Hammond, D., & Hartmann Boyce, J. (2017).
Tobacco packaging design for reducing tobacco use. The Cochrane Library. Issue 4. Art. No.:
CD011244, pages 4-5. I note that one of the reasons (but not the only reason) why the authors have
limited confidence on the results of some of these studies is that they analyse the effect of
standardised packaging using data from Australia, where enhanced pictorial health warnings were
implemented at the same time as standardised packaging, making it difficult to separate the effects of
these two measures on tobacco use. This is a critique that applies inevitably to all studies that use
observational data from Australia.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0069.png
packaging is associated with a reduction of smoking prevalence using survey data on self-
15,16
reported smoking status among Australian consumers.
3.11
The other four studies focus on the effects of standardised packaging on tobacco
consumption (rather than on smoking prevalence):
a. Scollo et al. (2015) (‘[2]’) use survey data on self-reported smoking status among
Australian consumers from April 2012 (eight months before the full implementation of
standardised packaging) to March 2014 (fifteen months after the full implementation of
standardised packaging) and find that tobacco consumption did not change in the year
after the implementation of standardised packaging, and that it decreased after the
17
12.5% tax increase on tobacco products of December 2013.
b. Miller et al. (2015) (‘[3]’) use post-implementation survey data, where a limited number of
cigar or cigarillo smokers (256) were asked to self-report changes in consumption since
a period before the implementation of standardised packaging. The authors find that
more smokers report a decrease than an increase in consumption, but acknowledge the
limitation of the study, especially with respect to the representativeness of the sample
and the accuracy of self-report measures.
18
15
Diethelm, P.A., and Farley, T.M., (2015) Refuting tobacco-industry funded research: empirical data
shows decline in smoking prevalence following introduction of plain packaging in Australia, Tob. Prev.
Cessation 2015; 1(November): 6.
The 2017 Cochrane Review also mentions, but does not review in depth, two additional unpublished
studies that analyse the impact of standardised packaging on smoking prevalence using the same
survey data as Diethelm and Farley (2015): a study by Kaul and Wolf (2014) that finds no evidence for
a plain packaging effect on smoking prevalence (Kaul, A., and Wolf, M. (2014) The (possible) effect of
plain packaging on smoking prevalence in Australia: A trend analysis); and a study by Dr Chipty (T.
Chipty (2016). Study of the Impact of the Tobacco Plain Packaging Measure on Smoking Prevalence in
Australia, Appendix A to the Australian Government’s Post-Implementation Review on Tobacco Plain
Packaging), which uses more updated survey data (to September 2015) and finds results similar to
Diethelm and Farley (2015). I note that this is the study relied upon by the Australian Government in its
Post-Implementation Review on Tobacco Plain Packaging.
16
17
Scollo, M., Zacher, M., Coomber, K., Bayly, M., and Wakefield, M. (2015). Changes in use of types of
tobacco products by pack sizes and price segments, prices paid and consumption following the
introduction of plain packaging in Australia. Tobacco control, 24 (Suppl 2).
Miller, C., Ettridge, K. A., and Wakefield, M. A. (2015). You're made to feel like a dirty filthy smoker
when you're not, cigar smoking is another thing all together. Responses of Australian cigar and cigarillo
smokers to plain packaging. Tobacco Control, 24.
18
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0070.png
c. The other two studies, Maynard et al. (2015) (‘[4]’)
19
and Moodie et al. (2013) (‘[5]’),
20
use UK experimental data with a limited number of participants (128 and 187,
respectively). Maynard et al. (2015) assign participants randomly to either of two groups
of smokers: one group is given branded cigarette packs and the other group is given
standardised cigarette packs; the authors analyse whether there is any difference in
cigarette consumption between the two groups. Moodie et al. (2013) use a different but
similar approach. Maynard et al. (2015) find that standardised packaging does not
affect consumption, while Moodie et al. (2013) find that standardised packaging reduces
consumption.
The contribution of this report
3.12
The data I have been provided with allows me to overcome many of the limitations of the
studies reviewed above and undertake a significantly more robust analysis – on the
questions that I have been asked – than any that has been published to date.
First, among studies that review the Australian experience ([1], [2], [3]), I use a materially
longer time series that provides information on cigarette consumption up to and including
December 2016, i.e.,
four years after the full implementation of standardised packaging
in Australia.
This compares to the use of one year of post-implementation data in [1] and
22
fifteen months of post-implementation data in [2]. [3] uses information collected in February
and March 2014 (i.e., approximately fifteen months after the full implementation of
standardised packaging) via a survey, and asks survey participants to self-report changes in
consumption since a period preceding the implementation of standardised packaging. I note
that reviewing a longer post-implementation period is an advantage because, as the
Australian Post Implementation Review states, “the
full effect of the tobacco plain packaging
23
measure is expected to be realised over time”.
21
3.13
19
Maynard, O. M., Leonards, U., Attwood, A. S., Bauld, L., Hogarth, L., and Munafò, M. R. (2015). Effects
of first exposure to plain cigarette packaging on smoking behaviour and attitudes: a randomised
controlled study. BMC public health, 15(1), 240.
Moodie, C. S., and Mackintosh, A. M. (2013). Young adult women smokers’ response to using plain
cigarette packaging: a naturalistic approach. BMJ open, 3(3), e002402.
In particular, participants to the Moodie et al. (2013) study self-reported their smoking habits during two
weeks: in one week they used plain cigarette packs provided to them, and in the other week they used
their own fully branded packs. Moodie et al (2013) analyse the differences in consumption between the
two weeks.
In her report relied upon by the Australian Government for its Post Implementation Review (see
footnote 16), Dr Chipty uses data up to September 2015. My report thus uses a fifteen month longer
post-implementation period to that used by Dr Chipty.
Australian Government (2016) Post-Implementation Review on Tobacco Plain Packaging, page 4.
20
21
22
23
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0071.png
3.14
Secondly, in this report I proxy cigarette consumption using scanner sales data rather than
self-reported smoking status like [1], [2], [3], and [5]. I consider this to be an advantage to the
extent that the academic literature has shown that smokers tend to under-report their
24 25
smoking status (a limit that is acknowledged by the authors of [3] ).
Thirdly, some of the studies reviewed above ([3], [4], and [5]) are based on a very limited
number of observations, which casts doubt on the representativeness of their results, as
26
expressly acknowledged by the authors of [3]. In contrast, in this report I use scanner sales
data that covers the vast majority of the Australian tobacco market.
Fourthly, in this report I use observational data from the Australian experience rather than
experimental data like [4] and [5]. In my opinion in this case actual rather than experimental
data are preferable. This is because experiments such as those performed in [4] and [5] (i.e.,
giving branded packages to one group and standardised packaging to another group of
smokers that are otherwise similar, and assessing whether there is any difference in
27
consumption among the two groups) may capture the ‘appeal’ effect (if any), but miss
important effects of standardised packaging on cigarette consumption, e.g., through its
impact on prices and down-trading. These effects are instead captured in analyses that use
actual data, such as that used in this report.
In conclusion, the academic literature on the effects of standardised packaging on tobacco
use is scarce and has some significant limitations. In this report I use data that allows me to
overcome many of these limitations and undertake a significantly more robust analysis – on
the questions that I have been asked – than any that has been published to date.
3.15
3.16
3.17
24
Miller, C., Ettridge, K. A., and Wakefield, M. A. (2015). You're made to feel like a dirty filthy smoker
when you're not, cigar smoking is another thing all together. Responses of Australian cigar and cigarillo
smokers to plain packaging. Tobacco Control, 24, page ii64.
Gorber, S. C., Schofield-Hurwitz, S., Hardt, J., Levasseur, G., and Tremblay, M. (2009) The accuracy
of self-reported smoking: a systematic review of the relationship between self-reported and cotinine-
assessed smoking status. Nicotine & Tobacco Research, 11(1).
25
26
Miller, C., Ettridge, K. A., and Wakefield, M. A. (2015). You're made to feel like a dirty filthy smoker
when you're not, cigar smoking is another thing all together. Responses of Australian cigar and cigarillo
smokers to plain packaging. Tobacco Control, 24, page ii64. The limitations of a small sample size are
also mentioned at paragraph 208 of
The Final Impact Assessment on standardised packaging of
tobacco products
by the UK Department of Health, dated 10 February 2015.
See paragraph 3.7.
27
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0072.png
Section 4
Empirical analysis
Introduction
4.1
In this section, I set out the empirical analysis I carried out to assess the impact of
standardised packaging in Australia.
Following this introduction, the rest of the section is organised in three subsections:
a. First, I describe the methodology used for my empirical analysis. In particular I explain
the difference-in-differences (”DID”) methodology, that allows me to analyse the impact
of standardised packaging in Australia using New Zealand as a benchmark comparator
(but taking account of differences between the two countries, such as different tax
policies or a different evolution of income per capita).
b. Secondly, I set out the results of the econometric analysis I carried out in order to
establish that New Zealand is a good comparator for Australia in relation to tobacco
consumption, and thus that the DID methodology is reliable.
c. Thirdly, I present the results of my econometric analysis on the effects of standardised
packaging on cigarette consumption, cigarette prices and down-trading.
4.3
I find that:
a. First, standardised packaging is associated with an increase in consumption relative to
the counterfactual of no standardised packaging.
b. Secondly, standardised packaging is associated with a reduction in prices relative to the
counterfactual of no standardised packaging.
c. Thirdly, while consumers have been shifting from premium to non-premium brands in
Australia since at least 2009, the adoption of standardised packaging is associated with
a significant acceleration of this down-trading trend.
4.2
Methodology
4.4
Any statement such as ‘standardised packaging is likely to have increased/decreased
consumption of cigarettes’ is inherently based on a view about what would have happened in
the absence of standardised packaging.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0073.png
4.5
In particular, any such statement implies that a comparison has been made between the
actual level of consumption (“the factual”) and the level of consumption that we would have
observed in a hypothetical scenario where standardised packaging had not been adopted
(“the counterfactual”).
While the factual is observable from post-implementation data, the counterfactual is not
(since the hypothetical scenario has not happened), and one thus needs to find a reliable
way of estimating it. This is not a trivial exercise. For instance, it would be wrong to assume
that – in the absence of standardised packaging – cigarette consumption in Australia would
have been the same as in the pre-implementation period: on the contrary, cigarette
consumption would have been likely to change due to some inertia from previous decreasing
trends (e.g., because, as healthier lifestyles becomes more widespread, people may decide
to quit or reduce smoking, and older generations of heavier smokers are replaced by
younger generations of no-smokers or lighter smokers) and also due to the effects of other
relevant policy measures (such as the excise tax increases in Australia, including the large
excise tax increases introduced in December 2013 and September 2014, 2015 and 2016).
In this section, I estimate the counterfactual level of cigarette consumption using a DID
approach. DID analyses are widely used in economics for impact evaluation of public
28
policies, including for the evaluation of tobacco-related policies. For instance, in their paper
“Recent developments in the econometrics of program evaluation”, Professors Imbens and
Wooldridge note that “since
the seminal work by Ashenfelter (1978) and Ashenfelter and
Card (1985), the use of Difference-In-Differences (DID) methods has become widespread in
29
empirical economics”.
In their simplest form, DID analyses are carried out as follows. Imagine if I were to test
whether the free distribution of mosquito nets is effective at reducing the incidence of malaria
in a malaria-plagued area. I would first randomly select two groups of families: those who will
receive the mosquito net (“the treatment group”) and those who will not receive it (“the
control group”). I would then measure malaria incidence in these two groups before the
distribution of the mosquito nets. I would then repeat the measurement (again, for the two
groups) after the distribution of the mosquito nets.
I would then compare the average reduction in the malaria incidence between the two
groups. If I were to find that malaria incidence had reduced more in the treatment group
(e.g., by 20%) than in the control group (e.g., by 15%), and that this difference was
4.6
4.7
4.8
4.9
28
For example see analysis carried out by the U.S. Food and Drug Administration on the impact of
graphic health warnings on cigarette packets which used the U.S. as a comparator for analysing the
effects of graphic health warnings introduced in Canada: U.S. Dept. of Health and Human Services,
Food and Drug Administration, Required Warnings for Cigarette Packages and Advertisements – Final
Rule, Federal Register, Vol. 76, No. 120, June 22, 2011, pp 36719-36721.
Imbens, G.M., and J.M. Wooldridge (2009), "Recent developments in the econometrics of program
evaluation"
Journal of Economic Literature 47, no. 1, page 64
29
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0074.png
statistically significant, I would conclude that the free distribution of mosquito nets reduces
the incidence of malaria (e.g., by 20% - 15% = 5%).
4.10
The DID analysis is more reliable than other, less sophisticated, statistical methodologies.
For instance, imagine if – between the first and the second measurement – a pesticide
programme was carried out that reduced the incidence of malaria by 15% in the relevant
area (which would explain why the incidence of malaria also decreased for the control
group). Had I not used a DID analysis (i.e., a comparison with a control group) but only a
simplistic approach of comparing the incidence of malaria for the treatment group before and
after the distribution of mosquito nets), I would have incorrectly attributed the full reduction of
30
malaria incidence of the group receiving free nets (i.e., 20%) to the mosquito nets.
Other econometrics methods exist in theory to control for the effect of confounding factors.
For example one could carry out a before-during regression that includes the pesticide
31
programme as an explanatory variable for the change in malaria incidence.
However,
these alternative methods are less reliable than the DID approach when the confounding
factors are not observable or are difficult to control for, as is the case for the decreasing
trend in cigarette consumption, e.g., as a result of healthier lifestyles becoming more
prevalent. In those cases, the DID approach is more reliable because – insofar as the
confounding factor has the same or similar effects in the treatment group and in the control
group – it automatically isolate the effect of the relevant variable (e.g., standardised
packaging, or mosquito nets) from the effect of the confounding factor (e.g., the decreasing
trend in cigarette consumption, or the pesticide programme).
As is clear from the above, a DID approach requires a reliable benchmark comparator, or
control group, which is similar to the treatment group but which is not affected by the policy.
Some authors – including the Australian Bureau of Statistics and the New Zealand Ministry
4.11
4.12
30
Gertler P.J., Martinez S., Premand P., Rawlings L.B., Vermeersch C.M.J. (2011) “Impact Evaluation in
Practice”, The World Bank, page 96: “The
difference-in-differences approach thus combines the two
counterfeit counterfactuals (before-and-after comparisons and comparisons between those who
choose to enrol [e.g., implemented standardised packaging] and those who choose not to enrol [e.g.,
did not implement standardised packaging]) to produce a better estimate of the counterfactual”
(emphasis added).
31
For a more detailed explanation of the before-during approach, see Section 5, and in particular
paragraphs 5.3a, 5.10-5.20. That section also sets out an alternative approach, known as prediction
approach (see paragraphs 5.3b, 5.21-0). This alternative approach is subject to the same limitations as
the before-during approach, i.e., it is less reliable than the DID approach in the presence of
confounding factors that also affect cigarette consumption but are difficult to control for.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0075.png
of Social Development – have used New Zealand as a comparator for Australia, and vice
32
versa.
4.13
In the next subsection, I carry out my own analysis and find that New Zealand is a good
comparator for Australia because there is a stable relationship between cigarette
consumption in Australia and in New Zealand.
New Zealand is a good comparator for Australia
4.14
As explained in the previous section, the DID analysis only provides reliable results if the
evolution of tobacco consumption in New Zealand is a good indicator of what the evolution of
tobacco consumption would have been in Australia but for the adoption of standardised
packaging.
In order to assess whether this is the case, I analyse long-term time series of tobacco
consumption in the two countries. In particular, I use OECD data for 1970-2010 on
consumption of tobacco products per capita among 15+ year old (Figure 1).
Figure 1: Evolution of tobacco consumption (grams per capita, 15+ year old) in
Australia and New Zealand (1970-2010)
3,500
3,000
2,500
2,000
1,500
1,000
500
1970
4.15
1975
1980
1985
Australia
1990
1995
New Zealand
2000
2005
2010
Source:
Compass Lexecon based on OECD data.
4.16
The figure suggests that consumption of tobacco products in Australia and New Zealand
followed similar trends and that New Zealand is thus a good candidate for a comparator.
32
See for instance Australian Bureau of Statistics (2001) “Australian Social Trends”, and New Zealand
Ministry
of
Social
Development
(2010)
“2010
Social
Report”
(http://socialreport.msd.govt.nz/2010/comparisons/australia.html).
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0076.png
4.17
In order to assess whether New Zealand is indeed a good comparator, I empirically tested
whether tobacco consumption in Australia and New Zealand co-move in the long-run using
co-integration analysis techniques. Co-integration analysis allows for testing empirically
whether two or more (non-stationary) series maintain a systematic equilibrium (or long-run)
33
relationship.
Intuitively, two non-stationary series that are co-integrated will not move too far apart over
34
time. At any one time they may be very close, while at another time they may be further
apart, but over the long run they will move together.
This is the reason why two co-integrated series are often regarded as maintaining a long
term equilibrium relationship over time. In other words, the difference between two co-
integrated series is relatively stable over time.
To test whether tobacco consumption in Australia and New Zealand are co-integrated (i.e.,
they maintain a long-run equilibrium relationship) I ran a regression of the (log of) Australian
consumption on (log of) New Zealand consumption and analysed whether the residuals of
such regression (i.e., the component of the consumption in Australia not explained by the
35
evolution of the consumption in New Zealand) are stationary or follow a trend. Intuitively, if
4.18
4.19
4.20
33
A time series is said to be non-stationary when its mean and variance are not independent of time. For
example, a series exhibiting an upward or a downward trend over time is non-stationary, since the
mean of the series changes depending on the time period analysed. Hence, the mean of a stationary
series is constant over time. In other words, non-stationarity, a property common to economic time
series, means that a variable has no clear tendency to return to a constant value. An important feature
of stationary time series is that they frequently cross their mean and exhibit a tendency to revert to it.
Non-stationary series, on the contrary, do not necessarily have a constant mean and do not cross the
mean line frequently, suggesting there is no such reversion to the mean value. Tobacco consumption
series in Australia and New Zealand both exhibit a clear downward trend over time, suggesting they
are not stationary. Formal statistical tests unambiguously indicate that tobacco consumption in
Australia and in New Zealand are non-stationary.
34
Formally, two non-stationary series are said to be co-integrated if there is a unique linear combination
of them that is stationary.
These types of tests are commonly referred to as residual-based co-integration tests and are widely
accepted and used in the analysis of non-stationary series to test for co-integration. Residual-based
co-integration tests consist of testing whether the residuals resulting from running a regression
between the two non-stationary variables of interest are stationary. If the residuals are stationary, this
means there is a linear combination of the variables that is stationary, and therefore the variables are
co-integrated. Residual-based tests are implemented in two stages. First, a regression between the
non-stationary variables of interest (in this case tobacco consumption in Australia and New Zealand) is
run. Second, unit root tests proposed in the literature are used to test whether the residuals resulting
from the regression in the first stage are stationary [I(0)]. If the residuals are found to be stationary, the
series included in the regression in the first stage are co-integrated. For a more detailed description of
residual-based co-integration tests see Hamilton, J. (1994) “Time series analysis”, Princeton University
35
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0077.png
the residuals of the regression are stationary, the difference between the two consumption
variables is stable in the long term, and therefore the consumption levels in New Zealand
can be used to forecast the level of consumption in Australia.
4.21
Figure 2 below shows the actual tobacco consumption in Australia and the fitted values
using the regression model. This figure shows that most of the variability in Australian
consumption of tobacco products can be explained (and thus forecasted) using data on New
Zealand consumption of tobacco products. In particular, above 97% of the observed
variability in the series of tobacco consumption in Australia is explained by the evolution of
the tobacco consumption in New Zealand.
Figure 2: Actual and fitted tobacco consumption in Australia (1970-2010)
3,500
3,000
2,500
2,000
1,500
1,000
500
1970
1975
1980
1985
1990
Actual
1995
Fitted
2000
2005
2010
Source:
Compass Lexecon based on OECD data.
4.22
Figure 3 below depicts the residuals of the regression model. These residuals – the
component of the consumption in Australia not explained by the evolution of the
consumption in New Zealand – exhibit no trend, indicating that tobacco consumption in
Australia and New Zealand do maintain a long-run equilibrium relationship (i.e., are co-
36
integrated). In other words, these results indicate that New Zealand tobacco consumption
can be used to predict the evolution of the Australian consumption, and therefore is a valid
Press, and Engle, R.F. and Granger, C.W.J. (1987) “Cointegration and error correction: representation,
estimation and testing”,
Econometrica,
Vol. 55.
36
Formal statistical tests (Engle-Granger and Philips-Ouliaris residual-based co-integration tests) confirm
that residuals are stationary [I(0)] at the 95% confidence level, indicating tobacco consumption in
Australia and New Zealand are co-integrated.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0078.png
benchmark to assess what the Australian consumption would have been had standardised
packaging not been introduced.
Figure 3: Residuals of the regression of the (log of) tobacco consumption in Australia
on the (log of) tobacco consumption in New Zealand (1970-2010)
.20
.15
.10
.05
.00
-.05
-.10
-.15
-.20
1970
Source:
1975
1980
1985
1990
1995
2000
2005
2010
Compass Lexecon based on OECD data.
The impact of standardised packaging on the Australian tobacco
market
Consumption
4.23
Having established that New Zealand is a good comparator for Australia in relation to
consumption of tobacco products, I proceed to analyse the effect of standardised packaging
on Australian consumption of cigarettes using New Zealand as a benchmark.
As a first step, I plot consumption in the two countries from January 2009 to November 2012
37
(i.e., before standardised packaging was fully implemented in Australia) in Figure 4.
4.24
37
Note that the consumption data that I analysed in the previous section was for a different time period,
i.e., 1970-2010.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0079.png
Figure 4: Cigarette consumption (in millions of sticks) in Australia and New Zealand,
January 2009 – November 2012
Source:
Compass Lexecon based on Nielsen and IRI-Aztec data.
4.25
The above chart shows that, before the implementation of standardised packaging in
Australia:
Consumption in Australia (measured on the left-hand side axis) was higher than
consumption in New Zealand (measured on the right-hand side axis);
Consumption in both countries was seasonal;
Consumption decreased in both countries; and
New Zealand is a good comparator for Australia.
4.26
In order to analyse the impact of standardised packaging on cigarette consumption in
Australia, I use an extended version of the DID approach, whereby I also control for
differences between Australian and New Zealand trends that are not due to standardised
packaging but rather to other factors such as different tax policies or a different evolution of
income in the two countries.
The different evolution of excise taxes in the two countries is shown in Figure 5.
4.27
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0080.png
Figure 5: Excise taxes in Australia and in New Zealand, January 2009 to December
2016, in purchasing power parity cents
Source:
Compass Lexecon using data from
www.comlaw.gov.au
and from
www.legislation.govt.nz.
4.28
To the extent that increases in taxes result in price increases, taxes affect consumption. If I
did not control for the different evolution of taxes in Australia and in New Zealand, I could
incorrectly attribute to standardised packaging a divergence in cigarette consumption among
these two countries that is in fact due to different tax policies.
In order to isolate the impact of standardised packaging on cigarette consumption from the
impact of other factors, such as excise taxes and income (as measured by GDP per capita),
I carry out a regression analysis using monthly data on per-capita cigarette consumption in
Australia and New Zealand from January 2009 to December 2016.
Table 1 reports the effects of standardised packaging on consumption per capita according
38
to my econometric analysis. In particular, it reports the average percentage increase in
4.29
4.30
38
As explained in Annex B, consumption per capita is computed as total consumption (from Nielsen and
IRI-Aztec) divided by 20+ year old population (from Australian Bureau of Statistics and from Statistics
New Zealand). Note that this definition of consumption per capita does not imply that I am focusing on
consumption among 20+ year old consumers: in fact, I am considering consumption across the whole
population (including consumption among e.g., teenagers), since the consumption data that I use come
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0081.png
consumption per capita (up to December 2016) due to standardised packaging relative to the
counterfactual (e.g., the coefficient 0.035 means that standardised packaging is associated
with an increase in consumption per capita of 3.5% relative to the counterfactual).
4.31
In both models of Panel A and Panel B of Table 1, I use year indicator variables and monthly
dummies to control for time effects common to New Zealand and Australia. I also control for
the effect of GDP per capita on consumption per capita. In Model 1, I control for the effect of
excise taxes on consumption per capita. Since consumers care about excise taxes only
insofar as these affect prices, in Model 2 I control for the indirect effects of taxes on
39
consumption via prices (for this I use an Instrumental Variable approach).
The difference between Panel A and Panel B is that, in the former, prices, excise taxes and
GDP per capita are expressed in local currencies, while in the latter they are expressed in
40
PPP.
Table 1: DID regression analysis on the effect of standardised packaging on cigarette
consumption per capita in Australia
Model 1
Controls for monthly
dummies, year indicator
variables and
Panel A: local currencies
Effect of standardised
packaging
Panel B: PPP
Effect of standardised
packaging
Notes:
4.32
Model 2
IV: Price (instrument Excise
tax and Inflation), GDP per
capita
0.031**
Excise tax, GDP per capita
0.035**
0.030*
0.022*
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level.
4.33
Coefficients are always positive and statistically significant, indicating that standardised
packaging is associated with an increase in Australian consumption per capita relative to the
counterfactual.
from scanner sales data. Thus, the denominator is only a scaling factor. Choosing a different
denominator – e.g., 15+ year old population rather than 20+ year old population – does not change my
results.
39
40
See Annex B for a more detailed explanation of the econometric models.
Since Australia and New Zealand have different currencies, I express prices and other variables in both
countries in purchasing power parity (“PPP”) terms. PPP eliminates price and value differences due to
differences in the general levels of prices in the two countries.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0082.png
4.34
These results are robust to whether prices, excise taxes and GDP per capita are expressed
41
in local currencies or in PPP terms.
In order to control for the fact that standardised packaging was rolled out progressively from
October 2012 although it only came into full force in December 2012, I have also carried out
robustness checks running the same analyses as those in Table 1 but using (i) 1 October,
and then (ii) 1 November as the cut-off point for the adoption of standardised packaging. I
have also carried out robustness checks using quarterly data (as opposed to monthly data).
42
All of the results presented above are robust to these different specifications.
Prices
4.35
4.36
I am not aware of sufficient, publicly available, time series data on cigarette prices being
available to allow me to establish a long-run relation between the price of cigarettes in
43
Australia and New Zealand. Nevertheless, on the strength of the findings above regarding
the co-movement of cigarette consumption in Australia and New Zealand, and given that
prices are a very important determinant of consumption (as also confirmed by my
consumption analysis), I proceed to analyse the effect of standardised packaging on prices
using the same DID methodology that I used for consumption.
As a first step, I carry out a DID regression analysis to estimate the effects of standardised
packaging on the average price paid by consumers. For this, I analyse the average post-tax
price across all segments. The results of this analysis are reported in Table 2. This table
reports the average percentage decrease in prices (up to December 2016) due to
4.37
41
I note that in previous reports where I carried out similar empirical analyses, I imposed the assumption
that the effect of GDP on consumption and on prices was the same in Australia and New Zealand.
Given that for this report I have been provided with a longer time series (with an additional one and a
half years’ worth of data than my most recent previous analysis) I decided to test the assumption by
including an interaction term between the variable GDP per capita and a dummy for Australia. I found
this interaction term to be statistically significant, and thus I decided to include it in my regressions.
This implies that I am no longer imposing that assumption and that I instead ‘let the data speak’. I
provide more technical details in Annex B and Annex C where I also show that, even if I were to apply
the same model that I previously used, the results would not change, i.e., standardised packaging
would still be associated with an increase in consumption and with a decrease in prices relative to the
counterfactual (with results being statistically significant in seven out of the twelve model
specifications).
42
43
The robustness checks and the full specifications of the model are presented in Annex B.
While I do have data that allow me to compute the average market prices for cigarettes both in
Australia and in New Zealand (the data I use for the price analysis in this section), these data cover a
period of seven years (2009-2016) and thus do not allow me to establish robustly whether there exists
a long-run relation between the prices of cigarettes in the two countries (for establishing the long-run
relationship between cigarette consumption in Australia and New Zealand, I used data that covered 40
years, see paragraphs 4.14-4.22).
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0083.png
standardised packaging relative to the counterfactual (e.g., the coefficient -0.020 means that
standardised packaging is associated with a decrease in prices of 2% relative to the
counterfactual).
4.38
For this analysis, I control for the effects on prices of excise taxes and GDP per capita. I also
analyse prices (and control variables) both in local currencies (Model 1) and in PPP (Model
2).
Table 2: DID regression analysis on the effect of standardised packaging on the
average cigarette price in Australia
Model 1
Controls for monthly
dummies, year indicator
variables and
Effect of standardised
packaging
Notes:
Model 2
Excise tax PPP, GDP per
capita PPP
-0.026**
Excise tax, GDP per capita
-0.020**
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level.
4.39
This table shows that the effect of standardised packaging on average prices is negative and
statistically significant, indicating that standardised packaging is associated with a decrease
in the average cigarette price paid by consumers (relative to the counterfactual).
As was the case for consumption, these results are robust to whether prices, excise taxes
44
and GDP per capita are expressed in local currencies or in PPP terms.
In order to control for the fact that standardised packaging was rolled out progressively from
October 2012, although it only came into full force in December 2012, I have carried out
robustness checks using (i) October 2012, and then (ii) November 2012 as the start date of
standardised packaging. I have also carried out robustness checks using quarterly data (as
opposed to monthly data). All of the results presented above are robust to these different
45
specifications.
The average price decline in Australia (relative to the counterfactual) may be the result of
down-trading (i.e., smoker switching from more expensive to cheaper brands), of lower
4.40
4.41
4.42
44
See footnote 41 and Annex C for an explanation of how these models compare to models I used in
previous reports.
The robustness check and the full specification of the model are presented in Annex C. For regression
with quarterly data, the coefficients of interest are negative (indicating that standardised packaging is
associated with a price decrease) but some of them are not statistically significant. Losing significance
can be expected when one reduces the number of data points by two thirds (I aggregate three monthly
data points into one quarterly data point) and with a short time period (given that I work with data from
the 2009-2016 period, for the quarterly data I use only 32 data points per each country).
45
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0084.png
prices (i.e., brands reducing prices) or a combination of the two. Since it is possible that
46
down-trading results in increased consumption (relative to the counterfactual), the results
presented in Table 1 and Table 2 would be sufficient for me to conclude that standardised
packaging is associated with an increase in consumption (relative to the counterfactual).
4.43
Nevertheless, as a second step in my price analysis, I proceed below to analyse the effect of
standardised packaging on brand-level prices for the subset of brands that are sold both in
Australia and in New Zealand, and conclude that standardised packaging is also associated
with a decrease in these prices (relative to the counterfactual).
For this analysis, I first identify the 21 brands that are sold both in Australia and in New
Zealand: Ashford, Benson & Hedges, Camel, Chunghwa, Davidoff, Double Happiness,
Dunhill, Easy, Holiday, Honeyrose, Horizon, JPS, Kent, Longbeach, Marlboro, Pall Mall,
47
Peter Jackson, Peter Stuyvesant, Rothmans, Vogue, Winfield.
I then analyse the effect of standardised packaging on each of these brands by using similar
regression analyses as those used for the average market price. The results of these
analyses are reported in Table 3. This table splits the brands into (i) brands whose price has
increased relative to the counterfactual (and the increase is statistically significant); (ii)
brands whose price has decreased relative to the counterfactual (and the decrease is
statistically significant); and (iii) brands whose price has neither increased nor decreased
relative to the counterfactual (i.e., those brands for which the model finds no statistically
significant change in price).
For each category, and for each model, the table reports (a) the weighted average
percentage change in price (relative to the counterfactual); (b) the number of brands in that
category; and (c) the 2016 volume in millions of sticks.
48
4.44
4.45
4.46
46
This is because consumers who down-trade from premium to cheaper brands can buy more cigarettes
while spending the same amount of money.
Given that New Zealand data is provided at the level of broad brands (e.g., Dunhill), I carry out the
analysis at this level (e.g., I analyse the evolution of the price of Dunhill cigarettes) rather than at the
brand variant level (e.g., analysing separately the evolution of the price of Dunhill, Dunhill Essence,
Dunhill International, Dunhill Fine Cut …). In some cases variants within a brand belong to different
market segments. Thus, a reduction of the average price of those brands in Australia may be partially
due to the acceleration of down-trading (see next section) or to brand repositioning. However, given
that this issue may arise for only three of the 21 brands analysed, I do not consider that it significantly
affects my results.
47
48
The models are explained extensively in Annex C.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0085.png
Table 3: DID regression analysis on the effect of standardised packaging on the
average cigarette price in Australia
Model 1
%
change
in price
(i)
(ii)
(iii)
Increase in price
Decrease in price
No effect on price
Weighted
average
Source:
Model 2
Number
of
brands
7
4
10
Million
sticks
(2016)
514
6,496
5,533
%
change
in price
5.0%
-16.3%
0.0%
-6.9%
Number
of
brands
9
3
9
Million
sticks
(2016)
1,213
5,670
5,659
6.1%
-15.4%
0.0%
-7.7%
Compass Lexecon analysis
4.47
The table shows similar results for both models. In Model 1, the average price decrease
(relative to the counterfactual) for brands in category (ii) (-15.4%) is larger than the average
price increase (relative to the counterfactual) for brands in category (i) (6.1%). Also, despite
the fact that the number of brands that increase price (relative to the counterfactual) exceeds
the number of brands that decrease price (relative to the counterfactual) (7 vs 4), the brands
that decrease price (relative to the counterfactual) account for more than ten times the
volume of the brands that increase price (relative to the counterfactual). As a result of a
stronger impact that affects larger volumes of consumption, the price decrease dominates
the price increase, and – on average – the price of the brands analysed decreased by 7.7%
relative to the counterfactual.
A similar analysis applies for Model 2, whereby, on average, the price of the brands
analysed decreased by 6.9% relative to the counterfactual.
The above results suggest that at least part of the average price decrease can be explained
by a decrease in the price of some brands.
Down-trading
4.48
4.49
4.50
As explained in Section 3, standardised packaging severely restricts the scope for branding.
In particular, by reducing consumers’ valuation of premium brands vis-à-vis lower-quality
brands, it may induce a shift from premium to lower-quality brands.
My before-after regression analysis on down-trading confirms that high-quality brands lost
market share following the introduction of standardised packaging in Australia. This analysis
is based on Nielsen’s scanner data for Australia from January 2009 to March 2012 and IRI-
49
Aztec retail data from April 2012 to December 2016.
4.51
49
Cigarette brands are classified by the industry into the following segments (in decreasing order of
value): Premium, Aspirational Premium, Value for Money (“VFM”), and Low Price.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0086.png
4.52
These data are reported in Figure 6, which shows that the increasing trend of low value
brands (the yellow line) at the expense of more premium brands (the other three lines) can
be dated back to at least 2009.
Figure 6: Market shares of different brand segments, Australia January 2009 –
December 2016
Source:
Compass Lexecon based on Nielsen and IRI-Aztec data.
4.53
In order to assess whether standardised packaging is associated with an acceleration of this
trend, I carry out a two-step regression analysis.
First, I assess whether the market share of each segment followed a linear or a quadratic
time trend before the adoption of standardised packaging. The results of this analysis are
reported in Table 4.
4.54
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0087.png
Table 4: Results of the regressions of market shares on a quadratic time trend
(1)
Premium
(Global)
Time trend
-0.027**
[0.012]
Quadratic term of the time trend
-0.000
[0.000]
Constant
Observations
Adjusted R-squared
Notes:
(2)
Aspirational
Premium
0.020
[0.014]
-0.001***
[0.000]
24.735***
[0.137]
47
0.682
(3)
VFM
(4)
Low Price
-0.118***
[0.019]
0.000
[0.000]
32.983***
[0.199]
47
0.924
0.125***
[0.019]
0.001***
[0.000]
25.059***
[0.173]
47
0.961
17.223***
[0.109]
47
0.753
Robust standard errors in brackets.
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level.
4.55
Each column of the table represents one segment. For instance column (1) reports the
results of the time-trend analysis for the Premium (Global) segment. The coefficients of this
regression show that this segment was following a decreasing time trend (-0.027**) before
the adoption of standardised packaging, and that this trend may have been linear since the
coefficient of the “Quadratic term of the time trend” is not statistically significant.
Table 4 suggests that:
a. The Premium (Global) and the VFM segments followed a decreasing linear trend;
b. The Aspirational Premium segment followed a decreasing quadratic trend where only the
quadratic term is significant; and
c. The Low Price segment followed an increasing quadratic trend.
4.56
4.57
Secondly, I analyse whether the above trends have changed after the adoption of
standardised packaging. For this, and following the results in Table 4, I use a linear trend for
the Premium (Global) and the VFM segments, and a quadratic trend for the Aspirational
Premium (without including a linear term) and the Low Price segment (including both a linear
and a quadratic term).
Table 5 shows that standardised packaging is associated with an acceleration of the down-
trading to the Low Price segment.
4.58
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0088.png
Table 5: Regression analysis on the effect of standardised packaging on down-trading
in Australia
(1)
Premium
(Global)
Time trend
-0.040***
[0.003]
Change in time trend
-0.094***
[0.007]
Quadratic term of the time trend
-0.001***
[0.000]
Change in the quadratic term of
the time trend
-0.000**
[0.000]
Dummy for standardised
packaging
3.878***
[0.422]
Constant
17.327***
[0.082]
Observations
Adjusted R-squared
Notes:
(2)
Aspirational
Premium
(3)
VFM
(4)
Low Price
-0.114***
[0.004]
-0.193***
[0.008]
0.125***
[0.019]
1.371***
[0.180]
0.001***
[0.000]
-0.008***
[0.001]
-0.504
[0.406]
24.917***
[0.068]
96
0.945
9.120***
[0.508]
32.958***
[0.131]
96
0.989
-49.388***
[6.305]
25.059***
[0.173]
96
0.991
96
0.962
Coefficients of interest in bold.
Robust standard errors in brackets.
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level.
50
4.59
The variables of interest are “Change in time trend” (for Premium, VFM and Low Price) and
51
“Change in the quadratic term of the time trend” (for Aspirational Premium and Low Price).
In particular:
50
This variable represents the interaction between the linear time trend and a dummy variable that takes
the value of 1 for December 2012 and for the following months (i.e., the months when standardised
packaging was fully in place), and the value of 0 for earlier months.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0089.png
a. For the Premium (Global) and the VFM segments, the coefficients of “Change in time
trend” are negative and significant, indicating that standardised packaging is associated
with an acceleration of consumers’ shift away from these segments;
b. For the Aspirational Premium segment, the negative and significant coefficient of the
variable “Change in the quadratic term of the time trend” indicates that standardised
packaging is associated with an acceleration of consumers’ shift away from this
segment; and
c. For the Low Price segment, the combination of coefficients of the “Change in time trend”
and “Change in the quadratic term of the time trend” terms indicates that standardised
52
packaging is associated with an acceleration of consumers’ shift to this segment.
4.60
As a robustness check, I have run similar regressions whilst taking into account the potential
effect of excise taxes on down-trading. Table 6 shows that results are robust to this different
53
specification.
51
This variable represents the interaction between the quadratic term of the time trend and a dummy
variable that takes the value of 1 for December 2012 and for the following months (i.e., the months
when standardised packaging was fully in place), and the value of 0 for earlier months.
In the Low Price segment, the coefficient of “Change in time trend” is positive (indicating an
acceleration of down-trading to this segment) and the coefficient to “Change in the quadratic term” is
negative (indicating a deceleration of down-trading to this segment). However, the overall effect of
these two coefficients implies that down-trading to the Low Price segment’s market shares has
accelerated. The reader can check this result by computing estimated values of market share with and
without the change in slopes. When the change in slopes is taken into account, the market share of the
Low Price segment is significantly higher than when the change in slopes is not taken into account.
The time trend is a stepwise variable that takes values from 1 (January 2009) to 96 (December 2016).
52
53
In the Low Price segment, the coefficient of “Change in time trend” is positive (indicating an
acceleration of down-trading to this segment) and the coefficient to “Change in the quadratic term” is
negative (indicating a deceleration of down-trading to this segment). See paragraph 4.59c and footnote
52 for an explanation of why the overall effect of these two coefficients implies that down-trading to the
Low Price segment’s market shares has accelerated.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0090.png
Table 6: Regression analysis on the effect of standardised packaging on down-trading
in Australia, after controlling for the effect of excise taxes
(1)
Premium
(Global)
Excise tax
4.942
[3.544]
Time trend
-0.053***
[0.010]
Change in time trend
-0.108***
[0.011]
Quadratic term of the time trend
-0.001***
[0.000]
Change in the quadratic term of
the time trend
-0.000**
[0.000]
Dummy for standardised
packaging
4.836***
[0.687]
Constant
16.090***
[0.873]
Observations
Adjusted R-squared
Notes:
(2)
Aspirational
Premium
0.216
[3.308]
(3)
VFM
(4)
Low Price
-6.694*
[3.520]
-0.097***
[0.010]
-0.173***
[0.012]
15.439**
[7.293]
0.040
[0.043]
1.455***
[0.178]
0.002***
[0.001]
-0.009***
[0.001]
-0.498
[0.397]
24.857***
[0.916]
96
0.944
7.822***
[0.733]
34.633***
[0.895]
96
0.990
-49.677***
[6.121]
21.563***
[1.665]
96
0.992
96
0.963
Coefficients of interest in bold.
Robust standard errors in brackets.
*** indicates significant
at 1% level, ** indicates significant at 5% level, * indicates significant at 10% level.
4.61
Consistent with economic theory, the results of this empirical analysis suggest that the
adoption of standardised packaging was associated with an acceleration of down-trading
from more premium to low price brands.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0091.png
Section 5
Alternative consumption analyses
Introduction
5.1
In Section 4, I have set out the results of my DID analysis on the effect of standardised
packaging on cigarette consumption, i.e., that standardised packaging is associated with an
increase in consumption per capita (relative to the counterfactual and up to December 2016)
of between 2.2% and 3.5%.
In this section, I set out alternative analyses of the effect of standardised packaging on
cigarette consumption. I have included them in addition to my preferred DID analysis, and on
the basis that similar approaches have been adopted by other authors for estimating the
effect of standardised packaging in Australia.
These analyses are:
a. A
before-during
approach, whereby I estimate the effects of income and other relevant
factors (including standardised packaging) on cigarette consumption per capita in
Australia using the full time series of data (from January 2009 through December
2016).
54
5.2
5.3
b. A
prediction
approach, whereby I estimate the effects of income and other relevant
factors on cigarette consumption per capita in Australia using the before-period only, i.e.,
the period preceding the implementation of standardised packaging. Using the results of
this estimation, I project the level of cigarette consumption per capita that would have
prevailed had standardised packaging not being implemented.
55
I finally estimate the
54
Other authors have also employed a before-during approach for estimating the effect of standardised
packaging in Australia. In particular, Chipty (2016) employs this approach in analysing the effect of
standardised packaging on smoking prevalence in Australia. I use retail sales data as opposed to self-
reported smoking status data used by Dr. Chipty. I also use a materially longer time series that
provides a fifteen month longer post-implementation period to that used by Dr Chipty.
55
For example, if I estimate that in the pre-implementation period a 1% increase in income results in an
average increase in cigarette consumption by 0.3%, and in the post-implementation period income
increased by 2%, I predict that – everything else equal – in the same period cigarette consumption
would have increased by 0.6% (i.e., 2 x 0.3%).
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0092.png
effect of standardised packaging to be the difference between the actual and the
predicted level of consumption.
5.4
These two analyses have different advantages and disadvantages, and thus they
56
complement each other.
Although, in my opinion, these analyses are carried out in a less rigorous framework, the
results are consistent with those of my DID analysis, and indicate that standardised
packaging is associated with an increase in cigarette consumption per capita relative to the
counterfactual.
57
5.5
The need to account for the declining trend in cigarette consumption
5.6
As explained in Section 4, cigarette consumption follows a downward trend, in part for
reasons not directly related to tobacco control policies. If this downward trend were
neglected, the statistical analysis would incorrectly attribute the associated reduction in
cigarette consumption to the implementation of standardised packaging, and thus the
estimated effect of standardised packaging on cigarette consumption would be unreliable.
This is acknowledged by experts engaged by regulatory authorities to analyse the effects of
59
standardised packaging.
My DID approach controls for the downward trend in cigarette consumption as follows:
a. It assumes that, in the absence of standardised packaging and save for the different
evolution of other modelled factors in the two countries, cigarette consumption in
58
5.7
56
The standard before-during approach is preferable if standardised packaging does not change the
effect of modelled variables (such as prices and income) on cigarette consumption. In this case, the
before-during approach estimates these effects more robustly because it uses a longer time series,
and thus more information. Instead, if standardised packaging does affect the impact of prices and
income on cigarette consumption (e.g., it makes consumers more sensitive to price increases), then
the prediction approach is preferable because – contrary to the standard before-during approach – it
attributes to standardised packaging not only its direct effect on consumption, but also these indirect
effects.
See paragraphs 5.28-5.37.
See paragraphs 4.6 and 4.11.
See, for instance, Chipty, T. (2016) “Study
of the Impact of the Tobacco Plain Packaging Measure on
Smoking Prevalence in Australia”,
Annex A to the Australian Department of Health’s “Post-
Implementation Review - Tobacco Plain Packaging 2016”, paragraph 23: “To
the extent there is a
societal trend causing a decline in smoking prevalence or important omitted factors that vary over time,
failure to include a time trend will falsely credit the packaging changes for the decline in prevalence that
would have otherwise occurred anyway”.
57
58
59
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0093.png
Australia would have followed a similar trend as cigarette consumption in New Zealand;
and
b. It captures the consumption trend in these two countries by mean of yearly indicator
variables,
5.8
60
i.e., allowing the trend to be non-linear.
The same approach of using yearly indicator variables to capture the downward trend in
consumption cannot be used in the alternative approaches set out in this section.
a. In the before-during approach, this is because the model would attribute any increase (or
decrease) in consumption in each year from 2013 to 2016 to the year indicator variables,
even if the increase (or decrease) was due to standardised packaging. As a result, such
a model would only attribute to standardised packaging the effect this policy had in
December 2012 (but not the effect it had from January 2013 onwards).
b. In the prediction approach, it would make no sense to include yearly indicator variables
because I could only estimate the effect of these variables for the pre-implementation
period but not for the post-implementation period.
5.9
As a result, the two alternative approaches to DID set out in this section control for the
61
downward trend in cigarette consumption differently (and less robustly ) than the DID
approach, as explained in more detail when setting out those approaches.
The before-during approach
5.10
As explained above, in this approach I use data from January 2009 through December
2016 to estimate the effects of income and other relevant factors (including standardised
packaging) on cigarette consumption per capita in Australia. The results of this approach
indicate that standardised packaging is associated with a statistically significant increase in
cigarette consumption per capita relative to the counterfactual. These results are
63
summarised in Table 7 and set out in more detail in Annex D.
62
60
61
62
63
See paragraphs B.3-B.4 for details of the DID model for cigarette consumption.
See paragraphs 5.28-5.37.
See paragraph 5.3a.
As explained in paragraph 5.12, Model 1 is included for completeness but cannot be relied on. Models
2 and 3 show a statistically significant increase in cigarette consumption per capita relative to the
counterfactual.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0094.png
5.11
Since cigarette consumption in Australia follows a downward trend and it is seasonal, if I
want to reliably estimate the effect of standardised packaging on consumption I need to
65
control for both features.
I use two methods for doing this: in the first method (Model 1 in Table 7) I use a linear trend
to capture the downward trend and monthly indicator variables to capture the seasonality. In
the second method (Models 2 and 3 in Table 7), and given that the downward trend and the
seasonality of cigarette consumption are features that are common to many countries, I use
information from cigarette consumption per capita in a different country to control for these
features. In both methods, and consistently with my DID approach, I also control for the
effect of income and excise taxes (or, alternatively, cigarette prices) on cigarette
66
consumption. As explained in the next subsection, the first method, at least as applied to
the data I use, is fatally flawed, but I set it out to explain why this is the case.
Table 7: regression analysis on the effect of standardised packaging on cigarette
consumption per capita in Australia – before-during approach
Model 1
OLS
Controls for GDP per
capita and
Standardised
packaging
Notes:
64
5.12
Model 2
OLS
New Zealand
consumption per
capita, excise taxes
0.026***
Model 3
IV
New Zealand
consumption per
capita, prices
0.038***
Linear trend, excise
taxes, month fixed
effect
-0.013
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level.
First method: linear trend and monthly indicator variables
5.13
Other authors have also employed the first method, i.e., a before-during approach with the
assumption of a linear downward trend and the use of monthly indicator variables to control
for seasonality. In particular, Chipty (2016) finds, using this approach, that standardised
64
65
See Figure 4.
As explained above (see paragraph 5.7), in my DID approach I control for the trend using yearly
indicator variables. I also control for seasonality using monthly indicator variables.
This implies that the linear trend method that I employ is not what Diethelm and Farley (2015) refer to
as a “crude” or “simple” linear model, but also attributes part of the decrease in cigarette consumption
to other variables (such as the increase in excise taxes and, to the extent that these are passed
through to consumers, to the increase in cigarette prices). In this sense, my methodology is similar to
those of Diethelm and Farley (2015) and of Chipty (2016) (see footnote 67).
66
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0095.png
packaging is associated with a statistically significant
reduction
in smoking prevalence in
67
Australia, i.e., the
opposite
of the result that I obtain using DID.
5.14
Although I analyse the effects of standardised packaging on cigarette consumption rather
68
than on smoking prevalence, I find – using a before-during approach with linear trend and
monthly indicator variables – that standardised packaging has no statistically significant
69
effect on cigarette consumption.
However, more importantly, I do not consider that any reliance can be placed on this
approach because the model confounds the effect of the linear trend variable with the effect
70
of standardised packaging variable on cigarette consumption. This is not surprising,
because these variables take a high value at the beginning of the period and a low value at
the end of the period (the linear trend), or the other way around (the indicator variable for
standardised packaging). Since cigarette consumption in Australia is higher at the beginning
of the period than at the end of the period, the model does not know whether this decline is
due to the trend or to standardised packaging.
Since the model cannot perfectly distinguish between the effect of the linear trend variable
and the effect of the standardised packaging variable on consumption, it assigns effects
arbitrarily to these two variables. In other words, this approach very likely attributes to
standardised packaging changes in cigarette consumption that are in fact due to the
decreasing trend, or
vice versa.
Second method: including consumption in a different country as an explanatory
variable
5.17
Since the downward trend and the seasonality of cigarette consumption are features that are
common to many countries, the second method I use to control for these features is to use
5.15
5.16
67
Chipty, T. (2016). Study of the impact of the tobacco plain packaging measure on smoking prevalence
in Australia. Appendix A, Post-implementation review tobacco plain packaging. Diethelman and Farley
(2015) also find a similar result using a linear trend but without controlling for seasonality (see
Diethelman, P. A., and Farley, T. M. (2015). Refuting tobacco-industry funded research: empirical data
shows a decline in smoking prevalence following the introduction of plain packaging in Australia.
Tobacco Prevention & Cessation, 1(November): 6).
I use retail sales data as opposed to self-reported smoking status data used by Dr. Chipty. I also use a
materially longer time series that provides a fifteen month longer post-implementation period to that
used by Dr Chipty.
68
69
70
See Model 1 in Table 7.
As explained in Annex D (paragraph D.6), which reports the full results of this model, the correlation
between the coefficient on the trend and the coefficient on the standardised packaging variable is very
high. The higher the correlation between these two coefficients, the higher the risk identified above,
i.e., that the model confounds the effect of the linear trend and the effect of standardised packaging on
cigarette consumption.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0096.png
information from cigarette consumption per capita in a different country. In particular, due to
data availability, I use data on cigarette consumption per capita in New Zealand (Models 2
and 3 in Table 7).
5.18
Two important remarks are as follows:
a. First, this method is not subject to the same problem as the previous one, i.e., that the
model confounds the effect of standardised packaging with the effect of the trend on
cigarette consumption. This is because New Zealand cigarette consumption captures
both the seasonality and the trend, and thus allows the model more easily to identify the
effect of standardised packaging on Australian consumption without confounding it with
the effect of the trend.
b. Secondly, while this method uses data from New Zealand, it is fundamentally different
from the DID approach. In particular, the DID approach assumes that cigarette
consumption per capita in Australia would have followed a similar trend as in New
Zealand, save for the different evolution of other modelled factors in the two countries.
This is not the case in this alternative method. In particular, in this method I ‘ask the data’
whether we can learn something about cigarette consumption per capita in Australia
from cigarette consumption per capita in New Zealand. I also ask the data whether, if this
is the case, the relationship between the two series (if any) is statistically significant.
Thus, even if one were sceptical about the assumption underlying the DID analysis, i.e.,
that New Zealand is a good comparator for Australia, one would not need to be sceptical
about this alternative method for controlling for trend and seasonality because it does not
use the same assumptions as the DID analysis. While this is a potential advantage of
this method over the DID approach, the DID approach is overall preferable, as explained
in detail in paragraphs 5.32 to 5.37.
5.19
In both Model 2 and Model 3, I control for the effect of GDP per capita on consumption per
capita. In Model 2, I control for the effect of excise taxes on consumption per capita. Since
consumers care about excise taxes only insofar as these affect prices, in Model 3 I control
for the indirect effects of taxes on consumption via prices (for this I use an Instrumental
71
Variable approach ).
71
The Instrumental Variable approach is a standard method to control for endogeneity issues, i.e., that if I
want to estimate the effect of prices on consumption, a problem arises because my analysis may
actually capture the opposite effect, i.e., the effect of demand (and thus consumption) on prices. The
Instrumental Variable methodology allows me to only analyse the effects of prices on consumption that
are due to general inflation and to excise tax increases, and thus to disregard as irrelevant any change
in price that is due to e.g., standardised packaging or change in cigarette demand. See paragraph D.3c
for a more detailed explanation of this method.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0097.png
5.20
Coefficients are positive and statistically significant in both models, indicating that
standardised packaging is associated with an increase in cigarette consumption per capita
relative to the counterfactual (of 2.6%-3.8% on average up to December 2016).
The prediction approach
5.21
As explained above, in the prediction approach I estimate the effects of relevant factors
(such as income and excise taxes) on cigarette consumption per capita in Australia using the
before-period only, i.e., the period preceding the implementation of standardised packaging.
Using the results of this estimation, I project the level of cigarette consumption per capita
that would have prevailed had standardised packaging not being implemented. I finally
compare the actual and the predicted level of consumption and attribute the difference
between the two to the effect of standardised packaging.
I again use two different methods to control for the declining trend and for the seasonality of
cigarette consumption: a linear trend combined with monthly indicator variables (Model 1,
Figure 7), and NZ cigarette consumption per capita (Model 2, Figure 8). Both models also
control for the effect of income (GDP per capita) and excise taxes on cigarette consumption
in Australia.
First method: linear trend and monthly indicator variables
5.23
Model 1, which implements the prediction analysis and uses a linear trend, is not subject to
the same limitations of the before-during analysis when employing a linear trend. In
particular, since it only estimates the effect of relevant factors (including a linear trend) on
cigarette consumption during the pre-implementation period, this analysis cannot – by
construction – confound the effects of the pre-implementation trend with the effect of
standardised packaging.
The results of Model 1 are summarised in Figure 7. The blue line indicates the evolution of
actual cigarette consumption per capita from January 2009 to December 2016, and the red
line represents post-implementation cigarette consumption per capita as predicted by the
73
model. Since actual consumption is higher than predicted consumption, this model
associates the implementation of standardised packaging to an increase in consumption per
capita relative to the counterfactual.
72
5.22
5.24
72
73
See paragraph 5.3b.
See Annex D for a more detailed explanation of the methodology for predicting consumption using pre-
implementation data only.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0098.png
Figure 7: Actual and predicted cigarette consumption per capita in Australia –
Prediction analysis – Model 1 (linear monthly trend + monthly indicator variables)
Source:
Compass Lexecon analysis
5.25
In particular, Model 1 estimates that standardised packaging is associated with a 4.0%
average increase in cigarette consumption per capita relative to the counterfactual and up to
December 2016. This increase is statistically significant: there is a 95% probability that
cigarette consumption, as predicted by Model 1, falls in the area between the dotted lines in
Figure 7. Actual consumption falls outside this area, indicating that the increase in
consumption is statistically significant at 5% level.
Model 1 also predicts that the post-implementation increase in consumption is more
pronounced as time goes by and the full effects of standardised packaging unfold,
consistently with the predictions of the Australian Post Implementation Review, according to
which “the
full effect of the tobacco plain packaging measure is expected to be realised over
74
time”.
5.26
74
Australian Government (2016) Post-Implementation Review on Tobacco Plain Packaging, page 4.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0099.png
Second method: including consumption in a different country as an explanatory
variable
5.27
The results of Model 2 are summarised in Figure 8. This figure indicates that standardised
packaging is associated with a 4.0% average increase in cigarette consumption per capita
relative to the counterfactual and up to December 2016. Again, the effect is statistically
significant, and the increase in post-implementation consumption is more pronounced as
time goes by, consistently with the predictions of the Australian Post Implementation Review.
Figure 8: Actual and predicted cigarette consumption per capita in Australia –
Prediction analysis – Model 2 (NZ cigarette consumption per capita)
Source:
Compass Lexecon analysis
Reasons for preferring the DID approach
5.28
In Section 4 I explained that the DID approach is more reliable than alternative methods,
such as the before-after and the prediction approaches, for estimating the effect of
standardised packaging on cigarette consumption as it enables me to better control for
75
confounding factors. In what follows, I also explain why the way my DID approach controls
for the downward trend in consumption is superior both to the use of a linear trend and to the
use of cigarette consumption per capita in New Zealand among the explanatory variables.
75
See paragraph 4.11.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0100.png
5.29
As explained above, my DID approach controls for the downward trend in cigarette
consumption in a particularly robust way. In particular:
a. Based on the strength of the results of my analysis of New Zealand as a good
comparator for Australia, I assume that – save for different evolution of factors that can
affect cigarette consumption, such as standardised packaging, excise taxes, income,
and cigarette prices – cigarette consumption in Australia would have followed a similar
trend as cigarette consumption in New Zealand; and
b. I capture the consumption trend in these two countries by means of yearly indicator
76
variables, which – as explained immediately below – is a very flexible approach to
accounting for the downward trend in cigarette consumption and it is superior to
alternative approaches, including those employed in the analyses set out in this section.
The way my DID approach captures the declining trend in cigarette consumption is
superior to the linear trend method
5.30
As explained above, the first method I use to capture the declining trend in consumption in
my before-during and in my prediction approaches assumes that – save for the influence of
factors that affect cigarette consumption, such as income, excise taxes, prices and the
implementation of standardised packaging – cigarette consumption in Australia would have
followed a linear trend, i.e., it would have decreased by the same proportion in each year.
My DID approach, instead, does not make such a strong assumption and rather ‘lets the
data speak’, i.e., it allows the model to capture year-on-year differences in the downward
trend of cigarette consumption. If the downward trend in consumption was actually linear, my
DID analysis would capture that linearity, and thus would not be inferior to the linear trend
method; if the trend was non-linear, the DID approach would capture the non-linearity, but
the linear trend method would not. The DID is thus more reliable.
The DID approach is superior to including New Zealand cigarette consumption per
capita as an explanatory variable
5.31
5.32
As explained above, the second method I use in my before-during and in my prediction
approaches to control for the downward trend and the seasonality of consumption is to
include cigarette consumption per capita in New Zealand among the explanatory variables,
because this variable is both seasonal and follows a declining trend.
However, as explained, while this method uses data from New Zealand, it is fundamentally
different from the DID approach. In paragraph 5.18b, I set out a potential advantage of this
method (the “New Zealand method”) over the DID approach, namely that – while the latter
assumes that Australian and New Zealand cigarette consumption would have followed a
similar trend, save for different evolution of other modelled factors in the two countries – this
5.33
76
See paragraphs B.3-B.4 for details of the DID model for cigarette consumption.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0101.png
is not the case for the former. I also explained that, while this is a potential advantage of the
New Zealand method over the DID approach, the DID approach is overall preferable, as I set
out in what follows.
5.34
First, I have tested with data the DID assumption that New Zealand is a good comparator for
77
Australia, and confirmed that the assumption is valid with 95%-99% probability. Therefore,
it is unclear that the potential advantage of the New Zealand method over the DID approach
is actually realised.
Secondly, the DID approach has material advantages over the New Zealand method. In
particular, the DID approach, but not the New Zealand method, allows me to control for
factors (other than the trend) that affect New Zealand consumption, such as increases in
excise taxes on tobacco product, or reduced consumer income.
For example, when excise taxes increase in New Zealand but not in Australia, the DID
approach recognises that such increase would reduce New Zealand consumption but not
Australian consumption. This approach thus attributes to the trend any decline in
consumption that is common to both countries, and attributes to the increase in New Zealand
excise taxes any reduction in New Zealand consumption in excess of the decline due to the
common downward trend.
Following the same increase in New Zealand excise tax, the New Zealand method would
instead expect Australian consumption to also fall. If, as expected, Australian consumption
does not fall, this method would look for alternative explanations of why that is the case. As a
result, it would incorrectly attribute the lack of the decline in Australian consumption to other
factors, such as a small increase in Australian income.
5.35
5.36
5.37
Conclusions
5.38
In this section, have set out alternative analyses of the effect of standardised packaging on
cigarette consumption. Although, in my opinion, these analyses are carried out in a less
rigorous framework, the results are consistent with those of my DID analysis, and indicate
that standardised packaging is associated with an increase in cigarette consumption per
capita relative to the counterfactual.
77
See paragraphs 4.14-4.22 and in particular footnote 36.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0102.png
___________________________
Neil A. Dryden
10 October 2017
Compass Lexecon
Davidson Building, 200 Aldersgate Street,
London, EC1A 4HD
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0103.png
Annex A
Curriculum vitae of Neil Dryden
A.1
I am an Executive Vice President in Compass Lexecon’s European competition policy
practice, based in the firm’s London office. I have worked as a professional economist for
over 20 years and during that time I have advised on numerous mergers, agreement cases,
dominance cases, damages and market investigations.
My significant cases since 2010 include acting as an expert in the pay TV and tobacco cases
(both at the Competition Appeal Tribunal), and advising in outdoor advertising (OFT), Sports
Direct/JJB (Competition Commission and Competition Appeal Tribunal), Asda/Netto (OFT)
and Level 3/Global Crossing (OFT).
In addition to numerous cases in the European Union, I have advised on cases in India and
South Africa. I also have extensive experience in regulatory economics, including a series of
projects for the UK postal regulator. I have prepared submissions in the context of a number
of UK government inquires including the Barker review of land use planning.
In my career I have advised on several important matters for government and NGOs
including for the Shareholder Executive in the Department for Business, Innovation and
Skills in relation to the future ownership of the Royal Mail and for the Department of
International Development on energy market reforms.
I have analysed cases in sectors including advertising, banking and financial services,
chemicals, energy, FMCG, grocery retailing, healthcare, manufacturing, media and
broadcasting, mining, petroleum, pharmaceuticals, postal services, publishing, scientific
instruments, sports, technology, telecommunications, tobacco, transport, and water.
I was educated at Oxford University where I obtained a B.A. in Philosophy, Politics and
Economics (first class) and an M.Phil. in Economics, and held a college lectureship for two
years. At King’s College, London, I obtained a postgraduate diploma in EC competition law
(with distinction). I co-authored “What makes firms perform well?” published in the European
Economic Review.
Prior to joining Compass Lexecon, I worked as a Director at LECG and prior to that as an
Associate Director in NERA’s European competition policy practice for seven years. I spent
the first six years of my career in Arthur Andersen’s economic and financial consulting
practice, where I was a Senior Manager.
My full CV can be found at http://www.compasslexecon.com/professionals/bio?id=209.
A.2
A.3
A.4
A.5
A.6
A.7
A.8
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0104.png
Annex B
Consumption analysis
B.1
In Section 4 I present the results of my econometric analysis on the effects of standardised
packaging on consumption. In this annex, I set out this econometric analysis in more detail
and also show that the results presented in Section 4 are reliable across a wide range of
robustness checks.
As explained in footnote 41, in prior reports, I used a slightly different model than the one I
use in this submission. In this Annex I explain the reasons why I now use a slightly different
model. For completeness, I also present the results of the econometric analysis using the
model I used in previous reports. I show that this model confirms the results presented in
Section 4, i.e., that standardised packaging is associated with an increase in cigarette
consumption per capita relative to the counterfactual.
B.2
Benchmark analysis
B.3
In order to analyse the effect of standardised packaging on cigarette consumption per capita,
I use the following DID model:
B.4
Where subscripts
c, j
and
k
refer to country, year and month, respectively;
is
measured as the log of cigarette consumption per capita;
is the constant;
is an
for New Zealand and
for Australia) which
indicator variable for Australia (
is an
captures the average difference between Australia and New Zealand consumption;
indicator variable for standardised packaging (
for pre-implementation period and
78
for post-implementation period) which captures the average change in consumption
(common to Australia and New Zealand) after the implementation of standardised
packaging; the interaction term
is an indicator variable for the Australian
standardised packaging (it takes the value of 1 for Australia during the post-implementation
period, and it takes the value of 0 for the pre-implementation period and for New Zealand);
78
PP stands for Plain Packaging, which has the same meaning as Standardised Packaging.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0105.png
is set of indicator variables – one for each year – which capture the difference between
the average consumption in that year and the average consumption in a year of reference
(2009);
is set of indicator variables – one for each month – which capture the
difference between the average consumption in that month and the average consumption in
and
fixed effects are common to Australia
a month of reference (January); both
is a set of control variables, which change from model to model.
and New Zealand.
includes:
In particular,
a. Model 1: logarithm of excise taxes in each country in local currencies, logarithm of GDP
per capita in each country in local currencies;
b. Model 2 (Instrumental Variables approach): This is the same as Model 1 but it uses
prices (in local currencies) instead of excise taxes as a control variable. For this analysis,
I use a two-stage-least-square approach where I first regress the logarithm of prices on a
set of instrumental variables which includes the logarithm of excise taxes, the quarterly
change in the consumer price index, its square, and the interaction of these variables
with the AUS indicator. I then use the predicted values of this regression as a control
variable; and
c. Models 3 and 4 are the same as Models 1 and 2, respectively, but the control variables
are expressed in PPP rather than in local currencies.
B.5
For the consumption analysis, I use the following data sources: for cigarette volumes
(measured in millions of sticks) I use Nielsen data from January 2009 to January 2012, and
IRI-Aztec data from February 2012 to December 2016 for Australia, and Nielsen data for
New Zealand. Nielsen and IRI-Aztec datasets are consistent: for the period in which the two
datasets overlap, from March 2012 through December 2013, the correlation coefficient
between the Nielsen data and the IRI-Aztec data is 0.999 and is highly statistically significant
at the 1% level. I thus combine Nielsen and IRI-Aztec data after omitting sales through
79
convenience independent stores from Nielsen data.
I compute cigarette per capita by dividing cigarette volumes by the adult population (20+
years), sourced from the Australian Bureau of Statistics and from Statistics New Zealand.
For excise taxes I use information from the Australian Government website
(www.comlaw.gov.au) and from the New Zealand Parliamentary Counsel Office
(www.legislation.govt.nz). For GDP per capita, I use data from the Australian Bureau of
B.6
79
I note that in previous reports where I analysed data from New Zealand Nielsen included sales through
the convenience independent channel. Since January 2016 Nielsen only provides scan data in New
Zealand so they have provided back data from 2008 based on the scanning component only which
excludes the convenience independent channel. I have found the evolution of prices and volumes in
the new data excluding the convenience independent channel to be very similar to the data I used
previously.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0106.png
Statistics and from Statistics New Zealand. For PPP I use annual data downloaded from the
OECD website.
B.7
The coefficient to the interaction term
(i.e.,
captures the effect of standardised
packaging on average prices. I use two standard econometric methods – ordinary least
square (‘OLS’) and Instrumental Variables (‘IV’) to estimate the regression coefficients and I
report robust standard errors.
The results of Models 1-4 are reported in Table 8. As a robustness check, in the following
two tables I also report the results of robustness checks where I have assumed different start
dates for the implementation of standardised packaging (November 2012 and October 2012,
respectively).
B.8
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0107.png
Table 8: Results of consumption – monthly – PP from December 2012
Log (consumption)
Model 1
AUS
PP
AUS*PP
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(Price)
Log(Price)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Log(PricePPP)
Log(PricePPP)*AUS
Year fixed effect
Month fixed effect
Constant
Observations
Adjusted R-squared
Notes:
Model 2
2.096***
[0.000]
0.004
[0.850]
0.031**
[0.014]
Model 3
1.389***
[0.009]
0.009
[0.650]
0.030*
[0.055]
Model 4
2.284***
[0.000]
0.009
[0.638]
0.022*
[0.081]
1.185**
[0.011]
0.005
[0.795]
0.035**
[0.021]
-0.199***
[0.000]
-0.083*
[0.067]
0.390***
[0.001]
0.039
[0.738]
0.447***
[0.000]
0.271**
[0.015]
-0.403***
[0.000]
-0.173***
[0.000]
-0.205***
[0.000]
-0.081*
[0.056]
0.430***
[0.000]
0.089
[0.462]
0.461***
[0.000]
0.304***
[0.007]
-0.419***
[0.000]
-0.169***
[0.000]
YES
YES
4.830***
[0.000]
192
0.998
YES
YES
5.051***
[0.000]
192
0.998
YES
YES
5.092***
[0.000]
192
0.998
YES
YES
5.123***
[0.000]
192
0.998
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0108.png
Table 9: Results of consumption – monthly – PP from November 2012
Log (consumption)
Model 1
AUS
PP
AUS*PP
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(Price)
Log(Price)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Log(PricePPP)
Log(PricePPP)*AUS
Year fixed effect
Month fixed effect
Constant
Observations
Adjusted R-squared
Notes:
Model 2
2.038***
[0.000]
-0.007
[0.635]
0.030**
[0.014]
Model 3
1.354**
[0.013]
-0.003
[0.837]
0.030**
[0.049]
Model 4
2.252***
[0.000]
-0.002
[0.855]
0.022*
[0.079]
1.124**
[0.019]
-0.006
[0.678]
0.035**
[0.018]
-0.199***
[0.000]
-0.079*
[0.073]
0.393***
[0.001]
0.023
[0.845]
0.447***
[0.000]
0.257**
[0.025]
-0.404***
[0.000]
-0.167***
[0.000]
-0.204***
[0.000]
-0.079*
[0.060]
0.434***
[0.000]
0.081
[0.517]
0.463***
[0.000]
0.297**
[0.012]
-0.417***
[0.000]
-0.166***
[0.000]
YES
YES
4.843***
[0.000]
192
0.998
YES
YES
5.053***
[0.000]
192
0.998
YES
YES
5.113***
[0.000]
192
0.998
YES
YES
5.134***
[0.000]
192
0.998
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0109.png
Table 10: Results of consumption – monthly – PP from October 2012
Log (consumption)
Model 1
AUS
PP
AUS*PP
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(Price)
Log(Price)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Log(PricePPP)
Log(PricePPP)*AUS
Year fixed effect
Month fixed effect
Constant
Observations
Adjusted R-squared
Notes:
Model 2
1.952***
[0.000]
-0.009
[0.418]
0.032***
[0.009]
Model 3
1.301**
[0.022]
-0.007
[0.631]
0.031**
[0.037]
Model 4
2.180***
[0.000]
-0.005
[0.633]
0.025*
[0.052]
1.050**
[0.034]
-0.010
[0.474]
0.036**
[0.013]
-0.197***
[0.000]
-0.076*
[0.078]
0.399***
[0.001]
0.005
[0.967]
0.452***
[0.000]
0.236**
[0.047]
-0.398***
[0.000]
-0.163***
[0.000]
-0.201***
[0.000]
-0.077*
[0.063]
0.441***
[0.000]
0.069
[0.594]
0.469***
[0.000]
0.281**
[0.022]
-0.410***
[0.000]
-0.164***
[0.000]
YES
YES
4.872***
[0.000]
192
0.998
YES
YES
5.079***
[0.000]
192
0.998
YES
YES
5.148***
[0.000]
192
0.998
YES
YES
5.172***
[0.000]
192
0.998
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0110.png
Quarterly regressions
B.9
Below, I also present the results of consumption regressions where I use quarterly rather
indicator variables with
than monthly data. For such regressions, I replace
indicator variables. The rest of the model is the same as the model set out at paragraphs
B.3-B.4. Since the full implementation of standardised packaging happened at the end of
2012Q4, meaning that some of the sales in 2012Q4 were still made under branded
packaging, I present the results of two regressions with different starting dates for
standardised packaging: Table 11 assumes that standardised packaging may already had
an effect in 2012Q4, while Table 12 assumes that standardised packaging may only had a
statistically significant effect from 2013Q1. Both models confirm the result that standardised
packaging is associated with an increase in consumption.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0111.png
Table 11: Results of average consumption per capita – quarterly – PP from 2012Q4
Log (consumption)
Model 1
AUS
PP
AUS*PP
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(Price)
Log(Price)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Log(PricePPP)
Log(PricePPP)*AUS
Year fixed effect
Quarter fixed effect
Constant
Observations
Adjusted R-squared
Notes:
Model 2
0.868
[0.111]
-0.012
[0.284]
0.049***
[0.000]
Model 3
0.711
[0.312]
-0.010
[0.516]
0.046**
[0.010]
Model 4
1.057*
[0.070]
-0.010
[0.377]
0.045***
[0.000]
0.519
[0.399]
-0.012
[0.431]
0.049***
[0.004]
-0.117**
[0.014]
-0.047
[0.344]
0.466***
[0.006]
-0.129
[0.393]
0.493***
[0.000]
-0.041
[0.759]
-0.189***
[0.004]
-0.092*
[0.076]
-0.127***
[0.004]
-0.047
[0.331]
0.492***
[0.004]
-0.070
[0.661]
0.505***
[0.000]
0.014
[0.916]
-0.208***
[0.001]
-0.094*
[0.060]
YES
YES
6.326***
[0.000]
64
0.999
YES
YES
6.462***
[0.000]
64
0.999
YES
YES
6.573***
[0.000]
64
0.999
YES
YES
6.615***
[0.000]
64
0.999
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0112.png
Table 12: Results of average consumption per capita – quarterly – PP from 2013Q1
Log (consumption)
Model 1
AUS
PP
AUS*PP
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(Price)
Log(Price)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Log(PricePPP)
Log(PricePPP)*AUS
Year fixed effect
Quarter fixed effect
Constant
Observations
Adjusted R-squared
Notes:
Model 2
1.250**
[0.014]
-0.192***
[0.000]
0.047***
[0.001]
Model 3
0.929
[0.154]
-0.169***
[0.000]
0.042**
[0.032]
Model 4
1.323**
[0.013]
-0.179***
[0.000]
0.042***
[0.005]
0.837
[0.147]
-0.235***
[0.000]
0.046**
[0.012]
-0.128***
[0.007]
-0.066
[0.194]
0.434***
[0.006]
-0.049
[0.727]
0.468***
[0.000]
0.054
[0.662]
-0.208***
[0.001]
-0.122**
[0.024]
-0.140***
[0.002]
-0.060
[0.211]
0.459***
[0.004]
-0.019
[0.894]
0.476***
[0.000]
0.076
[0.533]
-0.233***
[0.000]
-0.116**
[0.024]
YES
YES
6.167***
[0.000]
64
0.999
YES
YES
6.336***
[0.000]
64
0.999
YES
YES
6.389***
[0.000]
64
0.999
YES
YES
6.449***
[0.000]
64
0.999
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0113.png
Results with the model used in previous submissions
B.10
In previous reports where I have analysed Australian data, using data for a shorter time
period (with my most recent previous analysis using data up to May 2015, as opposed to
data up to December 2016 used in this report) I used a slightly different model from the one I
use in this submission.
The only differences are:
a. In the previous submission I presented the results of two models where I did not control
for the effect of income (as measured by GDP per capita) on consumption per capita.
Since I find that GDP per capita does affect consumption, I do not use those models
here.
b. The old model did not allow for the effect of GDP per capita on consumption per capita to
differ in the two countries. In practice this means that the old model assumed that the
elasticity of cigarette consumption to income was the same in the two countries. In this
report, I instead allow this elasticity to differ, i.e., I do not impose any assumption but I let
the data speak. I find that indeed the effect of GDP per capita on cigarette consumption
per capita differs in the two countries, which makes the model in this report more
reliable.
c. For the instrumental variable regression, I use a different set of instrumental variables.
The instrumental variables I used previously did not pass the standard tests used to
assess the validity of the instruments with the new data.
B.12
80
B.11
For completeness, however, I have also carried out the econometric analysis using the old
model. The results are reported in Table 13, and confirm that standardised packaging is
associated with an increase in consumption, with the effect being statistically significant in 6
out of the 8 models.
80
See paragraph B.4b.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0114.png
Table 13: DID regression analysis on the effect of standardised packaging on
consumption in Australia – models in the previous submission
Model 1
Controls for
monthly dummies,
year indicator
variables and
Excise tax
Model 2
Excise tax,
GDP per
capita
Model 3
IV: Price
(instrument
Excise tax)
Model 4
IV: Price
(instrument
Excise tax),
GDP per
capita
0.031**
Panel A: local currencies
Effect of standardised
packaging
Panel B: PPP
Effect of standardised
packaging
Notes:
0.022
0.036**
0.013
0.033*
0.034**
0.033**
0.033**
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0115.png
Annex C
Price analysis
C.1
In Section 4 I present the results my econometric analysis on the effects of standardised
packaging on the price of cigarettes. In this annex, I set out this econometric analysis in
more detail and also show that the results presented in Section 4 are reliable across a wide
range of robustness checks.
As explained in in footnote 41 and in Annex B, in prior reports where I analysed data from
Australia I used a slightly different model than the one I use in this report. For completeness,
I also present the results of the econometric analysis using the model I used in previous
reports. I show that this model confirms the results presented in Section 4, i.e., that
standardised packaging is associated with a decrease in average cigarette prices relative to
the counterfactual (with the effect being statistically significant in one of the models).
C.2
Average price
C.3
In order to analyse the effect of standardised packaging on average prices, I use the
following DID model:
C.4
Where subscripts c, j and k refer to country, year and month, respectively;
is
measured as the log of price per stick, in local currencies for Model 1, and in PPP for Model
2; is the constant;
is an indicator variable for Australia (
for New Zealand and
for Australia) which captures the average price difference between Australia and
is an indicator variable for standardised packaging (
for pre-
New Zealand;
81
implementation period and
for post-implementation period) which captures the
average change in price (common to Australia and New Zealand) after the implementation of
is an indicator variable for the
standardised packaging; the interaction term
Australian standardised packaging (it takes the value of 1 for Australia during the post-
81
PP stands for Plain Packaging, which has the same meaning as Standardised Packaging.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0116.png
implementation period, and it takes the value of 0 for the pre-implementation period and for
is set of indicator variables – one for each year – which capture the
New Zealand);
difference between the average price in that year and the average price in a year of
reference (2009);
is set of indicator variables – one for each month – which capture
the difference between the average price in that month and the average price in a month of
and
fixed effects are common to Australia and New
reference (January); both
is a set of control variables, which change from model to model. In
Zealand;
includes:
particular,
a. Model 1: logarithm of excise taxes in each country in local currencies, logarithm of GDP
per capita in each country in local currencies;
b. Model 2: logarithm of excise taxes in each country in PPP, logarithm of GDP per capita
in each country in PPP.
C.5
For this analysis, I use the following data sources: for cigarette volumes (measured in sticks)
and values (measured in local currencies) I use Nielsen data from January 2009 to March
2012, and Aztec data from April 2012 to December 2016 for Australia; and Nielsen data for
82
New Zealand. I compute prices by dividing revenues by volumes. For excise taxes I use
information from the Australian Government website (www.comlaw.gov.au) and from the
New Zealand Parliamentary Counsel Office (www.legislation.govt.nz). For GDP per capita, I
use data from the Australian Bureau of Statistics and from Statistics New Zealand. For PPP I
use annual data downloaded from the OECD website.
(i.e.,
captures the effect of standardised
The coefficient to the interaction term
packaging on average prices. I use OLS to estimate the regression coefficients and I report
robust standard errors.
The results of the regressions are reported in Table 14. As a robustness check, I also report
in Table 15 and Table 16 the results of additional regressions where I have assumed
different start dates for the implementation of standardised packaging (November 2012 and
October 2012, respectively).
C.6
C.7
82
As explained in paragraph B.5 above, Nielsen and Aztec data are nearly identical during the overlap
period from March 2012 through December 2013. As explained in footnote 36, the Nielsen data for
New Zealand that I am using in this report excludes sales through the convenience independent
channel.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0117.png
Table 14: Results of average price analysis – monthly – PP from December 2012
Model 1
Log(price)
AUS
PP
AUS*PP
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Year fixed effect
Month fixed effect
Constant
Observations
Adjusted R-squared
Notes:
Model 2
Log (pricePPP)
1.510***
[0.001]
0.005
[0.507]
-0.026**
[0.015]
1.534***
[0.000]
0.002
[0.762]
-0.020**
[0.043]
0.537***
[0.000]
-0.115***
[0.000]
0.089
[0.345]
0.422***
[0.000]
0.535***
[0.000]
-0.111***
[0.000]
0.045
[0.641]
0.387***
[0.000]
YES
YES
0.376
[0.383]
192
0.994
YES
YES
0.013
[0.978]
192
0.995
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0118.png
Table 15: Results of average price analysis – monthly – PP from November 2012
Model 1
Log(price)
AUS
PP
AUS*PP
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Year fixed effect
Month fixed effect
Constant
Observations
Adjusted R-squared
Notes:
Model 2
Log (pricePPP)
1.540***
[0.001]
0.005
[0.476]
-0.026**
[0.015]
1.579***
[0.000]
0.003
[0.616]
-0.020**
[0.034]
0.535***
[0.000]
-0.116***
[0.000]
0.083
[0.384]
0.433***
[0.000]
0.535***
[0.000]
-0.114***
[0.000]
0.040
[0.686]
0.394***
[0.000]
YES
YES
0.350
[0.428]
192
0.994
YES
YES
-0.015
[0.975]
192
0.995
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0119.png
Table 16: Results of average price analysis – monthly – PP from October 2012
Model 1
Log(price)
AUS
PP
AUS*PP
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Year fixed effect
Month fixed effect
Constant
Observations
Adjusted R-squared
Notes:
Model 2
Log (pricePPP)
1.549***
[0.001]
0.006
[0.450]
-0.025**
[0.027]
1.617***
[0.000]
0.005
[0.484]
-0.020**
[0.041]
0.535***
[0.000]
-0.119***
[0.000]
0.081
[0.411]
0.443***
[0.000]
0.536***
[0.000]
-0.118***
[0.000]
0.037
[0.712]
0.397***
[0.000]
YES
YES
0.338
[0.456]
192
0.994
YES
YES
-0.026
[0.958]
192
0.995
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
Quarterly regressions
C.8
Below, I report the results of the average price regressions where I use quarterly rather than
monthly data. For such regressions, I replace
indicator variables with
indicator variables. The rest of the model is the same as the model set out at paragraphs
C.3-C.4. Since the full implementation of standardised packaging happened at the end of
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0120.png
2012Q4, meaning that some of the sales in 2012Q4 were still made under branded
packaging, I present the results of two regressions with different starting dates for
standardised packaging: Table 17 assumes that standardised packaging may already had
an effect in 2012Q4, while Table 18 assumes that standardised packaging may only had a
statistically significant effect from 2013Q1. Both models find that standardised packaging is
associated with a decrease in average prices, although the effects are in some cases not
statistically significant (to that end, as explained in footnote 45, losing significance can be
expected when one reduces the number of data points by two thirds).
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0121.png
Table 17: Results of average price analysis – quarterly – PP from 2012Q4
Model 1
Log(price)
AUS
PP
AUS*PP
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Year fixed effect
Quarter fixed effect
Constant
Observations
Adjusted R-squared
Notes:
Model 2
Log (pricePPP)
1.270**
[0.021]
0.004
[0.657]
-0.024*
[0.093]
1.356***
[0.003]
0.004
[0.628]
-0.021
[0.118]
0.591***
[0.000]
-0.120***
[0.008]
0.070
[0.615]
0.375***
[0.001]
0.593***
[0.000]
-0.120***
[0.008]
0.023
[0.873]
0.331**
[0.010]
YES
YES
0.365
[0.570]
64
0.995
YES
YES
0.003
[0.997]
64
0.995
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0122.png
Table 18: Results of average price analysis – quarterly – PP from 2013Q1
Model 1
Log(price)
AUS
PP
AUS*PP
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Year fixed effect
Quarter fixed effect
Constant
Observations
Adjusted R-squared
Notes:
Model 2
Log (pricePPP)
1.121**
[0.027]
0.151***
[0.000]
-0.019
[0.221]
1.197***
[0.006]
0.225***
[0.000]
-0.016
[0.298]
0.604***
[0.000]
-0.118**
[0.017]
0.092
[0.479]
0.337***
[0.003]
0.603***
[0.000]
-0.115**
[0.015]
0.044
[0.742]
0.298**
[0.013]
YES
YES
0.479
[0.417]
64
0.995
YES
YES
0.124
[0.847]
64
0.995
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets
Brand level prices
C.9
In order to analyse the effect of standardised packaging on brand-level prices using a DID
approach I have first identified a list of 21 brands that are sold both in Australia and in New
Zealand: Ashford, Benson & Hedges, Camel, Chunghwa, Davidoff, Double Happiness,
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0123.png
Dunhill, Easy, Holiday, Honeyrose, Horizon, JPS, Kent, Longbeach, Marlboro, Pall Mall,
Peter Jackson, Peter Stuyvesant, Rothmans, Vogue, Winfield.
C.10
I then run the following regression:
C.11
Where subscripts c, j, k and i refer to country, year, quarter and brand, respectively;
, ,
,
,
are defined in paragraph C.4;
is set of indicator
variables (one for each brand) which capture the average price difference between that
is a set of indicator variables (one for each
brand and a reference brand (Ashford);
quarter) which capture the difference between the average price in that quarter and the
fixed effects are common to
average price in a quarter of reference (Q1). The
Australia and New Zealand.
For this analysis, I use the same data sources indicated in paragraph C.5.
Given the model, the coefficient
indicates the effect of standardised packaging on the
price of the brand of reference (Ashford). For the other brands, the effect of standardised
packaging is captured by
. The statistical significance of the coefficients for these
other brands is tested using a F-test on the joint significance of the two coefficients. The p-
values of these tests (where the null hypothesis is that the two coefficients are not
statistically significantly different from zero) are reported in the column next to the column
with the coefficient.
C.12
C.13
C.14
C.15
All regressions use OLS and robust standard errors.
The results of the regressions are presented below. The regressions in Table 19 assume the
for 2009Q1 to 2012Q3, and
start date for standardised packaging at 2012Q4 (i.e.,
from 2012Q4 onwards). However, since the full implementation of standardised
packaging only happened on 1 December 2012 (i.e., at the end of 2012Q4), the regressions
in Table 20 assume the start date for standardised packaging at 2013Q1 (i.e.,
for
from 2013Q1 onwards).
2009Q1 to 2012Q4, and
For simplicity, I have identified in green the brands whose price decrease is statistically
significant, and in red the brands whose price increase is statistically significant.
C.16
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0124.png
Table 19: Results of brand-level price analysis – PP from 2012 Q4
Model 1
Log
(price)
Joint
significan
ce p-
value
Model 2
Log
(pricePPP)
Joint
significan
ce p-
value
AUS
PP
AUS*PP
Benson & Hedges
Camel
Chunghwa
Davidoff
Double Happiness
Dunhill
Easy
Holiday
Honeyrose
Horizon
JPS
Kent
Longbeach
Marlboro
Pall Mall
-0.048
[0.878]
-0.045***
[0.010]
-0.096
[0.226]
0.221***
[0.000]
0.228***
[0.000]
0.000
[1.000]
0.296***
[0.000]
0.092***
[0.000]
0.233***
[0.000]
-0.024***
[0.005]
0.111***
[0.000]
-0.631***
[0.000]
0.107***
[0.000]
0.034***
[0.003]
0.258***
[0.000]
0.017**
[0.041]
0.224***
[0.000]
0.084***
-0.071
[0.847]
-0.048***
[0.005]
-0.089
[0.263]
0.221***
[0.000]
0.228***
[0.000]
0.255***
[0.000]
0.296***
[0.000]
0.093***
[0.000]
0.233***
[0.000]
-0.024***
[0.005]
0.111***
[0.000]
-0.629***
[0.000]
0.107***
[0.000]
0.034***
[0.004]
0.258***
[0.000]
0.017*
[0.051]
0.224***
[0.000]
0.084***
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0125.png
Model 1
[0.000]
Peter Jackson
Peter Stuyvesant
Rothmans
Vogue
Winfield
AUS*PP*Benson & Hedges
AUS*PP*Camel
AUS*PP*Chunghwa
AUS*PP*Davidoff
AUS*PP*Double Happiness
AUS*PP*Dunhill
AUS*PP*Easy
AUS*PP*Holiday
AUS*PP*Honeyrose
AUS*PP*Horizon
AUS*PP*JPS
AUS*PP*Kent
AUS*PP*Longbeach
AUS*PP*Marlboro
0.035***
[0.006]
0.223***
[0.000]
0.248***
[0.000]
0.431***
[0.000]
0.183***
[0.000]
0.123
[0.131]
0.135
[0.102]
0.016
[0.852]
0.174**
[0.035]
0.111
[0.176]
0.131
[0.106]
0.132
[0.112]
0.082
[0.321]
0.029
[0.762]
0.066
[0.415]
0.019
[0.813]
0.245***
[0.004]
0.117
[0.153]
0.175**
[0.033]
0.000
0.285
0.000
0.000
0.056
0.206
0.490
0.128
0.039
0.608
0.000
0.006
0.073
0.102
Model 2
[0.000]
0.035***
[0.000]
0.223***
[0.000]
0.248***
[0.000]
0.433***
[0.000]
0.183***
[0.000]
0.124
[0.125]
0.136*
[0.097]
0.019
[0.823]
0.175**
[0.034]
0.112
[0.173]
0.133
[0.102]
0.134
[0.107]
0.083
[0.310]
0.033
[0.732]
0.068
[0.403]
0.021
[0.798]
0.247***
[0.004]
0.119
[0.146]
0.176**
[0.031]
0.000
0.138
0.000
0.002
0.190
0.294
0.780
0.064
0.015
0.442
0.000
0.018
0.036
0.037
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0126.png
Model 1
AUS*PP*Pall Mall
AUS*PP*Peter Jackson
AUS*PP*Peter Stuyvesant
AUS*PP*Rothmans
AUS*PP*Vogue
AUS*PP*Winfield
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Year fixed effect
Quarter fixed effect
Brand*PP fixed effect
Brand*AUS fixed effect
Constant
Observations
R-squared
Notes:
Model 2
0.073
0.577
0.708
0.000
0.037
0.212
0.176*
[0.054]
0.084
[0.310]
0.092
[0.254]
-0.198*
[0.082]
0.162*
[0.059]
0.076
[0.347]
0.463
0.019
0.001
0.852
0.823
0.050
0.174*
[0.056]
0.083
[0.314]
0.091
[0.263]
-0.199*
[0.080]
0.160*
[0.063]
0.075
[0.358]
0.587***
[0.000]
-0.018
[0.624]
-0.004
[0.960]
0.023
[0.777]
0.615***
[0.000]
-0.033
[0.343]
-0.008
[0.917]
0.022
[0.801]
YES
YES
YES
YES
-0.125
[0.689]
1,261
0.971
YES
YES
YES
YES
-0.256
[0.455]
1,261
0.972
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0127.png
Table 20: Results of brand-level price analysis – PP from 2013 Q1
Model 1
Log
(price)
Joint
significan
ce p-
value
Model 2
Log
(pricePPP)
Joint
significan
ce p-
value
AUS
PP
AUS*PP
Benson & Hedges
Camel
Chunghwa
Davidoff
Double Happiness
Dunhill
Easy
Holiday
Honeyrose
Horizon
JPS
Kent
Longbeach
Marlboro
Pall Mall
-0.111
[0.717]
0.231***
[0.000]
-0.110
[0.189]
0.225***
[0.000]
0.233***
[0.000]
0.261***
[0.000]
0.299***
[0.000]
0.097***
[0.000]
0.237***
[0.000]
-0.016
[0.127]
0.116***
[0.000]
-0.628***
[0.000]
0.111***
[0.000]
0.041***
[0.001]
0.263***
[0.000]
0.022**
[0.024]
0.229***
[0.000]
0.089***
-0.212
[0.535]
0.210***
[0.000]
-0.097
[0.249]
0.225***
[0.000]
0.233***
[0.000]
0.263***
[0.000]
0.299***
[0.000]
0.098***
[0.000]
0.237***
[0.000]
-0.016
[0.127]
0.116***
[0.000]
-0.627***
[0.000]
0.111***
[0.000]
0.041***
[0.001]
0.263***
[0.000]
0.022**
[0.026]
0.229***
[0.000]
0.089***
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0128.png
Model 1
[0.000]
Peter Jackson
Peter Stuyvesant
Rothmans
Vogue
Winfield
AUS*PP*Benson & Hedges
AUS*PP*Camel
AUS*PP*Chunghwa
AUS*PP*Davidoff
AUS*PP*Double Happiness
AUS*PP*Dunhill
AUS*PP*Easy
AUS*PP*Holiday
AUS*PP*Honeyrose
AUS*PP*Horizon
AUS*PP*JPS
AUS*PP*Kent
AUS*PP*Longbeach
AUS*PP*Marlboro
0.043***
[0.002]
0.228***
[0.000]
0.253***
[0.000]
0.437***
[0.000]
0.187***
[0.000]
0.141*
[0.097]
0.160*
[0.063]
0.045
[0.609]
0.195**
[0.024]
0.127
[0.136]
0.151*
[0.077]
0.155*
[0.074]
0.099
[0.251]
0.053
[0.580]
0.084
[0.323]
0.044
[0.608]
0.272***
[0.002]
0.134
[0.118]
0.195**
[0.023]
0.000
0.230
0.000
0.002
0.100
0.240
0.599
0.061
0.017
0.564
0.000
0.038
0.021
0.059
Model 2
[0.000]
0.043***
[0.003]
0.228***
[0.000]
0.253***
[0.000]
0.439***
[0.000]
0.187***
[0.000]
0.143*
[0.092]
0.162*
[0.060]
0.048
[0.587]
0.196**
[0.024]
0.128
[0.123]
0.153*
[0.073]
0.157*
[0.070]
0.101
[0.242]
0.056
[0.559]
0.086
[0.312]
0.046
[0.594]
0.274***
[0.002]
0.136
[0.112]
0.197**
[0.022]
0.000
0.047
0.000
0.017
0.474
0.399
0.866
0.013
0.001
0.713
0.000
0.122
0.003
0.005
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0129.png
Model 1
AUS*PP*Pall Mall
AUS*PP*Peter Jackson
AUS*PP*Peter Stuyvesant
AUS*PP*Rothmans
AUS*PP*Vogue
AUS*PP*Winfield
Log(Excise)
Log(Excise)*AUS
Log(GDPpc)
Log(GDPpc)*AUS
Log(ExcisePPP)
Log(ExcisePPP)*AUS
Log(GDPpcPPP)
Log(GDPpcPPP)*AUS
Year fixed effect
Quarter fixed effect
Brand*PP fixed effect
Brand*AUS fixed effect
Constant
Observations
R-squared
Notes:
Model 2
0.043
0.898
0.990
0.000
0.011
0.277
0.203**
[0.033]
0.108
[0.215]
0.112
[0.187]
-0.200*
[0.086]
0.189**
[0.035]
0.093
[0.275]
0.807
0.002
0.000
0.342
0.654
0.020
0.201**
[0.035]
0.107
[0.218]
0.110
[0.195]
-0.202*
[0.083]
0.187**
[0.036]
0.091
[0.286]
0.602***
[0.000]
-0.022
[0.581]
0.017
[0.801]
0.008
[0.916]
0.635***
[0.000]
-0.043
[0.238]
0.014
[0.840]
-0.006
[0.945]
YES
YES
YES
YES
-0.019
[0.949]
1,261
0.972
YES
YES
YES
YES
-0.124
[0.705]
1,261
0.973
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0130.png
Results with the model used in previous submissions
C.17
Finally, as explained above in footnote 41 and in Annex B, in previous reports where I have
analysed data from Australia, using data for a shorter time period I used a slightly different
model from the one I use in this report. The differences and the reasons for employing a
different model are explained in paragraph B.11.
For completeness, however, I also carried out the econometric analysis using the old model.
The results are reported in Table 21 and Table 22, and confirm that standardised packaging
is associated with a decrease in (i) the average price of cigarettes, with results being
statistically significant for Model 1 (Table 21); (ii) the price of individual brands: the results of
the econometric analysis using the model in the previous submission (Table 22) are very
similar to the results in Table 3.
Table 21: DID regression analysis on the effect of standardised packaging on the
average cigarette price in Australia – model of previous submission
Model 1
Controls for
monthly dummies,
year indicator
variables and
Effect of standardised
packaging
Notes:
C.18
Model 2
Excise tax,
GDP per
capita
-0.012
Model 3
Excise tax
PPP
Model 4
Excise tax
PPP, GDP per
capita PPP
-0.006
Excise tax
-0.022**
-0.007
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0131.png
Table 22: DID regression analysis on the effect of standardised packaging on the
average cigarette price in Australia – model of previous submission
Model
Model 1
% change in price
Number of brands
Million sticks
(2016)
Model 2
% change in price
Number of brands
Million sticks
(2016)
Model 3
% change in price
Number of brands
Million sticks
(2016)
Model 4
% change in price
Number of brands
Million sticks
(2016)
Source:
Compass Lexecon analysis
Increase
in price
6.1%
8
3,274
6.1%
8
3,274
6.9%
9
3,436
6.9%
9
3,436
Decrease
in price
-12.7%
7
6,643
-12.7%
7
6,643
-11.7%
6
6,643
-11.7%
7
6,643
No effect on
price
0.0%
6
2,626
0.0%
6
2,626
0.0%
6
2,464
0.0%
5
2,464
Weighted
average
-5.1%
-5.1%
-4.3%
-4.3%
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0132.png
Annex D
Alternative consumption analyses
D.1
In Section 5, I present the results of my alternative (i.e., non-DID) econometric analysis on
the effects of standardised packaging on cigarette consumption. In this annex, I set out this
econometric analysis in more detail.
Before-during approach
D.2
The before-during approach is as follows:
D.3
Where subscripts j and k refer to year and month, respectively;
is measured as
the log of cigarette consumption per capita; is the constant;
is an indicator variable for
for pre-implementation period and
for post-
standardised packaging (
83
implementation period) which captures the average change in consumption after the
implementation of standardised packaging;
is a set of control variables which
change from model to model. In particular,
includes:
a. Model 1: the logarithm of excise taxes, the logarithm of GDP per capita, a linear monthly
84
trend, and monthly indicator variables.
b. Model 2: the logarithm of excise taxes, the logarithm of GDP per capita, the logarithm of
cigarette consumption per capita in New Zealand.
c. Model 3 (Instrumental Variables approach): This is the same as Model 2 but it uses
prices instead of excise taxes as a control variable. For this analysis, I use a two-stage-
least-square approach where I first regress the logarithm of prices on a set of
instrumental variables which includes the logarithm of excise taxes, the quarterly change
in the consumer price index, and its square. I then use the predicted values of this
regression as a control variable.
83
84
PP stands for Plain Packaging, which has the same meaning as Standardised Packaging.
These indicator variables do not change from one year to another.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0133.png
D.4
D.5
The data is the same used for the DID analysis and described in paragraphs B.5-B.6.
The results of this approach are reported in Table 23. The coefficient of interest, i.e., the
coefficients that indicates the effects of standardised packaging on cigarette consumption
per capita, is that on the variable PP. This coefficient is negative but not statistically
significant in Model 1, which uses a linear trend and monthly indicator variables.
I note that this model is problematic because it confounds the effect of the trend with the
effect of standardised packaging on cigarette consumption, making the result set out
immediately above unreliable. This can be seen in the last row of Table 23, which shows that
the correlation between the coefficient on the trend and the coefficient on the standardised
packaging variable is very high. In practice, this implies that the model is likely to attribute to
standardised packaging changes in cigarette consumption that are in fact due to the
decreasing trend, or vice versa. Most likely as a result of this inability to distinguish the two
effects, the model predicts that – save for the effects of other modelled factors – cigarette
consumption in Australia would have followed a slightly increasing trend (because the
85
coefficient on the ‘Linear monthly trend’ variable is positive ). The negative coefficient on
the GDP per capita variable, estimating that an increase in income would reduce
consumption, is a further warning that the model is unreliable.
D.6
85
The coefficient is very small (0.000) and possibly positive from the fourth decimal. If it were negative, it
would have a minus sign in front.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0134.png
Table 23: regression analysis on the effect of standardised packaging on cigarette
consumption per capita in Australia – before-during approach – PP from December
2012
Log (consumption)
Model 1
PP
Log(GDPpc)
Log(Excise)
Log(Price)
Linear monthly trend
Log(NZ consumption per
capita)
Constant
Month fixed effect
Observations
Adjusted R-squared
Beta correlation between
trend (or NZ consumption)
and PP
Notes:
Model 2
0.026***
[0.005]
0.225***
[0.002]
-0.251***
[0.000]
Model 3
0.038***
[0.000]
0.380***
[0.000]
-0.013
[0.311]
-0.156
[0.191]
-0.347***
[0.000]
-0.366***
[0.000]
0.000
[0.420]
0.593***
[0.000]
3.407***
[0.000]
YES
96
0.918
-0.7402
3.195***
[0.000]
NO
96
0.892
-0.079
0.558***
[0.000]
4.011***
[0.000]
NO
96
0.891
-0.119
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
D.7
The same problem does not arise for Models 2 and 3, which include cigarette consumption
per capita in New Zealand as an explanatory variable.
Although I do not present the results here, I have carried out sensitivity analyses where I
have assumed different start dates for the implementation of standardised packaging
(November 2012 and October 2012). These analyses confirm the results reported in Table
23, i.e., (i) that Model 1 is unreliable for estimating the effect of standardised packaging on
cigarette consumption; and (ii) that Models 2 and 3 indicate that standardised packaging is
associated with a statistically significant increase in cigarette consumption per capita in
Australia relative to the counterfactual.
D.8
Prediction approach
D.9
The prediction approach is as follows:
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0135.png
D.10
Where
is defined in paragraph D.3, and
which change from model to model. In particular,
is a set of control variables
includes:
a. Model 1: the logarithm of excise taxes, the logarithm of GDP per capita, a linear monthly
86
trend, and monthly indicator variables.
b. Model 2: the logarithm of excise taxes, the logarithm of GDP per capita, the logarithm of
cigarette consumption per capita in New Zealand.
D.11
The prediction model only estimates the effect of relevant factors on consumption in the pre-
implementation period, and – as a result – Model 1 is not subject to the same limitations as
the before-during model, as explained in paragraph 0.
The results of the prediction models are shown in Table 24.
Table 24: regression analysis on the effect of standardised packaging on cigarette
consumption per capita in Australia – prediction approach – PP from December 2012
Log (consumption)
Model 1
Log(GDPpc)
Log(Excise)
Linear monthly trend
Log(NZ consumption per capita)
Constant
Month fixed effect
Observations
Adjusted R-squared
Notes:
D.12
Model 2
0.305***
[0.001]
-0.298***
[0.000]
0.356*
[0.061]
-0.326***
[0.000]
-0.002***
[0.001]
0.632***
[0.000]
5.644***
[0.000]
YES
47
0.967
3.349***
[0.000]
NO
47
0.904
*** indicates significant at 1% level, ** indicates significant at 5% level, * indicates significant at 10%
level. p-values in brackets.
D.13
I use the results of this analysis to estimate the level of cigarette consumption that would
have prevailed in the absence of standardised packaging. For example, the coefficient of the
86
These indicator variables do not change from one year to another.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0136.png
GDP per capita (log) variable in Model 1 indicates that, in the pre-implementation period,
when GDP per capita increases by 1%, cigarette consumption increases by 0.356%. Using
this estimated effect (and that of other relevant variables) and the actual increases of GDP
per capita (and that of other relevant variables) in the post-implementation period, I estimate
the level of consumption that would have prevailed in the absence of standardised
packaging. For example, if GDP per capita increased by 3% in the post-implementation
period, I estimate that – everything else equal – cigarette consumption per capita would have
increased by 1.068% (i.e., 0.356% x 3) in the absence of standardised packaging.
D.14
I call this estimated level of cigarette consumption per capita in the post-implementation
period ‘predicted consumption’ and compare it to the actual cigarette consumption per capita
in Table 25 (Model 1) and Table 26 (Model 2). These tables show that actual consumption
per capita is on average 4.0% higher than predicted consumption per capita, indicating that
standardised packaging is associated with an increase in consumption relative to the
counterfactual up to the end of 2016. This increase is more pronounced as time goes by and
the full effects of standardised packaging unfold.
As explained in Section 5, this increase is statistically significant.
Table 25: Actual vs Predicted consumption – Model 1 – PP from December 2012
Year
Actual
consumption
[a]
951
904
873
828
3,556
Compass Lexecon analysis
D.15
Predicted
consumption
[b]
942
878
821
775
3,416
Difference
([a]-[b])/[a]
0.9%
2.9%
5.9%
6.4%
4.0%
2013
2014
2015
2016
Total
Source:
Table 26: Actual vs Predicted consumption – Model 2 – PP from December 2012
Year
Actual
consumption
[a]
951
904
873
828
3,556
Compass Lexecon analysis
Predicted
consumption
[b]
925
878
829
781
3,413
Difference
([a]-[b])/[a]
2.7%
2.9%
5.1%
5.7%
4.0%
2013
2014
2015
2016
Total
Source:
D.16
Although I do not present the results here, I have carried out sensitivity analyses where I
have assumed different start dates for the implementation of standardised packaging
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0137.png
(November 2012 and October 2012). These analyses confirm the results reported above,
i.e., that standardised packaging is associated with an increase in cigarette consumption per
capita relative to the counterfactual.
 L 61 - 2020-21 - Bilag 1: Høringssvar og høringsnotat, fra sundheds- og ældreministeren
2259709_0138.png