statistical analysis of large health warnings on … · the rationale for lhws is given by the who...
TRANSCRIPT
![Page 1: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/1.jpg)
STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON CIGARETTE PACKS IN MALAYSIA AND THAILAND
Report prepared for
JTI (Japan Tobacco International)
27 July 2018
© Case Associates. All rights reserved 27 July 2018
![Page 2: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/2.jpg)
2 | P a g e
ABSTRACT
Objective:
To determine whether the introduction or enhancement of large health warnings (LHWs) on cigarette packs has reduced the sales of cigarettes in Malaysia and Thailand.
Methods:
Using Nielsen monthly retail point of sales data for the period January 2010 to September 2017 for Malaysia and quarterly point of sale data for the period 2009 to 2017 for Thailand, various fixed effects panel and time series regressions have been estimated. The econometric results are supplemented by a general analysis of data trends.
Results:
The enhancement of LHWs during the periods concerned was not associated with a statistically
significant reduction in the volume of cigarette sales.
Conclusion:
There is no reliable statistical evidence using Nielsen panel data that LHWs have reduced cigarette sales in Malaysia and Thailand.
![Page 3: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/3.jpg)
3 | P a g e
TABLE OF CONTENTS I. INTRODUCTION ........................................................................................................................... 4
II. LARGE HEALTH WARNINGS ........................................................................................................ 5
Malaysia ............................................................................................................................. 6
Thailand ............................................................................................................................. 7
III. SHORT REVIEW OF EMPIRICAL RESEARCH ................................................................................ 8
IV. EMPIRICAL APPROACH .............................................................................................................. 9
Estimating techniques ........................................................................................................ 9
Caveats ............................................................................................................................. 10
V. MALAYSIA ................................................................................................................................. 14
Data .................................................................................................................................. 14
Empirical findings ............................................................................................................. 18
VI. THAILAND ............................................................................................................................... 19
Data .................................................................................................................................. 19
Empirical findings ............................................................................................................. 23
REFERENCES ................................................................................................................................. 24
ANNEX A: STATISTICAL APPROACHES ........................................................................................... 27
Basic estimating equation ................................................................................................ 27
Panel regressions ............................................................................................................. 27
Time series regressions .................................................................................................... 34
ANNEX B: MALAYSIA REGRESSION RESULTS ................................................................................ 35
Data .................................................................................................................................. 30
Panel regression ............................................................................................................... 31
Efficient dynamic panel estimators .................................................................................. 33
Time series analysis ......................................................................................................... 34
Main findings ................................................................................................................... 35
ANNEX C: THAILAND REGRESSION RESULTS ................................................................................ 35
Data .................................................................................................................................. 36
Panel regression ............................................................................................................... 37
Efficient dynamic danel estimators .................................................................................. 39
Time series analysis ......................................................................................................... 40
Main findings ................................................................................................................... 41
![Page 4: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/4.jpg)
4 | P a g e
I. INTRODUCTION
This report has been commissioned by JTI (Japan Tobacco International) to examine the impact, if
any, on cigarette sales of large health warnings (LHWs) placed on cigarette packaging in Malaysia
and Thailand.
The study uses a variety of multivariate panel and times series statistical techniques to analyse data
on cigarette sales to determine whether LHWs reduced cigarette sales.
The selection of Malaysia and Thailand from among South Eastern Asian countries was based solely
on the availability of suitable data to carry out a rigorous statistical study.
This report is an independent analysis, and the analysis and conclusions are those of its authors
alone.
Main Findings
Based on econometric analysis of Nielsen point of sale (POS) retail data no statistically reliable
evidence was found that LHWs reduced cigarette sales in either Malaysia or Thailand.
There was no statistically significant reduction in cigarette sales associated with the
introduction of enhanced LHWs in 2014 in Malaysia.
There was no statistically significant reduction in cigarette sales in Thailand associated
with the introduction of enhanced LHW in 2014.
Organisation of Report
The report is organised as follows:
A brief description of LHWs in Malaysia and Thailand (Section II)
Short survey of the empirical research on the impact of LHWs (Section III)
An overview of the statistical approaches used to assess the impact of LHWs on cigarette
sales (Section IV). These are explained in greater technical detail in the Annex A.
A description and summary results of the analysis of data for Malaysia (Section V)
A description and summary results of the analysis of data for Thailand (Section VI)
In order to make the study accessible to the general reader the technical aspects of the statistical
analysis have been placed in three annexes which provide more details on the statistical techniques
![Page 5: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/5.jpg)
5 | P a g e
used (Annex A), and the statistical results of applying these techniques to data for Malaysia (Annex
B) and Thailand (Annex C).
II. LARGE HEALTH WARNINGS
LHWs are text messages together with pictorial or graphic images which aim to warn of the health
risks associated with smoking, and which cover half or more of the exterior panels of tobacco
product packaging.
The World Health Organisation (WHO) Framework Convention on Tobacco Control (FCTC) suggests
that health warning labels cover at least 30% but ideally 50% or more of the main surface areas of
tobacco packaging. The FCTC goes further stating that ‘nothing shall prevent a country from
imposing stricter requirements’ than provided by the Treaty itself. The coverage of LHWs for SE Asia
countries mainly exceeded 50% in 2017 with Thailand having the largest coverage in excess of 80%
of a cigarette pack (Figure 1).
Figure 1 Coverage of LHW in SE Asia, 2017
Source: Sancelmec (2017)
LHWs were first introduced by the Canadian government in 2001. Since then they have become
widespread. The rationale for LHWs is given by the WHO (2013):
Large graphic health warnings on tobacco packaging are the most effective means of communicating the risks of tobacco use to users. There is conclusive scientific evidence indicating that large graphic health warnings contribute to quitting tobacco and avoiding starting its use among young people. Further, the larger the warning, the more effective it is in protecting health. Research indicates that both adult and youth smokers report graphic warnings to be a credible source of information. Graphic health warnings reach all tobacco users, help people understand the physical impact of tobacco use, cost governments little to implement and are understood even by those who are illiterate. Pack-a-
![Page 6: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/6.jpg)
6 | P a g e
day smokers are potentially exposed to the message of the health warnings over 7000 times per year. Today 1 billion people live in countries that have graphic warning labels 50% or larger.
The central policy issue is whether this approach has had a discernible medium to long term effect
in reducing cigarette consumptions and sales.
Malaysia1
LHWs were first introduced in Malaysia in January 2009. This required that 40% on the front panel
and 60% on the back panel of a cigarette pack display one of six authorized health warnings, with a
graphic picture and accompanying text. The text was in Malay on the front panel and English on the
back panel. Health warnings are not required on packaging for tobacco products other than
cigarettes.
In January 2014 the LHWs were enhanced. The coverage on the front panel was increased to 50%,
and six new pictorials were issued which had to appear on a similar number of packs. A new text
message describing the contents and emissions was required on one side-panel, and on the front or
back panel for cartons.
Table 1. Development of LHWs for Malaysia, 2009-2017
Regulation Date Requirements
Control of Tobacco Product (Amendment) Regulations 2008
Jan 2009 Pictorial health warnings required to cover 40% of the front and 60% of the back of all cigarette packages
Control of Tobacco Product (Amendment) Regulations 2013
Jan 2014
A second set of six pictorial health warnings adopted with size increased to 50% of the front and 60% of back panels. Warning text must be in Malay (front) and English (back). Text on constituents and emission are required on the side panel. Each of the six authorized health warnings to appear as far as possible on an equal number of packs of each brand and type of cigarette. Misleading packaging and labelling, including terms such as “light” and “low tar” and other signs, is prohibited. Misleading descriptors such as ‘light’, ‘ultra light,’ ‘mild’, ‘cool’, ‘extra’, ‘low tar’, ‘special’, ‘full flavour’, ‘premium’, ‘rich’, ‘famous’, ‘slim’, ‘Grade A’ or similar terms are prohibited from appearing on packages.
Source: Euromonitor (2017)
The Malaysian government has imposed many other controls on tobacco packaging, advertising and
products including a ban on selling cigarettes at discounted prices, a ban on the sale of single sticks
1 Based on information collated on https://www.tobaccocontrollaws.org/legislation/country/malaysia/summary; https://www.tobaccocontrollaws.org/legislation/country/malaysia/pl-health-warnings
![Page 7: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/7.jpg)
7 | P a g e
and permitting sales in packs of 20 sticks only (2013), and the extension of the smoking in public
places (SIPP) ban too outdoor areas such as rest areas and parks (2014 and 2017).
Thailand2
LHWs were introduced in Thailand in March 2005 and enhanced and modified periodically in
subsequent years (Table 2).
The current LHWs were implemented in September 2014. The law requires that pictorial health
warnings cover 85 per cent of the front and back panels of cigarette packs. This was a significant
increase I coverage from the 55% required under the 2005 regulation. Thailand has the largest
LHWs of any country in Southeast Asia.
Table 2. Development of LHWs in Thailand, 2004-2017
Provision Date Requirements
Pictorial Warning Labels Law, 2004. Ministry of Public Health’s Announcement (No.8) B.E. 2547 On Rules, Procedures and Conditions of Cigarette Labelling and label Content According to the Tobacco Products Control Act B.E.2535.
March 2005
Overall 55% of the package space be devoted to pictorial health warnings with a set of six pictorials to be rotated on cigarette packages.
2006 New set of 9 health warnings.
Notification of the Ministry of Public Health No. 13 of 2007.
2007
For cigars, picture and text warnings must occupy 50% of both main surfaces PDAs (30% if the package is not rectangular). The warnings shall be printed in four colours.
2009 New set of 10 health warnings
Ministry of Public Health Notification No. 17 of 2012.
2012
LHWs on packages of shredded tobacco and blended shredded tobacco. LHWs must occupy 55% of both main surfaces PDAs (30% if the package is not rectangular) and be printed in four colours. Four LHWs are
2 Based on information collated on https://www.tobaccocontrollaws.org/legislation/country/thailand/summary; https://www.tobaccocontrollaws.org/legislation/country/thailand/pl-health-warnings
![Page 8: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/8.jpg)
8 | P a g e
prescribed and rotation must be at the rate of 500 packs/cartons per image. .
Ministry of Public Health Notice of Rules, Procedures, and Conditions for the Display of Images, Warning Statements, and Contact Channels for Smoking Cessation on Cigarette Labels, 2013.
June 2014 (given to
Sept 2014 to comply)
New LHWs required to cover 85% of both front and back panels printed in four colours. Images and text must be placed at the top edge of both main surfaces of the pack or carton. Rotation of images at the rate of 5,000 packs/cartons per image together with text quit-smoking hotline number. Also pictorial on each main surfaces, each side panel of the unit (e.g., pack) and outside (e.g. Carton) packaging must display a text warning statement. Text must include details about toxins, carcinogens, or other substances in tobacco products and printed on both sides and must occupy no less than 60% of each side. If the unit or outside packaging is not rectangular, the packaging must bear the statement on two sides, each side with an area of at least 10% of the total surface area of the pack or container.
Ministry of Public Health Notification No. 18, 2015.
2015 New text warnings on side panels of cigarette packaging.
Source: www.tobaccocontrollaws.org
Pictorial health warnings are also required on cigars (2007) and loose tobacco typically used for
hand-rolled cigarettes (2012).
The Thai government has imposed other controls under the Tobacco Products Control Act 19923
such as bans on media advertising and SIPPs in 1997; prohibitions on misleading packaging and
labelling in 2007; and the extension of SIPP ban to workplaces, public transport and other areas in
2010 and 2012.
III. SHORT REVIEW OF EMPIRICAL RESEARCH
Research on health warnings on cigarette packaging is considerable. A number of different
approaches have been used to measure the impact, if any, on cigarette sales, consumption and
smoking prevalence ranging from surveys to econometric analysis.
The published studies vary considerably in quality and do not generate consistent or reliable
evidence, positive or otherwise, of the impact of LHWs on smoking prevalence and/or cigarette
sales.
3 The Tobacco Products Control Act 2017 (effective July 4, 2017) replaces the 1992 Act. The 2017 Act gives the government expanded powers to regulate all tobacco products’ packaging (their size, colour, symbol, labels, trademark, pictures and messages) including the possible introduction of plain packaging.
![Page 9: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/9.jpg)
9 | P a g e
Rather than review the research on LHWs we have drawn on the independent and scholarly survey
by Monárrez-Espino et al (2014) published in the American Journal of Public Health. This critically
reviewed 21 of the more credible of the nearly 2,500 studies published between 1993 and 2013
which attempt to measure smokers’ responses to LHWs or what they term pictorial warnings on
cigarette packaging (PWCPs).
Monárrez-Espino et al found that the quality of the published research was “generally low”:
We found very large heterogeneity across studies, poor or very poor methodological
quality, and generally null or conflicting findings for any explored outcome.
… we considered 57% of the reviewed studies (n=12) to be of poor or very poor quality, and
only 1 could be classified as being of good quality. (p. e15)
Monárrez-Espino et al also state that had they used stricter criteria to assess the quality of published
studies even fewer studies would have been considered for review because they adopted
unacceptable methodologies and lacked sufficient scientific rigour. Notwithstanding these
methodological limitations and weaknesses Monárrez-Espino et al concluded that:
The evidence for or against the use of PWCP is insufficient, suggesting that any effect of
PWCP on behaviour would be modest.4
Of the studies reviewed, three are for Malaysia and/or Thailand - Silpasuwan et al. (2008) for
Thailand; Fathelrahman et al. (2010) for Malaysia; Yong et al. (2013) for Malaysia and Thailand -
which were assessed as having an “overall quality” rated as “poor” and “poor methodological
quality”, “poor” and “fair” respectively.
In the light of these concerns we have undertaken an exhaustive and clearly explained analysis of
the Nielsen data.
IV. EMPIRICAL APPROACH
Estimating Techniques
In this study panel and time series regression analyses has been used to investigate whether LHWs
have had a statistically significant effect in reducing cigarette sales. These techniques are described
in Annex A.
To test whether LHWs affect cigarette sales the volume of cigarette sales on the price of cigarettes
has been regressed on variables representing LHWs, other regulations on tobacco consumption
such as SIPPs; and a number of control variables expected to influence the demand for cigarettes
4 Wakefield et al (2014) in a critical assessment of earlier research (Kaul & Wolf, 2014) found that “graphic health warnings in Australia had no impact on smoking prevalence”.
![Page 10: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/10.jpg)
10 | P a g e
such as GDP per capita. The choice of variables used in the statistical analysis has been determined
by the availability of data over a suitably long period.
The data for each country has been collated by Nielsen. It is monthly point of sale (POS) for each
brand and type of cigarette sold in Malaysia and Thailand. For each month Nielsen has collated data
on cigarette sales and prices by brand, region and type of retail outlet. This enables the statistical
analysis to control for product heterogeneity, location and other sales outlet over time. This pooled
timeseries and cross-sectional data is called ‘panel data’, on retail cigarette sales.
The Nielsen data has been statistically investigated using different types of panel econometric
techniques and also by a time series technique. These are described in Annex A. To our knowledge
this is the first time Nielsen aggregate panel data has been used to assess the impact of LHWs and
tobacco controls.
Panel data has a number of attractive features suited to the empirical analysis of the effects of
regulation. First, panel data greatly expands the sample size thereby increasing the statistical
efficiency of the estimated coefficients and their standard errors. This leads to more credible
statistical analysis. Secondly, by providing repeated observations over time for the same group of
products it is possible to control for missing or unobserved variables which may affect the cross-
sectional data. This minimises so-called “heterogeneity bias” arising from omitted variables when
cross-sectional data are correlated with the dependent variable.5 In the context of cigarette
sales/smoking studies these unobserved variables can be factors such as tastes, preferences, age
and so on; or the different characteristics of cigarettes which may result in a different relationship
between consumption and demand over time. Thirdly, panel data allows product heterogeneity
and regional differences to be taken into account.
As a further check on the statistical robustness of the panel regressions a timeseries technique has
been used. This is described in Annex A.
Caveats
Any statistical analysis of tobacco packaging, health and price regulation must contend with a
number of challenges, and the results treated with some caution. The more important of these, and
their implications and the way they have been handled in this study, are briefly addressed in this
section.
The first issue is how the qualitative changes brought about by packaging regulations are to be
incorporated into a quantitative empirical analysis. Specifically, how can changes to the exterior
appearance of a cigarette pack which can be described but cannot be reduced to a single metric like
the price or volume of cigarettes sales be represented in a regression analysis? This is routinely done
in statistical research on cigarette consumption by time-related dichotomous or dummy variables.
5 Baltagi and Levin (1986) were the first to use (regional) panel data to study cigarette sales.
![Page 11: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/11.jpg)
11 | P a g e
These are binary variables which take the value of zero (0) when the regulation under investigation
is not present; and unity (1) for the months or quarters following its introduction.6 To illustrate the
use of these dummy variables, one can think of the relationship between cigarette sales and other
factors as a statistically measured straight line. When a dummy variable is included it shifts this line
up, down, or not at all for the months when the packaging control is in place indicating that the
control leads to higher, lower or negligible effects respectively on cigarette sales compared to the
situation in the absence of the LHW.7
There are limitations to the dummy variable approach. For example, it does not easily allow for any
delayed or diffused reaction by smokers. It is possible to investigate some delay in the reaction of
smokers by lagging the dummy variable by several time periods (which has been explored in the
statistical analysis below). Dummy variables also assume that the impact of a regulation is similar
throughout the period of the implementation of a regulation. This may not be a valid assumption –
it can disguise and confound the initial impact which can be assumed to be greater; with the
subsequent diffusion effects when smokers have adjusted to the regulation (which may clawback
some of the reduction in monthly sales or consumption of cigarettes). Dummy variables also do not
take account of changes in the level and intensity of public enforcement and/or compliance.
Secondly, often a large number of different tobacco controls and other measures are implemented
at around the same time or operate over the same period which may affect cigarette sales. LHWs
are one aspect of a multipronged approach to deter smoking which includes advertising and point
of sale display bans, SIPP bans, minimum pack size, content regulation and so on.
Table 3. Tobacco control regulations
Public information campaigns
Health warnings
Advertising and promotion bans
Point of sale display bans
Bans on Smoke in Public Places (SIPPs)
Minimum age
Content Controls
Ban on cigarette types e.g. menthol; coloured; flavoured
Specification of cigarette contents
Tar caps
6 This is called a “shift dummy”. Other dummies can be used – ‘slope dummies’ to test whether the health warnings alter the slope of the line; and ‘interaction dummies’ to see if the coefficient for selected variables alter.
7 E.g. Norashidah, Mustapha & Mastura (2013) time series (yearly 1980 -2009) use a dummy to represent the Malaysian Government’s ‘TakNak’ or ‘Don’t Want’ national antismoking campaign launched on 9 Feb 2004 for five years with a budget of RM100m. It considers no other regulatory intervention.
![Page 12: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/12.jpg)
12 | P a g e
Minimum pack size
Fiscal & Price Controls
Taxes
Minimum Price Laws (MPLs) (Malaysia)
Discounting & price promotion bans
For example, in Malaysia (see Figure 2) during the period being considered in this study, there were
changes to LHWs, SIPP bans were extended, the tar ceiling was reduced, and cigarettes could only
be sold in packs of 20 sticks. It may therefore be difficult to disentangle and isolate the individual
effects of these different measures (and the statistical problem known as multicollinearity).
Figure 2 Tobacco controls and tax changes in Malaysia, 2013-2017
Most government tax cigarettes heavily to, at least in part, discourage smoking. These taxes are
usually high and constitute a significant percentage of the retail price of a pack of cigarette. Other
price controls have been used - such as minimum prices and bans on price discounts. For example,
In Malaysia during the period 2013 to 2017 excise and other taxes on cigarettes were increased four
times, minimum prices increased twice (August 2015 and 2016)8 and the discounting of cigarette
prices banned. The statistical analysis reported below adjusts for the effects of cigarette excise and
other taxes (see Annex A).
8 Liber et al (2015) examine the impact of minimum cigarette prices and find that it did not reduce cigarette consumption.
![Page 13: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/13.jpg)
13 | P a g e
Thirdly, sufficiently detailed, reliable and consistent data extending over a sufficiently long period is
generally difficult to collect. Nielsen, the source of the data used in this study, does not keep a
central database of historical data, and the data we have used has been especially collated for this
study at our request. Despite the considerable effort by Nielsen to retrieve their historical data it
has only been possible to collate a consistent and reliable series going back to 2009/10. However,
the introduction of LWHs took place before 2009 in both Thailand and Malaysia. It has not,
therefore, been possible to test whether the introduction of LWHs in each country affected cigarette
sales; only the effect, if any, of the enhancement of LHWs which took place in 2014 in both
countries.
This is not as a significant drawback as it might first appear. The analysis below addresses the
important question of whether the enhancement and refreshing of LHWs by increasing their
coverage and introducing new pictorials do or do not lead to behavioural responses from smokers
which reduce cigarette sales. After all the premise behind Governments’ and health authorities’
policies to increase the size LHWs, and introduce new and rotating pictorials, is that the impact of
earlier LHWs on smokers may have waned with familiarity and/or to boost the deterrence effect of
LHWs by these enhancements.
Fourthly, in common with other data sources, the Nielsen data is only for sales of licit cigarettes. It
does not include data on the sales of roll your own (RYO) tobacco, pipe tobacco, cigars and illicit
cigarette sales. The last are particularly significant in Malaysia where Euromonitor (2016, p. 26)
estimates that one in two cigarettes sold were illicit. Given the clandestine and illegal nature of illicit
cigarette sales, reliable data on their volumes and prices are difficult to collect and unreliable.
Estimates exist, such as Nielsen (2016), Euromonitor (2017) and Oxford Economics (2017), but they
are incomplete and lack the detail, frequency and consistency that allow them to be matched to the
Nielsen data. However, proxies can be used. In previous studies the lower prices of cigarettes in
contiguous countries/regions have been used to represent the incentive to bring in illicit cigarettes
e.g. Baltagi and Levin (1986); Gruber & Sen (2003); or measures of the level of corruption and law
enforcement in the country using indicators such as published by the World Bank. In this study an
attempt has been made to take account of smuggled cigarettes by using the price differences
between contiguous countries as a proxy.
The omission of other tobacco products and illicit cigarette sales from the Nielsen data are unlikely
to bias the statistical results against finding that LHWs reduced cigarette sales. If as a result of LHWs
smokers substituted to illicit cigarettes, then our results would capture this as a decrease in licit
cigarette sales. This reaction would therefore bias the empirical work to finding that LHWs
decreased cigarette sales. On the other hand, the presence of a large volume of cheaper illicit
cigarettes would tend to bias upwards the estimated price elasticity as sales would appear more
responsive than was the case if all cigarettes sales were taken into account and increase the
perceived impact of tax increases in reducing cigarette sales and smoking prevalence.
The final area is the high and increasing tax rates imposed on licit cigarette sales. Most of the large
price increases for cigarettes discussed in the next sections were due to increases the taxes imposed
on cigarettes. These taxes have been levied to make licit cigarettes more expensive, and hence
reduce their sales and raise government revenues (Laffer, 2014). Given the significance of these
![Page 14: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/14.jpg)
14 | P a g e
taxes will have had a major impact on licit (and possibly illicit) cigarette sales and because they are
exogenous (that is imposed by governments rather than generated by the market) they may swamp
the effects of other measures. Moreover, a failure to control for tax changes will tend to bias the
statistical results which are designed to show the impact of LHWs and other tobacco control
measures on sales. This has been handled in the statistical analysis by using an instrumental variable
approach as discussed in Annex A.
V. MALAYSIA
This section describes the data, provides an overview of trends and relationship discernible for the
raw Nielsen data for Malaysia to guide the statistical analyses, and summarises the results of the
statistical analysis set out in detail in Annex B.
Data
Monthly Nielsen data is available for Malaysia for the period January 2010 to September 2017. It
reports sales and prices of cigarettes by brand/type categorised by:
Region - Central, East Coast, Northern, Southern North, Southern, East Malaysia; and
Sales channel - CVS/Petrol/Mini, Entertainment, Hawker Stall, Other retail, Refreshment,
Sundry Prov, Supermarkets.
The Nielsen data is available by region and sales channel separately (not both together) for brand
owner, cigarette brand and price segment9 (Base, Value, Premium and Unknown10).
Trends and Relationships in Nielsen Data
Figure 3 plots monthly cigarette sales, and nominal and real average selling prices (ASPs)11, the last
deflated by the Consumer Price Index (CPI). As can be seen cigarette prices increased significantly
over the period with large ‘jumps’ attributable to tax increases accompanied by a slow decline in
sales’ volumes until November 2015. In November 2015 there was a significant increase in prices
and a consequent steep decline in cigarette sales (see below). Cigarette sales have nearly halved
from period from around 1 billion sticks/month in 2010 600 million sticks/month in 2017.
9 By region only. Not by sales channel.
10 Nielsen identifies six brand segments - Base, Value, Sub-premium, Premium, Prestige, Unknown. These have been reduced to four by combining Value and Sub-premium into the Value segment; and Premium and Prestige into the Premium segment.
11 The Nielsen data reports sales revenues. From this and sales volumes the average sales price or ASP has been calculated for each type of cigarette for each month.
![Page 15: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/15.jpg)
15 | P a g e
Figure 3. Monthly sales and average cigarette prices Malaysia
In Figure 3 a vertical line has been drawn on the date (January 2014) the new enhanced LWHs were
implemented in Malaysia. As can be seen there was no discernible reduction in cigarette sales
around that date. The statistical analysis reported below (Annex B) seeks to see whether adjusting
for other factors influence cigarette sales this conclusion is also true.
As has been noted above increases in cigarette taxes affect cigarette prices and sales. This assumes
that the taxes are mostly passed through in increased retail prices; which research and general price
analysis shows to be the case (Laffer, 2014).
Table 4. Development of taxes/duties in Malaysia, 2011-2016 Tax 2011 2012 2013 2014 2015 2016
Import duty (MYR/kg) 200 200 200 200 200 200
Specific taxes (MYR/1,000 sticks)
220 220 250 280 400 400
Ad valorem excise (%) 20 20 20 20 0 0
VAT/sales tax (%) 5 5 5 5 6 6
Source: Euromonitor, Passport - Cigarettes in Malaysia, July 2017, Table 1.
In Malaysia over the period of this study there were significant increases in the taxes imposed on
cigarettes (Table 4). Taxes were increased three times in 2013, 2014, and 2015. 12 If the price
increases in Figure 3 are compared to the tax increases it is clear that they have been primarily
12 Narashidah et al (2013) and Hum et al (2016).
.4
.5
.6
.7
.8
MYR
/sti
ck
400
600
800
1,000
1,200
Mill
ion
sti
cks
LHW enhancement
2010m1
2012m1
2014m1
2016m1
2018m1
Month
Volume (LHS) Price (RHS)
Price adjusted by CPI (RHS)
![Page 16: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/16.jpg)
16 | P a g e
responsible for the large price ‘jumps’ in 2013 and especially 2015. The 2015 tax increase from 280
MYR/1,000 sticks to 400MYR/1,000 sticks appears to have greatly reduced cigarette sales.
The cigarette industry classifies cigarette brands into Base, Value, Mid-price and Premium brand
segments based on their retail prices. These segments cater to smokers with different tastes and/or
income levels, who may react differently to price changes and tobacco controls, and specifically
‘downtrade’ from a higher priced segment to a lower priced segment when the relative prices of
cigarettes change.
The sales of Premium cigarettes, which are the largest segment by volume of sales in Malaysia,
declined significantly over the period. This was probably triggered by a large increase in their prices
from an average of 0.5 MYR/stick in 2010 to over 0.8 MYR/stick in 2017. However, there is no
evidence from the raw Nielsen data that smokers downtraded to cheaper brand segments since the
sales of the latter did not increase disproportionately. In any case, the sales pattern of Premium
cigarettes is significantly different from that of Value and Base brand segments.
Figure 4. Monthly volumes and average prices by price perception
Another possible explanation for the decline in the sales of Premium brands may have been an
increase in illicit cigarette sales. During the period studied cigarette prices in North Malaysia were
some 160% to 220% higher than those in the contiguous South region of Thailand. This is a large
inducement to smuggle cigarettes into Malaysia from Thailand (Figure 4).
0
200
400
600
800
1,000
Mill
ion
sti
cks
2010m1
2012m1
2014m1
2016m1
2018m1
Month
Unknown Premium
Value Base
.2
.4
.6
.8
MYR
/sti
ck
2010m1
2012m1
2014m1
2016m1
2018m1
Month
Unknown Premium
Value Base
![Page 17: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/17.jpg)
17 | P a g e
Figure 4. Malaysia North vs Thailand South average monthly prices
a) All cigarettes
b) By brand segment13
The data does not reveal any significant differences in regional cigarette price and sales patterns
with one exception. Regional cigarette sales and prices displayed similar patterns as for Malaysia as
a whole as shown for the Central region (Figure 5) with the exception of East Malaysia where
volumes increased to the end of 2013, then stagnated and then plummeted in response to the large
13 Comparable only to the extent that the same or similar cigarette brands are categorized under the different price segments in both countries.
.05
.1
.15
.2
USD
/sti
ck
2010m1
2012m1
2014m1
2016m1
2018m1
Month
Malaysia Thailand
.05
.1
.15
.2
.25
USD
/sti
ck
2010m1
2012m1
2014m1
2016m1
2018m1
Month
Malaysia Base Malaysia Value Malaysia Premium
Thailand Value Thailand Mid-price Thailand Premium
![Page 18: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/18.jpg)
18 | P a g e
price increase in November 2015. This suggests that different factors were at play in the East
Malaysia (Borneo) region (which represents only 5% of cigarette sales).
Figure 5. Monthly sales and average prices for Central and East Malaysia regions
Central East Malaysia (Borneo)
The analysis of the Nielsen data above suggests that any statistical analysis should:
Take account of regional differences.
Exclude Malaysia (Borneo) because of its different sales patterns from the other mainland
regions because it would otherwise introduce a bias in the statistical analysis.
Use the price differences between Thailand and Malaysia as a proxy for smuggled cigarette
sales/consumption from Thailand to Malaysia.
Where possible adjust for changes in cigarette taxes.
Empirical Findings
Using the panel and time series regression techniques discussed in Annex A, and the results reported
in Annex B for Malaysia the main finding are that:
The 2014 enhancement of LHWs did not reduce cigarette sales. This is supported by both
panel and time series regressions.
The estimated price elasticities were negative and statistically significant ranging from
between -0.4 and -1.7 depending on the statistical approach. This range is in line with price
elasticity estimates for other similar countries.
The explanatory power of the variables included in the model does not vary much across
different specifications indicating that the statistical results can be considered reliable.
.4
.5
.6
.7
.8
MYR
/sti
ck
150
200
250
300
350
Mill
ion
sti
cks
2010m1
2012m1
2014m1
2016m1
2018m1
Month
Volume (LHS) Price (RHS)
.4
.5
.6
.7
.8
MY
R/s
tick
25
30
35
40
Mill
ion
sti
cks
2010m
1
2012m
1
2014m
1
2016m
1
2018m
1
Month
Volume (LHS) Price (RHS)
![Page 19: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/19.jpg)
19 | P a g e
All regressions fail to find a statistically significant reduction in cigarette sales associated with the
enhancement of LHWs in 2014.
VI. THAILAND
The section describes the data, provides an overview of trends and relationships discernible in the
Nielsen data for Thailand which can guide the statistical analyses, and reports the main results of
the statistical analysis which is set out in greater detail in Annex B.
Data
Nielsen data for Thailand has been collected for the period January 2009 to September 2017. The
data includes information on cigarette sales volumes and revenues by brands, price segments
(Value, Mid-price, Premium), region (Central, North, South, Greater Bangkok, Northeast) and sales
channels (Convenience, Super/hypermarket, Traditional trade).
Data Trends and relationships
Figure 6 shows monthly and quarterly nominal and real monthly cigarette and sales; the quarterly
data providing a better picture of trends given the large variability in the monthly data.
Both cigarette sales and prices increased over the period, the former despite the very large price
increases in May 2009 and August/September 2012. This pattern changed after the large price
increase in February/March 2016 when cigarette sales declined significantly. Similar trends were
evident in all the administrative regions of Thailand.
Figure 6 show the date enhanced LHWs regulations were implemented as a vertical line. Again, this
does not appear to have led to a decline in cigarette sales.
![Page 20: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/20.jpg)
20 | P a g e
Figure 6. Total volumes and nominal and real average prices
A. Monthly
B. Quarterly
Cigarette taxes were (and are) very high in Thailand. Between 2007 and 2017 the excise tax rate
increased from 80% to 90% (Table 5). Again, the price ‘jumps’ correspond more or less to the dates
when these tax increases were implemented.14
14 White & Ross (2015), WHO (2011), Tittabut (2014), Chonvihorpan & Lewis (20xx)
2
2.5
3
3.5
THB
/sti
ck
2,200
2,400
2,600
2,800
3,000
Mill
ion
sti
cks
LHW enhancement
2008m1
2010m1
2012m1
2014m1
2016m1
2018m1
Month
Volume (LHS) Price (RHS)
Price adjusted by CPI (RHS)
2
2.5
3
3.5
THB
/sti
ck
7,000
7,500
8,000
8,500
9,000
Mill
ion
sti
cks
LHW enhancement
2009q3
2011q3
2013q3
2015q3
2017q3
quarter
Volume (LHS) Price (RHS)
Price adjusted by CPI (RHS)
![Page 21: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/21.jpg)
21 | P a g e
Table 5. Development of excise taxes in Thailand, 2007-2017
Excise tax based on ex-factory price
(%)
Effective from
80 29 Aug 2007
85 13 May 2009
87 22 Aug 2012
90 9 Feb 2016
2-tier system based on RSP Effective from
24 THB + 40% if RSP > 60 THB/pack
16 Sep 2017 24 THB + 20% if RSP <= 60
THB/pack
Figure 7 shows cigarette sales and average prices by brand segment. It provides some evidence that
Thai smokers have downtraded from Medium-price to Value segment (sales of Premium cigarettes
were insignificant). This is most likely attributable to the price increase of Mid-price segment rather
than to a change in the relative price of Mid-price and Value segments15 (see explanation below).
The impact of price increases on the Value segment was, if anything, masked by downtrading from
Medium to Value segments, except for the price shock in 2016 after which Value segment sales
were stable.
Figure 7. Monthly and quarterly volumes and price by brand segment.
15 Value segment cigarettes became less expensive over time compared to Mid-price segment brands, but
this was not a gradual change. First, between 2009 and 2011 the price premium of Mid-price cigarettes increased from 40% to 55%. Then (between 2011 and 2016) Mid-price cigarettes became relatively cheap compared to the Value segment (from 55% to 45% price premium). This changed again after 2016: by 2017 the price premium on Mid-price cigarettes increased again to 55%. Despite these changes in the relative price difference between Mid-price and Value segment, neither the decrease in the Mid-price segment volumes nor the increase in the Value segment volumes seem to have been affected.
2
3
4
5
6
Pri
ce p
er
stic
k (T
HB
), d
ash
ed
lin
e
0
500
1,000
1,500
2,000
Mill
ion
sti
cks,
so
lid li
ne
2008m1
2010m1
2012m1
2014m1
2016m1
2018m1
Month
Premium Mid-price Value
Premium Mid-price Value
2
3
4
5
6
Pri
ce p
er
stic
k (T
HB
), d
ash
ed
lin
e
0
2,000
4,000
6,000
Mill
ion
sti
cks,
so
lid li
ne
2009q3
2011q3
2013q3
2015q3
2017q3
quarter
Premium Mid-price Value
Premium Mid-price Value
![Page 22: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/22.jpg)
22 | P a g e
The Nielsen data does not show significant differences in price patterns across the regions (Figure
8). However, sales patterns were different – sales did not decline in the Central Thailand and Greater
Bangkok regions when there were significant price increases; whereas in the other regions they did.
This indicates different factors at play across Thailand.
Figure 8. Total volumes and prices by region (quarterly)
a) Northeast b) Greater Bangkok c) Central
d) South e) North
The decline in cigarette sales after the 2016 price increase may be due (at least partially) to increases
in illicit cigarette sales. The three border regions – the Northeast region borders Laos; the North
region borders Burma and Laos; the South region borders Malaysia - all experienced declining or
stagnant sales of Value segment cigarettes possibly due to increased cross-border smuggling.16
The data suggests that the statistical analysis should:
Use quarterly rather monthly data to avoid masking potential statistical relationships
due to variability in monthly data.
Control for regional differences.
16 Smuggling from Malaysia to Thailand is unlikely as relative prices in Malaysia were significantly higher throughout the period.
2
2.5
3
3.5
THB
/sti
ck
1,600
1,800
2,000
2,200
2,400
Mill
ion
sti
cks
2009q3
2011q3
2013q3
2015q3
2017q3
quarter
Volume (LHS) Price (RHS)
2
2.5
3
3.5
THB
/sti
ck
1,400
1,600
1,800
2,000
2,200M
illio
n s
tick
s
2009q3
2011q3
2013q3
2015q3
2017q3
quarter
Volume (LHS) Price (RHS)
2
2.5
3
3.5
THB
/sti
ck
1,400
1,600
1,800
2,000
2,200
Mill
ion
sti
cks
2009q3
2011q3
2013q3
2015q3
2017q3
quarter
Volume (LHS) Price (RHS)
2
2.5
3
3.5
THB
/sti
ck
900
1,000
1,100
1,200
1,300
Mill
ion
sti
cks
2009q3
2011q3
2013q3
2015q3
2017q3
quarter
Volume (LHS) Price (RHS)
2
2.5
3
3.5
THB
/sti
ck
800
900
1,000
1,100
1,200
Mill
ion
sti
cks
2009q3
2011q3
2013q3
2015q3
2017q3
quarter
Volume (LHS) Price (RHS)
![Page 23: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/23.jpg)
23 | P a g e
Seek to adjust analysis for smuggling activity.
Distinguish short and long-term impacts as they may reveal important market
characteristics.
Empirical Findings
The panel and time series statistical analysis for Thailand are reported in Annex C. In summary these
show that:
The enhancement of LHWs did not reduce cigarette sales.
The indoor SIPP ban had no effect on cigarette sales.
The estimated price elasticity was, as expected, negative and statistically significant with
ranges between -0.5 and -0.8.
The explanatory power of the variables used do not vary much across different
specifications of the statistical analysis, indicating that the results provide reliable
estimates.
The analysis for Thailand does not find a statistically significant reduction in cigarette sales
associated with the enhancement of LHWs introduced in 2014.
![Page 24: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/24.jpg)
24 | P a g e
References Arellano, M and S R Bond (1991) ’Some Specification Tests for Panel Data: Monte Carlo evidence and an application to employment equations’ Review of Economic Studies, Vol. 58, pp. 277–298. Baltagi, B H (2013) Econometric Analysis of Panel Data, 5th edn, Wiley.
Baltagi, B H and J Griffin (2001) ‘The Econometrics of Rational Addiction: The case of cigarettes’ Journal of Business Economics & Statistics, Vol. 19, pp. 449-454.
Baltagi, B H, J M Griffin & W Xiong (2000) ‘To Pool or Not to Pool: Homogeneous versus heterogeneous estimators applied to cigarette demand’ Review of Economics and Statistics, Vol. 82, pp. 117-126
Baltagi, B H & D Levin (1986) ‘Estimating Dynamic Demand for Cigarettes Using Panel Data: The effects of bootlegging, taxation and advertising reconsidered’ Review of Economics and Statistics, Vol. 48, pp. 148-155.
Becker, G S & K V Murphy (1988) “A Rational Model of Addiction" Journal of Political Economy, Vol. 96, pp. 675 -700. Becker, G S, M Grossman & K V Murphy (1991) ‘Rational Addiction and the Effect of Price on Consumption’ American Economic Review, Vol. 81, Papers & Proceedings pp. 237-241. Blundell, R & S Bond (1998) ‘Initial Conditions and Moment Restrictions in Dynamic Panel Data Models’ Journal of Econometrics, Vol. 87, pp. 115–144. Cebula, R J (2013) ‘An Exploratory Panel Data Analysis of Impacts of Cigarette Excise Taxes and State-wide Bans on Cigarette Consumption’ International Journal of Applied Economics, Vol. 10, pp. 22-28
Chonviharnpan, B & P Lewis (2015) ‘The Effects of Tax Changes on Tobacco Consumption in Thailand’ Singapore Economic Review, Vol. 60, pp. 1-18.
Czubek, M & S Johal (2010) Econometric Analysis of Cigarette Consumption in the UK HMRC Working Paper Number 9. Deloitte (2011) Tobacco Packaging Regulation – An International assessment of the intended and unintended impacts Report prepared for British American Tobacco, May. Euromonitor (2017a) Passport - Tobacco in Malaysia, Euromonitor International.
Euromonitor (2017b) Passport Tobacco in Thailand, Euromonitor International.
Fathelrahman AI, M Omar, R Awang et al. (2010) ‘Impact of the New Malaysian Cigarette Pack Warnings on Smokers’ Awareness of Health Risks and Interest in Quitting Smoking’ International Journal of Environ Res Public Health, Vol. 7, pp. 4089-4099. Gallet, C A & J A List (2003) ‘Cigarette Demand: A meta-analysis of elasticities’ Health Economics, Vol. 12, pp. 821–835.
![Page 25: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/25.jpg)
25 | P a g e
Gruber, J, A Sen & M Stabile (2003) “Estimating Price Elasticities when there is Smuggling: The sensitivity of smoking to price in Canada” Journal of Health Economics, Vol. 22, pp. 821–842. Laffer, A B (2014) The Handbook of Tobacco Taxation – Theory and practice, The Laffer Center at the Pacific Research Institute. Liber, A C, H Ross, M Omar & F J Chaloupka (2015) ‘The Impact of the Malaysian Minimum Cigarette Price Law: Findings from the ITC Malaysia survey’ Tobacco Control, Vol. 24, pp. ii83-iii87. Minsoo, J (2016) ‘Implications of Graphic Cigarette Warning Labels on Smoking Behavior: An international perspective’ Journal of Cancer Prevention, Vol. 21, pp. 21–25. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4819662/ Monárrez-Espino, J, B Liu, F Greiner, S Bremberg & R Galanti (2014) ‘Systematic Review of the Effect of Pictorial Warnings on Cigarette Packages in Smoking Behavior’ American Journal of Public Health, Vol 104(10), pp. e11-e30. Nielsen (2016) Illicit Cigarettes Study in Malaysia (ICS) Report for CMTM Member Companies, Annual Report 2016 (Annualised).
Norashidah M N, R A N Mustapha & Y Mastura (2013) ‘Cigarettes Demand and Tax Strategy in Malaysia’ Malaysia Journal of Social Science & Humanities Vol 21(S), pp. 99-114.
Oxford Economics (2017) Asia Illicit Tobacco Indicator 2016: Malaysia, prepared for Philip Morris International Management SA.
Ross, H, A Nabilla & M Al-Sadat (2007) ‘Demand Analysis of Tobacco Consumption in Malaysia’ Nicotine & Tobacco Research, Vol. 9, pp. 1163–1169. Sancelmec, N (2017) ‘The Era of Plain Packaging is Coming in Thailand’ Vidon, https://www.vidon.com/en/news/249-the-era-of-plain-packaging-is-coming-in-thailand.html SEATCA (2015) SEATCA Tobacco Tax Index - Implementation of WHO Framework Convention on Tobacco Control Article 6 in ASEAN Countries 2015, Southeast Asia Tobacco Control Alliance.
Silpasuwan P, N Yaowaluk, C Viwatwongkasem et al. (2008) ‘Potential Effectiveness of Health Warning Labels Among Employees in Thailand’ Journal Medical Association of Thailand, Vol. 91, pp. 551-558. Tittabut J. (2014) ‘Does Cigarette Tax Affect Smokers? The Thailand controversy’ Journal of Health Research, Vol. 28(2), pp. 71-6.
WHO (2011) Tax Policies on Tobacco Products in Thailand: The way forward.
WHO (2013) ‘Press Release on Current Tobacco Packaging Situation in Thailand’ 30 September 2013 Http://Www.Searo.Who.Int/Thailand/News/Tobacco_Packaging/En/
Wakefield, M A, K Coomber, S J Durkin, M Scollo, M Bayly, M J Spittal, J A Simpson & D Hill (2014) ‘Time Series Analysis of the Impact of Tobacco Control Policies on Smoking Prevalence Among Australian Adults, 2001–2011’ Bulletin of World Health Organisation, Vol. 92(6), pp. 413–422.
White J S & H Ross (2015) ‘Smokers’ Strategic Responses to Sin Taxes: Evidence from panel data in Thailand’. Health Economics Vol. 24(2), pp. 127–141.
![Page 26: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/26.jpg)
26 | P a g e
Wooldridge, J M (2014) Introduction to Econometrics, Europe, Middle East & Africa edition, Cenage Learning. World Bank (2001) Economics of Tobacco Toolkit: Tool 3 - Demand Analysis: Economic Analysis of Tobacco Demand by N Wilkins, A Yurekli and T-w Hu. Yong HH, G T Fong, P Driezen, et al. (2013) ‘Adult Smokers’ Reactions to Pictorial Health Warning Labels on Cigarette Packs in Thailand and Moderating Effects of Type of Cigarette Smoked: Findings from the international tobacco control Southeast Asia survey’ Nicotine Tobacco Research, Vol 15, pp. 1339-1347. https://europepmc.org/articles/PMC3715385
![Page 27: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/27.jpg)
27 | P a g e
ANNEX A: STATISTISCAL APPROACHES
BASIC ESTIMATING EQUATION
The panel regression equation used for both Malaysia and Thailand takes a double-log specification:
(1) ln Salesit = 𝑎𝑖 + β1 LHWt + β2 SIPPt + ln 𝑃𝑟𝑖𝑐�̂�it + β2 lnXt + eit
where variables i are collected for each time period t for each of
ln Salesit - the natural log of the sales of number of cigarette sticks.
ln 𝑃𝑟𝑖𝑐�̂�it - the natural log of the average sales price (ASP) of a cigarette stick (and for the IV
approach the instrumented price – see below)
LHWt - a dummy variable representing LHWs
𝑆𝐼𝑃𝑃 - dummies for SIPP ban or bans
𝑙𝑛 𝑋𝑡 is a vector of control variables such as GDP per capita
This specification is routinely in econometric analysis of the determinants of cigarette sales and
consumption e.g. World Bank (2001).
Two statistical approaches have been used - panel and time-series regressions.
PANEL REGRESSIONS
A fixed effects (FE) panel regression has been adopted to deal with the cross-section observations
(Wooldridge, 2014, Chap 14). This technique exploits the richness of the Nielsen panel data to take
account of the unobserved heterogeneity in the cross-sectional observations.
Three different specifications for have been estimated to take greater account of the structural
features of the cigarette market.
(1) Ordinary Least Squares (OLS) with regional dummies for fixed-effects (FE)
This is a static reduced-form equation of the relationship between cigarette sales and prices
and other variables which takes account of regional differences.
(2) Fixed-effects with instrumental variables (FE-IV)
Since market demand and supply are simultaneously determined, cigarette prices in
Equation (1) will be endogenously determined. Pooled OLS or fixed effects regression
estimates will therefore be biased and inconsistent. To adjust for this an instrumental
![Page 28: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/28.jpg)
28 | P a g e
variable (IV) approach has been used. This is a two-staged regression procedure. In the first-
stage prices are regressed on a set of instrumental variables that are expected to explain
changes in supply but not in demand conditions (‘price regression’):
(2) ln Priceit =β1i+ β2Excise𝑡 + β3ln PPIt+ eit
Based on the available data the following instruments have been used - cigarette excise
taxes (Excise𝑡) which are assumed to be exogenously set by the government and which
have a major impact on cigarette prices; the producer price index (PPI) as a proxy for the
costs of production; and for Thailand only the market share of the state-owned Thailand
Tobacco Monopoly (TTM) as a proxy for any potential market power it may have exerted
over cigarette prices.
The adjusted (or instrumented) cigarette prices estimated from Equation (2) (ln 𝑃𝑟𝑖𝑐�̂�it ) are
then used in Equation (1) above (‘sale regression’).
(3) Dynamic fixed effects with instrumental variables (Dynamic FE-IV)
To take account of the adjustment process the preceding panel FE-IV regression has been
re-estimated with cigarette sales lagged by one period to take account of a delayed
response of sales due to smokers’ habits.17 This partial adjustment model focuses on
estimating short run elasticities and speed of adjustment e.g. Baltagi & Levin, (1986 & 1992),
Baltagi et al (2000).18
Dynamic panel regression models can result in estimators being correlated with the error
term giving rise to biased and inconsistent estimates. As a rule of thumb where the number
of time periods exceed 50 the bias is small but still present. The Malaysian data consists of
103 time-periods (months) and the Thai data 38 time-periods (quarters). Many methods
are available to potentially correct for this problem. Below the two most common are used
- the Arellano-Bond (1991) GMM procedure and Blundell-Bond (1998) estimators. These
confirm the estimates derived from specification (3).19
17 The dynamic panel regressions are routinely used in cigarette demand studies e.g. World Bank (2001), Gallet & List (2003), Baltagi & Levin (1986).
18 Another dynamic approach frequently used is the ‘rational addiction model’ developed by Becker & Murphy (1988). This model assumes that current consumption is based on past and future. Since the model is deduced from long term disaggregated individual behaviour it is unsuited to aggregate panel data used in this study e.g. Chaloupka (1991), Becker et al (1991)..
19 See generally Baltagi (2013) Chap 8.
![Page 29: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/29.jpg)
29 | P a g e
TIME SERIES REGRESSIONS
As a check on the validity of the panel regression results the data has been analysed using a dynamic
time-series statistical technique known as the Engle-Granger two-step cointegration analysis.
The Engle-Granger approach has been used in previous cigarette demand studies such as by the UK
tax authorities to examine the impact of VAT on cigarette consumption (Czubek & Johal, 2010).
In time series data many variables exhibit trends which can give rise to ‘spurious regressions’ as they
are correlated to one another due to the trend while, in reality, there is no economic relationship
between them. This is a potential problem with cigarette data where prices and sales often exhibit
trends. However, if there is a linear combination of the variables that is stationary, the econometric
analysis will identify the true relationship between them by allowing standard econometric theory
to apply.
The presence of cointegration is tested by running an OLS regression on the level of the variables,
and if the residual (error term) is stationary then the time series are cointegrated. This is the first
step of the Engle-Granger cointegration test and also estimates the long-term relationship between
the time-series in the model (Wooldridge (2014) Chap 18).
If there is evidence of cointegration (first step), a dynamic error-correction procedure is carried out
in the second step, which estimate whether there are short-term effects that push the time series
back to its long-term equilibrium.
Although the delineation of long- and short- term effects is appealing, the Engle-Granger procedure
only allows for one observation per month/quarter. This greatly reduces the number of
observations to almost one-fifth of the number of observations in the panel regressions, and
adversely affects the explanatory power of the approach.
![Page 30: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/30.jpg)
30 | P a g e
ANNEX B: MALAYSIA REGRESSION RESULTS
DATA
The variables used in the regression analysis for Malaysia together with the source of the data if
other than from Neilsen are listed in Table B.1.
To examine for the impact of tobacco controls, a dummy variable has been for the LWH introduced
in January 2014; and separate dummy variables for the SIPP bans introduced in 2013, 2014 and 2017
respectively.
The regressions also include controls for regional factors by including separate regional dummy
variables for East Coast, Northern, Southern, Southern North; an Index of the percentage difference
between the average cigarette prices in Malaysia and the average cigarette price in Southern
Thailand (Malay-Thai Price diff) as a proxy for the smuggling cigarettes into Malaysia; and as a proxy
for demand GDP per capita.
Table B.1. Definition of variables used in Malaysia regression equations
Variable name Description* Interpretation
Cigarette variables
lnSales
Natural log of monthly sales volumes as the dependent variable
LnPrice
Natural log of weighted average monthly cigarette prices adjusted by CPI Source of CPI: IECONOMICS.
1% change in real prices leads to x% change in sales
Malay-Thai Price diff
Index of % difference between the average cigarette prices between Malaysia and Southern Thailand Source: Nielsen and FED
Proxy for smuggling.
East Coast Northern Southern Southern North
Regional dummy variables in pooled OLS regressions20
Sales, on average, are lower/higher in the region indicated, in comparison to the comparator Central region.
lnSales lag Cigarette sales volume lagged by one-month
Sales volumes in previous month explain x% of current volumes.
20 Regressions by sales channels and by region cannot be run due to absence of data.
![Page 31: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/31.jpg)
31 | P a g e
Large health warning dummies
LHW_2014
Binary variable which equals one in Jan 2014 after second set of LHWs introduced Source: Control of Tobacco Product (Amendment) Regulations.
Effect of LHW on sales from Jan 2010
Other tobacco controls
SIPP_ban 2013 SIPP_ban 2014 SIPP_ban 2017
Dummy variables for SIPP bans in 2013, 2014 and 2017 Source: Thai Government websites
Effect of SIPP bans on sales in comparison to previous sales
lnGDP/capita Natural log of real GDP per capita21 Source: World Bank.
1% change in GDP/capita leads to x% change in sales
Instrumental variables
Tax increase_2013 Tax increase_2014 Tax increase_2015
Dummy variables for specific tax increases Source: Euromonitor.
The effect of specific tax increases on prices, in comparison to sales before the tax change.
PPI Producer price index Source: IECONOMICS.
1 percentage point increase in PPI leads to 100% change in cigarette prices
* Unless otherwise indicated, the source of the data is Nielsen.
PANEL REGRESSIONS
Results Using OLS and FE Estimators
Three panel regressions have been estimated – OLS using regional dummies (1); fixed effects (2) and
dynamic fixed effects using regional panels (3).
Equations (2) and (3) also use an instrumental variable (IV) approach to take account of the
endogeneity of cigarette prices, with a first-stage price regression using separate dummies for the
2013, 2014 and 2015 tax increases, and the Producer Price Index (PPI) as a proxy for production
costs.
The Dynamic panel IV regression (3) uses the dependent variable (cigarette sales) lagged by one
period to represent a possible adjustment process. The dynamic panel regression (3) has also been
re-estimated using Arellano-Bond and Blundell-Bond approaches to generate more efficient
estimators (see next section).
The three sales panel regressions each explained over 90% of the variation in cigarette sales (Table
B.2 below).
Most of the non-tobacco control variables are statistically significant and have coefficients with the
expected sign i.e. have the expected impact on cigarette sales.
21 Only annual data. To avoid unrealistically sudden changes in the data, annual figures were converted to monthly figures by calculating 12-month moving averages. 2017 data are linear predictions.
![Page 32: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/32.jpg)
32 | P a g e
Table B.2. Cigarettes by sales regressions for Malaysia, 2010 -2017
OLS Regional dummies
(1)
Panel Regressions (Fixed-effects)
Panel IV
Dynamic Panel IV
(2) (3)
lnSales regression LHW_2014
0.005 0.010 0.020
(0.802) (0.595) (0.117)
lnPrice -1.635*** -1.730*** -0.430***
(0.000) (0.000) (0.000)
lnSales lag 0.787***
(0.000)
East Coast -0.585***
[Regional Panels]
[Regional Panels]
(0.000)
Northern -0.713***
(0.000)
Southern -0.242***
(0.000)
Southern North -1.244***
(0.000)
SIPP_ban 2013 0.019 0.022 -0.002
(0.323) (0.141) (0.862)
SIPP_ban 2014 0.038* 0.055*** 0.007
(0.061) (0.002) (0.537)
SIPP_ban 2017 -0.129*** -0.121*** -0.016
(0.000) (0.000) (0.143)
GDP_capita -0.240 -0.084 0.386
(0.572) (0.842) (0.148)
Malay-Thai Price Diff 0.294*** 0.298*** 0.013
(0.000) (0.000) (0.619)
lnPrice regression
Tax increase_2013 0.107*** 0.105***
(0.000) (0.000)
Tax increase_2014 [omitted] [omitted]
Tax increase_2015 0.181*** 0.177***
(0.000) (0.000)
PPI 0.000 0.000
(0.208) (0.259)
No. observations 465 465 460
R2 lnSales regression 0.971 0.900 0.967
Prob > F 0.000 0.000 0.000
R2 lnPrice regression 0.944 0.890
Prob > F 0.000 0.000
Note: The 2014 tax and SIPP ban 2014 dummies are perfectly correlated and therefore the 2014 tax dummy has been dropped from the regression due to multicollinearity.
![Page 33: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/33.jpg)
33 | P a g e
The IV price regressions for Equations (2) and (3) explain 90% to 98% of the variation in cigarette
prices. All the tax rate dummies had positive coefficients indicating that tax changes were passed
on in higher cigarette prices. The PPI had a negligible and statistically insignificant effect on cigarette
prices. Since the PPI is for all industry this is not surprising.
All three panel regressions show that the 2014 enhancement of LHWs did not reduce cigarette sales.
Indeed, the coefficients are positive in all regressions, implausibly suggesting that the LWH
‘increased’ cigarette sales although the estimates were not statistically significant.
The SIPP bans did not reduce cigarette sales.
EFFICIENT DYNAMIC PANEL ESTIMATORS
The dynamic panel regression has been re-estimated using the Arellano-Bond and Blundell-Bond approaches (Table B.3 and B.4 respectively).
The SIPP dummies have been excluded from the regression because the t-statistic cannot be calculated due to collinearity.
The Arellano-Bond estimates are in line with those reported for Equation (3) above confirming that the 2014 LWH enhancement did not have a statistically significant effect in reducing cigarette sales.
Table B.3. Arellano-Bond dynamic panel-data estimation
Group variable: region_num Number of groups = 5 Time variable: month Obs per group: min = 91; avg = 91; Max = 91 Number of instruments = 454 Number of obs = 455 Wald chi2(4) = 40.56 One-step results (SE adj for clustering on region_num Prob > chi2 = 0.0000
Variable Coef. Robust Std. Err.
z P>|z| [95% Conf. Interval]
LHW 2014 0.013 0.009 1.54 0.123 -0.004 0.030
LnSales lag(1) 0.827 0.022 36.99 0.000 0.784 0.871
lnPrice -0.347 0.068 -5.08 0.000 -0.480 -0.213
LnGDP capita 0.382 0.138 2.76 0.006 0.111 0.653
Malay-Thai Price Diff 0.006 0.011 0.57 0.566 -0.015 0.027
Constant -0.732 0.981 -0.75 0.456 -2.654 1.191
The Blundell-Bond estimators also show that the LWH dummy did not lead to a reduction in
cigarette sales (the dummy is significant but has the wrong sign suggesting that it increased cigarette
sales, which is implausible). The lagged cigarette sales coefficient is larger; hence the price elasticity
is smaller.
![Page 34: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/34.jpg)
34 | P a g e
Table B.4. Blundell-Bond dynamic panel-data estimation
Number of obs = 460 Group variable: region_num Number of groups = 5 Time variable: month Obs per group: Min = 92; Avg = 92; Max = 92 Number of instruments = 636 Wald chi2(4) = 80334.85 One-step results Prob > chi2 = 0.0000
Variable Coef. Robust Std.
Err. z P>|z| [95% Conf. Interval]
LHW 2014 0.019 0.008 2.55 0.011 0.004 0.034
LnSales lag(1) 0.970 0.016 61.84 0.000 0.939 1.000
LnPrice -0.081 0.031 -2.62 0.009 -0.141 -0.020
SIPP ban 2013 -0.008 0.002 -3.42 0.001 -0.013 -0.003
SIPP ban 2014 -0.011 0.004 -2.47 0.013 -0.019 -0.002
SIPP ban 2017 0.006 0.004 1.56 0.120 -0.002 0.014
LnGDP capita 0.387 0.147 2.63 0.009 0.098 0.675
Malay-Thai Price Diff -0.057 0.023 -2.44 0.015 -0.102 -0.011
Constant -2.170 0.886 -2.45 0.014 -3.905 -0.434
TIME SERIES ANALYSIS
The time series results indicate that there is no short-term effect for any of the variables included
(which were all statistically insignificant in the second-step table).
The first-step table indicates that price has a long-term strong negative impact on cigarette sales
with the LHW dummy having a small significant positive (unexpected) long-term effect on cigarette
sales.
The results are different from the panel regression estimates (and produce more unexpected
coefficients) which indicate the importance of panel variations vis-à-vis variations over time (see
comment on the reliability of time series models for short periods in Annex A).
Table B.5. Engle-Granger estimation, Malaysia- All variables; quadratic trend
Engle-Granger first-step regression (dependent variable: volume)
Coef. Std. Err. t P>t
[95% Conf. Interval]
LHW 2014 0.054 0.022 2.42 0.018 0.010 0.099
lnPrice -0.956 0.142 -6.73 0.000 -1.239 -0.674
SIPP ban 2013 0.030 0.019 1.6 0.114 -0.007 0.068
SIPP ban 2014 0.065 0.022 3.01 0.003 0.022 0.108
SIPP ban 2017 -0.001 0.032 -0.05 0.964 -0.065 0.062
Dif price Mal Tha 0.104 0.061 1.69 0.095 -0.018 0.226
lnGDP capita 0.105 0.929 0.11 0.910 -1.742 1.952
Trend 0.004 0.002 2.11 0.038 0.000 0.007
Trend2 0.000 0.000 -4.31 0.000 0.000 0.000
Constant 12.146 6.216 1.95 0.054 -0.218 24.510
![Page 35: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/35.jpg)
35 | P a g e
Engle-Granger second-step ECM (dependent variable: D.volume)
Coef. Std. Err. t
P>t [95% Conf. Interval]
L1.egresid -0.638 0.115 -5.56 0.000 -0.867 -0.410
LD.LHW 2014 -0.019 0.037 -0.51 0.609 -0.092 0.054
LD.lnPrice -0.208 0.241 -0.87 0.389 -0.687 0.270
LD.SIPP ban 2013 -0.018 0.037 -0.5 0.617 -0.092 0.055
LD.SIPP ban 2014 0.003 0.039 0.08 0.940 -0.075 0.081
LD.SIPP ban 2017 0.006 0.040 0.15 0.885 -0.074 0.086
LD.Dif price Mal Tha -0.032 0.089 -0.36
0.716 -0.209 0.144
LD.lnGDP capita 1.940 1.193 1.63 0.108 -0.433 4.313
Constant -0.007 0.004 -1.82 0.072 -0.015 0.001
Note: LD = lagged differences of the respective variable.
MAIN FINDINGS
The main findings of the panel and time series regressions are that:
There is no statistical evidence that the 2014 LWH enhancement led to reduction in licit
cigarette sales.
The SIPP bans also did not reduce cigarette sales.
The estimated price elasticities range fall between -0.4 and -1.7.
The tax rate change dummies in all the IV price regressions had a statistically positive effect
on cigarette prices as expected.
The smuggling proxy had a positive coefficient and was statistically significant in Equations
(1) and (2) but not in the dynamic panel regression (Equation (3)). This unexpected result
suggests that the price differential variable is not a good proxy for smuggling.
The coefficients of LHWs, SIPP bans and cigarette prices do not vary much across the three
panel regression specifications which indicate that the models are statistically robust.
![Page 36: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/36.jpg)
36 | P a g e
ANNEX C: THAILAND REGRESSION RESULTS
DATA
The variables used in the regressions for Thailand are listed in Table C.1.
Quarterly data has been used to adjust for the large variation in monthly data.
The LHW implemented in September 2014 is represented by the dummy variable LHW_2014.
In addition, a dummy has been included to capture the extension of the SIPP ban to indoor smoking
in September 2014 (SIPP Indoor).
Table C.1. Definition of variables used in Thailand regression equations
Variable name Description* Interpretation
Cigarette variables
lnSales Natural log of cigarette sales (quarterly22) – dependent variable
lnPrice
Natural log of weighted average (quarterly) price adjusted for regional CPI Source of CPI: Thailand Bureau of Trade and Economic Indices.
1% change in prices leads to x% change in sales
Greater Bangkok North Northeast South
Regional dummies23 in pooled OLS
regressions
Sales, on average, are lower/higher in the region indicated, in comparison to the comparator Central region.
lnSales lag Cigarette sales lagged by one quarter Sales volumes in previous month explain x% of current volumes.
Large health warning dummies
LHW 2014
Dummy equals 1 for each quarter after the
introduction of the Sep 2014 LHW which
increased minimum size to cover 85% of the
total packaging
Effect of LHW on sales from Q4 2014
22 2017 Q3 was removed from the data because new government regulations and taxes were introduced (see section II above).
23 Separate regressions for sales channels could be carried out but sales channel data are not split by region.
![Page 37: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/37.jpg)
37 | P a g e
Control variables
SIPP Indoor
Dummy variable equalling 1 for each quarter
after 2010q1,24 for ban on smoking indoors.
Source: Ministry of Public Health.
Effect of prohibition of smoking in indoor places from Q1 2010.
Ln GDP/capita Natural log of real GDP per capita25 Source: World Bank.
1% change in GDP/capita leads to x% change in sales
Instrumental variables
Tax Increase_2009 Tax Increase_2012 Tax Increase_2016
Dummies for excise increases Source: Euromonitor.
The effect of tax increases on prices
PPI Producer price index Source: IECONOMICS.
1 percentage point increase in PPI leads to 100x% change in cigarette prices
TTM Volume-based market shares of the Thailand Tobacco Monopoly (TTM)
1 percentage point decrease in TTM market share leads to 100x% change in cigarette prices
* Unless otherwise indicated, the source of the data is Nielsen.
Differences in regional effects are captured by dummy variables for Greater Bangkok, North,
Northeast, South.
As a proxy for smuggling an index of the per centage difference between the average price in
Thailand and the average yearly cigarette price in Laos has been used but in initial regressions this
generated and was dropped.
PANEL REGRESSIONS
Three panel regressions have been estimated – OLS using regional dummies (1); fixed effects (2) and
dynamic fixed effects (3); the latter two using instrumental variables (IV) to capture the endogeneity
of prices.
For Equations (2) and (3) three instrumental variables have been used in the price regression: (i) a
dummy variable for each of the tax increases in 2009, 2012 and 2016; (2) the PPI to take account of
changes in production costs; and (3) the market share of the state-owned Thailand Tobacco
Monopoly (TTM) to allow for the possibility that this state “monopoly” producer of cigarettes, albeit
one with a dwindling market share, had market power over the period that might have increased
cigarette prices.
The Dynamic panel IV regression (3) uses the dependent variable (cigarette sales) lagged by one
period to represent a possible adjustment process to changes in sales. The dynamic panel regression
has also been re-estimated using the Arellano-Bond and Blundell-Bond approaches.
Main results are reported in Table C.2.
24 It is assumed that indoor prohibition was implemented in January 2010.
25 Only annual data available. This has been smoothed in a 12-month moving average. Data for 2017 was estimated using a linear projection.
![Page 38: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/38.jpg)
38 | P a g e
Table C.2. Cigarettes sales regressions for Thailand, 2009-2017
OLS
Regional dummies
(1)
Panel Regressions (Fixed Effects)
IV Dynamic IV
(2) (3)
lnSales regression LHW 2014 0.038 0.035* 0.008
(0.117) (0.079) (0.578)
lnPrice -0.762*** -0.819*** -0.483***
(0.000) (0.000) (0.000)
lnSales lag(1) 0.616***
(0.000)
Greater Bangkok 0.085***
[Regional Panels]
[Regional Panels]
(0.000)
North -0.632***
(0.000)
Northeast 0.068***
(0.000)
South -0.493***
(0.000)
SIPP Indoor 0.073*** 0.071*** 0.030
(0.000) (0.000) (0.097)
lnGDP_capita 1.483*** 1.671*** 0.902**
(0.005) (0.001) (0.022)
lnPrice regression: Tax Increase_2009a 0.275*** 0.095***
(0.000) (0.000)
Tax Increase_2012 0.086*** 0.086***
(0.000) (0.000)
Tax Increase_2016 0.185*** 0.183***
(0.000) (0.000)
PPI -0.011*** -0.011***
(0.000) (0.001)
TTM 0.006*** 0.006***
(0.002) (0.003)
No. observations 170 170 165 R2 lnSales regressions 0.956 0.449 0.713
Prob > F 0.000 0.000 0.000
R2 lnPrice regression 0.894 0.831
Prob > F 0.000 0.000
Note: a: The tax effect in (3) is partial because the dynamic specification used omits the first quarter of the tax rise.
The sales panel regressions differ in explanatory power: with Equation (1) explaining over 90% of
the variation in cigarette sales; the static IV regression (2) only 45%; and the dynamic IV specification
around 71%.
![Page 39: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/39.jpg)
39 | P a g e
The coefficients for the price, GDP per capita, PPI, TTM, and the regional and tax dummy variables
all have the expected signs and are statistically significant at the 1 per cent level. They indicate that
GDP per capita had a large positive effect in cigarette sales.
The IV price regressions explain over 80 to 90 per cent of the variation in cigarette prices. Increases
in the PPI, excise tax rates and a higher market share of the domestic TTM cigarette producer all
increased cigarette prices and were statistically significant at the 1 per cent level.
The coefficients on (the instrumented) cigarette prices indicate price elasticities of between -0.48
to -0.82.
The LHW and SIPP dummies have, with one exception, positive coefficients which implausibly
indicate that they increased cigarette sales. The LHW was not statistically significant except in
Equation (2) but had the wrong sign.
EFFICIENT DYNAMIC PANEL ESTIMATORS
The Arellano-Bond and Blundell-Bond dynamic estimations show similar coefficients to the dynamic
panel specification. However, because of the singularity of the covariance matrix, standard errors
could not be estimated and therefore the results are not meaningful. They are reported in Table C.3
for completeness.
Table C.3. Arellano-Bond dynamic panel estimation
Arellano-Bond dynamic panel-data estimation Number of obs = 160 Group variable: region_num Number of groups = 5 Time variable: quarter Obs per group: min = 32; avg = 32; max = 32 Number of instruments = 159 Wald chi2(4) = . Prob > chi2 = . One-step results - (Std. Err. adjusted for clustering on region_num)
Blundell-Bond dynamic panel coefficients are in line with the above but again due to singularity in
the variance matrix, standard errors cannot be estimated. They are reported in Table C.4 for
completeness.
Variable Coef.
LHW 2014 0.012
Sales lag(1) 0.634
LnPrice -0.395
SIPP Indoor 0.036
LnGDP capita 0.618
Constant -0.404
![Page 40: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/40.jpg)
40 | P a g e
Table C.4. Blundell-Bond dynamic panel estimation
System dynamic panel-data estimation Number of obs = 165 Group variable: region_num Number of groups = 5 Time variable: quarter Obs per group: min = 33; avg = 33; Max = 33 Number of instruments = 223 Wald chi2(4) = . One-step results Prob > chi2 =
TIME SERIES ANALYSIS
The time series regressions (Table C.5) indicate that there is no short-term effect for any of the
variables included (insignificant coefficients in the second-step table).
The first-step results indicate that price has a strongly negative long-term impact on cigarette sales.
The LHW dummy is insignificant at 5% confidence level (but with the wrong sign). The other control
variables are insignificant. The results are in line with the dynamic panel regression results above.
Table C.5. Engle-Granger second-step estimation - first-step linear trend
Engle-Granger first-step regression (dependent variable: volume)
Coef. Std. Err. t P>t
[95% Conf. Interval]
LHW 2014 0.034 0.020 1.71 0.099 -0.007 0.074
lnPrice -0.568 0.089 -6.39 0.000 -0.750 -0.386
SIPP Indoor 0.023 0.018 1.28 0.212 -0.014 0.061
LnGDP capita -0.153 0.582 -0.26 0.795 -1.346 1.040
Trend 0.005 0.001 3.61 0.001 0.002 0.007
Constant 17.753 5.512 3.22 0.003 6.463 29.043
Engle-Granger second-step ECM (dependent variable: D.volume)
Coef. Std. Err. t P>t
[95% Conf. Interval]
L1. egresid -1.135 0.239 -4.75 0.000 -1.626 -0.643
LD.LHW 2014 -0.017 0.028 -0.6 0.552 -0.073 0.040
LD.lnPrice -0.183 0.126 -1.45 0.158 -0.443 0.076
LD.ISIPP Indoor 0.004 0.027 0.14 0.893 -0.052 0.059
LD.lnGDP capita 0.788 0.892 0.88 0.385 -1.045 2.621
Constant 0.004 0.005 0.71 0.482 -0.007 0.014
Note: LD = lagged differences of the respective variable.
Variable Coeff.
LHW 2014 0.008
LnSales lag(1) 0.891
LnPrice -0.205
SIPP Indoor 0.026
LnGDP capita 0.051
Constant 1.227
![Page 41: STATISTICAL ANALYSIS OF LARGE HEALTH WARNINGS ON … · The rationale for LHWs is given by the WHO (2013): Large graphic health warnings on tobacco packaging are the most effective](https://reader035.vdocument.in/reader035/viewer/2022071217/604ab16e84923e6fc024e0a9/html5/thumbnails/41.jpg)
41 | P a g e
To sense-check the results a separate regression has been run using volumes and prices only (Table
C.6). The Engle-Granger results confirm what the charts in Section VI show - there may be both
short-term and long-term effects of price changes on cigarette sales.
Table C.6. Engle-Granger 2-step estimation – sales & prices only; first-step linear trend
Engle-Granger first-step regression (dependent variable: volume)
Coef. Std. Err.
T P>t [95% Conf. Interval]
lnPrice -0.576 0.060 -9.66 0.000 -0.697 -0.454
Trend 0.006 0.000 12.66 0.000 0.005 0.007
Constant 16.307 0.050 325.68 0.000 16.205 16.409
Engle-Granger second-step ECM (dependent variable: D.volume)
Coef. Std. Err. T P>t [95% Conf. Interval]
L1. Egresid -1.017 0.208 -4.89 0.000 -1.443 -0.592
LD.lnPrice -0.211 0.120 -1.75 0.090 -0.457 0.035
Constant 0.006 0.005 1.21 0.235 -0.004 0.015
Note: LD = lagged differences of the respective variable.
MAIN FINDINGS
The main findings of the panel and timeseries regressions are that:
The LHW dummy is not associated with any reduction in cigarette sales.
The indoor SIPP ban had a positive statistically significant effect on cigarette sales. This is
implausible and is most likely to be due to spurious correlation.
The price elasticity estimates range between -0.5 (Dynamic FE-IV) and -0.8 (FE-IV) and are
all significant at the 1 per cent level.
The growth in GDP per capita increased cigarette sales and is statistically significant.
The tax rate changes increased cigarette prices in all regression.
The state TTM cigarette producer’s market share had a small but statistically significant
positive effect on cigarette prices indicating that the higher its market share the higher were
cigarette prices. (Or looked at the other way, greater competition from international
cigarette producers lowered cigarette prices all things equal)
The results do not vary much across the three panel regression specifications which indicate
that the panel models are robust.