Effects of the implementation of pictorial warnings on cigarette packages on current and expected health

(1)

Effects of the implementation of pictorial warnings on cigarette

packages on current and expected health

By

Timon Louwsma (s2773368)

A thesis submitted in fulfillment of the

requirements for the degree of

Master of Science in

Business Administration Specialization Health

at the

University of Groningen

22-6-2020

Supervisor: Prof. Gerard J. Van Den Berg

Co-Assessor: Prof. dr. M.J. Postma

(2)

2

Abstract

Introduction

Since May 2016, cigarette packages are obliged to have pictorial warnings on both sides in the European Union. These warnings are meant to better inform smokers about the negative health consequences of smoking. The change from text-only to pictorial warnings has been shown to increase the amount of initiated and successful quit attempts in smoking individuals, and reduce the intention to start smoking in non-smokers. Nevertheless, a large part of the smokers has continued smoking. This study has been conducted to investigate how the implementation of pictorial warnings affected the perceived risk of damage in these continuing smokers.

Methods

A difference in difference (DID) model was used to determine the effect of the pictorial warnings on the perceived risk of damage in continuing smokers. Subjective survival probabilities and self-perceived health were used to determine the effect on the perceived risk of damage in continuing smokers. Individuals that smoked in 2015 and 2016 were assigned to the treatment group, and individuals that did not smoke in both of these years were assigned to the control group. Using ordinary least squares and fixed-effects regression analyses, effects of the pictorial warnings on the scores of the subjective survival probabilities and self-perceived health were estimated. Furthermore, a logit regression was performed to find if a similar decrease in likelihood that an individual smokes was present.

Results

The implementation of pictorial warnings was found to have no effect on the subjective survival probability (0.054 ± 0.055, p > 0.05) and the self-perceived health (0.0076 ± 0.023, p > 0.05) of continuing smokers. Continuing smokers did not change in average scores after the implementation in 2016, which was also found in all subgroup analyses based on age and sex. The logit regression showed that the likelihood that an individual smoked decreased after the implementation of pictorial warnings 0.22 ± 0.019, p < 0.01). This decrease was stronger in women 0.24 ± 0.049, p < 0.01) than in men (-0.13 ± 0.059, p < 0.05) in the first year after implementation, but over the first three years after implementation, the decrease in likelihood in men and women was similar.

Discussion

(3)

3

Introduction

Smoking kills. The habit of smoking reduces the life span of an individual with an average of 10 years (Jha & Peto, 2014). In 2017, almost 20.000 people died in the Netherlands as a result of smoking. (Rijksinstituut voor Volksgezondheid en Milieu, 2018), and more than one in five people in the Netherlands still smokes (Trimbos Instituut, 2019). Even though this is a large part of the population, a decrease in smokers of 3,3% has been observed in recent years (Centraal Bureau voor de Statistiek, 2019). In 2017, 22,4% of the Dutch population smoked, compared to 25,7% in 2014. Nevertheless, this reduction is far from the goal of the Dutch government and 70 other organizations, like healthcare organizations and municipalities. The national government of the Netherlands has set the following goal for the year 2040, in which there has to be a smoke free generation and where merely 5% of the population smokes (Ministerie van Volksgezondheid, Welzijn en Sport, 2018). Most actions to discourage smoking behaviors are scheduled to be carried out in 2020, but some measures have already been implemented (Rijksoverheid, 2018).

Creating awareness about the negative effects of smoking on an individual’s health is one of the priorities of many interventions to reduce smoking behavior (Winefield, 2006). In order to create this awareness, the Dutch government introduced a regulation in 2002, stating that all cigarette packages should include textual warnings about the negative consequences of smoking behavior. Because an average Dutch smoker smokes 13 cigarettes a day, the exposure to warning messages on a cigarette package would be around 4700 times a year (Nationaal Experticecentrum Tabaksontmoediging, 2016). The World Health Organization (WHO) followed the Dutch government by introducing the WHO Framework Convention on Tobacco Control in 2003. This framework requires all suppliers of cigarettes to put textual warning messages about the consequences of smoking on their packages (Pötschke-Langer et al., 2015). However, due to a lack of variation in the warning messages, the effectiveness of this seemingly good intervention has reduced after the implementation. (Nationaal Experticecentrum Tabaksontmoediging, 2016).

(4)

4

2016). By the end of 2016, already 100 countries worldwide had adopted warning pictures on cigarette packages. By the end of 2018, this number had grown to 118 and more countries will probably adopt pictorial warnings in the near future (Canada Cancer Society, 2018).

Literature Review

Before the implementation of pictorial warnings on cigarette packages in May 2016 in Europe, textual warnings were used to inform smokers about the negative consequences of smoking (Nationaal Expertisecentrum Tabaksontmoediging, 2016). The effectiveness of these text-only warnings has been thoroughly studied. A study from Japan showed that text-only warnings have a weak effectiveness on the intention to quit in smokers (Chung-Hall et al. 2020). Furthermore, the International Tobacco Control Policy Evaluation Project (2015) has studied the overall effectiveness of the text-only warnings in multiple countries. They also found that the textual warnings had a limited effect on the perception and behavior of smokers and that the effectiveness of text-only warnings further declined over time. Out of the 19 countries they examined, the Netherlands was the country that had the smallest percentage of smokers who thought about their health due to the warnings and noticed the warnings least often. Hitchman et al. (2012) showed similar results. They found that the effectiveness of text-only warnings differed between countries and that they had the least impact on smokers in the Netherlands. Therefore, they suggested that the introduction of new warnings such as pictorial warnings could especially be beneficial for the Netherlands. Moreover, several studies demonstrated that the introduction of new warnings like pictorial warnings could be beneficial for European countries (Agaku et al. 2014; Mannocci et al. 2012). A couple of years before the actual implementation, these studies showed that pictorial warnings may stimulate smokers to quit and inform smokers better about the negative consequences of smoking.

(5)

5

Since other countries already introduced pictorial warnings on both sides of cigarette packages in 2001, a considerable amount of research has been conducted on the direct effects of this intervention on smoking behaviors. For Canada, the first country to introduce the pictorial warnings, promising results have been found on the effectiveness of the pictorial warnings compared to textual warning messages. Azagba and Sharaf (2013) showed that, as a result of the graphical warnings, smoking prevalence reduced and quit attempts increased. Moreover, in 2018, Ratih and Susanna published a literature review about the effect of pictorial warnings in Asia. The 14 studies they included also demonstrated that pictorial warnings were perceived as more effective in discouraging individuals to start smoking and that the pictorial warnings stimulate smoking cessation. In America, similar effects have been shown in a research carried out in California by Brewer et al. (2016). In their study, they compared the amount of intentioned, initiated and successful quit attempts of smokers exposed to the pictorial warnings with smokers exposed to textual warnings. They found that smokers who were exposed to the pictorial warnings intentioned, initiated and succeeded more quite attempts than smokers who were not exposed to the pictorial warnings. The findings that individuals reduced their intention to start smoking and had a higher intention to quit were also found in the meta-analysis performed by Noar et al. in 2016, which was performed using data from 37 different studies from 16 countries. In their subsequent study, they also confirmed the finding of their meta-analysis that enhanced warnings increase the attention that is drawn towards it, the way in which warnings are processed and the effectiveness of the warning (Noar et al. 2017).

In Europe, also mostly positive results have been found after the implementation of pictorial warnings in 2016. In their study with participants from ten European countries, Woelbert and d’Hombres (2019) found that pictures had a positive effect on the health warnings. Pictorial warnings improved warning salience and the motivation to quit smoking. Besides, they found that the pictures have to be renewed in order to cope with a wear-out effect. In line with the previous results, the pictorial warnings were found to have a positive effect on quit attempts and smoking cessation in the Netherlands (Van Mourik et al. 2019). Van Mourik et al. stated that this was caused by an increase in health worries, a social pressure to quit and a more positive attitude towards quitting. On the other hand, a study in Austria showed only limited effect of the pictorial warnings, even on quitting (Mayerl et al. 2018). In this study, it was suggested that the effect of pictorial warnings was small because smokers denied the negative health effects of smoking.

(6)

6

Tabaksontmoediging, 2016). Little is known about this effect of the pictorial warnings in continuing smokers. To find out if the implementation of pictorial warnings enhanced the knowledge did enhance the knowledge of continuing smokers, the perceived risk of damage of continuing smokers can be used. The perceived risk of damage is an individual’s perception of the likelihood of developing negative outcomes and the severity of those outcomes (Van Der Pligt, 2001). Continuing smokers are aimed to improve their knowledge about the risks of smoking due to pictorial warnings and therefore their perceived risk of damage should increase as well. So far, the perceived risk of damage in continuing smokers has not been shown to change as a result of the pictorial warnings (Brewer et al. 2019; Hall et al. 2018). However, these studies looked at the differences in perceived risk of damage between a group of smokers exposed to textual warnings and a group exposed to pictorial warnings. No study has investigated how the implementation of pictorial warnings affected the perceived risk of damage in continuing smokers who were previously exposed to textual warnings.

Research Question

The current study is conducted to further investigate if the implementation of pictorial warnings reached its aimed effect on continuing smokers. This study differentiates itself because the effects of the pictorial warnings on continuing smokers are studied within the same individual, whereas previous studies focused on a difference between individuals exposed to textual and pictorial warnings. With the use of longitudinal data on the same individuals over multiple years, the effect of the implementation of the pictorial warnings on the perceived risk of damage in continuing smokers can be studied thoroughly. As a measure for the perceived risk of damage, subjective survival probabilities are used. The subjective survival probability is an individual’s assessment of the probability of surviving to a certain target age (Rappange et al. 2015). Questions about subjective survival probability are used in multiple household surveys and have been shown to yield reliable measures of expectations about an individual’s survival (De Bresser, J. 2019). Moreover, Hurd and McGarry showed in 2002 that subjective survival probability is a respectable predicter of actual survival. They also found that new information, such as the commencement of a disease, causes respondents to adjust their survival probabilities. Pictorial warnings improve the information about the risks of smoking that is given to smoking individuals. So, if the policy produces a desired effect, smoking individuals should be more aware of the life-shortening risks smoking. Therefore, a reduction in subjective survival probability of smoking individuals from 2016 is expected if the pictorial warnings achieve their intended effect. Hence, this study uses longitudinal data on the subjective survival probability to determine if the implementation did have an aimed effect on the perceived risk of damage in continuing smokers, and aims to answer the following question.

“To what extend has the change from textual warnings to pictorial warnings on cigarette

(7)

7

Furthermore, self-perceived health is used to determine short-term effects on the experienced health of smokers. One of the first health impairments individuals experience as a result of smoking is a deterioration of the self-perceived health (Rius et al. 2004). Self-perceived health is a subjective measure of an individual’s perspective on his/her health status (Shields and Shooshtari, 2001). It can be measured using a questionnaire, asking individuals how they would rate their current health status and is proven to be a reliable measure for the overall health status of an individual (Martikainen et al. 1999). Different research has shown that heavy smoking negatively influences self-perceived health (Szklo and Coutinho, 2009; Shields and Shooshtari, 2001; Rius et al. 2004). Because pictorial warnings can evoke greater emotional response, it is interesting to find out if this had a short-term effect on the experienced health of continuing smokers. However, since the self-perceived health is a measure of in individual’s experienced health at that moment in time and most negative health consequences due to smoking are developed later in life, the self-perceived health is not expected to change. The following question is aimed to be answered.

“To what extend has the policy affected the self-perceived health of continuing smokers?”

Questions about the subjective survival probability can be difficult for respondents to answer, but give a rather good indication about the perceived risk of damage. Questions about the self-perceived health are easier to comprehend, but yield less clear results about the perceived damage of smoking, since smoking produces insidious health effects that are visible later in life. Thus, information about the future and present are used complementary in this study. In order to give a broader perspective, the following sub questions were asked.

“How does the effect of the implementation of the pictorial warnings change over time?”

“Is there a difference of the effectiveness of the pictorial warnings on the subjective survival probabilities and self-perceived health different based on sex or age?”

Lastly, this study will investigate if a similar effect of the implementation of the pictorial warnings was present on quit attempts in the used longitudinal data, as found in previous studies. Using longitudinal data about smoking, this study aims to answer the following question.

(8)

8

Methodology

Data Sample

In this study, the data of the LISS (Longitudinal Internet Studies for the Social sciences) panel administered by CentERdata (Tilburg University, The Netherlands) was used. The LISS panel is a representative sample of Dutch individuals who participate in monthly internet surveys. The panel is based on a true probability sample of households drawn from the population register. Data from the longitudinal Health data and the demographic information of the participants was combined and further used for the analysis. Variables that were not needed for further analysis were left out during the merging of the data. The data collection period for health data each year covers the month November. In 2014, no data has been collected. This resulted that the data collection has been advanced to the month July in 2015. From 2016, data has been collected during the month November again.

Because the policy about the implementation of pictorial warnings has come into effect the 16th of May 2016, individuals that did not give answers on the questions about the subjective survival probability in the years around the policy, 2015 and 2016, were excluded from the dataset. Also, individuals that were younger than 18 years old in 2015 were excluded. In total, 2560 individuals were included. Since 2015 is the last year of measurement before the implementation of the policy, this year will be referred to as the baseline. In order to create correct interpretable results from regression analyses and logical baseline characteristics, variables that corresponded to ‘yes’ with ‘1’ and ‘no’ with ‘2’ were renamed so that a ‘no’ corresponds to a ‘0’. In this way, incidences of variables yield representative values in the regression model.

Subjective survival probability and self-perceived health

Multiple questions from the dataset were chosen to function as indicator for subjective survival probability. Five questions regarding the possibility of reaching a certain age were used to determine the change in perceived risk of getting life shortening complications. One of these questions was “How

would you rate your chance of living to be 75 years old or older? Please rate your chance on a scale from 0 to 10, where 0 means ‘no chance at all’ and 10 means ‘absolutely certain’.” In the other

questions, the prospected age is changed into 80, 85, 90 and 95. Unfortunately, no respondents submitted answers to the question about age 85, 90 and 95. Thus, only the subjective survival probabilities for the age of 75 and 80 were used in the analysis. Answers to these questions from 2013 to 2018 were used to determine the effect of pictorial warnings on the subjective survival probability and in this way on the perceived risk of damage.

(9)

9

speaking?”. The answer is measured on a scale from 1 to 5, where 1 means poor health, 2 suggests

moderate health, 3 implies good health, 4 very good health and 5 is associated with excellent health. In order to determine how the health compares to the self-perceived health of the previous year, the following question was used: “Can you indicate whether your health is poorer or better, compared to

last year?”. This question was used to determine if individuals that indicate a similar self-perceived

health to the previous year, did actually feel slightly better or worse this year. It is again measured on a 1 to 5 scale, where individuals choose 1 if their health is considerably poorer, a 2 if their health is somewhat poorer, 3 if their health stated the same, 4 if their health became somewhat better and 5 if their health is considerably better. Scores on both questions can thus be interpreted as the higher the score, the better the self-perceived or changed health.

Statistical analysis

Difference in difference

For the first part of the analysis, a Differences-In-Difference (DID) method was used (Wooldridge, 2013). Because the implementation of the policy is an exogenous event, the data can be treated as a natural experiment to determine the direct effect of the change to pictorial warnings on perceived risk of damage and self-perceived health. With the use of DID, analysis is controlled for time-invariant unobserved heterogeneity (Wooldridge, 2012). The law reform has been implemented on the 20th of May 2016 (EUR-Lex, 2015). This was the time cutoff in the analysis. Data from 2012, 2013 and 2015 were used as pre-reform periods (T=0) and data from 2016, 2017 and 2018 were used as post-reform periods (T=1). The DID analysis was performed with two groups, where an individual was allocated to the treatment group (S=1)1_{if that individual smoked in 2015 and 2016. Individuals that did} not smoke in both these years were allocated to the control group (S=0). Respondents that changed their smoking behavior between 2015 and 2016 were left out of the DID analysis to best estimate the effect of the pictorial warnings on continuing smokers compared to continuing non-smokers. These respondents were not included in the analysis because it is unspecified when exact month each individual quitted or started smoking, making it unclear how much these individuals were exposed to the pictorial warnings. Absence of a policy effect was assumed in both groups at baseline (T = 0|S = 0, 1). Since non-smokers do not buy or use cigarettes packages, they are not exposed to the new pictorial warnings. One possible effect that the pictorial warnings can have on non-smokers is to discourage them to start smoking, which is not related to their subjective survival probability and self-perceived health. Therefore, the implementation of pictorial warnings was expected to have no direct effect on the subjective survival probability and the self-perceived health in the control group (T = 1|S = 0). Furthermore, a negative effect of the policy on subjective survival probability was expected in the

1_{Note that no actual treatment is given. Treatment refers to the possibility to come into contact with pictorial}

(10)

10

treatment group (T = 1|S = 1). The population DID effect for the outcome variable is given using the average difference in subjective survival probability between smokers and non-smokers before and after the policy using the following formula

DID = [Ê (Y | T = 1, S = 1) – Ê (Y | T = 1, S = 0)] –

[Ê (Y | T = 0, S = 1) – Ê (Y | T = 0, S = 0)]

The linear estimation of the population DID of the formula above is provided by

Y = α + β · policy

+ γ

+ δ + ε

In this formula, Y represents the outcome variable, for instance the subjective survival probability. The variable ‘policy’ captures the implementation of the pictorial warnings on cigarette packages in 2016. The effect of this implementation is represented by β, which is the estimate of interest. Moreover, γ is the group difference and δ denotes the period difference. The error term is denoted as ε. With the Ordinary Least Squares (OLS) regression performed, standard errors were clustered in order to cope with possible within-observation autocorrelation (Bertrand, Duflo, & Mullainathan, 2004).

Several variables were added to the regression model to control for possible systematic differences between the groups. An additional effect of these control variables is that they reduce the error variance and diminish the standard error of our estimate of interest β (Wooldridge, 2013). The control variables (X) that were added to the extended DID model were age, sex, BMI, number of children, level of education and the natural logarithm of net income. Age was a discrete variable, indicating the age of an individual. The variable BMI (Body Mass Index) was a continuous variable indicating the body mass of an individual. It was calculated using the weight of an individual in kilograms divided by the individual’s squared length in meters. Furthermore, sex was a binary variable, where ‘0’ equals a female and ‘1’ equals a man. The variable number of children was a discrete variable indicating the number of children living in the household, ranging from 0 to 6. Further, the variable ‘level of education’ was a categorical variable about the highest level of education an individual completed in CBS (Statistics Netherlands) categories (1 = primary school, 2 = vmbo (intermediate secondary education), 3 = havo/vwo (higher secondary education), 4 = mbo (intermediate vocational education), 5 = hbo (higher vocational education), 6 = wo (university)). Last, the natural logarithm of net income in Euros was used to control for the effect of wage. The logarithm was applied in order to ensure that outliers and extremes did not have an excessive influence on the results. If an individual did not submit its net income, this net income was estimated based on the personal gross income. The new DID setting for this estimation is given by

DID = [Ê(Y

| T = 1, S = 1, X) - Ê(Y | T = 0, S = 0, X)] –

(11)

11 The resulting linear estimation for these settings is

Y = α + β · policy

+ γ

+ δ

+ Ω · X

+ ε

For the next part of the analysis, both previous mentioned estimations were done using a fixed effects model in order to minimize omitted variable bias due to time-invariant unobserved factors (Verbeek, 2012). The main reason to apply the fixed-effects model is to account for a possible omitted variable bias as a consequence of a selection issue. Individuals that continue smoking all have a time-invariant characteristic which ensured that they do not stop smoking. Therefore, the selected group of continuing smokers is not a complete random sample. To control for this selection issue and additional endogeneity, the fixed-effects model is applied. With the use of a fixed effects model, all unobserved factors that do not change over time but possibly affect the outcome variable are captured (ρi). The

resulting error of the estimation then only consists of time-varying unobserved factors (uit). For these models, presence of the policy over time is denoted by P, where t = 0 determines the years before policy implementation and t = 1 the years after the policy. It was also expected that there is an absence of a policy effect in both groups at baseline (Pi,t=0 = 0|Si = 0, 1), no effect of the policy in the control group after implementation (Pi,t=1 = 1|Si = 0) and a negative effect of the policy after implementation in the treatment group (Pi,t=1 = 1|Si = 1). For this model, the DID settings are given by

DID = [Ê(Y

i,t=1

| P

i,t=1

= 1, S

i

= 1) - Ê(Y

i t=1

| P

i,t=1

= 1, S

i

= 0)] –

[Ê(Y

i,t=0

| P

i,t=0

= 0, S

i

= 1) - Ê(Y

i t=0

| P

i,t=0

= 0, S

i

= 0)]

With the linear estimation

Y

it

= α + β · policy

it

+ γ

i

+ δ

t

+ ρ

i

+ u

it

Last, some time-varying control variables were added to the fixed effects model. These variables were age, BMI and the natural logarithm of net income in Euros. For this last analysis, the equivalent DID settings and linear estimation are

DID = [Ê(Y

i,t=1

| P

i,t=1

= 1, S

i

= 1, X

i

) - Ê(Y

i t=1

| P

i,t=1

= 1, S

i

= 0, X

i

)] –

[Ê(Y

i,t=0

| P

i,t=0

= 0, S

i

= 1, X

i

) - Ê(Y

i t=0

| P

i,t=0

= 0, S

i

= 0, X

i

)],

Y

it

= α + β · policy

it

+ γ

i

+ δ

t

+ Ω · X

it

+ ρ

i

+ u

it

(12)

12

the second control group. A similar DID analysis was applied to these groups as described in the beforementioned paragraph.

Logit regression

The second part of the statistical analysis focused on the direct effect of the policy on smoking. Since smoking is a variable with only two outcomes, a binary choice model is used (Verbeek, 2012). To begin, a logit regression was performed at baseline to determine which characteristics indicate that an individual smokes or not. After this estimation, multiple years of observation were combined and logit regression with clustered standard errors was performed. This logit regression was done with and without the control variables age, sex, BMI, number of children, level of education and the natural logarithm of income. Finally, a logit conditional fixed effects regression model was performed. With the logit conditional fixed-effect regression model, omitted variable bias due to time-invariant unobserved factors was minimized. In this way, the effect of the policy implementation on smoking behavior could be estimated. The following formula was used to estimate the effect of the policy implementation (Chamberlain, 1980)

Pr (y

i

= 1) = (exp^(α

i

+ x’

it

β)/(1 + exp^(α

i

+ x’

it

β))

This logit regression model estimates the effect (coefficient β) of the new policy about pictorial warnings (xi) on the probability that someone smokes (yi = 1). Coefficient αi is the fixed-effect. The calculated coefficients are difficult to interpret. In order to interpret the results of this regression model, the marginal effects were calculated. The average marginal effects at the means were estimated using the predicted probabilities, assuming that the fixed-effect is zero.

Subgroup analysis

(13)

13

Results

Baseline characteristics

Four groups were created in order to perform the analyses. The first groups consisted of a treatment group with individuals who smoked in both 2015 and 2016, and a control group with individuals that did not smoke in these years. The baseline characteristics for these groups are described in Table 1. Furthermore, a second treatment group and a second control group were created to account for a possible delayed effect of the policy. This second treatment group included individuals that smoked in 2015, 2016 and 2017. The second control group consisted of individuals who reported that they did not smoke in all of these years. For these groups, the baseline characteristics are provided in Table 2.

Table 1: Means and Standard Deviations of the baseline characteristics of smokers and non-smokers in 2015 & 2016.

Non-smokers N = 2,023

Smokers N = 392

Variables of interest

Subjective survival probability for the age of 75 (0-10)

7.36 ± 0.037 6.72** ± 0.091

6.54 ± 0.043 5.72** ± 0.11

Self-perceived health (1-5) 3.18 ± 0.017 2.94** ± 0.035

Self-perceived changed health (1-5) 3.01 ± 0.014 3.00 ± 0.031

Demographics

Age 45.88 ± 0.29 46.54 ± 0.64

Males (in %) 43.70 ± 1.10 51.79 ± 2.53

BMI 25.69 ± 0.12 25.47 ± 0.37

Background

Number of children living in the household 1.03 ± 0.26 0.86* ± 0.54

Level of education (0-6) 3.92 ± 0.031 3.40** ± 0.069

(14)

14

Table 2: Means and Standard Deviations of the baseline characteristics of smokers and non-smokers in 2015, 2016 & 2017.

Non-smokers N = 2,005

Smokers N = 285

Variables of interest

7.36 ± 0.038 6.71** ± 0.11

6.54 ± 0.044 5.75** ± 0.13

Self-perceived health (1-5) 3.18 ± 0.017 2.90** ± 0.042

Self-perceived changed health (1-5) 3.01 ± 0.015 3.01 ± 0.034

Demographics

Age 45.90 ± 0.29 47.93 ± 0.72

Males (in %) 43.79 ± 1.11 52.28 ± 2.96

BMI 25.70 ± 0.12 25.51 ± 0.48

Background

Number of children living in the household 1.03 ± 0.26 0.79** ± 0.61

Level of education (0-6) 3.88 ± 0.030 3.35** ± 0.081

Net income 1508.76 ± 24.57 (N = 1,871) 1426.68 ± 67.98 (N = 269) Note: */** indicate significant differences based on the mean differences between the groups, at respectively a 5%/1% alpha level, derived with an independent sample t-test.

Observed trendlines

Before the creation of trendlines, histograms were made to determine if the answers of the respondents in all years were reliable (see Appendix A). Answers were normally distributed, indicating that individuals understood the questions and answered honestly. Then, trendlines were created for the scores on multiple questions of interest. These trendlines were created in order to visualize a possible effect of the policy implementation. First, it was explored if there was a stable difference in trend of scores on the variables of interest between smokers and non-smokers before the implementation of the policy. If a stable difference was present, a continuation of this trend could be expected for later years without any interventions. A possible difference in change of scores after the implementation between smokers and non-smokers could indicate the aimed effect of the policy. In every figure, a designation is added to indicate the year of the baseline and last measurement before the policy implementation. Second, for the graphs, additional groups have been created. Because a part of the smokers intended to stop smoking after the policy implementation, it is interesting to see how that impacted their scores on subjective survival probabilities and self-perceived health. For this reason, a group was created of individuals that stopped smoking after 2015 and a second group for individuals that stopped after 2016. All groups mentioned in the previous sections and the groups of quitters are displayed in one figure for each question. These figures are showed below2. For the questions about subjective survival

2_{Note that in all figures, average scores of non-smokers in 2015 and 2016 are almost similar to the average scores}

(15)

15

probabilities and the question about self-perceived health, smokers scored significantly lower than non-smokers in all years.

Figure 1: Average subjective survival probability scores for the age of 75 years (1-10 scale) over the period of 2012-2018

Figure 1 shows the average scores on the subjective survival probability for the age of 75. As can be seen, the trend before the policy is nearly similar for smokers and non-smokers. After the policy implementation, a decrease in average scores was expected in the treatment group and no change in scores was expected in the control group. Contrary to these expectations, there seems to be a very slight decrease in average score of non-smokers, and an even small increase in average scores of smokers. Furthermore, longer after the implementation, scores for smokers continue to increase slightly, but these years scores for non-smokers also increase. These trendlines, also because changes in average scores are minimal, suggest that the hypothesized effect of the pictorial warnings on the subjective survival probability is absent or even contradictory.

Figure 2: Average subjective survival probability scores for the age of 80 years (1-10 scale) over the period of 2012-2018

(16)

16

implementation of the policy, comparable changes in the scores are present as in the probability to reach the age of 75. The average scores of continuing smokers slightly improve, whereas the average scores of non-smokers slightly drop. After a year, scores of both groups slightly increase. These changes in average scores are again in contrast with the hypothesized effects that the policy would negatively influence the subjective survival probability of smokers.

Figure 3: Average self-perceived health scores (1-5 scale) over the period of 2012-2018 In Figure 3, the average scores on the self-perceived health question are displayed over the years. For these scores, there is no presence of a clear consistent difference in trend between smokers and non-smokers before the policy implementation. Furthermore, the self-perceived health of continuing non-smokers does not seem to be affected by the policy, since there is no drop after 2015. One remarkable part of this graph is the peak in 2017 for individuals that quitted smoking after 2015. Since these individuals stopped smoking after 2015, the peak would be expected in 2016 and not in 2017 and therefore seems not to be related to the policy implementation. Additionally, the groups of quitters were too small, which can be seen in the rather large standard deviations, to draw actual conclusions based on these observations.

(17)

17

2015, the question has been found to yield invalid results. An additional analysis has been performed in order to determine the validity of this question. Since the self-perceived health has been proven to be a valid measure for the overall health of an individual, answers to this question were used as a golden standard (Martikainen et al. 1999). The score of one’s actual self-perceived health of the prior year was compared to one’s score in the subsequent year. If an individual scored higher the subsequent year compared to the year before, one’s relative self-perceived changed health should also indicate this improvement. However, more than two third of the individuals that reported a higher self-perceived health in one year than in the year before, did not report this improvement in their answer on the question about relative self-perceived changed health. Similar deviations were found in individuals that reported a lower score on their self-perceived health compared to the previous year, where on average two third did not indicate this decline in their answer on the question about self-perceived changed health. Thus, this question has been shown to yield non-valid answers and was therefore not included for further analyses.

Additionally, the direct effect of the policy on the quantity of smokers was assessed and used to visualize the effect of the policy on smoking behavior. For each year, the percentage of respondents stating that they smoke was calculated. In 2015, the baseline year, the percentage of smokers in the data sample was 19.24%. This is comparable to the Dutch population, of which 19.30% smoked daily in 2015 (Zantinge E.M., Plasmans M.H.D.,& Zomer, C., 2020). Figure 5 shows the percentages of smokers in 2012 until 2018, with a designation at 2015 indicating the implementation of the policy. This graph shows a steady decline in the percentage of respondents that smoke after the implementation of the policy.

Figure 5: Percentage of respondents smoking over the years

(18)

18 Statistical analysis

Difference in difference

Subjective survival probability was used to determine the perceived risk of damage of smoking in continuing smokers. With the DID analysis, the effect of the policy on the subjective survival probability for the age of 75 was estimated. The effect was first estimated using an OLS regression without control variables for smokers and non-smokers in the years 2015 and 2016. After this regression, a similar model was applied, including the control variables. The results of both OLS regressions are represented in Table 3.

Table 3: OLS DID estimates on the subjective survival probability for the age of 75 for smokers and non-smokers in 2015 and 2016

Without control variables With control variables

Number of observations 13,312 10,910 Number of groups 2,415 2,115 R-squared 0.0167 0.0368 Period difference 0.018 ± 0.025 0.025 ± 0.030 Group difference -0.66** ± 0.092 -0.52** ± 0.099 Policy effect 0.067 ± 0.062 -0.00033 ± 0.071 Age -0.0068* ± 0.0028 Sex -0.18* ± 0.073 BMI -0.012* ± 0.0060 Number of children 0.060* ± 0.028

Natural logarithm of net income 0.12* ± 0.057

Level of education 0.11** ± 0.028

OLS with clustered standard errors on the subjective survival probability for the age of 75 Significance level: ** p < 0.01; * p < 0.05

(19)

19

Table 4: Fixed-effects DID estimates on the subjective survival probability for the age of 75 for smokers and non-smokers in 2015 and 2016

Number of observations 13,312 10,919

Number of groups 2,415 2,117

Overall R-squared 0.0049 0.0038

Period difference 0.011 ± 0.020 -0.051 ± 0.043

Group difference Omitted because of collinearity

Policy effect 0.10* ± 0.050 0.054 ± 0.055

Age 0.019 ± 0.010

BMI -0.0018 ± 0.0018

Natural logarithm of net income 0.055 ± 0.043

Fixed-effects regression on the probability of reaching the age of 75 Significance level: ** p < 0.01; * p < 0.05

The second analyses contained the subjective survival probability for the age of 80 as dependent variable. These results were no difference to the results of the subjective survival probability for the age of 75 (see Appendix B). In Table 5, the results of the OLS regression with self-perceived health as dependent variable are displayed. Table 6 holds the results of the fixed-effects regression for this variable. The results of these models show that, also for self-perceived health, no effect of the policy has been found.

Table 5: OLS DID estimates for self-perceived health for smokers and non-smokers in 2015 and 2016 Without control variables With control variables

Number of observations 13,531 11,112 Number of groups 2,415 2,126 R-squared 0.0109 0.104 Period difference -0.0022 ± 0.010 0.040** ± 0.012 Group difference -0.22** ± 0.035 -0.18** ± 0.036 Policy effect -0.0041 ± 0.023 -0.0081 ± 0.026 Age -0.012** ± 0.0012 Sex 0.081** ± 0.029 BMI -0.015** ± 0.0042 Number of children 0.029** ± 0.011

Natural logarithm of net income 0.095** ± 0.021

(20)

20

Table 6: Fixed-effects DID estimates on the self-perceived health for smokers and non-smokers in 2015 and 2016

Period difference 0.00086 ± 0.0085 0.0014 ± 0.018

Policy effect 0.0036 ± 0.021 0.0076 ± 0.023

Age -0.0020 ± 0.0043

BMI -0.0073** ± 0.0017

Natural logarithm of net income 0.059** ± 0.017

Fixed-effects regression on the self-perceived health Significance level: ** p < 0.01; * p < 0.05

Moreover, the analyses have been performed for the second groups of smokers and non-smokers to account for a possible delayed effect of the policy. These analyses gave almost identical results as the analyses with the first groups, showing no effects of the policy on any of the dependent variables (see Appendix C).

Logit regression

A logit regression was performed to get an indication which individual characteristics determine if one smokes at baseline or not. Mean marginal effects of the estimates were calculated with the results of this regression and are listed in table 7. As can be seen, multiple variables can be associated with an increased or decreased probability to smoke. Being a male is associated with an increase in the likelihood to smoke, whereas a higher level of education, more children living in the household, and a higher BMI are associated with an increase in the likelihood that an individual smokes.

Table 7: Average marginal effect estimates of characteristics indicating if an individual smokes No control variables Number of observations 2,081 LR chi^2 52.73 Age -0.0003 ± 0.00076 Sex 0.049** ± 0.019 BMI -0.0039* ± 0.0019 Number of children -0.023** ± 0.008

Natural logarithm of net income -0.16 ± 0.015 Level of education -0.023** ± 0.0067 Marginal effects after logit regression

Significance level: ** p < 0.01; * p < 0.05

(21)

21

subsequent performed fixed-effects model (Appendix D). During this logit fixed-effects regression, 12,029 observations (2,152 groups) were dropped because these individuals did not change their smoking behavior over the years. With the results of the regression, the average marginal effect of the policy was calculated. The coefficient can be found in Table 8. The coefficient is shown to be negative and significant. Therefore, the policy implementation can be associated with a reduction in the likelihood that an individual smokes.

Table 8: Average marginal effect estimate of the policy effect after logit regression with fixed effects No control variables

Number of observations 2,273

Number of groups 408

LR chi^2 111.41

Policy -0.22** ± 0.019

Marginal effects of the policy after logit regression with fixed effects Significance level: ** p < 0.01; * p < 0.05

Subgroup analysis

The beforementioned analyses were repeated based on sex and age to determine if there was an effect of the policy on one of these subgroups. First, the DID analyses were done based on age, performing the analyses for adults younger than fifty and adults of fifty years old and older. Again, for both groups, no effect of the policy on the subjective survival probabilities and self-perceived health was found. Also, when the analyses were performed based on sex, the policy was estimated to have no effect on the subjective survival probabilities and self-perceived health in men or women. Since these results were no different than the results of the initial analyses, full tables are not reported.

Second, the logit conditional fixed-effects model was performed based on age and sex. Also for this analysis, no different results were found compared to the initial analysis and between the groups. What can be stated after these findings is that the policy can be associated with a reduction in the likelihood that an individual smokes, as well for males as for females and for adults younger than fifty and adults older than fifty. These results were based on the effect on the policy on the first three years after the implementation. However, an extra analysis showed that from 2015 to 2016, that women were more likely to quit smoking than men. In the years before the policy, on average fifty percent of the quitters was male and fifty percent was female. In the year direct after the policy implementation, almost sixty percent of the quitters was female and forty percent was male. Logit conditional fixed-effects models3 for data from 2015 and 2016 also show a higher estimated policy effect in females than in males. The results of these models can be found in Table 9. Based on these results, it seems that the

3_{OLS models were performed in before the fixed-effects analysis and showed similar results as the fixed-effects}

(22)

22

policy can be associated with a higher reduction in the likelihood that a women smokes direct after the implementation of the policy than that for male.

Table 9: Average marginal effect estimate of the policy effect after logit regression with fixed effects based on sex

Model based on females Model based on males

Number of observations 156 134

Number of groups 78 67

LR chi^2 19.33 4.36

Policy -0.24** ± 0.049 -0.13* ± 0.059

Marginal effects of the policy after logit regression with fixed effects based on sex Significance level: *** p < 0.01; ** p < 0.05

Discussion

In this study, the effect of the implementation of pictorial warnings on the perceived risk of damage has been investigated within continuing smokers. This was done by comparing scores on questions about the subjective survival probability and the self-perceived health between and within smokers and non-smokers. The results show that scores on the perceived risk of damage and self-perceived health in continuing smokers were not affected by the policy. However, results from the logit regression analyses show that the policy implementation can be associated with a reduction in the likelihood that an individual smokes. Furthermore, right after the implementation of pictorial warnings, the likelihood that an individual smoked decreased more in women than in men. This difference in effect was not present anymore longer after the policy implementation or between age groups.

(23)

23

smoking, but denies his own vulnerability for the risks of smoking (Mayerl et al. 2018). Smokers have been shown to often deny risks, stating that they smoke too few cigarettes or for a too short period of time to be susceptible for negative health effects (Peretti-Watel et al. 2007). Brewer et al. (2019) also addressed the role of the addicting effect of nicotine within risk denial. Some smokers are too addicted to smoking to stop, even though they know smoking is not without risk. Because individuals cannot cope with this inner inconsistency of performing a behavior which they know is harmful but cannot stop, they change their perception towards the behavior in order to eliminate this inner inconsistency (Festinger, L. 1957). Improving information about the risks of smoking may not be the best way to help current continuing smokers to quit smoking, since this groups seems to deny the newly introduced information about the negative and possible life shortening health consequences of smoking. Further research should which future policy implementations, like plain packaging or increasing the price, work best to stimulate these continuing smokers to quit smoking.

Another possible explanation is that the subjective survival probability is not consistent enough to simulate the effect of the policy implementation. The subjective survival probability has been shown to be a good predictor of actual survival and to change with respect to new information (Hurd and McGarry, 2002). However, some participants of the used LISS household survey thought questions about the subjective survival probabilities were difficult to answer. The host of the survey, the Tilburg University, emphasized the importance and relevance of these questions on January 2019 in order to improve quality of the collected data. Data from before 2019 could thus contain answers from respondents that found the question too difficult and were not able to correctly respond to these questions. The fact that there was no stable difference in trend between smokers and non-smokers could also be a consequence of noise in the data as a result of respondents that did not understand the questions.

(24)

24

The results of the logit regression analysis show the direct effect of the policy on the likelihood that an individual smokes. This statistical analysis showed that the implementation of the policy can be associated with a reduction in this probability. These results can be supported with findings from previous studies. Multiple studies have shown that the introduction of pictorial warnings on cigarette packages has stimulated the amount of initiated and successful quit attempts (Azagba, S. & Sharaf, M. F. 2013; Brewer et al. 2016; Noar et al. 2016, Noar et al. 2017). Moreover, these studies showed that non-smokers have a lower intention to start smoking due to the pictorial warnings. Another study that looked at the impact of the pictorial warnings on Dutch smokers also found that Dutch smokers initiated more quit attempts due to the pictorial warnings (Van Mourik et al. 2019). With the combination of previous studies and the current results, there has been shown that the implementation of the pictorial warnings can be associated with a reduction in the likelihood that an individual will smoke. Moreover, right after implementation, the pictorial warnings were associated to reduce the likelihood that a woman smoked farther than the likelihood that a man smoked. Mannocci et al. 2012 also showed that women were more impressed by the pictorial warnings and were more likely to change cigarette brand to avoid the pictorial warnings. Thus, women seem to be more sensitive to the pictorial warnings on short notice. However, this difference in effect was only present right after implementation and disappeared in the subsequent years, resulting in a similar average effect of the pictorial warnings in men and women in the first three years after implementation.

A limitation of this study was that data from 2014 was missing. Therefore, periods between collection of data were not consistent and collection of 2015 had to be performed earlier in the year. It is possible that this change in collection period has affected scores of respondents. People may submit a lower score on their self-perceived health in November, the month of normal collection, compared to the month July, because of differences in weather and temperatures. If data was collected yearly in the same month, these factors would not have been able to influence the data and results of this study. Also, compared to other studies, no direct questions about the perceived risk of damage or other smoking related health effects were asked to the respondents. Since these questions tend to improve the visibility of the effect of the policy implementation, absence of questions specified to perceived risk of damage of smoking can be seen as a limitation of the current study as well. Future studies are recommended to use data from specific tobacco related surveys, for instance the International Tobacco Control survey. On the other hand, a strength of the current study is the use of supplementary variables in order to map the effects of the pictorial warnings. With the combination of questions about the subjective survival probability and self-perceived health, effects of the implementation of pictorial warnings on the long term as well as on the short term perception towards developing negative health in continuing smokers were studied.

(25)

25

to determine which actions best discourage current continuing smokers to quit smoking. One of these interventions is plain packaging, which may improve the warning salience of the messages on cigarette packages. The effectiveness of this intervention on quit attempts should be studied, as well as its effect on the perceived risk of damage in continuing smokers. Besides the change to plain packaging, prices for a pack of cigarettes will rise in the Netherlands in 2020. Studies should look at the effect of this increase in prices on smoking behaviors. Furthermore, the two interventions should be looked into simultaneously in one study to find which intervention had more influence on smoking behavior and perceived risk of damage and possibly determine which strategy works best to discourage smoking behavior. Lastly, the effects of COVID-19 on smoking behavior could be studied. Since COVID-19 can affect the lungs and lead to respiratory diseases (Singhal, 2020), smokers could have become more aware of the risks of smoking and improved their perceived risk of damage from smoking behavior. It is interesting to study if COVID-19 has affected smoking behavior.

Finally, it was not possible to perform any more analyses than performed using this data. The chain reaction of events due to one exogenous policy implementation is so big that was not possible to determine all these effects with the present data.

Conclusion

(26)

26

References

Aryal, U. R., & Bhatta, D. N. (2015). Perceived benefits and health risks of cigarette smoking among young adults: Insights from a cross-sectional study. Tobacco Induced Diseases, 13(1), 22-9. eCollection 2015. doi:10.1186/s12971-015-0044-9 [doi]

Azagba, S., & Sharaf, M. F. (2013). The effect of graphic cigarette warning labels on smoking behavior: Evidence from the canadian experience. Nicotine & Tobacco Research : Official Journal of the Society for Research on Nicotine and Tobacco, 15(3), 708-717. doi:10.1093/ntr/nts194 [doi]

Bertrand, M., Duflo, E., & Mullainathan, S. (2004). How much should we trust differences-in-differences estimates? The Quarterly Journal of Economics, 119(1), 249-275. doi:10.1162/003355304772839588

Brewer, N. T., Hall, M. G., Noar, S. M., Parada, H., Stein-Seroussi, A., Bach, L. E., . . . Ribisl, K. M. (2016). Effect of pictorial cigarette pack warnings on changes in smoking behavior: A randomized clinical trial. JAMA Internal Medicine, 176(7), 905-912. doi:10.1001/jamainternmed.2016.2621 [doi]

Brewer, N. T., Parada, H., Hall, M. G., Boynton, M. H., Noar, S. M., & Ribisl, K. M. (2019). Understanding why pictorial cigarette pack warnings increase quit attempts. Annals of Behavioral

Medicine, 53(3), 232-243. doi:10.1093/abm/kay032

Canadian Cancer Society. (2018). Cigarette package health warnings: International status report - sixth edition

Centraal Bureau voor de Statistiek. (2019). Lichte daling aantal rokers onder volwassenen. Retrieved from https://www.cbs.nl/nl-nl/nieuws/2019/12/lichte-daling-aantal-rokers-onder-volwassenen

Chamberlain, G. (1980). Analysis of covariance with qualitative data. The Review of economic

studies, 47(1), 225-238. Retrieved from http://www.econis.eu/PPNSET?PPN=384651526

Chung-Hall, J., Fong, G. T., Meng, G., Yan, M., Tabuchi, T., Yoshimi, I., Mochizuki, Y., Craig, L. V., Ouimet, J., & Quah, A. (2020). Effectiveness of Text-Only Cigarette Health Warnings in Japan: Findings from the 2018 International Tobacco Control (ITC) Japan Survey. International journal of

environmental research and public health, 17(3), 952. https://doi.org/10.3390/ijerph17030952

de Bresser, J. (2019). Measuring subjective survival expectations – do response scales matter? Journal of Economic Behavior and Organization, 165, 136-156. doi:10.1016/j.jebo.2019.06.018

EUR-Lex. (2015). Directive 2014/40/EU of the european parliament and of the council of 3 april 2014 on the approximation of the laws, regulations and administrative provisions of the member states concerning the manufacture, presentation and sale of tobacco and related products and repealing directive 2001/37/EC (text with EEA relevance). Retrieved from https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:02014L0040-20150106

Festinger, L. (1957). A theory of cognitive dissonance. Stanford University Press,

Hall, M. G., Sheeran, P., Noar, S. M., Boynton, M. H., Ribisl, K. M., Parada Jr, H., . . . Brewer, N. T. (2018). Negative affect, message reactance and perceived risk: How do pictorial cigarette pack warnings change quit intentions? Tobacco Control, 27(e2), e136-e142. doi:10.1136/tobaccocontrol-2017-053972

Hammond, D. (2011). Health warning messages on tobacco products: A review. Tobacco Control, 20(5), 327-337. doi:10.1136/tc.2010.037630

(27)

27

Hurd, M. D., McGarry, K. (2002). The predictive validity of subjective probabilities of survival Oxford University Press on behalf of the Royal Economic Society. Retrieved from https://www.jstor.org/stable/798539

Institute of Medicine. (2015). Public health implications of raising the minimum age of legal access to tobacco products. Chapter 4. Retrieved from https://www.ncbi.nlm.nih.gov/books/NBK310413/

Israel T. Agaku, Filippos T. Filippidis, & Constantine I. Vardavas. (2015). Effectiveness of text versus pictorial health warning labels and predictors of support for plain packaging of tobacco products within the european union. European Addiction Research, 21(1), 47-52. doi:10.1159/000366019

ITC Project. (2015). ITC netherlands national report, Findings from the wave 1 to 8 surveys (2008-2014). Waterloo, Ontario, Canada: University of Waterloo

Jha, P., & Peto, R. (2014). Global effects of smoking, of quitting, and of taxing tobacco. The New England Journal of Medicine, 370(1), 60-68. doi:10.1056/NEJMra1308383

Mannocci, A., Antici, D., Boccia, A., & La Torre, G. (2012). Impact of cigarette packages warning labels in relation to tobacco-smoking dependence and motivation to quit. [Impatto delle avvertenze riportate sui pacchetti di sigarette in funzione della dipendenza dal fumo di tabacco e del desiderio di smettere in un campione di fumatori] Epidemiologia E Prevenzione, 36(2), 100-107. doi:1306 [pii]

Mayerl, H., Stolz, E., & Freidl, W. (2018). Responses to textual and pictorial cigarette pack health warnings: Findings from an exploratory cross-sectional survey study in austria. BMC Public Health, 18(1), 442. doi:10.1186/s12889-018-5342-8

Ministerie van Volksgezondheid, Welzijn en Sport. (2018). Nationaal preventieakkoord : Een gezonder nederland. (). Retrieved from https://www.captise.nl/Zorg-Jeugd/DigiBib/Details?entryId=18060

Nagelhout, G. E., Willemsen, M. C., de Vries, H., Mons, U., Hitchman, S. C., Kunst, A. E., Guignard, R., Siahpush, M., Yong, H. H., van den Putte, B., Fong, G. T., & Thrasher, J. F. (2016). Educational differences in the impact of pictorial cigarette warning labels on smokers: findings from the International Tobacco Control (ITC) Europe surveys. Tobacco control, 25(3), 325–332. https://doi.org/10.1136/tobaccocontrol-2014-051971

Nationaal Expertisecentrum Tabaksontmoediging. (2016). Factsheet: Waarschuwende afbeeldingen op tabaksverpakkingen Retrieved from https://natlib-primo.hosted.exlibrisgroup.com/primo-explore/search?query=any,contains,993007353502836&tab=catalogue&search_scope=NLNZ&vid=N LNZ&offset=0

Nationaal Expertisecentrum Tabaksontmoediging. (2018). Kerncijfers roken 2017

Noar, S. M., Francis, D. B., Bridges, C., Sontag, J. M., Brewer, N. T., & Ribisl, K. M. (2017). Effects of strengthening cigarette pack warnings on attention and message processing: A systematic review. Journalism & Mass Communication Quarterly, 94(2), 416-442. doi:10.1177/1077699016674188 [doi]

Noar, S. M., Hall, M. G., Francis, D. B., Ribisl, K. M., Pepper, J. K., & Brewer, N. T. (2016).

Pictorial cigarette pack warnings: A meta-analysis of experimental studies of experimental studies

doi:10.1136/tobaccocontrol-2014-051978

Peretti-Watel, P., Constance, J., Guilbert, P., Gautier, A., Beck, F., & Moatti, J. (2007). Smoking too few cigarettes to be at risk? smokers’ perceptions of risk and risk denial, a french survey. Tobacco Control, 16(5), 351-356. doi:10.1136/tc.2007.020362

(28)

28

Pötschke-Langer, M., Schotte, K., & Szilagyi, T. (2015). The WHO framework convention on tobacco control. The tobacco epidemic (pp. 149-157). Basel, Switzerland: S. Karger AG. doi:10.1159/000369441 Retrieved from https://www.karger.com/Article/FullText/369441

Putte, B. v. d., Thrasher, J. F., Willemsen, M. C., Vries, H. d., Guignard, R., Siahpush, M., . . . Fong, G. T. (2016). Educational differences in the impact of pictorial cigarette warning labels on smokers: Findings from the international tobacco control. Tobacco Control, 25(3), 325.

Rappange, D., Exel, J., & Brouwer, W. (2017). A short note on measuring subjective life expectancy: Survival probabilities versus point estimates. The European Journal of Health Economics, 18(1), 7-12. doi:10.1007/s10198-015-0754-1

Rijksinstituut voor Volksgezondheid en Milieu. (2018). Sterfte door roken. Retrieved from https://www.volksgezondheidenzorg.info/onderwerp/roken/cijfers-context/gevolgen#node-sterfte-door-roken

Rijksoverheid. (2016). Waar moet ik op letten bij de nieuwe regels voor tabak- en rookwaren? Retrieved from https://www.rijksoverheid.nl/onderwerpen/roken/vraag-en-antwoord/waar-moet-ik-op-letten-bij-de-nieuwe-regels-voor-tabak--en-rookwaren

Rijksoverheid. (2018). Maatregelen overheid om roken te ontmoedigen. Retrieved from https://www.rijksoverheid.nl/onderwerpen/roken/roken-ontmoedigen

Rius, C., Fernandez, E., Schiaffino, A., Borràs, J. M., & Rodríguez-Artalejo, F. (2004). Self perceived health and smoking in adolescents. Journal of Epidemiology and Community Health, 58(8), 698-699. doi:10.1136/jech.2003.008516

Sambrook Research International. (2009, May 18). A review of the science base to support the development of health warnings for tobacco packages. The Times

Shields, M., & Shooshtari, S. (2001). Determinants of self-perceived health. Health Reports, 13(1), 35-52.

Singhal, T. (2020). A review of coronavirus disease-2019 (COVID-19). Indian Journal of

Pediatrics, 87(4), 281-286. doi:10.1007/s12098-020-03263-6

Strecher, V. J., Kreuter, M. W., & Kobrin, S. C. (1995). Do cigarette smokers have unrealistic perceptions of their heart attack, cancer, and stroke risks? Journal of Behavioral Medicine, 18(1), , 45– 54. Retrieved from https://doi.org/10.1007/BF01857704

Szklo, A. S., & Coutinho, E. S. F. (2009). Vulnerability and self-perceived health status among light and heavy smokers: The relationship to short-term fear appeal tobacco control messages. Cadernos De Saude Publica, 25(7), 1534-1542. doi:10.1590/S0102-311X2009000700011

Tannenbaum, M. B., Hepler, J., Zimmerman, R. S., Saul, L., Jacobs, S., Wilson, K., & Albarracín, D. (2015). Appealing to fear: A meta-analysis of fear appeal effectiveness and theories. Psychological Bulletin, 141(6), 1178-1204. doi:10.1037/a0039729

Trimbos Instituut. (2019). Cijfers roken. Retrieved from https://www.trimbos.nl/kennis/cijfers/cijfers-roken

Trimbos-instituut. (2016). Roken onder volwassenen en jongeren in nederland. kerncijfers 2015. Factsheet, Retrieved from https://natlib-primo.hosted.exlibrisgroup.com/primo-explore/search?query=any,contains,9916005903502836&tab=catalogue&search_scope=NLNZ&vid= NLNZ&offset=0

(29)

29

control (ITC) netherlands surveys. International Journal of Environmental Research and Public Health, 16(21), 4260. doi:10.3390/ijerph16214260

Verbeek, M. (2012). A guide to modern econometrics. United Kingdom: John Wiley & Sons Inc. Winefield, H. R. (2006). Behavioural science learning modules: Encourage people to stop smoking. Retrieved from https://trove.nla.gov.au/

Woelbert, E., & d’Hombres, B. (2019). Pictorial health warnings and wear-out effects: Evidence from a web experiment in 10 european countries. Tobacco Control, 28(e1), e71-e76. doi:10.1136/tobaccocontrol-2018-054402

(30)

30

Appendices

Appendix A

Frequency distribution of scores on the four questions of interest, in histograms

0 1000 2000 3000 4000 5000 0 1 2 3 4 5 6 7 8 9 10 Fr equ ency

Subjective survival probability for the age of 75

0 500 1000 1500 2000 2500 3000 3500 4000 0 1 2 3 4 5 6 7 8 9 10 Fr equ ency

Subjective survival probability for the age of 80

0 2000 4000 6000 8000 10000

Poor Moderate Good Very good Excellent

Fr equ ency General health 0 2000 4000 6000 8000 10000 12000

Considerably poorer Somewhat poorer The same Somewhat better Considerably better

Fr

equ

ency

(31)

31 Appendix B

Additional tables for smokers and non-smokers in 2015 and 2016.

Table B1: OLS DID estimates on the subjective survival probability for the age of 80 for smokers and non-smokers in 2015 and 2016

Number of observations 13,524 11,112 Number of groups 2,415 2,126 R-squared 0.0188 0.0398 Period difference 0.020 ± 0.027 0.045 ± 0.033 Group difference -0.81** ± 0.11 -0.68** ± 0.11 Policy effect 0.086 ± 0.070 0.018 ± 0.078 Age -0.0096** ± 0.0032 Sex -0.30** ± 0.086 BMI -0.012 ± 0.0064 Number of children 0.074* ± 0.032

OLS with clustered standard errors on the subjective survival probability for the age of 80 Significance level: ** p < 0.01; * p < 0.05

Table B2: Fixed-effects DID estimates on the subjective survival probability for the age of 80 for smokers and non-smokers in 2015 and 2016

Policy effect 0.12* ± 0.054 0.082 ± 0.060

Age 0.025* ± 0.027

BMI -0.0047 ± 0.0044

(32)

32 Appendix C

Tables with the results of the performed analyses for smokers and non-smokers in 2015, 2016 and 2017

Table C1: OLS DID estimates on the subjective survival probability for the age of 75 for smokers and non-smokers in 2015, 2016 and 2017

Number of observations 12,689 10,382 Number of groups 2,290 2,001 R-squared 0.0149 0.0305 Period difference 0.017 ± 0.025 0.020 ± 0.030 Group difference -0.70** ± 0.11 -0.54** ± 0.11 Policy effect 0.11 ± 0.071 0.044 ± 0.081 Age -0.0060* ± 0.0029 Sex -0.17* ± 0.075 BMI -0.011 ± 0.0058 Number of children 0.051 ± 0.029

Natural logarithm of net income 0.14* ± 0.060

OLS with clustered standard errors on the subjective survival probability for the age of 75 Significance level: ** p < 0.01;* p < 0.05

Table C2: Fixed-effects DID estimates on the subjective survival probability for the age of 75 for smokers and non-smokers in 2015, 2016 and 2017

Policy effect 0.13* ± 0.056 0.075 ± 0.061

Age 0.019 ± 0.010

BMI -0.0012 ± 0.0041