Investigating the relationship between smoking and ...

Download PDF
4 downloads 0 Views 919KB Size Report
Comment
Using data from the 2014 Health Survey for England (HSE), we also address the .... impact of tobacco control policies and smoking bans on subjective wellbeing ...
Investigating the relationship between smoking and subjective welfare

Sefa Awaworyi Churchill School of Economics, Finance & Marketing RMIT University Email: [email protected]

Lisa Farrell School of Economics, Finance & Marketing RMIT University Email: [email protected]

Abstract Using data from the Health Survey for England, we examine the effect of smoking behavior and smoking addiction (considering the frequency and intensity of smoking) on happiness and depression, two well-known measures of subjective wellbeing. We find that smoking and smoking addiction are associated with lower levels of happiness and higher levels of depression. This finding is robust to alternative ways of measuring smoking behavior, as well as to the methodological approach to estimation that addresses endogeneity. Specifically, comparing the Ordinary Least Squares (OLS) results with the two-stage least square (2SLS) results, we find that although the OLS results overstate the effects of smoking status and addiction on wellbeing, the emerging conclusion of a negative effect of smoking behavior is still valid in the case of 2SLS. Keywords: smoking; health; subjective wellbeing; addiction; depression; happiness JEL codes: I31, I12

1

1. Introduction The costs of smoking have been well-documented for many western economies. In the United States, the cost is estimated at more than $300 billion each year (Xu et al., 2014), of which nearly $170 billion accounts for direct medical care for adults and more than $156 billion accounts for lost productivity (of which $5.6 billion is attributable to lost productivity resulting from second hand smoke exposure). A relatively large body of literature thus examines the impact of smoking (see, for example, Akl et al., 2010; Lakier, 1992; Lyvers et al., 2009; Rooke et al., 2013). One strand of this literature examines the impact of smoking on subjective wellbeing (see, for example, Kaliterna Lipovcan et al., 2013; Shahab and West, 2012), given that wellbeing has become well-established as an important measure of quality of life (Diener and Chan, 2011; Okun et al., 1984).1

Our study aims to contribute to the literature that examines the effect of addictive behaviors on subjective wellbeing. Specifically, we seek to examine the effect of smoking behavior on happiness and depression using two common measures of subjective wellbeing. Understanding the implications of smoking is a major issue among public health and policy debates. It motivates various policies including government tariff decisions and bans on smoking (Barone-Adesi et al., 2006; Cesaroni et al., 2008; Thomas et al., 2008). The public health debate, however, has focused predominantly on the physical health implications of smoking (Akl et al., 2010; Lakier, 1992; Rooke et al., 2013).

The primary objective of our paper is to broaden our knowledge of the effects of smoking behavior on subjective wellbeing using English data. Specifically, we address the question: what is the effect of smoking and associated addictive behaviors on wellbeing in England? Using data from the 2014 Health Survey for England (HSE), we also address the endogenous nature of the relationship between smoking behavior and wellbeing, which has not received a great deal of attention in the existing literature. Smoking and wellbeing are, to a large extent, endogenous as some individuals smoke because they are stressed and less happy. Put differently, reverse causality may be an issue here as it is likely that happier people may be more inclined to give up or avoid smoking, while less happy people are more likely to take up 1

In this study, our measure of subjective wellbeing is happiness drawn from individuals’ evaluation of their happiness. Thus, we use the term subjective wellbeing and happiness interchangeably. Further, our measures of smoking primarily capture addictive behaviors rather than addiction itself. Accordingly, proxies for addiction used in this study primarily capture various addictive behaviors.

2

smoking. Without controlling for the endogenous nature of this relationship, the direction of the effect of smoking behavior and the reported effect sizes may be biased. Thus, we address the endogenous nature of the relationship by adopting the novel Lewbel (2012) instrumental variable (IV) technique. We contribute to the existing literature by using new, high quality data and by thoroughly addressing issues of endogeneity, to provide a new perspective on the effects of smoking on subjective wellbeing. We use a wider range of measures of smoking behaviors than have been used by previous studies in this area. More importantly, our study further contributes to the literature by exploring the role of smoking addiction on subjective wellbeing. Although our data does not contain a clinically defined addiction scale, we adopt measures of smoking intensity and frequency as proxies for levels of addiction and examine the effect of smoking and tobacco dependence on wellbeing. When attempting to measure tobacco addiction, a number of existing scales could be utilized, such as the Fagerstrom Test for Nicotine Dependence (Heatherton et al., 1991). Unfortunately, such an instrument is not available in our data; however, we consider a number of measures related to the frequency of tobacco consumption and the quantity/intensity of tobacco consumption as proxies for tobacco dependence in our analysis. Drawing on measures of individual self-reported happiness and depression, our results suggest that smoking status and smoking addiction negatively influence happiness and increase depression. The remainder of the paper is structured as follows. Section 2 provides a brief overview of the existing literature. Section 3 presents a description of the data and variables, as well as an overview of the empirical strategy for the analysis. Section 4 presents results and examines the robustness of these results. Section 5 presents a brief discussion and conclusion.

2. Overview of existing literature and hypothesis Existing literature has investigated the relationship between smoking and subjective wellbeing. Although many studies show that smoking has significant physical health implications and can ultimately lead to death (see, for example, Akl et al., 2010; Lakier, 1992; Lyvers et al., 2009; Rooke et al., 2013; Weinhold and Chaloupka, 2016), smokers nevertheless continue to smoke. This implies that, from a rational agent’s perspective, they gain utility from the act of smoking (despite the potential negative health effects). Becker and

3

Murphy (1988), in their theory of rational addiction, present a model where smokers behave rationally given their level (stock) of addiction.

Alternative rationalizations of smoking behaviors exist; for example, it is often argued that the use of tobacco can help regulate stress and mood, and thus in the short term, smoking could deal with stress, depressive symptoms and promote happiness (Weinhold and Chaloupka, 2016). That is, smoking may allow individuals to deal with stress thus increasing their ability to socialize and elevate overall life satisfaction. However, in the long term, this may not be the case, as smoking could lead to various psychological and quality of life issues.

Behavioral models for understanding cigarette consumption also exist. For example, in an extension of the rational addiction framework, Orphanider and Zervos (1995) include learning and regret, whereby an individual’s subjective beliefs about their potential to become addicted play a significant role in addictive behaviors, and experimentation is important for informing that subjective belief.

In the broader addiction literature, it is common to find explanations for smoking that relate to loss of control due to the effects of nicotine on the human body, which result in cravings. In such instances, the individual can no longer be assumed to make optimal consumption decisions and this can lead to negative outcomes in terms of subjective wellbeing. Loss of control can lead to time inconsistent preferences (Laux, 2000) and the introduction of selfcontrol mechanisms to mitigate these factors (Gruber and Koszeqi, 2001).

A more recent, fully behavioral, approach to understanding addictive consumption can be found in the work of Bernheim and Rangel (2004). Here, the ideas of mistaken consumption, environmental triggers and addicts attempting to self-regulate are central to the theoretical framework. A common theme in the departures from rational behavior is the inclusion of a set of observed behaviors into a formal optimization framework. The difficulty then becomes knowing which key behaviors to consider, and each author has presented arguments for their choice of behaviors to be modeled. The models either predict that once we account for addictive behaviors, smoking is utility generating, or that behavioral factors imply that even though smoking does not generate utility, individuals nevertheless may continue to smoke. Hence overall, the theoretical literature has focused on understanding smoking behaviors but remains inconclusive about whether smoking decreases or increases subjective wellbeing, 4

thus making the determination of the relationship between smoking and wellbeing an empirical task (as noted by Weinhold and Chaloupka, 2016).

Empirically, the association between smoking and wellbeing has not received a great deal of attention in the existing literature. The literature on the effects of smoking focuses mainly on other outcomes and often comes to the conclusion of a negative effect. For instance, studies have shown an association between smoking and psychiatric issues (Glassman, 1993; Leonard et al., 2001); smoking and anxiety (Beckham, 1999); smoking and general health (Akl et al., 2010; Lakier, 1992; Rooke et al., 2013); and smoking and psychological health (Choi et al., 1997; Lyvers et al., 2009). These studies differ significantly from our study in various ways. For instance, Beckham (1999) focuses only on smoking among veterans, although their study examines anxiety, a phenomenon similar to depression. Further, their study focuses on a clinical sample of veterans who suffer from post-traumatic stress disorder. Other studies, such as those mentioned above, focus on outcome variables that are not measures of subjective wellbeing. However, at the aggregate level, McCann (2010) shows a negative relationship between smoking prevalence and wellbeing using state level variation in smoking prevalence. These results conflict with findings from Chang et al. (2016), which suggest that smoking makes people happy in a sample of five countries using country-level data on happiness and per capita cigarette consumption.

At the individual level, relatively little work has been conducted to understand the effects of smoking on people’s happiness and subjective wellbeing. Related literature examines the impact of tobacco control policies and smoking bans on subjective wellbeing (see, for example, Hinks and Katsaros, 2012; Odermatt and Stutzer, 2015) rather than the direct effects of smoking and addiction to smoking. The few studies that have examined the possible link between smoking and wellbeing are faced with issues of endogeneity (see, for example, Kaliterna Lipovcan et al., 2013; Shahab and West, 2012). For instance, Shahab and West (2012), using a sample of the British population, provide evidence to suggest that smokers and recent ex-smokers report lower levels of life satisfaction compared to ex-smokers who have abstained from smoking for a period beyond 12 months. However, consistent with other studies, this study suffers from an omitted variable bias, and more importantly, endogeneity. One recent study that controls for endogeneity is that of Weinhold and Chaloupka (2016), who use changes in smoking-related policy over time as exogenous variation from which to identify the relationship between smoking status and subjective wellbeing, using a rich Dutch 5

longitudinal dataset. The focus of their study is the difference in subjective wellbeing according to smoking status (i.e., never smoked, ex-smokers and current smokers). We go beyond this to use a comprehensive set of measures of smoking behavior to examine how smoking affects wellbeing.

In summary, the theoretical literature on the effects of smoking remains inconclusive and suggests that a priori, the effect of smoking on wellbeing could be positive or negative. However, given the large body of empirical literature that reports negative effects of smoking and smoking-related behaviors, and the theoretical discussions that support a negative effect of smoking (see, for example, Becker and Murphy, 1988), we formulate the following hypothesis: H1: Smoking-related behaviors negatively affect wellbeing. 3. Data and Empirical Strategy Our data comes from the latest release of the HSE, which was conducted in 2014. The HSE is a series of annual surveys that started in 1991 and since then has presented data about changes in the lifestyles and health status of people living in England. The series was designed to monitor trends in the nation’s health, to estimate the proportion of people in England who have specified health conditions, and to estimate the prevalence of risk factors associated with these conditions. The surveys provide regular information, which cannot be obtained from other sources, on a range of aspects concerning the public’s health. Data from the HSE are regarded as reliable and have produced several publications (see, for example, Colhoun et al., 1998; Jarvis and Feyerabend, 2015; Mindell et al., 2012; Oyebode and Mindell, 2013; Primatesta et al., 2001). 2 Overall, 2014 HSE interviewed 10,080 respondents resident in England including 8,077 adults and 2,003 children. However, accounting for missing observations (item non-response), our largest estimation sample for the regression analysis includes data on 5,827 respondents. 3.1. Dependent variables: Happiness and depression

2

For details on the HSE data, see https://data.gov.uk/dataset/health_survey_for_england We focus on the 2014 dataset given that it is the latest version at the time this research was conducted. Additionally, some previous waves of the HSE did not include a measure of subjective happiness, which is crucial for our analysis.

6

The dependent variables used in this study are measures of subjective wellbeing capturing happiness and depression. For our measure of happiness, the HSE provides information on respondents’ happiness through their answers to the question: “Have you recently been feeling reasonably happy, all things considered?”. We adopt a four-point scale to measure happiness, such that 1 means “much less than usual”; 2 means “less so than usual”; 3 means “about same as usual”; and 4 means “more so than usual”.3 For depression, the subjective wellbeing question in the dataset asks respondents, “Have you been feeling unhappy and depressed?”. Responses are coded such that 1 means “not at all”; 2 means “no more than usual”; 3 means “rather more than usual”; and 4 means “much more than usual”. Although this is a single item measure of depression, this item is identical to one of the items included in the Kessler Psychological Distress Scale (K10) (Kessler et al., 2002), the General Health Questionnaire (GHQ-12) (Goldberg, 1978) and the Patient Health Questionnaire (PHQ-9) (Spitzer et al., 1999), as well as other commonly used multi-item depression and anxiety scales. The use of such single item subjective wellbeing measures is consistent with the existing literature, although in some cases seven-point or 10-point scales are adopted (see, for example, Awaworyi Churchill and Mishra, 2017; Pinquart and Sörensen, 2000). The distribution of responses is shown in Figures 1 and 2. The mean value for happiness is 2.97 (standard deviation 0.51), while for depression the mean value is 1.78 (standard deviation 0.77). Both the happiness and depression questions are free of a specified timeframe; instead they ask the respondent to rank current/recent feelings in reference to usual feelings. This allows us to benchmark the question according to the individual’s norm rather than a collective norm. It also allows us to minimize current mood bias from the data. However, the fact that we are looking at subjective wellbeing in the context of ‘recent’ experience increases our concerns regarding reverse causality. This provides further motivation for including the Lewbel (2012) IV estimation results that are part of our analysis. As subjective wellbeing encompasses emotions as well as cognitive judgments, it includes both positive and negative affect. Thus, it is important to explore both happiness (positive affect) and depression (negative affect) to fully understand the impact of the consumption of addictive products on subjective wellbeing. Moreover, while negative and positive feelings are likely to be correlated, it is not the case that they are perfectly correlated. It is possible for an individual to experience an increase in feelings of happiness without seeing a change in 3

The original survey uses the scale in the opposite direction. For instance, 4 means “much less than usual”. We reverse the scale to allow for ease of interpretation of results.

7

their normal level of negative feelings. Positive and negative emotions are not a zero sum game; that is, happiness is not simply the inverse of depression, as has been shown by Zheng (2016). It is worth noting that our measures of subjective wellbeing are both global context free measures. Although both measures are single item measures as opposed to multiple item scales, they are nevertheless commonly used to operationalize concepts of happiness and depression by empirical researchers. Single item measures have the disadvantage that the internal consistency reliability statistic (Cronbach’s alpha) cannot be computed and they are vulnerable to differences in interpretation across respondents. That said, they are less cognitively challenging for respondents and less monotonous than multi-item scales, which are repetitive in terms of questions relating to a single domain of behavior/personality. Single item measures are also not susceptible to common method variance, where spurious correlations are observed between items due to a common response format rather than being due to a true response to the question content. However, we accept the theoretical shortcomings of using a single question to capture complex global constructs in a general setting. We also note that single item questions minimize respondent burden and are less likely to lead to an urge to engage in a particular behavior, and so have strong ethical merit in the context of collecting data on happiness, depression and tobacco consumption, and dependence and addiction, at the individual level. 3.2. Independent variables Our key covariates are measures of smoking behavior, and in this study are captured using three indicators. The first measure captures the smoking status of respondents (Smoking status), and thus is a dummy variable that equals to 1 if the respondent smokes cigarettes. We loosely define our next two measures as capturing domains of smoking behaviors that are correlated with tobacco dependence/addiction. The second (Addiction proxy 1) is a measure of the frequency of tobacco consumption. The HSE asks the question: “How soon after waking do you usually smoke your first cigarette of the day?”. Response to this question captures various bands. In the seven-point scale, we code 0 if respondent is a non-smoker, 1 if respondent smokes 2 hours or more after waking, 2 if respondent smokes 1 hour but less than 2 hours after waking, 3 if respondent smokes 30 minutes but less than 1 hour after waking, 4 if respondent smokes 15 to 29 minutes after waking, 5 if respondent smokes 5 to 14 minutes after waking, and 6 if respondent smokes in less than 5 minutes after waking. 8

Tobacco dependence can be assessed by a range of measures that tend to be frequency- or quantity-based. The above question is a frequency-based question concerned with the time from waking to the consumption of the first cigarette of the day. This measure has been examined extensively in the literature and has been shown to be related to nicotine levels in the body and to capture behaviors that can be described as heavy, uninterrupted, and automatic smoking (see Toll et al., 2007 and Baker et al., 2007 among others who use this question as a proxy for tobacco dependence). This question is also one of the items on the Fagerstrom Nicotine Dependence Scale and as such is an accepted behavioral indicator that is correlated with tobacco dependence. The last indicator of smoking behavior is the number of cigarettes smoked per day, which is also used as a measure of smoking addiction (Addiction proxy 2) and is an intensity indicator. This measure has the advantage that smokers usually are able to respond accurately and is a measure that is widely used in health promotion literature to help smokers make informed choices about how their smoking impacts their health. These two single item measures (Addiction proxy 1 and Addiction proxy 2) represent commonly used proxies for capturing tobacco dependence. Multi-item scales might be thought of as more psychometrically robust; however it is unclear where such scales, which are designed for clinical diagnosis, are appropriate in population-based samples. The scales tend to discriminate between those who are clinically defined as addicted and those who are not. Given the level of clinical addiction in population samples tends to be small, this discrimination can be problematic. The majority of observations are defined as non-addicts and so there is very little variation across the population. Using measures such as the number of cigarettes smoked per day or time to the first cigarette of the day allows us to think about addiction as a continuous spectrum within the population rather than as a discreet threshold, which is the approach used by clinical scales. Our two addiction proxy variables have a large degree of variation across the population-based sample, from which the impact of addictive behaviors on happiness and depression may be empirically identified. Consistent with the literature on subjective wellbeing, we also control for other relevant factors that may affect respondents’ reported wellbeing: gender, age, educational status, marital status, employment status, income, health conditions, household type, and race (see, for example, Awaworyi Churchill and Mishra, 2017; Diener and Oishi, 2000; Helliwell and Wang, 2011; Tay et al., 2014). For gender and employment status, we include dummy

9

variables for respondents who are female and unemployed, respectively. For educational status, excluding other qualifications as base, we include dummies for respondents with a degree, non-degree tertiary qualification and A-level qualification. For marital status, we include dummies for respondents who are married, single and widowed, with divorced as the base category. We control for the age of respondents as well as for age squared. Measures of race are captured by three dummy variables for respondents who are of African ethnic origin, Asian origin, mixed race or other ethnic origin. We exclude respondents who are White as the base category. As a proxy for social class, we control for household tenure by including a dummy variable for respondents who own a house either by paying outright or via a mortgage, and another for respondents who live in households with children. We also include dummy variables to capture whether respondents drink alcohol. Income is measured using a scale that represents five income categories. To account for regional-level fixed effects, we also include dummy variables for London, North, East and South of England, excluding the Midlands as the base category. Table A1 in the Appendix presents a list and description of variables used in the analysis, including summary statistics. Table A2 presents mean wellbeing according to levels of smoking addiction using the seven-point smoking addiction scale (Addiction proxy 1). 3.3. Empirical strategy To examine the association between smoking behavior and subjective wellbeing, we estimate the following equation: ∑ where

(1)

indexes the individuals,

depression);

is the measure of wellbeing (i.e., happiness and

is the measure of smoking behavior (i.e., smoking status and smoking

addiction proxies);

is a set of control variables described earlier;

and

are

parameters to be estimated; and is the random error term. 4. Results Table 1 present results for the association between our measures capturing smoking behavior and wellbeing. Columns 1 to 3 present results for the effects of smoking behavior on happiness, while Columns 4 to 6 present results for the effects on depression. The results

10

suggest that smokers show statistically lower levels of happiness than do non-smokers. Similarly, smokers tend to feel more depressed than non-smokers. Specifically, from Columns 1 and 4, we find the coefficient on the smoker status dummy is -0.061 and 0.103, respectively, implying 0.061 lower subjective happiness and 0.103 increase in the level of depression, on a scale of 1-4, if the respondent smokes. Here, a standard deviation increase in smoker status is associated with a decrease of 0.046 standard deviations in happiness and an increase of 0.051 standard deviations in depression. Results also suggest that increased frequency of smoking (Addiction proxy 1) is associated with depression and lower levels of happiness. Specifically, in terms of the seven-point indicator of the time to first cigarette of the day, we see a negative association between happiness with one’s life and an individual’s smoking frequency. Here, a standard deviation increase in smoking frequency is associated with a decrease of 0.053 standard deviations in happiness. Thus, the sooner an individual smokes in the day, the less happy they are. With regards to depression, we find that a standard deviation increase in smoking frequency is associated with an increase of 0.067 standard deviations in depression. Thus, the sooner an individual smokes each day, the more depressed they are. Last, we find that smoking intensity (Addiction proxy 2) is associated with depression and lower levels of happiness. Specifically, result suggests that the more cigarettes smoked per day, the less happy and more depressed respondents are. The standard coefficients suggest that a standard deviation increase in smoking intensity is associated with a decrease of 0.046 standard deviations in happiness and an increase of 0.051 standard deviations in depression.4 4.1. Endogeneity Lewbel (2012) proposes an IV approach, which is similar to the conventional two-stage least square (2SLS) method, to deal with endogeneity using internally generated instruments, where external instruments are either unavailable or weak. With a precondition of heteroskedasticity, Lewbel shows that internal instruments can be constructed from the 4

Other control variables reveal that, on average, females are more depressed and less happy than males. Consistent with the existing literature (see, for example, Awaworyi Churchill and Mishra, 2017; Helliwell and Wang, 2011), happiness and depression have a U-shaped relationship with age. Specifically, we observe a negative (positive) relationship in the happiness (depression) regressions and the opposite relationship is observed with age squared. Being married generates happiness and less depression, while unemployment does not. Higher levels of income are associated with lower levels of depression, while the absence of long-term illness generates happiness and less depression. Depression declines with house ownership and the presence of children at home. Other covariates are statistically insignificant.

11

residuals of auxiliary equations, which are multiplied by included exogenous variables ( ) in mean-centred form ( ̅ ) (Baum et al., 2012; Lewbel, 2012). In the context of our current study, in addition to reverse causation, one could argue that unobserved social factors promote individuals’ wellbeing are also responsible for smoking behavior. The estimation problem resulting from the current study can thus be summarized as: (2) (3) such that

is an individual’s subjective wellbeing and

smoking behavior.

is the measure of an individual’s

denotes the individual’s unobserved social environment that affects

smoking behavior and wellbeing.

and

are idiosyncratic errors.

is a vector of controls

variables. In the absence of traditional instruments for a 2SLS analysis, Lewbel argues that given some heteroskedasticity in the data, internally generated instruments ( ) given by: (

̅)

can be used, where has zero covariance with regressors and represents a vector of first-stage regression residuals of each endogenous regressor on all exogenous regressors. Thus, the mean of each internally generated instrument will be zero. The Lewbel (2012) approach is often used in the literature where conventional instruments are unavailable or weak, and as a robustness check on findings from conventional 2SLS regressions (see, for example, Awaworyi Churchill, Ocloo, et al., 2016; Awaworyi Churchill, Valenzuela, et al., 2016; Belfield and Kelly, 2012; Emran and Shilpi, 2012; Mishra and Smyth, 2015). This approach has also been used in the subjective wellbeing literature to address endogeneity (Awaworyi Churchill and Mishra, 2017). The Lewbel 2SLS results are presented in Table 2. As shown, the Breusch and Pagan (1979) test

for heteroskedasticity is

highly significant

throughout,

indicating that the

heteroskedasticity assumption for Lewbel (2012) is fulfilled. Further, consistently, across all columns, the first stage F-statistics are greater than 10, which suggests that the internally generated instruments are not weakly correlated with our measures of smoking behavior 12

(Stock and Yogo, 2005). We also fail to reject the null hypothesis for the Sargan-Hansen overidentifying restriction (OIR) tests, which suggests that the internally generated instruments used in first-stage regressions were not overidentified. We find that endogeneity generates a considerable upward bias in ordinary least squares (OLS) estimates across all columns. Results show that the coefficients on our measures of smoking behavior are considerably lower in the 2SLS specifications. We find that a standard deviation increase in smoker status is associated with a decrease of 0.022 standard deviations in happiness and an increase of 0.040 standard deviations in depression. Also, a standard deviation increase in smoking frequency is associated with a decrease of 0.032 standard deviations in happiness and an increase of 0.044 standard deviations in depression. Last, for smoking intensity, 2SLS results show that a standard deviation increase in smoking intensity is associated with a decrease of 0.019 standard deviations in happiness and an increase of 0.022 standard deviations in depression. Overall, comparing the OLS results with the 2SLS results, we find that across all columns, the OLS results overstate the effects of smoking status and addiction on wellbeing5. However, the emerging conclusion of a negative effect of smoking behavior remains valid. 4.2.Robustness checks In this section, we conduct two major exercises to examine the robustness of our results. First, we examine the robustness of our results to an alternate estimation technique used in the subjective wellbeing literature. In addition to OLS, one strand of the literature often adopts ordered logit estimation techniques given the ordinal nature of subjective wellbeing measures (Awaworyi Churchill and Mishra, 2017; Portela et al., 2013). To ensure the robustness of our results to both estimation methods, we also present ordered logit estimates in Table 3. We find that the nature of the association between smoking behavior and wellbeing is not altered by the estimation method used. Consistent with the OLS results, we find that all three smoking indicators are negatively associated with happiness and positively associated with depression. However, the ordered logit estimates appear to be consistently larger in magnitude compared to the OLS estimates.

5

We also examine the sensitivity of our Lewbel 2SLS result using different combinations of as . Lewbel results reported in the main text use . However, in sensitivity tests we try different choices of that are subsets of . Sensitivity results are reported in Table A3 in the Appendix and overall, are consistent with our main results. For brevity, we only report coefficients of smoking behavior in Table A3.

13

Next, we examine the robustness of our results to other measures of smoking behavior. Within this category, we first consider unsuccessful attempts at quitting as a proxy for addiction. Thus, we include an additional measure of addiction, which is a dummy variable that captures respondents who smoke and unsuccessfully have attempted to stop. This is based on the HSE question: “Have you ever tried to give up smoking?”. Second, in addition to the variable used in our main regressions that captures the number of cigarettes smoked per day, for robustness, we generate a new measure that excludes non-smokers, thus dropping the zeros from the sample. Third, among respondents who smoke, we include a dummy for respondents who smoke more cigarettes now compared to a year ago. We derive this dummy from the survey question: “Would you say that you are smoking about the same number of cigarettes as a year ago, or more than a year ago or fewer than a year ago?”. Fourth, we include a continuous variable that captures the number of years respondents have been smoking, and fifth, a dummy variable that captures whether respondents in the smoking category smoke regularly as opposed to occasionally. The fifth dummy capturing regular smokers is based on the survey question: “Do you regularly smoke a cigarette, at least one a day?”. Given that we exclude all non-smokers from this sample, the control category here is considered occasional smokers (i.e., those who smoke less than once a day). We also derive two measures of smoking behavior based on the data at hand. First, we transform our smoking frequency variable by coding the variable as minutes. We exclude the non-smokers from the sample and then take the middle of the bands in the seven-point scale to transform the smoking frequency. Last, we build a stock measure of smoking based on the quantity of cigarettes smoked in the past, based on the information we have on the average number of cigarettes smoked a day as well as the number of years an individual has been smoking. Results for the effects of these alternate measures of smoking behavior are reported in Table 4. Quite consistently, results from our robustness checks suggest that smoking behaviors are associated with a decline in feelings of happiness and an increase in feelings of depression. Specifically, smokers who have unsuccessfully attempted to stop smoking tend to have lower levels of happiness and higher levels of depression. Similarly, the alternate measure of number of cigarettes smoked confirms the observed negative and positive effect on happiness and depression, respectively. Smokers who smoke more cigarettes now compared to a year ago, on average have, lower levels of happiness and higher levels of depression. The

14

coefficient of years of smoking is statistically insignificant in the happiness regression but positive and significant in the depression regression. Regular smokers, as opposed to occasional smokers, have lower levels of happiness. Also, the alternate measure of smoking frequency and quantity of cigarettes smoked in a lifetime support the existing conclusion of a negative effect on happiness and positive effect on depression. 5. Conclusion Our study provides new evidence on the effects of smoking and smoking addiction on subjective wellbeing in England. The study of this relationship remains an important issue for governments and policy makers who seek to introduce smoking bans, tariffs on tobacco and other barriers to reduce smoking. Clearly, the implications of smoking for physical health have been established as negative; however, apart from the difficulty of quitting (i.e., due to addiction), the question remains as to why people still smoke despite the negative health implications. A reason hypothesized in the literature pertains to the possibility of smoking leading to happiness (Chang et al., 2016; Moore, 2009). That is, the mental health benefits might outweigh the physical health costs and so individuals might rationally choose to smoke. Focusing on two well-established measures of subjective wellbeing, happiness and depression, we examined the effect of smoking and smoking addiction (considering both the frequency and intensity of smoking) on subjective wellbeing. We found that smoking and smoking addictive behaviors are associated with lower levels of happiness and higher levels of depression. After controlling for endogeneity, we found that although this conclusion remained valid across a number of indicators of smoking behavior, the effect sizes were significantly lower than OLS results. Moreover, this general finding was robust to the methodological approach to estimation and to the variety of smoking behavior variables employed in this study. These results conflict with findings from Chang et al. (2016) that suggest that smoking makes people happy in a sample of five countries using country-level data on happiness and per capita cigarette consumption. The observed differences could be associated with sample differences and perhaps endogeneity bias. As previously mentioned, without control for endogeneity, the direction and size of effect may be biased. We also drew on various alternate measures of smoking addiction to examine the robustness of our results. Contrary to the findings from Shahab and West (2012), we found a significant 15

association between wellbeing and quit attempts—smokers who have unsuccessfully attempted to stop smoking tend to have lower levels of happiness and higher levels of depression. These results are thus similar to those reported by Weinhold and Chaloupka (2016), which suggests that there is an increase in wellbeing from quitting smoking. This increase is statistically significant and of a meaningful magnitude. In summary, our findings suggest that the relationship between smoking and mental health is negatively correlated. This implies that smoking has both physical and mental health costs. Hence, quit strategies may benefit from a dual approach where smokers are supported through the physical and mental health aspects of their behavior change. Our results are limited in that the pathways from smoking to poor mental health are not considered in our analysis—the data is simply not powerful enough to answer this question. Nevertheless, our results do tell a statistically robust story of the negative association between smoking and depression and life satisfaction more generally. This highlights a broader spill-over of smoking behaviors on wellbeing (in terms of both positive and negative affect) than has previously been illustrated in the literature.

16

References Akl, E. A., Gaddam, S., Gunukula, S. K., Honeine, R., Jaoude, P. A., & Irani, J. (2010). The effects of waterpipe tobacco smoking on health outcomes: a systematic review. International Journal of Epidemiology, 39(3), 834-857. Awaworyi Churchill, S., & Mishra, V. (2017). Trust, Social Networks and Subjective Wellbeing in China. Social Indicators Research, 132(1), 313-339. Awaworyi Churchill, S., Ocloo, J. E., & Siawor-Robertson, D. (2016). Ethnic Diversity and Health Outcomes. Social Indicators Research, 1-36. doi: 10.1007/s11205-016-1454-7 Awaworyi Churchill, S., Valenzuela, M. R., & Sablah, W. (2016). Ethnic diversity and firm performance: Evidence from China’s materials and industrial sectors. Empirical Economics, 1-21. doi: 10.1007/s00181-016-1174-5 Baker, T.B., Piper, M.E., McCarthy, D.E., et al.( 2007). Time to first cigarette in the morning as an index of ability to quit smoking: Implications for nicotine dependence. Nicotine & tobacco research : official journal of the Society for Research on Nicotine and Tobacco. Vol 9(Suppl 4), S555-S570. doi:10.1080/14622200701673480 Barone-Adesi, F., Vizzini, L., Merletti, F., & Richiardi, L. (2006). Short-term effects of Italian smoking regulation on rates of hospital admission for acute myocardial infarction. European Heart Journal, 27(20), 2468-2472. Baum, C. F., Lewbel, A., Schaffer, M. E., & Talavera, O. (2012). Instrumental variables estimation using heteroskedasticity-based instruments. Paper presented at the United Kingdom Stata User’s Group Meetings. Becker, G. S., & Murphy, K. M. (1988). A Theory of Rational Addiction. Journal of Political Economy, 96(4), 675-700. Beckham, J. C. (1999). Smoking and anxiety in combat veterans with chronic posttraumatic stress disorder: A review. Journal of psychoactive drugs, 31(2), 103-110. Belfield, C. R., & Kelly, I. R. (2012). The Benefits of Breast Feeding across the Early Years of Childhood. Journal of Human Capital, 6(3), 251-277. doi: 10.1086/667415 Bernheim, B.D., Rangel, A., (2004). Addiction and cue-triggered decision processes. American Economic Review 94 (5), 1558-1590. Breusch, T.S. and Pagan, A.R. (1979). Simple test for heteroscedasticity and random coefficient variation". Econometrica ,47 (5), 1287–1294.

17

Cesaroni, G., Forastiere, F., Agabiti, N., Valente, P., Zuccaro, P., & Perucci, C. A. (2008). Effect of the Italian smoking ban on population rates of acute coronary events. Circulation, 117(9), 1183-1188. Chang, T., Chu, H.-P., Deale, F. W., & Gupta, R. (2016). The causal relationship between happiness and smoking: a bootstrap panel causality test. Journal of Happiness Studies, 17(3), 1327-1336. Choi, W. S., Patten, C. A., Gillin, J. C., Kaplan, R. M., & Pierce, J. P. (1997). Cigarette smoking predicts development of depressive symptoms among US adolescents. Annals of Behavioral Medicine, 19(1), 42-50. Colhoun, H. M., Dong, W., & Poulter, N. R. (1998). Blood pressure screening, management and control in England: results from the health survey for England 1994. Journal of hypertension, 16(6), 747-752. Cummins, R. A. (2000). Personal income and subjective well-being: A review. Journal of Happiness Studies, 1(2), 133-158. Diener, E., & Chan, M. Y. (2011). Happy people live longer: Subjective well‐ being contributes to health and longevity. Applied Psychology: Health and Well‐ Being, 3(1), 1-43. Diener, E., & Oishi, S. (2000). Money and happiness: Income and subjective well-being across nations. Culture and subjective well-being, 185-218. Emran, M. S., & Shilpi, F. (2012). The extent of the market and stages of agricultural specialization. Canadian Journal of Economics/Revue canadienne d'économique, 45(3), 1125-1153. Glassman, A. H. (1993). Cigarette smoking: implications for psychiatric illness. American Journal of Psychiatry, 150(4), 546-593. Goldberg, D.P. (1978). Manual of the General Health Questionnaire. Windsor, England NFER Publishing. Heatherton, T.F., Kozlowski, L.T., Frecker, R.C. and Fagerstrom, K.O. (1991). The Fagerstrom Test for Nicotine Dependence: a revision of the Fagerstrom Tolerance Questionnaire. British Journal of Addiction, 86, 1119-27. Helliwell, J. F., & Wang, S. (2011). Trust and well-being. Journal of Wellbeing, 1(1), 42-78.

18

Hinks, T., & Katsaros, A. (2012). Smoking ban and life satisfaction: Evidence from the UK. Economic Issues, 17(1), 23-48. Jarvis, M. J., & Feyerabend, C. (2015). Recent trends in children's exposure to second‐ hand smoke in England: cotinine evidence from the Health Survey for England. Addiction, 110(9), 1484-1492. Kaliterna Lipovcan, L., Brkljacic, T., & Tadic, M. (2013). Tobacco consumption, alcohol intake frequency and quality of life: results from a nationally representative croatian sample study. Drustvena Istrazivanja, 22, 627+. Kessler, R.C., Andrews, G., Colpe, et al (2002). Short screening scales to monitor population prevalences and trends in non-specific psychological distress. Psychological Medicine, 32, 959-956. Lakier, J. B. (1992). Smoking and cardiovascular disease. The American journal of medicine, 93(1), S8-S12. Leonard, S., Adler, L., Benhammou, K., Berger, R., Breese, C., Drebing, C., . . . Olincy, A. (2001). Smoking and mental illness. Pharmacology Biochemistry and Behavior, 70(4), 561-570. Lewbel, A. (2012). Using heteroscedasticity to identify and estimate mismeasured and endogenous regressor models. Journal of Business & Economic Statistics, 30(1), 6780. Lyvers, M., Hall, T., & Bahr, M. (2009). Smoking and psychological health in relation to country of origin. International Journal of Psychology, 44(5), 387-392. McBride, M. (2001). Relative-income effects on subjective well-being in the cross-section. Journal of Economic Behavior & Organization, 45(3), 251-278. McCann, S.J.H.(2010) Subjective well-being, personality, demographic variables, and American state differences

in

smoking prevalence. Nicotine and

Tobacco

Research, 12 (9), 895-904. doi: 10.1093/ntr/ntq113 Mindell, J., Biddulph, J. P., Hirani, V., Stamatakis, E., Craig, R., Nunn, S., & Shelton, N. (2012). Cohort profile: the health survey for England. International Journal of Epidemiology, 41(6), 1585-1593. Mishra, V., & Smyth, R. (2015). Estimating returns to schooling in urban China using conventional and heteroskedasticity-based instruments. Economic Modelling, 47, 166173. doi: http://dx.doi.org/10.1016/j.econmod.2015.02.002

19

Molnar, D. S., Busseri, M. A., Perrier, C. P., & Sadava, S. W. (2009). A longitudinal examination of alcohol use and subjective well-being in an undergraduate sample. Journal of studies on alcohol and drugs, 70(5), 704-713. Moore, S. C. (2009). The nonpecuniary effects of smoking cessation: happier smokers smoke less. Applied Economics Letters, 16(4), 395-398. Odermatt, R., & Stutzer, A. (2015). Smoking bans, cigarette prices and life satisfaction. Journal of Health Economics, 44, 176-194. Okun, M. A., Stock, W. A., Haring, M. J., & Witter, R. A. (1984). Health and Subjective Well-Being: A Meta-Analyis. The International Journal of Aging and Human Development, 19(2), 111-132. doi: 10.2190/qgjn-0n81-5957-haqd Oyebode, O., & Mindell, J. (2013). Use of data from the Health Survey for England in obesity policy making and monitoring. Obesity Reviews, 14(6), 463-476. Pinquart, M., & Sörensen, S. (2000). Influences of socioeconomic status, social network, and competence on subjective well-being in later life: a meta-analysis. Psychology and Aging, 15(2), 187. Portela, M., Neira, I., & del Mar Salinas-Jiménez, M. (2013). Social capital and subjective wellbeing in Europe: A new approach on social capital. Social Indicators Research, 114(2), 493-511. Primatesta, P., Brookes, M., & Poulter, N. R. (2001). Improved hypertension management and control results from the health survey for England 1998. Hypertension, 38(4), 827-832. Rooke, S. E., Norberg, M. M., Copeland, J., & Swift, W. (2013). Health outcomes associated with long-term regular cannabis and tobacco smoking. Addictive behaviors, 38(6), 2207-2213. Shahab, L., & West, R. (2012). Differences in happiness between smokers, ex-smokers and never smokers: cross-sectional findings from a national household survey. Drug and Alcohol Dependence, 121(1), 38-44. Spitzer, R.L., Kroenke, K., Williams, J.B., et al. (1999). Validation and utility of a self-report version of PRIME-MD: the PHQ primary care study. JAMA. 282 (18), 1737– 44. doi:10.1001/jama.282.18.1737 Tay, L., Li, M., Myers, D., & Diener, E. (2014). Religiosity and Subjective Well-Being: An International Perspective Religion and spirituality across cultures (pp. 163-175): Springer.

20

Thomas, S., Fayter, D., Misso, K., Ogilvie, D., Petticrew, M., Sowden, A., . . . Worthy, G. (2008). Population tobacco control interventions and their effects on social inequalities in smoking: systematic review. Tobacco Control, 17(4), 230-237. Toll, B.A., Schepis, T.S., O’Malley, S.S., McKee, S.A., Krishnan-Sarin, S. (2007) Subjective Reactivity to the First Cigarette of the Day as a Predictor of Smoking Relapse: A Preliminary

Study. Drug

and

alcohol

dependence.

Vol.89(2-3),

302-305.

doi:10.1016/j.drugalcdep.2007.01.011 Weinhold, D., & Chaloupka, F. J. (2016). Smoking status and subjective well-being. Tobacco Control. doi: 10.1136/tobaccocontrol-2015-052601 Xu, X., Bishop, E.E., Kennedy, S.M., Simpson, S.A., and Pechacek, T.F. (2014) Annual Healthcare Spending Attributable to Cigarette Smoking: An Update. American Journal of Preventive Medicine, 48(3), 326–33. Young, K. S., & Rogers, R. C. (1998). The relationship between depression and Internet addiction. CyberPsychology & Behavior, 1(1), 25-28. Zheng, X. (2016) A Study on Relationship between Depression and Subjective Well-Being of College Students. Psychology, 7, 885-888.

21

Table 1 - Effects of smoking on happiness and depression (OLS regressions) VARIABLES

(1)

Happiness (2)

(3)

Smoker status (Smoker) -0.061*** (0.021) [-0.046] Smoking frequency -0.019*** (Addiction proxy 1) (0.006) [-0.053] Smoking intensity -0.004*** (Addiction proxy 2) (0.002) [-0.046] Female -0.024* -0.022* -0.024* (0.013) (0.013) (0.013) Age -0.016*** -0.016*** -0.016*** (0.003) (0.003) (0.003) Age squared 0.017*** 0.016*** 0.016*** (0.002) (0.002) (0.002) Degree 0.009 0.008 0.008 (0.017) (0.017) (0.017) Below degree 0.030 0.030 0.028 (0.021) (0.021) (0.021) A level 0.004 0.004 0.004 (0.021) (0.021) (0.021) Married 0.038** 0.038** 0.039** (0.018) (0.018) (0.018) Single -0.023 -0.024 -0.022 (0.026) (0.026) (0.026) Widowed -0.019 -0.020 -0.019 (0.032) (0.032) (0.032) Unemployed -0.075*** -0.071*** -0.074*** (0.018) (0.018) (0.018) Income Band 2 0.040 0.041 0.041 (0.027) (0.027) (0.027) Income Band 3 0.064*** 0.065*** 0.063** (0.024) (0.024) (0.024) Income Band 4 0.076*** 0.077*** 0.077*** (0.024) (0.024) (0.024) Income Band 5 0.062** 0.063** 0.062** (0.025) (0.026) (0.025) Black 0.052 0.052 0.051 (0.052) (0.052) (0.052) Asian 0.035 0.035 0.035 (0.035) (0.035) (0.035) Other/mixed race 0.024 0.032 0.022 (0.054) (0.054) (0.054) Children 0.032* 0.030* 0.032* (0.017) (0.017) (0.017) Drinker -0.053*** -0.053*** -0.052*** (0.019) (0.019) (0.019) House owner 0.032* 0.030* 0.033* (0.017) (0.017) (0.017) Constant 3.225*** 3.213*** 3.210*** (0.070) (0.070) (0.069) Regional dummies Observations R-squared

Yes 5,821 0.032

Yes 5,792 0.032

Yes 5,815 0.031

(4)

Depression (5)

(6)

0.103*** (0.030) [0.051] 0.036*** (0.008) [0.067]

0.042** (0.020) 0.027*** (0.004) -0.031*** (0.004) -0.029 (0.027) -0.056* (0.032) -0.029 (0.030) -0.073*** (0.027) -0.005 (0.037) 0.027 (0.050) 0.150*** (0.027) -0.094** (0.039) -0.130*** (0.035) -0.177*** (0.036) -0.176*** (0.037) -0.142** (0.071) -0.006 (0.050) -0.102 (0.078) -0.088*** (0.025) 0.057** (0.028) -0.089*** (0.026) 1.533*** (0.101)

0.042** (0.020) 0.026*** (0.004) -0.031*** (0.004) -0.023 (0.027) -0.052 (0.032) -0.025 (0.030) -0.071*** (0.027) -0.007 (0.037) 0.031 (0.050) 0.147*** (0.027) -0.088** (0.039) -0.126*** (0.035) -0.173*** (0.036) -0.170*** (0.038) -0.138* (0.071) -0.004 (0.049) -0.108 (0.078) -0.086*** (0.025) 0.059** (0.028) -0.083*** (0.026) 1.531*** (0.102)

0.007*** (0.002) [0.051] 0.042** (0.020) 0.026*** (0.004) -0.031*** (0.004) -0.026 (0.027) -0.053* (0.032) -0.027 (0.030) -0.073*** (0.027) -0.003 (0.037) 0.030 (0.050) 0.150*** (0.027) -0.096** (0.039) -0.129*** (0.035) -0.177*** (0.036) -0.175*** (0.037) -0.140** (0.071) -0.007 (0.050) -0.100 (0.077) -0.086*** (0.025) 0.056** (0.028) -0.090*** (0.026) 1.548*** (0.101)

Yes 5,827 0.051

Yes 5,798 0.053

Yes 5,821 0.051

Notes: Robust standard errors adjusted for heteroskedasticity in parentheses. Standardized coefficients in square brackets represent changes of standard errors. *** p