The economics of happiness: does smoking behaviour bring life (dis)satisfaction

Размещено на http://www.allbest.ru/

ФЕДЕРАЛЬНОЕ ГОСУДАРСТВЕННОЕ АВТОНОМНОЕ ОБРАЗОВАТЕЛЬНОЕ УЧРЕЖДЕНИЕ ВЫСШЕГО ОБРАЗОВАНИЯ

«НАЦИОНАЛЬНЫЙ ИССЛЕДОВАТЕЛЬСКИЙ УНИВЕРСИТЕТ

«ВЫСШАЯ ШКОЛА ЭКОНОМИКИ»

Международный институт экономики и финансов

Выпускная квалификационная работа

Экономика счастья: приносит ли курение (не)удовлетворение

The Economics of Happiness: Does Smoking Behaviour Bring Life (dis)satisfaction

Сидорина Мария Витальевна

Москва 2020



Abstract

This work investigates the association between smoking behavior and life satisfaction based on the cross-sectional data from Russian Longitudinal Monitoring Survey for the period of 2016. By analyzing this RLMS data, the next findings are received: current smokers have a tendency to be less satisfied with the quality of their life in contrast to those who do not smoke. Several methodological approaches were applied such as ordered probit and OLS estimation and also Lewbel's instrumental variable approach is used to handle the potential endogeneity in the framework of tis study. Moreover, the method of cardinalization of life satisfaction is applied to be able to use OLS kind of estimation. Overall, from the results of this study it might be stated that nicotine addiction has a substantial negative effect on life satisfaction and this connection may be used to contribute to the current measures attempting to reduce smoking in Russia like anti-tobacco limitations, smoking bans and government health strategies.

В данной работе исследуется связь между курением и удовлетворенностью жизнью на основе кросс-секционных данных российского лонгитюдного мониторингового исследования за период 2016 года. Анализируя эти данные RLMS, можно сделать следующие выводы: современные курильщики имеют тенденцию быть менее удовлетворенными качеством своей жизни в отличие от тех, кто не курит. Было применено несколько методологических подходов, таких как упорядоченная оценка probit и OLS, а также инструментальный вариативный подход Левбеля, который используется для преодоления потенциальной эндогенности в рамках исследования. Кроме того, метод кардинализации удовлетворенности жизнью применяется для того, чтобы иметь возможность использовать оценку типа OLS. В целом, из результатов данного исследования можно сделать вывод, что никотиновая зависимость оказывает существенное негативное влияние на удовлетворенность жизнью, и эта связь может быть использована для содействия текущим мерам по сокращению курения в России, таким как антитабачные ограничения, запреты курения и государственные стратегии здравоохранения.



Table of contents

Introduction

1. Literature review

1.1 Life satisfaction or subjective well-being concept and measurement

1.2 Smoking and life satisfaction

1.3 The formulation of hypothesis

2. Data and Model

2.1 Data description

2.2 Selection of variables

2.3 Descriptive statistics

2.4 Model

3. Empirical framework

3.1 Empirical strategy

3.2 Empirical methodology

4. Estimation and outcomes

Conclusion

List of references

Appendix



Introduction

The cigarette manufacturing sector endows significantly the world market and renders tobacco consumption a significant commercial activity. For instance, in Russia the tobacco industry adds approximately 600 billion of rubles to the annual excise tax receipts. Along with its financial impact, smoking has had disastrous ecological as well as health consequences, and therefore has been deeper observed and researched recently. Actually, smoking is one of the major concerns affecting health of the people around the world. Particularly, in Russia tobacco consumption is an important health problem that has a serious influence on smokers, non-smokers and their satisfaction with life. According the World Health Organization survey conducted in 2016, more than 36 million of Russian residents are current smokers and daily smokers encompass about 31 million of people. Those who are subject to passive smoking constitute around 13 million individuals. However, among the adults almost all of them (90.8%) are sure that nicotine dependence leads to the development of serious health issues. More than the half the current current smokers actually plan to quit smoking.

The analysis of satisfaction with life has been extremely prevalent in the human research and economic theory since the turn of the new millennium. This is apparent in the expanded work in this area and in socioeconomic literatures with an emphasis on life satisfaction combined with human welfare. A great majority of activities such as smoking behavior tend to be reviewed and updated in financial theory with a higher importance placed on attitudinal welfare as opposed to strictly financial happiness. The association regarding nicotine addiction and life satisfaction has aroused some attention in academic research inasmuch as the latter metrics offers a different framework for the measurement of general quality of life which might be attributed to pecuniary and health background.

There exists multitudinous academic examination of the impact of smoking status and quality of life with the great majority finding the adverse effect of smoking habit on being content with life (Wilmans & Rashied, 2020; Awaworyi Churchill & Farrell, 2017). Notwithstanding the crescent studies investigating the interaction amidst tobacco consumption and life satisfaction, the existing publication explicitly oriented towards the emerging nations including Russia tends to be immensely restricted. The findings received by Stickley et al. (2015) in the research of nine post-Soviet Union nations analyzed the link of smoking attitude towards happiness that is determined as a subjective variable. Apart from this work, there is a significant shortage of research specifically evaluating this association in unadvanced economies that ought to be compensated.

As present study in this field of academic work of not so high-income nations is extremely scarce and since identifying the interaction with tobacco consumption and quality of life may bring visibility into the public usefulness of various government prevention strategies towards smoking, the question arises whether smoking brings about satisfaction or rather dissatisfaction with life. The purpose of this study is to determine the connection involving nicotine dependence and life satisfaction in Russia as well as the relevance of other determinants affecting life satisfaction by applying various methods of estimation for robustness check of the findings in cross-section data analysis with the benefit of using additional controls to handle the obstacle of possible confounders. Moreover, a novel Lewbel (2012) IV approach as an attempt to cope with possible endogeneity is applied here which contributes a lot to the existing research.

This thesis encompasses the following arrangement. The literature review will cover diverse descriptions of the concept of life satisfaction applied by scholars, numerous tools of gauging life satisfaction moving on to the elaboration on the link between cigarette smoking and satisfaction with life involving the reasons for smoking cessation and continuing with this habit on the ground of various research papers dedicated to this topic. It is followed by enunciating the hypothesis. What is more, several models will be proposed for the estimation on the cross-section data of the connection amidst life satisfaction and smoking behavior which is followed by empirical methodology highlighting the methods used for measurement of satisfaction with life as well as some approaches of estimation applied in this paper. After which the estimation outcomes with the discussion of them will be provided. Hereupon, the conclusion, the probable limitations of the research, the list of references and appendix are going to be given.



1. Literature review

1.1 Life satisfaction or subjective well-being concept and measurement

Scholars during centuries perceived satisfaction as the greatest and primary incentive for human behavior. Although social scientists have for years mainly neglected meaningful human well-being, while personal misery was extensively observed. This condition has been ameliorated by psychological and behavioral areas of science over the past years with rapidly accelerated growth in scientific research.

Some psychologists and social scholars seek to interpret satisfaction or personal well-being. Life satisfaction is a slightly more complicated term than it might initially appear. Satisfaction with life is frequently applied as a synonym for happiness and subjective well-being. Nevertheless, there is no particular definition which is used for satisfaction with life. Thereby, a comprehensive analysis of subjective well-being was published by Warner Wilson in 1967. Owing to the constrained information accessible at that point in time Wilson (1967) states that a satisfied individual is someone who is an adolescent, accomplished, well-mannered, outgoing, enthusiastic, active, in good health ,content with their work, committed person of faith with a sense of self-worth, even-tempered ambitions, of either gender and of spacious mind. Diener (2000) describes satisfaction with life as the cognitive and affective assessment of person's existence. The scholar also mentions that Individuals are really pleased when they have a lot of joyful and not much negative feelings, when they perform entertaining actions as well as encounter less discomfort and more pleasant emotions and are fulfilled with their life as a whole. As opposed to Diener (2000), Veenhoven (2008) gives another definition to happiness which is represented as the ultimate enjoyment of existence in its entirety, particularly, the way one enjoys their existence.

Though the notions are not quite identical, the fundamental meaning remains that satisfaction with life is general perceptions of a person regarding their life. Similarly, happiness is a universal assessment instead of one that is concentrated on some particular point in time or in some restricted field.

The feelings and moods of individuals indicate their current emotions. Everybody takes greater presumptions about his or her own life as well as on aspects like relationship and employment. Although there are many specific elements of SWB such as happiness with living, contentment with crucial areas of life, positive affect and weak negative affect (which are termed as moods and emotions). (Diener, 2000) Such variables are distinct and can be assessed and studied independently. The existence of a pleasant affect does not necessarily imply that there is no unpleasant affect and vice versa. So, subjective well-being is a vast variety of factors, including the cognitive reactions of individuals, fulfillment in diverse domains of life and general quality of life levels.

Thereby the question arises as to what methods should be used in order to measure subjective well-being. Thus, Andrews and McKennell (1980) highlighted the impact of cognition which is a rational indicator and affect that is an irrational aspect as the major measures of subjective well-being. Cognitive appraisal explains how we perceive our overall contentment with life and satisfaction in particular fields such as personal life and employment, whereas affective appraisal takes into account our moods, so that greater SWB includes the repeated as well as strong pleasant attitude as happiness and optimism and the insufficiency of unpleasant ones like rage, envy or deception (Kashdan, 2004). The affective and cognitive elements of SWB are inspected through various measures and techniques. They also require self-reporting methods, however, may also be analyzed by matching the findings with certain factors like reports from friends or plausible welfare results (Layard, 2010).

Scientists have utilized alternative methods of measurement during the last years to achieve a clearer understanding of long-run emotions. For instance, in the Naturalistic Experience-Sampling Method (ESM), investigators evaluate the SWB of participants at different points in time of their daily living, typically for a span up to four weeks using repeated specific questions regarding the contentment of the current time (Stone et al., 1999). Other evaluation techniques might involve mates and family union's accounts (Sandvik et al., 1993) and reminiscence concerning pleasant and unpleasant experiences in living (Seidlitz, Wyer and Diener, 1997). Apart from these ones there are also self-report methods with application of multiple questions, such as life satisfaction surveys including Satisfaction With Life Scale (SWLS) which is a convenient survey for assessing global life satisfaction. (Diener et al., 1985). Social scientists who study subjective well-being focused largely on self-reporting metrics to measure satisfaction with life, particularly on how individuals evaluate their own degree of satisfaction using questionnaires. In these surveys respondents indicate their level of satisfaction, positive feelings, and lack of negative feelings. The Riverside Life Satisfaction Scale is one of the newly created tools for measuring life satisfaction. The scholars note that this technique is superior to SWLS as it raises the applicability of the metric and reduces the likelihood of acquiescence bias which implies a propensity to admit something easily.

There might be uncertainties as to the efficacy of self-report to determine life satisfaction owing to the fact that participants could be consistently biased, can grasp their feelings or the words used to express them in a restricted way and thus may be fickle in their replies. The current empirical evidence suggests, however, that a simple straight-forward query and evaluations from the perspective of comprehensive surveys give fairly accurate results.

1.2 Smoking and life satisfaction

The use of tobacco can lead to dependency on nicotine and serious health issues. Stopping smoking decreases the incidence of smoke-related illnesses and deaths dramatically. Even though there are effective treatment options and services available for cessation, cigarette addiction is a disorder which sometimes demands frequent interventions and a lot of effort from those smokers who attempt quitting.

One of the main fears of smokers who are pondering over quitting is that they forego an essential contributor of delight and an ability to handle the pressures and weight in their daily living, thus by ceasing smoking they become less content with their life. In addition, smokers have a shared misconception that reducing the frequency of cigarette usage would significantly lower SWB levels, disrupt their social interaction and the capability to tackle tension. Some tobacco users who seek to quit for the purpose of healthcare or the treatment plan claim that pleasure is a crucial aspect for their addiction (West, 2006). Still, there are a great many people who smoke that become disillusioned with cigarette smoking and most of smokers repent that they have begun initially.

The majority of studies examining potential dangers and advantages with respect to smoking have concentrated on the consequences of continuing smoking and the profit of cessation. Empirical studies that explore specific risk beliefs correlated with smoking cessation centered on gender disparities to further understand why females handled quitting worse relative to males. Particularly, it is more common for women than men to be worried about the gain in weight following quitting and to perceive weight gain as the reason for continuing smoking. (Swan et al., 1993). Moreover, the signs of smoking removal might involve bad-temper, anger and disturbance, difficulty concentrating, affinity for nicotine and more starving (Piper et al., 2012).

Generally, a worry for health is one of the most significant reasons for quitting cigarette consumption. Illnesses connected with smoking or issues with health are among the causes that make smokers quit smoking. Many people who smoke sought to strengthen their health and ease the diseases linked with tobacco. Beyond that, the smokers are encouraged to leave the smoking habit because of the possibility of having a disease even though they do not

suffer from any pain at the moment. It is aligned with the research, which indicates that problems with health are a significant incentive to quit. More specifically, surveys of ex-smokers who attempted quitting to preserve their health and active smokers who suggested health issues as a key reason for smoking cessation are positive of this. (McCaul et al., 2006; Riedel, Robinson, Klesges & McLain-Allen, 2002)

Gestation and feeding the baby were both sound reasons for quitting smoking. Many inquiries have found out that the majority of females avoid smoking, particularly during the state of being pregnant. Most of them tend to stop just after their pregnancy has been verified. The true purpose is to ensure that the baby grows properly. In fact, tobacco consumption and happiness have an impact on health in the long run (Steptoe et al., 2005).

However, in recent research some studies have concentrated on smoking in certain groups of individuals such as those with mental illnesses. Specifically, mentally disabled people may have greater difficulties in quitting smoking as they consume tobacco to cope with psychiatric issues, which assumes that mental disorder results in tobacco consumption. Nevertheless, later studies pose concerns regarding the causality direction. The smoking background was correlated with an elevated likelihood of distress, although not the reverse, in the children and teenage sample. (Wu & Anthony, 1999)

The aftereffect of tobacco control strategies on people's welfare may be fraught with obstacles for evaluation. Nevertheless, due to tobacco prevention programs, smoking bans and healthcare organizations, substantial declines in the consumption of cigarettes have been made in several countries lately. What is more, consuming tobacco is frequently forsaken as a result of cigarette expenses, which suggests that the incentive to quit smoking is enhanced by the rise in prices for cigarettes. Odermatt and Stutzer (2015) in their study attempt to exercise judgement in the reaction of subjective well-being to smoking bans and tobacco prices. They discover the finding that greater cigarette costs lower the life contentment of those who smoke. According to their analysis, smoking bans as self-control devices are not related to SWB but enhance satisfaction with life of the individuals who smoke and as well do not attempt smoking cessation. Checking the impact of anti-tobacco regulation policies and utilizing the time varying complexity of these actions to explore the reason of exogenous heterogeneity, Brodeur (2013) investigates the response of SWB to smoking regulations and the prevalence of smoking habit in the US. He does not receive any confirmation of the fact that that smoking bans might actually plunge smoking prevalence. However, policies affect wellbeing: smoking bans may induce cigarette consumers that do not try tobacco cessation be more content with their life. Overall, it is considered that tobacco users might be regarded as those who take advantage of smoking restrictions. Both these papers come to conclusion that smoking bans lead to the enhanced levels of satisfaction with life among current smokers (Brodeur, 2013; Odermatt and Stutzer, 2015).

The above observations create uncertainty of the effect on the overall frame of mind, purported health condition, subjective wellbeing and life satisfaction over the protracted period when the outcomes of cessation are subsided. Smokers can also report improvements in satisfaction with life as crucial reasons for stopping smoking as minimizing the probability of illness occurrence.

There is a lot of research on the link between life satisfaction and smoking, however, studies on changes in subjective well-being due to nicotine addiction have been rather limited by reverse causality or other diverse biases.

Most smokers worry that they would forego a significant means of pleasure in case they quit smoking and thus become less satisfied. Nevertheless, the long-run effects of quitting have been poorly researched. Shahab and West (2009) analyzed ex-smokers' accounts of improvement in satisfaction after quitting and the reasons correlated with such findings using cross-sectional analysis of households. In their study they use standard control variables such as age, gender, region, employment status and number of children and come to conclusion that former smokers predominantly state being more content at present compared to the time they were addicted to smoking. In their following article on the basis of the same survey Shahab and West (2012) used the same set of control variables but here account for more control groups which include current smokers and never smokers to avoid plausible recall bias (systematic error that happens when respondents cannot correctly recall information about past events) and find that former smokers enjoyed greater levels of subjective well-being (SWB) as opposed to current smokers and showed comparable SWB rates as those who never smoke.

In spite of the overall limitations, scientific study has launched to develop discovering the connection involving satisfaction and health, typically indicating a correlation of increased pleasure with improved healthcare results (Pressman & Cohen, 2005).

The risks of smoking and the positive results of ceasing smoking are well known. Nevertheless, many smokers continue to smoke to settle their nerves and pull through stressful emergencies. Some studies find that vulnerability to stress might be a fundamental frequent occurrence among those who smoke, or else it may be a mechanism that connects tobacco consumption with satisfaction with life. (Piper et al., 2012; Weinhold and Chaloupka, 2017)

As a contribution to the previous research, Wang et al (2014) in their research attempted to study the cross-sectional association between satisfaction with life and smoking among adults in China. In their article they use drinking alcohol (ever or never) as part of control since alcohol consumption is frequently connected with lower levels of life satisfaction and drinking is linked with smoking. Likewise, Stickley et al (2015) examine a broad sample from nine states with the benefit of heavy episodic drinking (described as consumption of at least 200 grams of strong alcoholic beverages) as a control but they expand a set of their controls to self-rated health (from very poor to very good) and apart from standard socio-economic and demographic ones to determine that former smokers report greater levels of life satisfaction compared to those who smoke now. Similar controls were used in the study of Weinhold and Chaloupka (2017) that incorporate a tendency towards habitual use of external substances which is defined as daily drinking and self-reported health status (from poor to excellent) to analyze the relationship between subjective well-being and smoking status.

Recent studies have been directed towards exploring the connection of changes in smoking status and alterations in life satisfaction over a lasting period of time. Moore (2009) used data from time series survey to explore the connection between changes in cigarette consumption and changes in happiness utilizing the standard independent variables (controls) except for number of GP attendances as proxy for general health. The later research of Weinhold and Chaloupka (2017) also analyze the relationship between changes in subjective well-being and alterations in smoking status over a period of time within individuals with the help of socio-economic, demographic and health controls.

The mentioned above papers explicitly investigated the discrepancies in life satisfaction of people with distinguishing attitude towards smoking and observed that current smokers tend to be less content with their life.

1.3 The formulation of hypothesis

This paper adds to the research by taking advantage of a rich quantitative sample that helps resolve possible issues of missing factors and constrained data in prior smoking behavior and life satisfaction works. In fact, a broader variety of control variables is going to be applied. Furthermore, distemper vulnerability and alcohol dependence which are possible confounders are going to be used. Lastly, the abundance of the sample enables to minimize the threat of endogeneity problem.

The results show that smokers appear to acknowledge being in a poorer mental state in contrast with non-smokers over the span of a day and display lower rates of satisfaction compared to non-smokers, irrespective of their prior experience. (Koivumaa-Honkanen et al., 2003)

H1: smoking significantly negatively affects life satisfaction



2. Data and Model

2.1 Data description

In order to investigate the relationship between smoking and life satisfaction the data gathered from the combined database of individuals from RLMS-HSE (Russian Longitudinal Monitoring Survey-Higher School of Economics will be used. Starting from 1994 RLMS-HSE has been an ongoing first and only non-governmental survey of the socio-economic circumstances and health status of the people in Russia. It is nearly the first and only representative survey in Russia that has a considerable panel composition, and questionnaires that very same individuals over a prolonged period of time. The element framework of the sample questionnaires follows the world practice norm, enabling the measurement of commonly used metrics and the accuracy of cross-country distinctions. Eventually, the RLMS-HSE provides a substantial collection of value assessments that enable us to complement knowledge on the patterns of transitions in the objective features of living of the Russians with a subjective evaluation of changes taking place in the nation and to provide a deeper understanding of the aspirations and intentions of diverse socio-economic categories.

For the part of the cross-sectional analysis the data from RLMS-HSE on individuals for the year 2016 is going to be utilized for the research purposes.

2.2 Selection of variables

Among control variables there are: employment status, income, age together with age squared, educational level, marital status, number of children in the household, self-reported health level, population, gender, drinking and susceptibility to stress.

Income. As a control for personal income the logarithm of the household income per head is going to be used which is calculated as total income per household to which an individual belongs divided by the household size which is the number of people in the household. A great majority of research works have indicated that income positively influences life satisfaction (Clark et al., 2001).

Age. It is found in lots of research papers that life satisfaction has a U-shaped link with age (Hayo & Seifert, 2003). Thus, this non-linear connection should be taken into account by accounting for age squared in the model used apart from age in years.

Educational level. It is represented by the presence of higher education either complete or incomplete which takes on the value one if there is higher education and value zero otherwise (dummy variable ). Some research papers find that a higher educational background tends to put imprint on satisfaction with life in an advantageous way (Diener et al., 1997).

Employment status. The employed are defined as those who had a job, were on a paid or unpaid leave, particularly paternity/maternity leave at the time of the interview (dummy variable . The employed are found to have greater levels of happiness compared to the unemployed (Gerlach and Stephan, 1996).

Number of children takes on integer numbers from zero which corresponds to no children in the household if a person answers “no” to the corresponding question about having kids in the survey to positive numbers which include those respondents who have answered “yes” to the question about having children. According to Becker (1981), individuals are anticipated to have a positive effect from having kids since they are encouraged to join the family and having children is seen as a benefit-enhancing practice.

Marital status. Such groups as single, widowed, divorced, and not cohabiting are taken into the unmarried denomination indicating a detrimental impact on life satisfaction. Married individuals are those who are officially registered and cohabiting (dummy variable . Wed people show greater satisfaction with life than individuals in other marital status (Diener et al., 1997).

Gender is a dummy variable for . Several empirical investigations including Shyam and Yadev (2006) find that men are more satisfied with their lives in contrast to women, so the expected sign is positive.

Self-reported health level includes 5 categories: excellent (1), good (2), satisfactory (3), poor (4), very poor (5). So, in the model there will be one dummy variable that will take on the value 1 for those who are in good (2) and excellent (1) health and value 0 for those who are in poor, satisfactory and very poor health (dummy variable . According to Veenhoven (2008), the higher the level of health, the greater the level of satisfaction with life.

Drinking will be controlled for and this variable implies whether an individual has consumed any alcoholic beverages for the past 30 days, so it would take on the value 1 if yes and value 0 otherwise (dummy variable). The consumption of alcoholic beverages is often linked with reduced levels of satisfaction with life. (Mentzakis et al., 2013)

Population is going to be controlled for and for this purpose a dummy variable is created which takes on the value 1 if a person lives in a type of settlement with population at least 1 million people and value 0 otherwise. The sign of this variable is anticipated to be negative as those who live in large cities tend to have lower levels of satisfaction.

Vulnerability to stress. As a proxy for sensitivity to stress a dummy variable is going to be controlled for which in the questionnaire is an answer to the question: “Are you characterized by attacks of aggression or irritability?”. It will take on value 1 if yes and 0 otherwise (dummy variable ). It is envisioned to detrimentally operate upon the state of being content with living. Perception to stress occurrences can be regarded as an intrinsic aspect of propensity to smoke, thus plunging the fulfillment with life. Conversely, such vulnerability may also reflect the process by which smoking and life satisfaction are connected, for instance, whether smoking makes people less stress-prone (Weinhold & Chaloupka, 2017).

The main variable of interest includes those who currently smoke.

Current smoker. An individual is identified as a current smoker if they answer “yes” to the question of whether they smoke now or not. Thus, the current smoker variable takes on the value 1 if an interviewee consumes tobacco at the instant of the survey and value 0 otherwise (.

2.3 Descriptive statistics

Table 1. Variables and their anticipated signs (Source: author's own calculations received from STATA MP)

Variable	Frequency/mean and st. dev.	Anticipated sign
	Completely satisfied-8.3% More likely satisfied-41.7% Yes and no-24.8% Not very satisfied-18.6% Completely dissatisfied-6.6%	Dependent variable
	Currently smokes 27.3% Does not smoke 72.7%	-
	Unemployed or out of labor force 45.5%, Employed 54.5%	+
	Male 42.2% Female 57.8%	+
	Married or cohabiting-61.9% Divorced, single, widowed or not cohabiting-38.1%	+
	Good or excellent health-38.6% Average, poor or very poor health-61.4%	+
	Live in a settlement with population over 1 mil. -21.9% Lower than 1 mil. -78.1%	-
	Higher education 32.2% No higher education 67.8%	+
	Drink alcohol recently 39.9% Do not drink 60.1%	-
	Stressed 10.8% No stress 89.2%	-
	Mean: 9.6 St. dev.: 0.6	+
	Mean: 1.3 St. dev.: 1.1	+
	Mean: 45.4 St. dev.: 18.6	-
	Mean: 2404.5 St. dev.:1827.5	+

2.4 Model

The model of the effect of the smoking status on satisfaction with life used on cross-section is the following:

Where is satisfaction with life that takes on the value 1 if completely satisfied, value 2 if more likely satisfied, value 3 if yes and no, value 4 if not very satisfied, value 5 if completely dissatisfied. specifies a variable on a respondent , is the respective coefficient for the control . indicates the intercept and represents the disturbance term is a dummy variable for current smoker for individual .

The cross-sectional model will include the following controls: (=1 for those who are employed is the log income in the household per head, (equals 1 for those who are male), , which is age squared, is equal to 1 for married and is a number of children in the household, (= 1 for those who reside in a type of settlement with population >=1 mil. people) , (equals 1 for those who have a degree), , (=1 for experiencing stress) and variables.

3. Empirical framework

3.1 Empirical strategy

The main strategy of this investigative work is to scrutinize the imprint of smoking habit on being content with life. There are various methods of estimation that are used in diverse research papers dedicated to life satisfaction. In this thesis the dependent variable which is satisfaction with life is ordinal, thus some transformation of it like cardinalization is needed in order to be able to apply the OLS estimation. The OLS method is used initially for interpretation purposes as it helps us to observe directly the effects of the main variables of interest on the dependent variable. Then after running the OLS regression with cardinalized life satisfaction I apply the instrumental variable approach which involves selecting a relevant set of instruments in place of smoking status and assists in coping with the probable endogeneity problem which will be discussed in greater detail later in this paper. Followed by these two techniques, the ordered probit method is employed which takes into account the ordered nature of life satisfaction, however, it might be rather complicated to interpret the results of this method of estimation, so the direction of the link of explanatory variables with the dependent one would be taken into consideration. The ordered probit is utilized for the comparison with the OLS and IV results to see how coefficients and signs of our variables might change with the alteration of an approach used for estimation.

3.2 Empirical methodology

The measurement of satisfaction with life

Numerous research papers use diverse methods to measure life satisfaction. This work is going to apply several approaches which are going to be considered in this section. Actually, life satisfaction is recognized to be an ordinal variable, in our case it follows the scale between 1 and 5, in which 1 represents completely satisfied and 5 indicates completely dissatisfied. One technique used is to make the life satisfaction variable a binary variable which would take on the value 1 if a person either completely satisfied (1) or satisfied (2) and would take on the value 0 if a respondent is either (3) yeas and no, (4) not much satisfied or (5) completely dissatisfied. This method is frequently utilized to simplify the comprehension of the estimation outcomes. This life satisfaction measure will look like the following:

Another tool to cope with the ordinal nature of happiness or satisfaction with life is to use the technique of cardinalization (van Praag, Frijters & Ferrer-i-Carbonell, 2003; Jilcova, 2017). This procedure allows for applying standard OLS estimation to an ordinal variable. It uses normal distribution's quantile values from 5 answer groups which belong to the portion of the sample. This method is going to be used for the cross-section analysis.

Therefore, our life satisfaction in the cross-sectional analysis for 2016 would have the following probabilities: the probability that life satisfaction is equal to 1 is 8.3%, the probability that life satisfaction is equal to 2 is 41.71%, the probability that life satisfaction is equal to 3 is 24.77%, the probability that life satisfaction is equal to 4 is 18.61%, the probability that life satisfaction is equal to 5 is 6.61%. While cumulative probabilities are the next: the probability that life satisfaction is equal to 1 is 8.3%, the cumulative probability that life satisfaction is less or equal to 2 is 50.01%, the cumulative probability that life satisfaction is less or equal to 3 is 74.79%, the cumulative probability that life satisfaction is less or equal to 4 is 93.39%, the cumulative probability that life satisfaction is less or equal to 5 is 100%.

At the next step we need to calculate z scores which are , , and .

Then we build other values for our dependent variable () using the formula for the conditional expectation of the fact that a z-score is in the range from one particular z-score to another that is aligned with a certain level of our dependent variable, so that:

(4)

(5)

(6)

* is probability density function and is cumulative density function that are used for normal distribution with zero mean and standard deviation equaling to one. This way we obtain the following values for our variable of interest:

Ordered probit estimation

As life satisfaction is an ordinal variable it is also possible to apply ordered response models that include probit or logit models which are used in a number of studies examining life satisfaction. These models' widespread application in sociology stems from McKelvey and Zavoina, (1957). These methods of estimation are frequently applied for qualitative variables. The fundamental concept of the ordered probit lies in the fact that there exists a latent continuous measure that encompasses the ordinal reactions detected by the observer. This variable is a linear mixture of some factors as well as an error term with zero mean and standard deviation equal to one (Della Lucia et al., 2013). The model looks like the following:

, where

Here, thresholds divide the true line into a number of sectors that relate to the different ordinal groups. This procedure can be applied to our case of life satisfaction which lies in the interval from 1(completely satisfied) to 5 (completely dissatisfied).

,

,

,

,

,

Where represent undisclosed thresholds that are uniquely selected to each respondent.

(10)



Where F () or c.d.f. is the cumulative density function of normal distribution.

Overall, we obtain the following:

Probit models are fitted by Maximum likelihood estimation (MLE). For this method we need log-likelihood function:

Then we need to obtain marginal effects with the object of illuminating the results of the fitted model. They are calculated as follows:

Endogeneity and IV approach

As there might be a problem of endogeneity connected with smoking and life satisfaction because people might smoke as they experience stress and not so satisfied with life. In other words, the obstacle of reverse causality might be present since those individuals who are less content with their life could embark upon smoking or it may be the case that those who are more content tend to stay away from cigarette smoking. If the endogeneity is not dealt with, it might bring about the reversed signs of the link between smoking and life satisfaction or it may occur that the magnitude of coefficients will be actually biased. Therefore, in order to handle the endogeneity in case of when adequate instruments are unavailable or rather weak there is an IV strategy recommended by Lewbel (2012) that is comparable with a traditional 2SLS procedure that might be helpful to handle this issue. His key idea lies in designing internal instruments that could be obtained from the multiplication of the residuals of auxiliary equations by exogenous variables that are present in the model extracting their mean but keeping in mind the heterogeneity problem. Moreover, some undetected parameters might influence one's contentment with living that for its part put imprint on the attitude toward smoking. This concept associated with the estimation of the cross-sectional model can be provided in the following form: variable life satisfaction

, where

, where

Where is life satisfaction level for a respondent , is a dummy for smoking for an individual , indicates unobserved factors for a person that might have an influence on smoking status and thus life satisfaction, whereas and represent some idiosyncratic disturbance terms and denotes a vector consisting of controls.

Due to the lack of conventional two-stage least squares instruments, the internal to the model instruments can be created as follows:

,

where and embodies a vector of the residuals generated from the first-stage regression of an endogenous factor (on every exogenous one. This way we get that the mean of all the instruments generated is equal to 0. Lewbel (2012) also notes that even a subset of variables and not all the exogenous ones from the regression can be chosen as the instruments for the endogenous one according to the rule )=0.

Moreover, residuals from the first-stage regression (15) should be heteroskedastic by Breusch-Pagan test for heteroskedasticity. (Lewbel, 2012)



4. Estimation and outcomes

Here I use the cardinalized life satisfaction from -1.84 which stands for completely satisfied to 1.95 which stands for completely dissatisfied in order to be able to apply the OLS estimation method to an ordinal variable like life satisfaction. As the initial scale of life satisfaction was in a descending order from 5 depicting completely dissatisfied to 1 indicting completely satisfied, the signs will be reversed when interpreting the effect of controls as well as that of smoking behavior on life satisfaction, however, the significance of coefficients will not be influenced.

Table 2. Cross-sectional analysis of the effect of smoking on satisfaction with life (Source: accounts received from STATA MP).

(dependent)=	OLS	Lewbel's IV
(smoker=1)	0.1537829*** (0.0186352)	0.1178387** (0.0482491)
(employed=1)	-0.1435859*** (0.018781)	-0.1423158*** (0.0188457)
(male=1)	-0.000975 (0.0163164)	0.0091054 (0.0203127)
(married=1)	-0.2156087 *** (0.0175526)	-0.2160102*** (0.0175611)
(health=1)	-0.2709325*** (0.0172101)	-0.2719905*** (0.0172575)
(bilcity=1)	0.0962182*** (0.0187123)	0.0985149*** (0.0188952)
(highereduc=1)	-0.0272654* (0.0165741)	-0.0320195* (0.017602)
(drink=1)	-0.0666965*** (0.0158815)	-0.0608084*** (0.0174298)
(stress=1)	0.3287299*** (0.0243566)	0.3303645*** (0.0244292)
	-0.322349*** (0.0139562)	-0.323258*** (0.014028)
	-0.0649391*** (0.0084505)	-0.0650733*** (0.0084559)
	0.052919*** (0.0025771)	0.0535549*** (0.0027152)
	-0.0004789*** (0.0000264)	-0.0004862*** (0.0000281)
	2.206326*** (0.1404211)	2.207982*** (0.1405678)

For the Table 2: ***significant at 1 %, **significant at 5%, *significant at 10% and robust standard errors are given in parentheses. Note: the signs are reversed when interpreting the effect of variables on life satisfaction, screenshots of estimation results are given in Appendix 2-3.

Figure 1. Correlation matrix (Source: calculated in STATA MP).

According to the Figure 1 given above, the variables taken for the analysis are not highly correlated. Though and depict a rather high correlation between them, that is not an obstacle for us as should be taken into consideration in order to account for its link with our dependent variable

Considering the OLS estimation, the majority of the variables turned out to be of the expected sign and most of them proved to be significant. Particularly, the significant impact of smoking on life satisfaction was obtained and it occurred to be negative as it was anticipated. Turning to the controls, we obtain the following results: the positive significant influence of employment on satisfaction with life was received as expected. As for the dummy, its link with life satisfaction also proved to be positive and significant. Those who are in good or excellent health (health=1) have a higher satisfaction with life compared to those who are in average, poor or very poor health, which is consistent with our expectations. Moreover, males have a higher satisfaction with life in comparison with females, but the effect of this dummy on life satisfaction occurred to be insignificant. Furthermore, those who live in a large city with population 1000000 proved to be not so content with their life in contrast to those who reside in smaller types of settlement, so the effect of dummy is negative as well as significant. The impact of dummy for on satisfaction with life is negative as expected and significant. What is more, the logarithm of income per head in the household positively affects satisfaction with life and it is significant. Also, the influence of the number of children on being satisfied with life is positive and significant as anticipated. Turning to age it proved to negatively affect life satisfaction, whereas squared age has a positive impact on satisfaction with life, both occurred to be significant, so we can assume that there is U-shaped connection between age and life satisfaction. Dummy for those who have been drinking alcoholic beverages for the past 30 days proved to positively impact being satisfied with life and to be significant which is not as expected but that may be due to the fact that it includes both those who are heavy drinkers and those who drink from time to time and the effect of light drinkers prevails. Also, the positive effect of drinking on satisfaction with life is consistent with the research conducted by Oksanen and Kokkonen (2016).

Lewbel's IV approach base on the generation of internal instruments was applied so as to overcome the presumable threat of endogeneity in the OLS estimation. For Lewbel's IV the following set of instruments for were created: , , , , , where the controls were defined earlier in this work, while is residuals received from regressing on all the exogenous variables except for from OLS. Apart from that these instruments were selected so that , where is a vector of control variables and are the residuals calculated from OLS. This relation can be seen in Figure 2 given below.

Figure 2. Correlation of with the controls (Source: received from STATA MP)

Then we should be sure that the residuals from first-stage regression () are heteroskedastic. This is done by performing Breusch-Pagan test for heteroskedasticity with the null hypothesis of the constant variance versus the alternative hypothesis of heteroskedasticity.

Figure 3. Breusch-Pagan test for heteroskedasticity (Source: received from STATA MP)

So, as p-value=0 we reject H0 in favor of H1 and get that there is heteroskedasticity presence in the first-stage regression. This is one of the conditions that should be sustained according the Lewbel's 2SLS method in case of the absence of available or quite weak instruments.

Then standard Two-stage least squares (2SLS) procedure was applied and the results for IV were received. As can be seen from the given estimation results, the OLS actually overestimates the coefficient of the dummy the IV. Such an outcome might stem from the occurrence of endogeneity in the OLS estimation that produces such an overstatement of this statistical outcome. All the coefficients of other variables just slightly change, however, the signs stay the same as well as the significance of these controls. Then we turn to applying Sargan-Hansen test for overidentification of restrictions from the first-stage regression with H0: overidentifying restrictions are valid. As p-value is greater than 10% significance level, we can conclude that our internally designed instruments for are not overidentified.

Figure 4. Sargan-Hansen test for overidentification of restrictions (Source: received from STATA MP)

Also, I applied a test with H0 the instruments are weak. As minimum eigenvalue exceeds the critical value even at 5% that means that we reject H0 in favour of Ha indicating that there is no weak instrument problem in internally generated instruments. Moreover, F statistic exceeds 10, which also indicates that the chosen instruments are relevant.

Figure 5. Test for weak instruments (Source: received from STATA MP)



Given that the applied instruments are valid, then we may conduct Hausman test to choose between OLS and IV with H0: OLS estimates are consistent versus Ha: IV estimates are consistent.

Figure 6. Test of endogeneity (Source: received from STATA MP)

As we obtain p-value greater than 10%, then we do not reject H0 and have enough statistical evidence to conclude that there is no big difference between OLS and IV estimates, so we should prefer OLS as it gives more efficient estimates. However, this test actually has its limitations, so we cannot be totally sure that OLS will actually be a better model.

The next model to be estimated on the cross-sectional data is the ordered probit model. Here, we do not cardinalize life satisfaction and just exploit its ordered nature. This technique of estimation is applied in order to investigate whether the previous outcomes of Lewbel's 2SLS and OLS approaches are actually robust. We should notice that life satisfaction is in a descending order from 1 that states for fully satisfied to 5 that indicates fully dissatisfied, so the signs are reversed when we interpret the effect of these variables on life satisfaction.

Table 3. The ordered probit estimation of life satisfaction (Source: received from STATA MP).

(dependent)
(smoker=1)	0.1879904*** (0.0227173)
(employed=1)	-0.1724763*** (0.0227718)
(male=1)	-0.0011105 (0.0199488)
(married=1)	-0.2609177*** (0.02132)
(health=1)	-0.3309871*** (0.021145)
(bilcity=1)	0.1196378*** (0.022861)
(highereduc=1)	-0.0331084 (0.0202127)
(drink=1)	-0.080855*** (0.0193341)
(stress=1)	0.3937728*** (0.029433)
	-0.3952594*** (0.0175481)
	-0.0781972*** (0.0102643)
	0.0640575*** (0.0031726)
	-0.0005791*** (0.0000324)

For the Table 3: ***significant at 1 %, **significant at 5%, *significant at 10% and robust standard errors are given in parentheses. Note: the signs are reversed when interpreting the effect of variables on life satisfaction, screenshots of estimation results are provided in Appendix 4.

...