Effects of Self-Efficacy on Individual Earnings: Study in the Thai Labour Market

Transcription

1 Effects of Self-Efficacy on Individual Earnings: Study in the Thai Labour Market Abstract This study estimated the earning differences in the Thai labor market due to workers non-cognitive skills, a psychological characteristic of self-efficacy. Self-efficacy is defined as a person s belief about his or her ability to organize and execute courses of action necessary to achieve a goal. The Thai Mental Health Survey provided both basic socioeconomic factors and psychological indicators of self-efficacy and affective state, drawn from 15 items of the Thai Mental Health Indicator. The effects of self-efficacy on individual annual earnings were estimated from various proxies for self-efficacy, using both OLS and IV techniques. From the selected sample, the study confirmed that the Thai labor market valued this worker s characteristic, even though it was endogenous. Self-efficacy was independent from gender, religion and marital status, and its effects on earnings were unrelated to years of schooling or work experience. Omission of self-efficacy proved underestimation of earnings in sub-occupations, underestimation of female earnings, and overestimation of earnings in the rural areas. Self-efficacy probably contributed to extra earnings beyond wages or salaries, such as in-cash overtime and bonuses, and other welfare more than the traditional human capitals did. This study also made brief policy recommendations to improve self-efficacy and recommendations for further studies. Keywords: self-efficacy, earning differences, non-cognitive skills, Thai Mental Health Survey 1

2 Effects of Self-Efficacy on Individual Earnings: Study in the Thai Labour Market Introduction In the traditional thought of labour economics, it is believed that earnings differences result from individual cognitive skills or how smart people are. The differences are explained by the human capital theory. People accumulate their skills through formal education in schooling and through the individual s work experiences. These skills increase the workers productivity and hence their earnings. However, since the 1970s a number of studies on earnings differences have highlighted the importance of personal psychological characteristics, such as behaviour, personality, and attitude. These characters, collectively called non-cognitive skills, have been being questioned concerning whether they have effects on both economic and social outcomes. Furthermore, if they do have effects, can they be improved or invested and how? The following are examples of research on the effect of psychological characteristics on economic outcomes. Andrisani (1977) examined the influences of locus of control 1 (LOC) on six different outcomes in the American labour market; namely, occupational attainment, average hourly earnings, annual earnings, occupational advancement, growth of average hourly earnings and growth of annual earnings. The study found that LOC significantly influences all outcomes but growth of annual earnings among young men, and all outcomes but growth of average hourly earnings among middle-aged men. These influences were independent from skills, abilities and demographic characteristics. Goldsmith, Veum and Darity (1997) examined the influences of self-esteem 2 on monthly wage. Based on the U.S. National Longitudinal Survey for Youth (NLSY), the study found that both wages and selfesteem were determined simultaneously by the traditional human capitals, e.g. years of completed education and weeks of work experience. Interestingly, individual wages were more responsive to changes in self-esteem than changes in human capital. Semykina and Linz (2007) investigated how LOC and the need for challenge or affiliation 3 (C-A) explained the gender wage gap in Russia. Based on surveys from workplaces in eight cities between 2000 and 2003, the study confirmed that the Russian labour market rewards those that are internal 4 and exhibit preference for challenge, particularly women. Controlling for personality traits also reduced the unexplained part in the gender wage gap. Furthermore, Heckman, Stixrud and Urzua (2006) confirmed that both cognitive and non-cognitive skills do co-determine economic and social outcomes, for example, wages, probabilities of college graduation, and probabilities of employment, as well as the probability of wrongdoing such as involvement in illegal activities, being imprisoned, and teenage pregnancy. Heckman (2008: 296) commended that as is intuitively obvious and commonsensical, much more than smarts is required for success in life. In the case of Thailand, however, studies on the outcomes of non-cognitive skills in the Thai labour market are very rare. Almost all of them are case studies of the work 1 Previously called the internal-external attitude, locus of control was defined by Julian Rotter as an individual s perception of events as determined internally by own behavior vs. fate, luck or external circumstances (Rotter, 1966). 2 Self-esteem is defined by Morris Rosenberg as a favorable or unfavorable attitude toward oneself or simply a sense of personal worth (Rosenberg, 1965). 3 The need for challenge is linked to getting ahead while the need for affiliation is linked to getting along (Linz and Semykina, 2009: 72). 4 From locus of control, those that believe themselves to be masters of their own lives are called internal and those that do not are external (Rotter, 1966). 2

3 performance of school teachers and university lecturers. Sukin (2008) investigated the teaching performance of engineering lecturers at twelve leading Thai universities, and found that the psychological characteristics of creative personality 5 and perceived control 6 influenced their effective teaching behaviour. Boonprakob and Boonprakob (2007) explored the working behaviour of the government officials that had retired from Srinakharinwirot University in 2003 and found that the retirees with higher self-efficacy continued working as they were. Mohan (2007) examined the job well-being of the teachers at international schools in Bangkok, and found that work locus of control directly affected the level of stress from work and indirectly affected job well-being significantly. It can be seen that non-cognitive skills are indeed as important as cognitive skills in determining economic outcomes. However, the knowledge of how non-cognitive skills influence the outcomes in the Thai labour market is still inadequate. From the Thai Mental Health Survey, this study identified the psychological characteristic of generalized selfefficacy and investigated its contribution to individual earnings. Self-Efficacy Introduced by psychologist Albert Bandura in 1977, self-efficacy is defined as a person s belief about his or her ability to organize and execute the courses of action necessary to achieve a goal (Bandura, 1997). Self-efficacy has a significant impact on personal goals and accomplishments, as it determines how people feel, think, motivate themselves and behave. Self-efficacy directly influences the level of effort and persistence that people demonstrate when facing obstacles. The stronger the perceived self-efficacy is, the more effort individuals make. Self-efficacy is also related to persistence, as Bandura (1977: 194) states that those who persist in subjectively threatening activities that are in fact relatively safe will gain corrective experiences that reinforce their sense of efficacy. Classification of Self-Efficacy Self-efficacy can be classified into three levels of generality of assessment (Bandura, 1997: 49). At the most specific level, perceived self-efficacy is rated for a particular performance under specific situations. For example, in the study done by Mavis (2001), the second year medical students were asked to rate their confidence in their abilities, such as the ability to conduct an interview with a female patient, the ability to perform an abdominal examination, and the ability to identify a heart murmur, priori to OSCE 7. At the intermediate level, perceived self-efficacy is rated for a class of performances of the same activity under a class of situations with the same conditions. For example, in the study done by Ayuppa and Kong (2010), the full-time employees at supermarkets and department stores were asked to rate their work competency through statements such as I feel confident that my skills and abilities equal or exceed those of my colleagues, and I feel that I am overqualified for the job I am doing. At the most general and global level, perceived self-efficacy is rated for performances without specifying the activities or the situations. For example, in the study done by Judge, Erez, Bono and Thoresen (2002), the participants were asked to respond to such statements When I make plans, I am certain I can make them work, and When I decide to do something, I go right to work on it. 5 Personality relates to independence, non-conformity, a wide set of interests, openness to new experiences, flexibility and risk taking (Sukin, 2008: 4). 6 Personality relates to participative decision making and job autonomy (Sukin, 2008: 5). 7 Objective Structured Clinical Examination 3

4 Sources of Self-Efficacy Self-efficacy is not a gifted trait. To recognize and realize their own ability, individuals must rely on information of self-knowledge. According to Bandura (1997), four sources of information are as follows. Enactive Mastery Experiences or Performance Accomplishments. A person s own experiences of successes and failures form his or her expectations of his or her ability. Perceived ability may be delivered in upcoming situations, which may be similar to or substantially different from past experiences. High expectations of one s own ability are developed through repeated success of a behavior while low expectations result from failures Vicarious Experience or Social Modeling. By observing other performances of challenging or threatening activities without adverse consequences, individuals start to compare the similar attributes in others and form expectations of their own ability. Without one s own direct experiences, individuals are still able to calculate which activities are achievable and how much effort is needed. The more similar the attributes are, the higher are the expectations formed. Verbal Persuasion or Social Persuasion. Individuals that are verbally persuaded that they are capable of achieving given activities are likely to mobilize greater effort and sustain it than those that doubt their own ability. However, verbal persuasion is considered as a weaker inducer of efficacy when one s own experiences of past failures dominate self-belief. Physiological and Affective States or Emotional Arousals. An individual s expectations of his or her own ability can be distorted by negative emotional arousals such as stress or fear and physiological state of fatigue since they directly decrease performance and lead to other avoiding behaviors. The more sensitive to these arousals a person is, the lower are the expectations that result. Consequently, by training or practicing to cope with stress and fear, individuals become less sensitive and can maintain their performances. Relationship between Effort, Self-Efficacy, and Affective States Self-efficacy is the worker s psychological characteristic that signals the skills or abilities in providing individual effort. Individual effort is as important as the worker s cognitive skills or abilities, since the effort contributes to the worker s productivity. Once productivity rises, higher wages can be paid. However, previous research on the relationship between mental states or emotions and labour market outcomes has concluded that positive or desired mental states have direct effects on the outcomes without mentioning how the effort was induced. For example, Mohanty (2009: 884) examined the effects of positive attitude on happiness and wages and found that happiness also affects the worker s earnings both directly and indirectly. Based on two psychological theories, this section explains why mental states or emotions do not directly influence labour market outcomes, but rather influence them indirectly through improving self-efficacy, inducing effort, and hence raising productivity and wages. To begin with, Goldsmith, Veum and Darity (2000: 111) acknowledged that the level of effort (E) depends on the level or strength of individual motivation (T). The individual effort function can be defined as E = e(t), for 0 E 1. According to Atkinson s theory of achievement motivation, the strength of motivation is influenced by both pre-determined and situational factors. First, the motives to achieve success (M) are the existing characteristics of an individual in approaching or avoiding certain behaviours (Dunifon and Duncan, 1998: 33), and these are pre-determined and continue from one situation to another (Atkinson, 1964: 242). The other two are situational factors. The strength of expectancy or probability of success (P) is how strongly individuals believe in the situation they are confronting that their performances will lead to the goal. 4

5 Lastly, the incentive value of success (I) is how attractive the success appears to them in a particular situation (Atkinson, 1964: 241). Atkinson (1964: 242) described this relationship in a multiplicative form as T = M P I. Hence the individual motivation function is defined as T = t(m, P, I(R)), (1) where R is the goal, e.g. a vector of the foreseen rewards which will be awarded at the end of an activity. Furthermore, performance expectation and outcome expectation are not the same, according to Bandura s theory of self-efficacy. Bandura explained that expectations influence action focused almost exclusively on outcome expectations (1997: 19), and an outcome is the consequence of an act, not the act itself (2002: 94). Figure 1 illustrates Bandura s explanation Person Behaviour Outcome Efficacy Expectations Outcome Expectations Figure 1 Diagrammatic Representation of the Difference between Efficacy Expectations and Outcome Expectations (Bandura, 1977: 193). Therefore, the belief that performances will lead to the goal or expectancy (the strength of expectancy in Atkinson s theory of achievement motivation) is influenced by both the belief concerning whether one can perform required actions or perceived self-efficacy (S), and the belief as to whether the actions will lead to the desired outcomes or outcome expectation (O). Efficacy belief accounts for most of the variance in expected outcomes when outcome is determined by one s own performance (Bandura, 1997: 24), not by fate, luck, external circumstances or unknown factors; therefore, the individual motivation function and individual effort function can be redefined as T = t(m, P(S, O), I(R)) and E = e(t(m, P(S, O), I(R))). (2) In explicit form, the two functions become T = T(M, S, O, I(R)) and E = E(M, S, O, I(R)). (3) Therefore, the strength of individual motivation and level of effort is determined by four factors: motive, the existing individual characteristic to approach or avoid certain behaviours; self-efficacy, the belief whether the person can perform the required actions; outcome expectation, the belief whether the actions will lead to desired outcomes or rewards; and incentive value, the attractiveness of the foreseen outcomes or rewards. Since the individuals with stronger self-efficacy provide more effort than those with weaker self-efficacy, the individual effort function constitutes a non-decreasing function of self-efficacy, E/ S 0. At this point mental states or mood characteristics have a direct effect neither on individual motivation nor level of effort. Recalling that one factor which determines self-efficacy is affective states (A) or emotional arousals, the individual effort function may be defined as E = E(M, S(A), O, I(R)). (4) Therefore, individual effort is an implicit function of affective states or emotions. Negative emotional arousals can distort the individual s expectations of his or her own ability; in other words, the positive or desired mental states strengthen the individual s perceived ability, and self-efficacy is also a non-decreasing function of affective states, S/ A 0, so that mental 5

6 states or emotions indirectly influence the individual s effort through changing self-efficacy as E/ A = E/ S S/ A. (5) The positive or desired mental states such as happiness are necessary but not sufficient to induce the effort. In other words, even if the individuals are happy but they weakly believe in their own ability, they are likely to provide less effort, less productivity, and earn a smaller wage than those with a stronger belief in their own ability. Therefore, affective states or emotions are a proxy for self-efficacy, while self-efficacy is a proxy for effort, productivity and wages. What Should Determine Affective States and Emotions? Even though economic research on affective states or emotions is rare, research on happiness is more common. Since happiness is a basic emotion and also in the studies done by economist Richard Easterlin, the terms happiness, subjective well-being, satisfaction, utility, well-being, and welfare [were used] interchangeably (Easterlin, 2001: 465), this study assumed that the factors that determine happiness should also influence affective states or emotions similarly. According to Easterlin (2001: 466), in most people s lives everywhere the dominant concerns are making a living, family life, and health, and it is these concerns that ordinarily determine how happy people feel. The same paper also cited The Pattern of Human Concerns (Cantril, 1965 quoted in Easterlin, 2001: ), which suggests that material circumstances, especially level of living are of the most concern in people s lives about three quarters, followed by family concerns such as a happy family life and good relations with children and relatives about a half, then one s personal or family health about one third, and matters relating to one s work and to personal character equally around one fifth of the population. Besides total household income, Easterlin (2001) found that socioeconomic circumstances, such as education and unemployment, which influence income, determine an individual s happiness. In addition, Easterlin (2003a, 2003b) found that disability and disable conditions, health conditions and ageing, and family structures such as marriage and children, also determine happiness. Hypothesis The following hypothesis to be tested in this study is: Holding other factors fixed, individuals with stronger self-efficacy provide more effort and earn more than those with weaker self-efficacy. Empirical Procedures Data and Variables The Thai Mental Health Survey (MHS) was conducted for the first time by the National Statistic Office of Thailand (NSO) in During the routine Socio-Economic Survey (SES), one member of each household was randomly selected to reply to 15 questionnaire items of the Thai Mental Health Indicator (TMHI-15), resulting in a nationwide sample of 81,019 observations. Therefore the data provide both basic socio-economic factors as well as useful psychological variables. TMHI-15 is a short version of a complete version of 55 items (TMHI-55), both of which were developed in 2007 by the Department of Mental Health, the Ministry of Public Health (Mongkol et al., 2009). Psychological Variables. TMHI-15 consists of 15 items in which the subjects are to assess their mental health condition in four domains mental state, mental capacity, mental quality, and supporting factors. Responses are based on a 4-point Likert-type scale ranging 6

7 from 0 (No) to 3 (Very Much). From the mental capacity domain, this study identified three items; namely, Do you accept hard-to-solve problems (when a problem occurs)?, Are you confident in controlling yourself in bad or serious situations?, and Are you confident in facing extremely bad situations in your life?, revealing the psychological characteristic of generalized self-efficacy. A self-efficacy index (S index ), which is a proxy for generalized selfefficacy, was created from a linear combination of these three items. Therefore the S index ranged between 0 and 9. An affective state index (A index ), which is a proxy for momentary emotion, was also created from a linear combination of five items; namely, Are you happy with your life?, Do you feel relaxed?, Do you feel bored with your daily life?, Do you feel disappointed with yourself?, and Are you depressed? from the mental state domain. 8 Therefore the A index ranged between 0 and 15. Dependent Variables. This study adopted three different scopes for annual earnings as labour market outcomes of interest. The narrowest Earn1 is a maximum value between wages or salaries (in cash) received in the previous month multiplied by 12 and the wages or salaries (in cash) received in the past 12 months. Earn2 is equivalent to Earn1 plus overtime, bonus and others (in cash) received in the past 12 months. The broadest Earn3 is equivalent to Earn2 plus the total value of welfare received in the past 12 months. This paper mainly uses Earn1 unless specified otherwise. Demographic Variables. Gender was assigned as a dummy variable, 0 for male and 1 for female. Marital status was assigned as a dummy variable of four categories never married, married, widowed, and split-up (including divorced, separated, and unknown marital status). Religion was assigned as a dummy variable for three categories Buddhist, Muslim, and Christian. Geographical Variables. Each location was assigned two sets of dummy variables. The first one was for the area of residence, which fell into urban (municipal areas) and rural (non-municipal areas). The second one was for the country s region, which fell into five regions: the central (25 provinces), the north (17 provinces), the northeast (19 provinces), the south (14 provinces), and Bangkok. Traditional Human Capital Variables. This study converted the grade of completed schooling to the year of completed schooling. 9 Work experience was calculated from the present age less six years of pre-schooling age and years of completed schooling. Those samples obtaining a negative value of work experience were discarded. Work Characteristics. Work status was categorized into 14 groups 10 seven for economically active and the other seven for economically inactive. Business or industry was categorized into 17 types 11 based on the International Standard Industrial Classification of All Economic Activities (ISIC: Revision 3). Occupation was recorded as a three-digit code of ten 8 Mental state domain is made of two subdomains general well-being positive and negative affects. Positive affect ranges from 0 to 3 while negative affect ranges from 3 to 0. 9 Primary as 6 years, lower secondary or vocational as 9 years, upper secondary or vocational as 12 years, post-secondary or equivalent as 14 years, higher vocational or equivalent as 15 years, bachelor, master and doctoral degrees as 16, 18 and 21 years respectively, unless specified otherwise 10 Economically active employer, own-account worker, contributing family worker, government employee, state enterprise employee, private company employee, and member of producers cooperative, economically inactive housewife, students, elderly person, illness and disabled person, looking for a job, unemployed, and others. 11 See Table 1 7

8 major groups 12, based on the International Standard Classification of Occupations (ISCO-88). Based on the first two digits, occupation was reorganized with a two-digit code, reduced from 114 to 28 categories 13. This increased the number of observations in each occupational category while occupational variation was maintained. All work status, business/industry and occupation were assigned as dummy variables. Mastery Experiences. Mastery experiences were used to pre-estimate self-efficacy. Mastery experiences included the years of completed schooling and work experience, work status, business or industry, and occupation (in ten major groups). Physiological States. Physiological states were used to pre-estimate self-efficacy. Physiological states consisted of two sets of dummy variables disability and disable conditions. The first variable was whether one was non-disabled, disabled at birth, or disabled later. The second variable was whether one can both take care of him/herself and go out without assistance; one can only take care of him/herself without assistance; one can only go out without assistance; or one can neither take care of him/herself nor go out without assistance. Exclusion Restrictions for Affective State. Exclusion restrictions for affective state were used to pre-estimate self-efficacy, when affective state was treated as an endogenous factor. This study primarily followed Cantril (1965 quoted in Easterlin, 2001) and oragnised the variabls into three sub-groups. First, material circumstances or household economic conditions consisted of ten sets of dummy variables four 14 for current household economic status and six 15 for present household financial problems. Secondly, personal health was approximated only from ageing or birth cohort 16 since disability and disable conditions were already oragnised into physiological states. Lastly, apart from marital status, family concerns were approximated from three items 17 of family structure and the health of family members. Wage Equation and Selected Sample This study primarily followed Mincer s wage equation with an estimation method of ordinary least squares (OLS) technique. Besides the traditional human capitals, the additional controlled variables included demographic variables, geographical variables, and work 12 Armed forces (Occ0), legislators, senior officials and managers (Occ1), professionals (Occ2), technicians and associate professionals (Occ3), clerks (Occ4), service workers, shop and market sale workers (Occ5), skilled agricultural and fishery workers (Occ6), craft and related trades workers (Occ7), plant and machine operators and assemblers (Occ8), elementary occupations (Occ9) 13 See Table 1 14 First, household socio-economic status was categorized into 20 groups. The next two were the values of household physical properties and financial assets, each of which was assigned a dummy variable of nine categories. Fourthly, the average monthly total income per household was assigned as a dummy variable of eleven categories. 15 The household debt problems were binary dummy variables responses to the six following situations; Had many problems (over 3 months) paying house rent, water/electricity rates or school fees, Could not borrow money for operating business or farm, Could not borrow money for emergency payment, Had debt at present, Had borrowed or owed any bills from the formal sector (previous month), and Had borrowed or owed any bills from the informal sector (previous month). 16 A dummy variable of nine sub-categories of a ten-year gap, starting from years to years. 17 The number of children age less than 15 years old, the number of elderly greater than 60 years old, and the numbers of disabled persons living in the same house. 8

9 characteristics. Due to the lack of availability, the controlled psychological characteristic was only self-efficacy, while the influences of motives, outcome expectation, and incentive value were included in the error term. They are unrelated theoretically. The effect of self-efficacy on earnings was investigated through a comparison of an inclusive model, in which selfefficacy was controlled, with an exclusive model, in which self-efficacy was omitted and its effect was embedded in an error. Hence the size of the effect and the model s explanatory power were estimated. The exclusive and inclusive models are specified as W = β D D + β G G + β H H + β F F + u, (6) W = β D D + β G G + β H H + β F F + β S Ŝ + v, (7) where W is logarithmic annual earnings; D, G, H, F are the vectors of demographic variables, geographical variables, traditional human capitals, and work characteristics, while β s are the corresponding coefficient vectors; Ŝ and β S are the proxy for self-efficacy and its coefficient; and u and v are error terms. This study expected a positive and statistically-significant value of β S. Since this study hypothesizes that self-efficacy positively contributes to individual earnings through intense effort and increasing productivity, the factors which may misrepresent this contribution have been removed. First, the study discarded all classes of disabled persons. Even though self-efficacy positively contributes to their productivity, their disabled conditions are still highly likely to limit their productivity. So keeping disabled persons in the same sample with non-disabled persons would result in misinterpretation. Secondly, the study discarded all those that changed their jobs or that had second jobs. According to Bandura (1997: ), self-efficacy influences how individuals interpret situations, anticipate scenarios, and visualize the futures they construct. Those with strong efficacy view the situations as realizable opportunities, while the weak may not or find it difficult. Therefore if other things were equal, those with strong efficacy are likely to realize better opportunities from the new better jobs and decide to move; or realize better opportunities from extra income or more challenging tasks and have second jobs, while the weak are likely to hesitate to change their jobs or have second jobs. However, there are many reasons why people decide to change jobs or have second jobs. These reasons cannot be totally controlled and had to be treated as unobservable factors. So keeping those that changed their jobs or that had second jobs in the same sample with those that remained in one single job would also result in misinterpretation. Consequently, the empirical analysis was limited to only earnings from wages and salaries of the non-disabled persons that continue working with only a single employer from the three sectors; namely, the government sector, state enterprise, and the private sector, in the past 12 months without a second job. Hence the size of the selected sample was reduced from the original 81,019 to 18,913 observations. Estimation of Self-Efficacy When estimating the coefficients in equation (7) with the OLS technique, self-efficacy must be assumed to be exogenous, i.e. self-efficacy must be uncorrelated with the error term (Wooldridge, 2006: 838). However, the literature insists otherwise. Bandura (1997) states clearly that self-efficacy is not an inborn trait and already highlights four sources of information on self-efficacy; namely, enactive mastery experiences, social modelling, verbal persuasion, and physiological and affective states. The OLS technique then would be inconsistent and produce biased estimators due to the omitted variables, in which selfefficacy is correlated with other variables contained in an error term, i.e. the endogeneity problem. This study came up with several methods to generate the proxy for self-efficacy. 9

10 Estimation from the Selected Sample. In the beginning, self-efficacy was treated as an exogenous factor, named model A. The self-efficacy index was directly substituted into equation (7) as Ŝ = S index. (8) However, the estimated effect of self-efficacy on earnings should not be accurate because this estimation of self-efficacy is not consistent with the literature, as previously explained. Alternatively, the coefficients in equation (7) were re-estimated using an instrumental variable or IV technique for endogenous self-efficacy, named model B. Since MHS provides information on self-efficacy: enactive mastery experiences, physiological states, and affective state, the new proxy for self-efficacy can be pre-estimated from these variables. The new proxy is no longer correlated with the error term and produce unbiased estimators, i.e. the unbiased coefficients in equation (7). With the selected sample, however, some constraints exist. This technique cannot adopt years of schooling, work experience or work characteristics as the exclusion restriction of the mastery-experience element since all already appear in equation (7). This technique cannot adopt disabled condition as the exclusion restriction of the physiological-state element either since the selected sample was limited to only the non-disabled. From five questionnaire items of TMHI-15, this study finally generated and adopted the affective state index as the exclusion restriction of the affective-state element. In model B, the first-stage regression to pre-estimate variable Ŝ in equation (7) can be revealed as Ŝ = α D D + α G G + α H H + α F F + α A A index, (9) where Ŝ is expected self-efficacy; D, G, H, F are the vectors of demographic variables, geographical variables, traditional human capitals, and work characteristics similar to those in equation (7), while α s are the corresponding coefficient vectors; A index and α A are the affective state index and its coefficient. In the second-stage or the main regression equation (7), the proxy for self-efficacy was replaced by the expected self-efficacy from equation (9). The re-estimated effect of self-efficacy on earnings should be more accurate than the OLS since this estimation is consistent with the literature. Pre-estimation from the Original Sample. The better proxy for self-efficacy can be drawn from the original sample of 81,019 observations. The selected sample for wage estimation has a limited number of items which can be adopted as the exclusion restriction, but the original sample provides all of the available sources of self-efficacy according to the literature. From the original sample, the proxy for self-efficacy can be pre-estimated from mastery experiences years of completed schooling and work experience, work status, business or industry, and occupation; physiological states disability and disable conditions; and affective state the affective state index, independently from the wage estimation and then substituted into equation (7). In model C, the proxy for self-efficacy was generated with an OLS technique from the original sample where affective state was treated as an exogenous factor. Close to equation (9), the regression to estimate the proxy for self-efficacy can be revealed as Ŝ = γ D D + γ G G + γ M M + γ P P + γ A A index, (10) where Ŝ is expected self-efficacy; D, G, M, P are the vectors of demographic variables, geographical variables, worker s mastery experiences, and physiological states, while γ s are the corresponding coefficient vectors; A index and γ A are the affective state index and its coefficient. However, this predicted value of self-efficacy directly from the affective state index should not be accurate, because the affective state itself is also endogenous. As with exogenous self-efficacy in model A, the OLS technique would be inconsistent due to omitted variables, where affective state is correlated to an error term. 10

11 In model D, another proxy for self-efficacy was generated with an IV technique, in which equation (10) was the main regression and affective state was treated as an endogenous factor. The exclusion restriction for the affective state included ten sets of dummy variables for current household economic status and present household financial problems, a dummy variable of birth cohort, and the numbers of children, elderly, and disabled persons living in the same house. The first-stage regression to pre-estimate variable A in equation (10) can be revealed as  = δ D D + δ G G + δ M M + δ P P + δ A A, (11) where  is expected affective state; D, G, M, P are the vectors of demographic variables, geographical variables, worker s mastery experiences, and physiological states similar to those in equation (10); A is the vector of the exclusion restriction for the affective state; and δ s are the corresponding coefficient vectors. In the second-stage equation (10), the affective state index was replaced by the expected affective state from equation (11). The predicted value of self-efficacy from the endogenous affective state should provide the most accurate estimated impact of self-efficacy on earnings, since the estimation of self-efficacy and affective state was consistent with the literature. Figure 2 illustrates how the four proxies for self-efficacy were estimated. Selected Sample (18,913 obs.) Model A: W = W(S Index ) Model B: W = W(S(A Index )) Model C & D: W = W(Ŝ) Original Sample (81,019 obs.) Ŝ = S(A Index ) Ŝ = S(A(A)) Figure 2 Estimation Techniques for the Proxy for Self-Efficacy Irrelationship between Individual Annual Earnings and Affective States. Even if the MHS has records of wages, salaries, overtime, and bonuses earned in the previous month, this study did not adopt them as the dependent variables intentionally. Instead, this study adopted the annual earnings in order to avoid the relationship between the dependent variable and the exclusion restriction, i.e. affective state. The following are the justifications why individual annual earnings and affective states are not related. First, research by psychologists has indicated that subjective well-being (SWB) is not a permanent characteristic. Unlike personality, which is relatively more stable, well-being keeps changing according to life events. Kennedy-Moore, Greenberg, Newman, and Stone (1992) and Egloff, Tausch, Kohlmann and Krohne (1995) found that the moods of the same person keeps changing across the time of the day and the day of the week. Moreover, Suh, Diener and Fujita (1996) reported that only life events during the previous 3 months influenced life satisfaction and positive and negative affect (1996: 1091), and events that had occurred more than 7 months previously failed to add significant increments to the prediction of current SWB level (1996: 1096). If the wage rate were raised in the previous month, the individual earnings in the previous month would definitely influence the worker s well-being. As a result, this study adopted individual annual earnings as the dependent variable. Secondly, Easterlin (2001, 2003a, 2003b) highlighted that happiness responded to household incomes. Based on basic economics, households own four different factors of 11

12 production and acquire four different returns from them. Households obtain rent from their land and property; obtain wages or salaries from their labour; obtain interest from their financial capital; and obtain profits if they were entrepreneurs (Tucker, 2003: 3-4). Moreover, each observation in the selected sample was of one and only member that was randomly selected from each household. Therefore, individual earnings from wages or salaries, which are a portion of the household income, were not related to affective states. Thirdly, workers do not consider only wages, salaries, or earnings when applying for a job. Besides the labour market outcomes from economists points of view, Bandura also suggested that the range of [labour market] outcomes include such things as salary, security, social status, freedom to exercise initiative and use one s special abilities, variety in work assignments, chance to learn new competencies, opportunity for advancement and leadership, congenial associates, and the social benefits of the particular line of work (1997: 426). This is supported by Frank (1984), as this study showed that some workers trade off some money for higher status in the workplace by accepting wages below their marginal products. This means that workers satisfaction with their job is not completely based on money and in-kind incomes. In this study, the selected sample was limited to only the workers that had been working continuously with the same employer for 12 months. If these workers voluntary stayed with their current jobs, this means that they were satisfied with the overall job attributes, not particularly wage, salary or earnings. If these workers were not satisfied with their current jobs, they would have changed their jobs. If they could not find a better job and had to stay with their current job, their dissatisfaction would be a result of the failure to find a new job, not a result of earnings from their current job. If these workers were not satisfied with their earnings from their current job and could not find a better one, they could have had another job. They might work as an employee with other employers or work as part-time self-employed. Since the selected sample excluded the workers that had more than one job, this study assumed that the workers in the sample were satisfied with the earnings from their current job. Therefore, in this selected sample, working at the same job for 12 months became a habit and changes in the worker s well-being resulted from other factors, not from individual annual earnings. Lastly, one may argue that when the workers earn more they will definitely be better off or feel happy. This is true when comparing the earnings of the same worker at different times, i.e. when using longitudinal or panel data in studying the relationship between earnings and well-being. However, this study used cross-sectional data in which affective state represented a measurement of emotion at a single point in time. Earnings differences across observations are different from worker to worker, not different at different times for the same worker, and cannot explain the well-being differences of different workers. Results To interpret the results, two coefficients years of schooling and work experience approximated the proportional additional earnings due to one more year of schooling or work experience (Wooldridge, 2006: 707), while the coefficients of the dummy demographic variables, geographical variables, and work characteristics approximated the proportional additional earnings differently from the reference groups (Wooldridge, 2006: 232). However, it is slightly complicated for generalized self-efficacy. The workers with the strongest selfefficacy should have replied to all three questions from the mental capacity domain with Very Much and scored 9, while those with the weakest self-efficacy should have replied to all of the same questions with No and scored 0 for the self-efficacy index. Hence, the coefficient of generalized self-efficacy approximates the proportional additional earnings due to selfefficacy improvement, one level out of nine. 12

13 Preliminary Study Table 1 confirms that the worker s generalized self-efficacy is rewarded in the Thai labour market, no matter how this variable is generated. Overall, the coefficients of selfefficacy remained statistically significant at 1%, even though demographic variables, geographical variables, and work characteristics were controlled along with the traditional human capitals. Between the exclusive model and all inclusive models, an inclusion of selfefficacy cannot improve the R 2, which was around An inclusion of self-efficacy does not alter the effects of the traditional human capitals on earnings either, as the coefficient of years of schooling was around and that of work experience was around This suggests that self-efficacy does not correlate with years of schooling or work experience and confirms that self-efficacy is rewarded independently from these two variables. However, self-efficacy may partially correlate with the additional controlled variables. Between the exclusive model and model A, in which the effect of exogenous selfefficacy on earnings was estimated, the coefficient of exogenous self-efficacy was estimated at 0.010, while an inclusion of self-efficacy barely made a change in the coefficients of the additional controlled variables. In model B, in which the effect of self-efficacy was estimated endogenously, the coefficient of endogenous self-efficacy rose to 0.108, while the coefficients of the additional controlled variables changed considerably. This confirms the partial correlations between self-efficacy and these variables. Moreover, when substituting the proxies for self-efficacy pre-estimated from the original sample, the results still support the endogeneity of self-efficacy. From model C, in which the proxy was pre-estimated from the exogenous affective state and model D, in which the proxy was pre-estimated from the affective state endogenously, the coefficient of endogenous self-efficacy rose further to and respectively. The coefficients of the additional controlled variables in these two models were similar but slightly different from those estimated in model B. Considering the additional controlled variables by groups, an inclusion of selfefficacy had diverse influences over the coefficients of these variables. An inclusion made no change in the coefficients of religion, marital status, area of residence (urban-rural), or employment sector, confirming that self-efficacy did not correlate with them. An inclusion improves the coefficients of female as they became less negative by 5-6 percentage points, underlining the underestimation of female earnings. An inclusion improved most of the coefficients of sub-occupations, as they became more positive or less negative, and were almost all statistically significant at 1%. Only the coefficients of the sub-occupation of physical, mathematical and engineering, science professionals (Occ21), life science and health professionals (Occ22), and teaching professionals (Occ23) barely changed. This also underlines the underestimation of earnings in the other major occupational groups and positive correlation between self-efficacy and occupations. All coefficients of business or industry also improved similarly but they were not statistically significant. An inclusion worsened all of the coefficients of region as they became more negative and all were statistically significant at 1%. This underlines the overestimation of earnings in non- Bangkok regions and negative correlation between self-efficacy and regions. 13

14 Table 1: Preliminary Study Dependent Variable: Log Earn1 a Model Exclusive Model A Model B Model C Model D β (1) (2) (3) (4) (5) (5)-(1) Efficacy *** *** *** *** Schooling *** *** *** *** *** Experience *** *** *** *** *** Experience *** *** *** *** *** Female *** *** *** *** *** Religion Muslim *** *** *** *** *** (Buddhist) Christian Marital Status Married *** *** *** *** *** (Single) Widow *** *** *** *** *** Break Up Rural *** *** ** *** *** Region Central *** *** *** *** *** (Bangkok) North *** *** *** *** *** Northeast *** *** *** *** *** South *** *** *** *** *** Employed Sector State enterprise *** *** *** *** *** (Government) Private company *** *** *** *** *** Business or Buss01, Agriculture, hunting and forestry * * Industry Buss02, Fishery (Extra-territorial Buss03, Mining and quarrying organization and Buss04, Manufacturing bodies, Buss17) Buss05, Electricity, gas and water supply Buss06, Construction Buss07, Wholesale and retail, repairing motor vehicles/ motorcycles/ personal goods/ household goods Buss08, Hotel and restaurant * * Buss09, Transport, storage and communication Buss10, Financial intermediate Buss11, Real estate, renting and business activities Buss12, Public administration and defence, compulsory social security * * Buss13, Education * * Buss14, Health and social work * * Buss15, Other activities related to community/ social/ personal service ** ** * * * Buss16, Private households with employed person ** ** * * *

15 Table 1 (Cont.) Model Exclusive Model A Model B Model C Model D β (1) (2) (3) (4) (5) (5)-(1) Sub-Occupation Occ11, Legislators and senior officials *** *** *** *** *** (Armed forces, Occ12, Corporate managers *** *** *** *** *** Occ01) Occ13, General managers *** *** *** *** *** Occ21, Physical, mathematical and engineering, science professionals *** *** *** *** *** Occ22, Life science and health professionals *** *** *** *** *** Occ23, Teaching professionals *** *** *** *** *** Occ24, Other professionals Occ31, Physical and engineering science associate professionals ** ** * ** ** Occ32, Life Science and health associate professionals *** *** *** *** *** Occ33, Teaching associate professionals *** *** *** *** *** Occ34, Other associate professionals * Occ41, Office clerks *** *** *** *** *** Occ42, Customer services clerks *** *** *** *** *** Occ51, Personal and protective services workers *** *** *** *** *** Occ52, Models, sales persons and demonstrators *** *** *** *** *** Occ61, Market-oriented skilled agricultural and fishery workers *** *** *** *** *** Occ62, Subsistence agricultural and fishery workers Occ71, Extraction and building trades workers *** *** *** *** *** Occ72, Metal, machinery and related trades workers *** *** *** *** *** Occ73, Precision, handicraft, printing and related trades workers *** *** *** *** *** Occ74, Other craft and related trades workers *** *** *** *** *** Occ81, Stationary-plant and related operators *** *** *** *** *** Occ82, Machine operators and assemblers *** *** *** *** *** Occ83, Drivers and mobile-plant operators *** *** *** *** *** Occ91, Sales and services elementary occupations *** *** *** *** *** Occ92, Agricultural, fishery and related laborers *** *** *** *** *** Occ93, Laborers in mining, construction, manufacturing and transport *** *** *** *** *** Constant *** *** *** *** *** R Adjusted R RMSE N 18,913 18,913 18,913 18,913 18,913 Note. *** p<0.01, ** p<0.05, * p<0.1; Sub-category in blanket is a baseline. a Earn1 - salaries (in cash) received in the past 12 months 15