A Study of the Relationship between Family Income and Worker Compensation Measured as Wage and Fringe Benefits

Transcription

1 A Study of the Relationship between Family Income and Worker Compensation Measured as Wage and Fringe Benefits Thesis Submitted to Kent State University Department of Economics in partial fulfillment of requirements for the Degree of Master of Arts By Eren Cifci July 2016 Thesis Committee: Dr. Emmanuel Dechenaux, Chair Dr. C. Lockwood Reynolds Dr. Donald R. Williams

2 DEDICATION I would like to dedicate this study to Alan Kurdi, and children who face the harm of wars in their early ages like Alan Kurdi did. Sometime, these children lose their life because of the devastations of wars. Alan Kurdi was a three-year-old child from Kurdish decent, Syria when he got drowned in the Mediterranean Sea on September 2, He had to leave his home to escape from the war. However, life was not very easy for him out of the war, either. So, he decided to leave. I also would like to dedicate this study to children who are being abused by unfavorable conditions and facing poverty in their early ages. i

3 ACKNOWLEDGMENTS I would like to thank my advisor Dr. Emmanuel Dechenaux for his help and encouragement during the process of this study. I also would like to thank Dr. C. Lockwood Reynolds for his useful comments and help during the writing and the data collection process of the thesis. Finally, I would like to thank Dr. Donald R. Williams for his useful comments along the way. ii

4 ABSTRACT In this study, I investigate the relationship between family (or parental) income and worker compensation. I consider compensation in the form of wage and fringe benefits. Most existing studies of intergenerational linkages focus solely on wage income. Therefore, I add to the literature by examining the relationship between parental income and a broader measure of compensation. The data used for the analysis are from the National Longitudinal Survey of Youth (NLSY-79). I report results separately for (i) the intergenerational income elasticity, (ii) the relationship between family income and the number of benefits provided by an individual s employer and (iii) the likelihood that each of nine specific fringe benefits is available to the individual. For much of the thesis, I consider a single year of compensation data (2008), but I also extend the analysis to a measure of long term compensation. Focusing solely on income, I find that the intergenerational income elasticity (IGE) is 0.313, which is broadly consistent with the literature. Using 2008 compensation data, the results show that the wage inequality across family income quartiles is substantial. Importantly, patterns that are found for wage income also hold for fringe benefits. This suggests that studies which focus only on wage income underestimate the role of family income on workers returns from the labor market. Compared to the results using a single year of income, the income inequality across family income quartiles increases slightly when an individual s long term income is used as a dependent variable. However, the opposite result is true for fringe benefits and the effect of family income on the availability of fringe benefits is less pronounced than in the single-year case. Finally, for both measures of income and fringe benefits single-year and long term the estimated effect of family income falls sharply when ability and education levels are controlled for. However even with these covariates, in most cases, family income remains a statistically significant determinant of worker compensation. iii

5 TABLE OF CONTENTS CHAPTER 1: INTRODUCTION... 1 CHAPTER 2: RELATED LITERATURE Intergenerational income mobility and family background characteristics Theoretical literature Empirical estimates of the intergenerational elasticity Recent studies Total compensation and fringe benefits Total compensation Studies focusing on specific fringe benefits Relationship between income and fringe benefits Intergenerational transmission of occupational status CHAPTER 3: HYPOTHESES Intergenerational transmission of income Facts behind Fringe Benefits CHAPTER 4: DATA AND EMPIRICAL MODELS Data used in the analysis Family income and other explanatory variables Compensation data for the year Permanent compensation data Empirical Models iv

6 4.2.1 Models for wage income and fringe benefits Specifications with four-year averages to capture permanent income CHAPTER 5: RESULTS Results using 2008 compensation data Wage income and total number of benefits Number of fringe benefits Specific fringe benefits Results using measures of permanent compensation Wage income Number of fringe benefits Limitations of the study CHAPTER 6: CONCLUSION REFERENCES APPENDIX OLS estimation results using 2008 data Probit estimation results for each fringe benefit using 2008 data OLS estimation results using four-year averages to capture permanent compensation v

7 LIST OF FIGURES Figure 1: Shares of total income by income quartile...25 vi

8 LIST OF TABLES Table 1: Some findings for the IGE as shown in Solon (1999)...8 Table 2: IGE for men and women in different countries...9 Table 3: Summary statistics for the family income quartiles...25 Table 4: Summary statistics for the explanatory variables...26 Table 5: Education levels of respondents by family income quartile...27 Table 6: Summary statistics of wage and fringe benefits using 2008 data...29 Table 7: Average wage and fringe benefits by income quartile using 2008 data...30 Table 8: Summary statistics of wage income and fringe benefits using four-year averages...31 Table 9: Average wage and fringe benefits by income quartile using four-year averages...32 Table 10: OLS estimation results using 2008 data without (top) and with (bottom) covariates.39 Table 11: Probit estimation results using 2008 data for each fringe benefit without (top) and with (bottom) covariates...43 Table 12: OLS estimation results using four-year averages without (top) and with (bottom) covariates...50 Table 13: OLS estimation results using 2008 data without covariates...60 Table 14: OLS estimation results using 2008 data with covariates...61 Table 15: Probit estimation results using 2008 data for each fringe benefit without covariates.63 Table 16: Probit estimation results using 2008 data for each fringe benefit with covariates...64 Table 17: OLS estimation results using four-year averages without covariates...66 Table 18: OLS estimation results using four-year averages with covariates...67 vii

9 CHAPTER 1: INTRODUCTION In current times, standards of living are much higher compared to a century ego. Nowadays a large fraction of the world population lives a relatively affluent life, with good access to healthcare and nutrition, high quality housing and more education opportunities, especially in developed countries (Easterlin, 2000). Concurrently, due to technological innovation, we are witnessing a transition from traditional blue collar jobs to jobs requiring a high level of specialized skills, which is referred to as skill-biased technological change (Goldin and Katz, 2007). Although current generations have been fortunate to experience such progress, the gains have not been equally distributed. This phenomenon has been studied by many researchers. Although some people can take advantage of the benefits afforded by economic and social progress, others still suffer from poverty. The major reason for this inequality is the inequality in income. While individuals with high incomes can afford to consume more goods and services as a result of development, individuals at the bottom of the income distribution do not gain as much. There are several reasons why income inequality exists. In this study, I focus on one aspect, namely how income levels are transmitted from one generation to the next. Overall, no one can control or choose what type of family they were born in. Accordingly, I investigate whether or not parental income creates some differences in individual incomes in the current generation. Moreover, if such differences do exist, I wish to measure how substantial they are. I propose to study these questions by analyzing a worker s compensation, which is made up not only of wage income but also fringe benefits. Many studies have investigated the relationship between parental characteristics and various children outcomes. While some studies seek to measure the Intergenerational Income Elasticity (IGE), others focus on the relationship between specific types of family characteristics 1

10 (parental education level, family structure, etc.) and the child s income (Black and Devereux, 2010). A small sample of this research includes Wilson and Wilson (1992) who investigate the influence of the environment on a child s desire towards education; Shea (2000) who investigates the effects of family income on human capital; Mendolia and Siminski (2002) who measure the effect of family background on income by applying mediation analysis. Recently, Duncan et al. (2010) look at the effect of parental poverty on child income, behavior, and health; Korupp et al. (2002) consider the relationship between occupational status and sex-typing focusing on a child s first job; Karagiannaki (2012) investigates the impact of parental wealth during early adulthood on a child s later outcomes; and Lovenheim and Reynolds (2013) measures the impact of home price on the type of college attended. The size of the literature focusing on the relationship between parental characteristics and a form of the child s outcome illustrates the significance of the topic. Thanks to many prior studies, we have learned quite valuable and applicable knowledge in this area. But while the current literature sheds light on our understanding of how family income and children outcomes correlate, in most cases, the available findings apply to child income. In particular, researchers have not considered broader measures of compensation that include fringe benefits. There exists a sizable literature with regard to comprehensive measures of compensation that cover both wage income and non-wage benefits from employers. Findings from this literature suggest that using only wage income as a proxy for compensation might underestimate the effect of family characteristics on a child s returns from the labor market. For instance, Duncan (1976) indicates that 13% of total compensation comes from non-wage benefits using the Quality of Employment Survey (QES) and this number is 8% using the Panel Study of Income Dynamics (PSID); the percentage is 9.9% in Smeeding s (1983) study using a combination of the Current Population Survey (CPS) and the Employment Cost Index (ECI), and is a somewhat surprising 40% in Solberg and Laughlin (1995) who use the 1991 data from the NLSY. More recently, using the CPS and the Chamber of commerce s Employee Benefits Survey (EBS), Chung (2003) suggests that the fraction of non-wage benefits is 16.2%; and Pierce (2010) reports 26% using a combination of CPS and ECI data. The point of citing these numbers is to provide evidence that fringe benefits can account for a substantial fraction of a worker s compensation. Also, we can see that from these studies the share of non-pecuniary earnings in the total compensation seems to increase over time. There are also some studies (e.g. 2

11 Benedict and Shaw, 1995, Levy, 2006) that focus on specific types of benefits (pension, health insurance) rather than total compensation. Importantly, when these researchers include the benefits as an addition to wage income, the inequality in the distribution of compensation across workers increase compared to if only wage income is used. Therefore, in the present study, to better capture the effect of family income on compensation, I consider both wage income and fringe (or non-wage ) benefits from the job. This is the contribution of my study. First, I find a positive relationship between family income and child compensation measured as wage income. Then I ask: do we observe a similar relationship when it comes to employer-provided fringe benefits? The results show that although the relationship between family income and benefits are smaller than the relationship between family income and wage income, it is still positive and statistically significant. Therefore it appears that family income level does influence an individual s full compensation beyond just income. I find that the differences in fringe benefits across family income quartiles mostly come from life insurance, health insurance, retirement, maternity/paternity leave and training and educational opportunities. On average, individuals who come from the top income category families have the highest wage income and highest levels of fringe benefits, while individuals whose families are in the bottom income category have the lowest levels in both dimensions of compensation. Compared to the results using a single year of income, the income inequality across family income quartiles increases slightly when an individual s long term income is used as a dependent variable. However, the opposite result is true for fringe benefits and the effect of family income on the availability of fringe benefits is less pronounced than in the single-year case. Finally, for both measures of income and fringe benefits single-year and long term the estimated effect of family income falls sharply when ability and education levels are controlled for. However even with these covariates, in most cases, family income remains a statistically significant determinant of worker compensation. Regarding the structure of the thesis, Chapter 2 provides an overview of the literature both on the IGE and on the existence of inequalities in fringe benefits coverage. Chapter 3 describes the hypotheses and Chapter 4 covers the data and empirical models; while Chapter 5 covers the results. In Chapter 6, I offer concluding remarks. 3

12 CHAPTER 2: RELATED LITERATURE 2.1 Intergenerational income mobility and family background characteristics While some studies measure inequality between children in terms of the IGE, some measure it by other family background characteristics such as parental education level, neighborhood, etc. Considering the similarities between the purposes of these studies, I devote part of my research for mentioning the studies related to both IGE and effect of other family background characteristics on earning. Not only the ones directly related to the measuring of inequality by IGE but also the ones related to measuring the inequality between children regarding the differences in background or other family characteristics Theoretical literature Becker and Tomes (1979) lead us to understand the importance of the intergenerational income correlation and they set up a basic model to estimate the correlation. Their studies open a door for understanding the intergenerational transmission of inequality and intergenerational mobility. Most of the recent studies about the intergenerational mobility are based on Becker and Tomes (1979) ideas. In their 1979 study, the authors assume that parents distribute their income between consumption and investment in their children, and family utility depends on their consumption and the children s income in the next period. The authors also consider family endowment to children, race, ability and other characteristics. Therefore, investment in children depends on family income, family endowment, and family propensity to invest. Even if all families are identical, because of luck, some children are getting more investment which also leads to more inequality in earnings. Another mechanism which affects the inequality is government intervention. Government intervention is a potential cause way to decrease inequality by taxes and subsidies, but it may also increase the inequality by inducing some parents not to invest in 4

13 their children since the rate of return of investing would go down after taxations. It is also indicated that the degree of IGE decreases and disappears in the later generations. Becker and Tomes s (1994) study is based on their 1979 study. The relationship between children income and parental income is shown following Markov model. (1) where is the child s income, is the parental income and is the part that is not explained by parental income, is constant and shows the degree of relationship between parental income and children income. The biggest difference between this model and the recent IGE models like Black and Devereux (2010) is that the log of income variable has been used instead of income itself, which measures IGE in addition to showing only the relationship. In my study, I also use log income model as a starting point of my analysis. Authors assume that parents invest children when the marginal rate return to invest in children is high. Parents borrow to invest in their children. Poorer families have difficulties to borrow. A limited resource for poorer families causes them not to invest human capital of their children. Therefore, children from wealthier families have a greater chance of access to a high level of education and technologies. In their study, it is also stated that intergenerational mobility in earnings is negatively correlated with the number of children in the families. Assuming families invest in each child and more children require more investment. Because of limited resources, families need to cut investment on one child to invest in the other. Also, investment in children has a decreasing marginal rate of utility. The earnings of children are related to parents earning through the channel of endowments and bequeaths. Solon s (1999) handbook chapter also explains the theory of the intergenerational income elasticity. The author states families divide their income between their consumption and child investment which is also stated in Becker and Tomes studies. The functional form: (2) where is family consumption, is the investment on child, and family income. Then, child permanent earning: (3) where is the return to the investment, is other determinants of earnings. Secondly, families try to maximize following Cobb-Douglas utility function: (4) 5

14 while in assumed to be the families taste between their consumption and children earning, it is a direct representative of family altruism level to invest their children in Solon (2004). As the altruism level and income level of families increases, they invest more in their children which causes IGE to go up. Higher IGE means the stronger effect of family income on child s income. Also, higher IGE means less intergenerational mobility from the reverse relationship between IGE and the mobility. When IGE is zero then, there is a full intergenerational mobility and the effect of family income on child income is zero. After combining equations 3 and 4 and taking the first order condition with respect to, it yields following model: (5) where is the intergenerational income correlation, and is the child s endowment and market luck. 1 Corak (2013) mentions intergenerational mobility and inequality. In this study, the author mentions The Great Gatsby Curve, the more inequality of income in the country, the more inequality transmission to the next generation. Countries like Finland, Norway, and Denmark are the most equal and countries like the UK and the USA are the least equal in terms of the income distribution. In the paper, the author also depicts that Canada seems to be more equal than the US. Corak also explains how some children from richer families earn more because of how higher income families invest in their children more. The relationship between family income 1 The parameter is the intergenerational income elasticity, Where is child s endowment and is the child s market luck. Here Solon differs from Becker and Tomes. In Becker and Tomes studies, it is assumed that is independent from family income. However, Solon states that since child s endowment ( with parents endowment ( from first order-autoregressive: is also correlated with family income. From there, the intergenerational income elasticity is given by ). For additional details, the reader is referred to (Solon 1999, 2004 or 2013). Solon also has done some studies about the intergenerational mobility in 2004 and 2013 as well. Solon (2004) and (2013) studies are improved based on his 1999 study except, in this study he also includes taxes and government investment to children. Solon (2013) takes out government investment and taxations out which make this study similar to his 1999 study. Solon defines child s lifetime earnings also as a function of parent s and grandparent s lifetime earnings. Most of the studies declare that intergenerational correlation goes down as a geometrically. In contrast, Solon states that intergenerational correlation goes down more than a geometrical series. Solon shows a negative insignificant sign on grandparent s earning coefficient which is consistent with Becker and Tomes (1979) study. On the other hand, some studies find a positive and significant coefficient on grandparent s earning variable. 6

15 and postsecondary income is higher in the US, children of richer families are more likely to attend postsecondary education compared to low income families. The author states that the income distribution program and policy in the US support top income families more than bottom income families. While in the developed countries, higher skilled teachers work in less advantaged schools, this is the opposite in the US Empirical estimates of the intergenerational elasticity Black and Devereux (2010) survey recent developments in intergenerational mobility. Their article updates Solon s (1999) handbook chapter. In Black and Devereux (2010), a basic intergenerational function is determined as follows: (6) where is the degree of intergenerational elasticity and is the intergenerational mobility. In addition to the degree of intergenerational mobility, the authors also show provide a functional form for the intergenerational correlation which is an alternative to the IGE: The intergenerational correlation is (7) where is the intergenerational correlation of earnings, stands for the standard deviations of parents ( ) and child s earnings ( ). While the correlation typically ranges from 0 to 1 in empirical estimates, the elasticity can be greater than one. Back and Devereux (2010) also state that using the elasticity can provide an unbiased estimate of the correlation between parent and child income and is easy to estimate. In order to give some ideas about the degree of IGE in the US, I report findings of some studies from Solon (1999) handbook chapter. For brevity, the references to the studies in Table 1 do not appear in the list of references of the thesis. The reader is referred to Solon (1999). The IGE changes when different measures of earnings are used. Panel Study of Income Dynamics (PSID): is a longitudinal survey. It is first begun at In the survey 18,000 individuals are followed from prenatal to the late age of. Since individuals are followed over time in the survey, many studies have used PSID also while measure the IGE. Solon (1992) measures the degree of intergenerational by using data from PSID. Solon claims that previous studies about the IGE are biased because of the homogenous sample and the uses of short-term data. He claims that using PSID data would help to increase the consistency of 7

16 estimation. The author finds IGE as 0.41 between father and son, which is higher than what the previous studies find. When he uses father education as an instrument of father income, the IGE gets even bigger. Solon claims that in terms of IGE, the US is less mobile than it is thought to be. Table 1: Some findings for the IGE as shown in Solon (1999). Sample of studies measuring IGE Estimate of IGE Data sources Altonji and Dunn (1991) 0.18 NLS Bjorklund and Jantti (1997) 0.39 PSID Couch and Lillard (1994) 0.37 NLS Minicozzi (1997) 0.42 PSID Mulligan (1997) 0.32 PSID Lillard and Reville (1996) 0.28 PSID Wilson and Wilson (1992) measure the effect of environment on a child s desire towards to education. The authors measure the effects of home and school environment on child s education aspiration. They use data from High School and Beyond Longitudinal Study of National Center for Educational Statistics (NCES, 1985). They use a subsample of data which includes the equal number of male and female. Results show that the effect of a home environment has a bigger impact on children in the early age while the effect of school has a bigger effect during the adolescent period. The effect is bigger for black individuals, although this finding may not be reliable considering that only approximately 6 percent of the sample is black. Therefore, it might be hard to see the difference between Black and Whites. Shea (2000) investigates effects of family income on human capital by using PSID data. In the study, it shows that the relationship between father earning and child s earnings are 0.356, which is close to Solon (1992). Although the author does not find a high effect of family income on human capital variables (education, skill, etc.), he finds a significant effect of family income on child skill for those whose father has less than 12 years of education. The author argues that the findings do not show a strong impact of family income on human capital accumulation. Shea explains why the impact might be slow due to public aid to poor children, or the fact that richer families are not as altruistic as they thought to be. Lastly there might be some issues with the 8

17 estimation. Fertig (2001) study measures the trend in intergenerational mobility across cohorts and earnings quintiles by using PSID s data. The findings show that there is an increase in intergenerational mobility between the father and son while there is a stability between motherson and mother-daughter. Mendolia and Siminski (2002) measure the effect of family background on earning by applying mediation analysis. To do that, they investigate the relationship between education and earning, and education and family background. Mediation analysis has not been used in economics much. Therefore, this makes the present study unique for applying the Mediation technique on the models. Authors use data from Australia (Household, Income and Labour Dynamics in Australia (HILDA)), and also from the UK (British Cohort Study (BCS)) to compare the two data sets. Findings show that the family background has a significant effect on earnings in both Australia and the UK. Around 24%-39% of this effect is explained by education level. The effect of family background on education is smaller compared to earnings and is even smaller in the UK in comparison to Australia Recent studies Jantti et al. (2006) compare the intergenerational mobility across countries. In Table-2, the IGE for the US, the UK, and Nordic countries are shown. Table 2: IGE for men and women in different countries. Results are based on Jantti et al. (2006). Countries Men Women Denmark Finland Norway Sweden United Kingdom United States

18 The results in from the table are also shown in Black and Devereux (2010). It can be seen that the US has the highest IGE, while the UK is lower than the US but higher than Nordic countries. Denmark has the lowest IGE. Authors use NLSY-79 data for the US. Moreover, they use the father s 1978 earnings and measure the IGE for son and daughters. Regarding degree of the intergenerational elasticity, it is stated that low income families might face credits constraint to invest in their children, which makes the IGE higher. Moreover, studies show a strong correlation between parental and children education, which is important because higher education might lead to higher income. In the study, it is also shown that in general, the US has a relatively high IGE and higher correlations between parental and child characteristics compared to Nordic countries. Duncan et al. (2010) look at the effect of poverty on child income, behavior, and health. It is stated that in the US, 4.2 million children were in the poverty threshold according to the 2006 data. They use the data from the (PSID). They look at the parental income during the prenatal, childhood and later periods of childhood, and observe the effect on education level, work hours, government program participation, health, crime, and the impact of having children out of wedlock by using the OLS model. Several observable variables are controlled to avoid the omitted variable bias problem like age, gender, parent live together or not, region, years of completing schooling, etc. Findings show that parental income has a more effect on children during the childhood and as children grow the effect is decreasing. During the prenatal and before the fifth birth year of a child, increasing annual income by $10,000 (in terms of 2005 dollars) for families with an income less than $25,000 leads to a 77.4% increase in earnings during the adulthood. It only increases 2.7% for families with an income higher than $25,000 per year. Karagiannaki (2012) investigates the parental wealth impacts during the age of early adulthood on child s outcome. This study is unique considering that it is one of the few studies that have looked at parental wealth s impacts on child outcome in early adulthood using UK data. There are several studies that have been done related to this topic. However, the US data has been used the most. In this study, the author uses data from the British Household Panel Survey (BHPS). The BHPS is a national longitudinal survey that was first implemented in Karagiannaki uses both the OLS model and the Probit model to check the effect of parental wealth on child education, employment, wage, and homeownership. Findings show that 10

19 there is a strong correlation between parental wealth and child education level. The overall effect of parental wealth on child outcomes education, employment, wage, and homeownership are significant. The effect is even stronger for families of lesser wealth. The author also separates parental net worth by financial - which is more liquid and more effective in the short run, and housing wealth - which is more illiquid and effective in the long run. Housing wealth effects look stronger than the financial wealth effect. The issue with this study is the small sample size. The number of children who gave the full interview is 492, which makes it difficult to assume that these results apply to the whole population. Also, the author looks at child outcome when they were 25. This age is quite early to measure the real earning. One can be just graduated the college and looks for a job. Lovenheim and Reynolds (2013) investigate the effect of a change in home price on the type of college children attend. They control for the student s cognitive ability, mother-father education level, ethnicity, and some other income and non-income variables. They categorize family into three income groups; lower income, middle income, and top income families. The findings show that a $10,000 increase in home price change leads to a 2% percent increase in the chance of attending a flagship public university compared to non-flagship university. Also, it increases the chance of completing the college by 1.8%. Effects are higher for the low income families sample. 2.2 Total compensation and fringe benefits Although following studies do not show the relationship between parents income or background and child s outcome, they are considerable in terms of showing the importance of using the total compensation and non-pecuniary earnings rather than only wage income Total compensation Duncan (1976) measures effects of human capital variables on the wage and nonpecuniary earnings. Duncan uses data from Quality of Employment Survey and PSID. He restricts the sample only for the age of white workers between the age of 21 and 65 and workers who work more than 35 hours per week. He describes fringe benefits like insurance, paid vocational, sick leave, free meals, etc. He converts the fringe benefits into the dollars values and then adds them to the wage to create a measure of the total compensation. The author uses 11

20 canonical correlation technique to create an index of wage, fringe benefits, and other nonpecuniary benefits. After including other non-pecuniary earning into the dependent variables, he finds that coefficients on human capital variables are changing. Especially, the effect of education on the index is bigger compared to the effect of education only on wage. In both cases, education has the biggest explanatory power on the dependent variable. Duncan (1979) indicates that 13% of the total compensation comes from non-wage benefits using Quality of Employment Survey and it is 8% when using (PSID) data. Smeeding (1983) uses a combination of wage and non-wage benefits from the works while measuring the employer cost versus employers value to these benefits. The author uses data from Employee Compensation Survey (EEEC), the 1977 Employment Cost Index Survey (ECI), and Consumer Population Survey (CPS). The author indicates that 90.1% of the total compensation comes from payments, the rest come from other types of the benefits for 1979 year. Smeeding includes payments for vacation and holiday, health insurance, life insurance, pension and retirement, sick and severance, nonproduction bonus, legally required contributions (social security, railroad retirement contributions and other required contributions, such as unemployment insurance and other minor legally required payments) to measure the dollar values for the fringe benefits. Results show that considering all types of workers; males have a higher salary and more fringe benefits than females. When restricting the sample with both health insurance and pension benefits, the difference is decreasing. Also, the experience and education level increase the wage and benefits received. Considering the fringe benefits the gap between college and non-college degree is widening, the author indicates that 9.9% of the total compensation comes from non-wage benefits. Solberg and Laughlin (1995) uses total compensation as the dependent variable to measure the men and female earning gap rather than using wage as the dependent variable. They use data from NLSY-91. The authors construct an index of the total compensation income from fringe benefits and wage. For creating the index of the compensation, they follow canonical correlation analysis of Duncan (1976). They look at the gap by 7 different occupation categories; Craft, Operatives, Sales, Managers, Services, Professional and Technical. The categories are ranked from the most men dominated to the most female dominated. They first use log-wage as dependent variables. Findings from using log-wage as dependent variables show that the wage gap between men and female are quite high. Men earn higher in almost all categories. The 12

21 difference between men and female are higher in more men dominated categories and decreases in more female-dominated categories. The differences are statistically significant in almost all categories. On the other hand, when they use the index as a proxy for total compensation as the dependent variables, the gap and significant level of differences shrink. The index includes wage and same nine types of fringe benefits with my study. These non-pecuniary fringe benefits are 40% of total compensation. Including the index as dependent variables, the lead earning gap shrinks and the difference in earning between men and female is no longer significant except in operative s jobs which are not because of the discrimination, but the nature of the jobs. Operatives are in the more men eligible occupation category. Using the log-wage illustrates that the mean of female s age was 87.4% of the mean of men s wage, this number increases to 96.4% when the index is used. Chung (2003) finds that using wage alone to measure the growth in inequality from 1987 to 1994 underestimates the inequality below the median of the wage distribution and overestimates it above the median. It is also stated that the share of fringe benefits in the total compensation increase from 13.7% to 16.2% during the eight years. To measure this, the author uses a combination of data from Current Population Survey (CPS), Employee Benefits Survey (EBS), and National Medical Care Expenditures Survey (NMCES). This study is unique regarding measuring the fringe benefits by using individual-level data. In the study, it is stated that 80% of the sample from CPS are also covered by EBS Studies focusing on specific fringe benefits Benedict and Shaw (1995) measure the effect of pension benefits on earnings inequality. Findings show that considering pensions increases the inequality of earnings distribution between full-time workers. Also, the study shows a positive relationship between earningpension and union-pension. Farber and Levy (2000) states that considering health insurance exacerbates the inequality while measuring the difference in earning between college and noncollege degree. Levy (2006) investigates wage and insurance coverage gap between women-men, Black- White, Hispanic-White. She uses data from the Current Population Survey (CPS). Findings show that the gap is bigger in wage compared to insurance. White has a higher average rate comparing Black and Hispanic. Men have an average wage compared to Women. The gaps are 13

22 getting smaller when insurance coverage is used. Insurance coverage rate is 74.1% for men and 68% for women; it is 74.1% for Whites, 70.1% for Blacks and 55.8% for Hispanics. The author explains these results as being that Black and Hispanics have a lower level of education, and big populations of Hispanics are immigrant and non-citizens. Thus, they have the lowest rate of insurance. The reasons for the gaps between Whites-Black, White-Hispanics are mostly because of observable characteristics. However, the gap between men and women is not explained by observable characteristics. Women overall have a higher education level, but they also have a lower insurance coverage rate and this gap increases when using the wage as a dependent variable. Also, the rate of being a citizen for women is slightly higher. The reason why women have a lower wage and less insurance coverage rate than men is not easy to explain by observable characteristics. Selén and Ståhlberg (2011) use data from Sweden to measure the wage and compensation earnings inequality. Results show that the inequality between male-female and between bluecollar and white-collar workers is increasing while including pension benefits. Mok and Siddique (2011) measure the distribution of fringe benefits between gender by using data from NLSY-79 and NLSY-97. The health insurance, life insurance, and pension benefits are available more for White compared to Black and Hispanic. The difference shrinks while controlling for education, cognitive ability, age, and family background Relationship between income and fringe benefits In this section, we discuss studies that have looked at the correlation between wage and non-wage benefits. For instance, Woodbury (1983) measures the substitution between wage and non-wage benefits. Woodbury indicates that the workers utility function is determined by wage and benefits. The author measures the benefit in two different forms. One form only includes health insurance and life insurance; while the second form includes health insurance, life insurance, and pension benefits. Results show that the first form of the benefits gives the elasticity of substitution between wage and non-wage benefits as 1.6, while the second form of the benefit which also includes pension benefits gives the elasticity as 7.7. In both cases, we can see that there is a substitution between wage and non-wage benefits and the substitution s level increases dramatically while considering the pension benefits. The authors also indicate 14

23 that bigger firms offer more benefits which can be because of how the big firms have the ability to offer the benefits with a lower cost. Hamermesh (1999) measures inequality in terms of wage and disamenities. Hamermesh assumes workers maximize the utility which is a function of wage and amenities from the work. The author uses the work injury and evening/night time works schedule as a measure of disamenities. Results show that using the only wage understates the changing inequality. Monheit and Vistnes (1999) indicates that young age and low wage workers prefer jobs with no health insurance, but with relatively higher wage compared to jobs with the insurance available to them. Olson (2002) investigates the tradeoff between fringe benefits and wage. He specifically tests tradeoff between health insurance and wage using data from Current Population Survey (CPS) March-June Findings from this study show that workers are willing to give up average 20% of their wage to have a health insurance. Converting it into dollar value people on average are willing to give up $4,000 (1990) annually to have the health insurance. Although, there are some noises with this study like only using married and full workers from a women sample. However, the author uses OLS and both instrumental variable models with more than one instrumental variable to obtain an unbiased estimate. Pierce (2010) looks at the correlation between fringe benefits-wage and the distribution of the fringe benefits over wage percentiles from The author uses data from the Employment Cost Index (ECI), and from Current Population Survey (CPS). Results show that on average the share of each fringe benefit in total cost rises, especially with health insurance and retirement benefits. In average during the given period, benefits cover 26% of the total compensation. 2.3 Intergenerational transmission of occupational status Although there are no studies which look the relationship between family wealth and total compensations, some authors measure the relationship between children and parents occupational status. It would worth to look at these studies for better understanding the correlation between family characteristics and child s outcome. Korupp et al. (2002) look at the relationship in occupational status and sex-typing (female-dominated or male-dominated jobs) between parents and children s first job by using 15

24 data from Netherlands. The authors look at the correlation between father-son and daughter, and mother-son and daughter. Findings show that there is a strong intergenerational transfer of occupational status and sex-typing. The effect of occupational status transfer is higher than the effect of the sex-typing. Also, although in general the effect of fathers is higher, daughters tend to follow mothers while sons tend to follow fathers. In Hellerstein and Morril (2011) investigates the relationship between fathers and daughters occupational status by using PSID data. Results show that probability of a daughter would enter her father occupation status increased between 22% to 44% overtime, the probability to enter her father-in-law occupation status increased between 16% to 36% overtime. The probability a son would enter his father-in-law occupation is around 25%. 16

25 CHAPTER 3: HYPOTHESES study. In this chapter, I briefly explain the theory and hypotheses underlying my empirical 3.1 Intergenerational transmission of income Many studies have estimated the equation where,, is a constant, is an error term and due to the log linear formulation, is the IGE. Such studies have found a high level of for the US: 0.41 in Solon (1992), in Shea (2000), for the son-father relationship and for the daughter-father relationship in Jantti et al. (2006). Based on these studies and the US GINI coefficients (40.8), which is also high compared to other countries like Nordic countries, I expect a high value of. Although a high IGE is a sign of inequality, it does not directly show how big the gap in income is between children from high income families and those from low income families. There are many studies that measure the degree of IGE by income quintile or quartile. However, this does not answer the question directly. Measuring the IGE by income quartile gives the fluctuation in IGE for different income levels. However, it does not show how much children from low family income quartile earn less compared to ones from higher family income quartiles. To see the difference in child s earning because of the family income, I categorize family by their income into four different groups by income quartiles. I represent each quartile by a dummy which shows what income level each child belongs to. Creating the dummy variable for each quartile enables me to show the size of the inequality in individuals earnings. By looking at the coefficients on dummies, one will be able to see differences in earnings between quartiles which are attributed to family income levels. The functional form of the model is shown in equation (10) in the empirical models section, where is the coefficient on dummy variable i. Dropping the first (bottom) quartile, takes value from 2 ( second ) to 4 ( fourth ). In this case, 17

26 the null hypothesis is. I am expecting to reject the null. Since the IGE is high, it is very likely that income level of families would cause substantial differences in income across quartiles. Until now, I look at the relationship between family income and child s income without controlling any other family or child characteristics. It is not surprising to have high coefficients without holding constant the child s endowment and parental investment in human capital. Recalling from Solon s (1999) study, we have (8) where is the intergenerational income elasticity. In equation 8, stands for child s endowment and child s market luck. I assume child s market luck is constant. For child s endowment, I consider race, ability, and gender. Becker and Tomes (1979) describe child s endowment as race, ability, and other family credits like family authority, family connection, etc. However, because it s difficult measuring the family connection, reputation, I only control for race, gender, and ability as part of child s endowment. Race, ability and gender, which pass to the next generation through gene, affect child s earning. Without controlling for them, the estimation would be upward biased since the effect of child s endowment is endogenous and correlated with earnings. Being male, white, and having high ability are advantages in many case, and would help to do better in the labor market. Although these variables would affect the family income, family income does not affect them. In another words there are not simultaneous causality between family income and child s endowment. Hence, controlling for them would increase the preciseness of the estimation. After controlling for a child s endowment variables, the way family income affects earning would be through family investment to child s human capital. Recalling from Solon (1999), the families distribute their income between their consumption and the child s human capital. As altruism level and income level of families increase, the child gets more investment which leads better earning in the future. From Solon (2013), the child human capital is a function of parental income. The functional form is shown as follow: (9) where is the child s human capital, is the investment made toward the child by the parents in the previous period, is a parameter and is the child s initial endowment that I explained in the previous paragraphs. 18

27 The variable captures any investment made in children. Parents would probably put education as a priority when they are investing in their children human capital. One of the best ways to accumulate human capital is by investing in education. Although the educational level and ability are strongly correlated and children with higher ability very likely would obtain postsecondary education compared to lower abilities, it does not guarantee that children with high ability would get higher education since the tuition and many other factors also affect the schooling levels. Considering the education is not free and even expensive in the US, it is very likely that children from high income families would have a better chance of obtaining a higher education. On the other hand, there is financial aid available for needy students and student loans that people can also borrow to obtain a higher education. Although these opportunities might help to offset the inequality, there still would be several challenges for lower income family children to take advantages of these benefits. One challenge can be the credit constraint; it might not be possible for some families to borrow for their children education. Also, because of the uncertainty, if they would be able to pay the loan back. As a result, some families might decide not to borrow. Another difficulty can be that the taking advantages of financial aid benefits might not be that easy. It is possible that very few individuals take benefits from the aid since sometime it is available only for a certain number of individuals. Therefore, only people who take advantage of financial aid might be the first applicants or the ones in the lowest level of the income distribution. The remaining needy families still would face the income constraint to invest in children. Since this is the case, high income children would obtain more education which leads better outcome. The next research question: is the education level the only mechanism that families invest in their children human capital? If yes, then controlling for the education level of children in addition to race, gender, and ability must close the gap between children s earning or at least reduce it dramatically. In other words, after controlling the level of the education, the children from the high income family and low income families should come up with a similar outcome in terms of the earnings assuming that they put the same amount of affording for it. To see if the differences in earnings of children are significant after holding child s endowment variables and education level constant, I estimate the effect of family income level on worker compensation condition on race, gender, ability, and the educational level. The 19