As workers’ health insurance contracts are generally renewed in January each year, employers can identify workers who choose a family plan. However, workers’ wages or employment contracts might not be renewed every year. Thus, there could be some lagged effects of the ACA-DM on workers’ labormarketoutcomes. To test this conjecture, we estimate the yearly impact of the ACA-DM using the regression specification (3). Figure 3 shows the estimated labormarket impact, 𝛽̂ 𝑘 s, with 95% confidence intervals. We standardize the estimated labormarket impact in 2010 to zero, as indicated by the horizontal lines. Panel A shows the estimated effects of the ACA-DM on parents’ wages. Before the ACA-DM, 𝛽̂ 𝑘 s are statistically insignificant during the pre-reform period except for 2008, which provides additional evidence for the parallel pre-reform trends assumption. After the ACA-DM, 𝛽̂ 𝑘 s are negative and gradually decrease, implying lagged adjustments to workers’ wages. Panels B–D show the effects of the ACA-DM on parents’ labor inputs such as the probability of employment, weekly working hours, and the probability of full-time work. Similar to the findings of the baseline analyses, 𝛽̂ 𝑘 s are generally close to zero and statistically insignificant, providing little evidence of lagged responses to labor inputs.
imately 20 percentage points. In columns (2) to (5) we add characteristics to the regression model that are arguably exogenous, including the child’s gender and age, and measures of parents’ cognitive and non-cognitive skills. In column (2) we find that adding child characteristics does not alter the esti- mated network coefficient. The estimates for gender and age are nonetheless statistically significant and similar to those found in most other wage regres- sions; that is, the earnings score is lower for women and concave in age. In column (3) we also find that including personality traits does no change the friendship effect. Of the five personality traits, only agreeableness and open- ness to experiences affect the child’s earnings score in a statistical significant way. In column (4) we add parental IQ and find that the total number of friendship connections continues to have a small but marginally significant effect on the child’s earnings score. Parental IQ itself has a significantly pos- itive impact, which suggests that high IQ parents have, on average, more high-school friends as well as more children who are more successful on the labormarket.
The selection of the control variables to estimate the propensity score is crucial in order to make the conditional independence assumptions credible. Here we can make use of the major strength of high quality survey data: the abundance of individual level vari- ables that potentially affect both treatment (paths) and potential outcomes as well as their changes over time. While a drawback compared to many administrative data sets is the comparably small sample size, the advantage is the widespread information on topics such as socio-economic background, health, or preferences that are usually not available in administrative data sets. We consider this crucial for the identifying assumptions. In deciding for informal care provision one might have three basic blocks of prerequisites in mind. Individuals decide to provide care if (i) they need to, if (ii) they are willing to, and (iii), they are able to provide care. As of (i), individuals are only in the position to decide for care provision if someone close becomes care dependent. We model the intra- social environment by using indicators whether parents are alive, parents’ age as well as the number of the potential caregiver’s siblings. The latter can reduce the need to provide care for frail parents as siblings could step in.
Recent empirical research has shown that social interactions at school can affect individual academic achievement and ultimately labormarketoutcomes. Two measures of school peer characteristics that have attracted attention are the share of girls and average parental background in the class/grade/school attended by an individual. 1 There is evidence that a higher share of girls positively affects the learning outcomes of both girls and boys in Israel (Lavy and Schlosser, 2011) and that the average earnings of peers’ fathers matter for the education and labormarketoutcomes of boys in Norway (Black, Devereux and Salvanes, 2013; see also Ammermueller and Pischke, 2009). Understanding the effects of school peers is important for the design of school admission and class formation policies. On the one hand, these effects help informing the ongoing debate about single-sex schools. A report by the UK Department of Education has suggested that boys should be taught separately (see Doris et al, 2013). According to the European Association of single-sex education, co-education should be challenged, especially for girls. On the other hand, the presence and nature of school peer effects shed light on the consequences of de-segregating schools. Assuming that parental education is a relevant measure of family background, they also inform about the intergenerational social returns to education, an area where relatively little is known to date (see Oreopoulos and Salvanes, 2011).
Several previous studies have shown the effects of childhood SES on earnings at a specific age in adulthood. Some of the first studies in addressing this issue focus on parental education to investigate the so-called family background bias in returns to education in earnings equations (Behrman and Wolfe 1984, Heckman and Hotz 1986, Lam and Schöeni 1993), usually identifying a larger association with mother’s education than with father’s education (see also Behrman and Rosenzweig 2004). Parental income is also widely known to be positively associated with individual’s later life earnings in the literature on intergenerational mobility in earnings. Moreover, these intergenerational elasticities in earnings vary across gender, time, country and age of both the parents and offspring (see Black and Deveraux (2011) for a survey). More recently, two studies explore the role of income shocks induced by changes in the labormarket status of a parent (due to firm closures), providing evidence for Canada
Fig 1: Estimated AFQT “Effects” on Annual Labor Income
Note: Figure shows estimated effects of 0.1 standard deviation change in AFQT score for representative NLSY79 sample. Horizontal axis shows results for two-year age groupings beginning with age listed (e.g., 30 refers to 30-31 year olds). Estimates are obtained from the coefficient on the computed AFQT Z- score regressed on annual labor incomes (including zeros). These are converted to percentage changes by dividing the estimated effect by the dependent variable mean at the specified age. Background covariates include: gender, race/ethnicity (black, Hispanic vs. Non-Black Non-Hispanics), magazines, library cards and newspapers in home at age 14 (3 variables), urban residence at 14, respondent and respondent’s parents foreign born (3 variables), mother and father’s education (< high school graduate, some college, college graduate vs. high school graduate), number of siblings (0, 1, 2, 3, 4, ≥5), older siblings (0, 1, 2, 3, ≥4) and survey year dummy variables. Time-varying covariates include values of: marital status
The current work adds value to Teal (2001) in five ways. Firstly, we examine the role of education in facilitating entry into lucrative occupations, by means of multinomial logit models of occupational attainment. This is important because, as noted before, education plays a role in labour market success not only directly by increasing earnings in any given occupation but also indirectly by promoting entry into the well-paying occupations. Secondly, we examine the role of cognitive skills in labour market success, both in terms of occupational outcome and earnings. Thirdly, we estimate returns to education along the earnings distribution by means of quantile regression analysis to ask whether the marginal return to education is greater at lower levels of earnings, i.e. whether education ameliorates economic inequality or exacerbates it. Fourthly, we estimate returns to education by age group to examine whether the labour market rewards education differentially for younger and older
Micklewright et al., 1990).
Labormarket conditions at the time of graduation can affect high- and low-ability individuals differently. First, it could be more difficult for low- than high-ability individuals to find a high-quality job, or any job for that matter, if the unemployment rate is high at graduation time. Such initial differences could could persist or become amplified in the long run if low-ability individuals struggle to escape the low-quality job, or if a low-quality job hampers their human capital accumulation. We may also expect differential effects on long-term labormarketoutcomes if, for example, high-ability individuals facing poor labormarket entry conditions are better than low-ability individuals at catching up with individuals within the same ability category facing good labormarket entry conditions, through a greater search intensity and job mobility: thus resulting in a differential effect on long-term earnings even if there were no initial differences. (see e.g. Oreopoulos et al., 2008). Second, the replacement rate of benefits from welfare programs are often a concave and capped function of prior earnings, giving a higher ratio for low-income workers than that for high-income workers. This could provide low-ability individuals who graduate in bad times stronger incentives to leave the labor force. 3 Finally, the loss in the opportunity cost of schooling might depend on the ability type, providing differential incentives to continue with education.
While Autor, Dorn, and Hanson (2013), Autor, Dorn, Hanson, and Song (2012) and Ebenstein, Harrison, McMillan, and Phillips (forthcoming) focus on more aggregate la- bor marketoutcomes, several recent papers analyze how firms adjust in response to in- creased import competition. Bernard, Jensen, and Schott (2006) show that American plant survival and growth are negatively correlated with industry exposure to imports from low wage countries. 1 Iacovone, Rauch, and Winters (2013) find that Chinese im- port penetration reduces sales of smaller Mexican plants and more marginal products and they are more likely to cease. 2 Bloom, Draca, and Van Reenen (2012) use the num- ber of computers, the number of patents, or the expenditure on R&D as measures of innovation and find that Chinese import penetration correlates positively with within- plant innovation in the UK. 3 Finally, using Belgian firm-level data, Mion and Zhu (2013) find that industry-level import competition from China reduces firm employment growth and induce skill upgrading in low-tech manufacturing industries. For a survey of recent firm-level empirical research on trade, see Harrison, McLaren, and McMillan (2011). In summary, there has been a revival in studies looking at firm-level outcomes, but none of these papers focus on wages as the outcome. In this paper we attempt to fill this gap.
Many of these studies use the weight of a biological relative as the instrument. This appears to be a valid instrument, since it is a source of variation in weight due to genetics (roughly half the variation in weight across people is genetic in origin) and ought to be unrelated to an individual’s labormarketoutcomes. While its validity would be compromised if many of the genes responsible for obesity were also responsible for other factors that affect labormarketoutcomes, such as willingness to delay gratification (time discount rate), most studies have been unable to detect any effect of a common household environment on body weight . Some more recent studies have begun to use an individual’s own genetic information as instruments. Genetic risk factors are an appealing instrument for obesity because they are determined at birth and are not under the control of the individual so in theory they make excellent IVs. However, since certain genes that determine obesity may also be correlated with other risky behaviors, genetic instruments may not be legitimately excluded from the labormarket outcome and it is fair to say that their use, while promising, is still in its infancy. Other instruments have also been used, but their validity is questionable. For example, studies have used the average BMI and the proportion of obese people who live in the same area as the study subjects as an instrument. But because people choose where to live, this instrument could be related to occupational choices and earnings, rendering it invalid. Instruments used in yet other studies have included the presence of other obese people in the household, being an oldest child, having only sisters, or having a parent who has been treated for obesity. Their validity is questionable, however, as they are probably correlated with an individual’s labormarket outcome independent of their association with obesity. A particularly innovative approach to estimating the causal impact of BMI on employment exploits the results of a randomized experiment in which obese individuals are assigned to a treatment group that is offered a financial reward for weight loss or a control group that is not. The researchers find that a decline in BMI leads to an increase in the probability of remaining employed for women, with no significant impact on employment for men.
and Lahey 2010). Authors have described the effects of the GI Bill on educational attainment, earnings, and employment (Angrist and Chen 2011).
In theory, veterans’ labormarketoutcomes might be better than nonveterans’ outcomes because veterans have received military training that may be transferrable, incentive policies encourage employers to hire veterans, and legislation gives veterans priority to receive civilian workforce development services. However, little research has focused exclusively on veterans in the workforce development system and their associated labormarket experiences, despite the prioritization of veterans within the system. The most closely related research to the current paper is an evaluation of the Priority of Service provision of the Jobs for Veterans Act, commissioned by the U.S. Department of Labor. It finds that service receipt within the WIA Adult and Dislocated Worker programs was similar for veterans and nonveterans, that
Next, we use an employment indicator as the measure of the labormarket outcome. Table 5
presents the results. We find that the main conclusions for the intensive labor supply margin from Table 4 hold. First, we observe that the effect of a health shock on employment is negative and statistically significant at the conventional levels in specifications (1) to (4). However, the effect becomes larger in absolute values as we add more interactions. Hence, the bias is in the opposite di- rection compared with the earnings results, and neglecting to account for possible pathways leads to an underestimation of the employment effects of a health event. Second, we also observe a protective effect of education on the negative impact of health shocks on employment. The attenuating effect of education is, however, more convex on employment than on earnings. We observe in columns (1) to (4) that workers with a post-secondary education lose between 3.2 and 1.8 percentage points less in the employment rate with respect to workers with no education. On the other hand, the negative effect of health shocks on employment for workers with a high school degree is statistically similar to that for workers with no formal education under most of the specifications presented in Table 5 . Third, we also find a heterogeneous health shock impact on the extensive labor supply margin across industries.
The impact of the regional unemployment rate on wages (Table 3) is remarkably in line with the wisdom established by the previous literature. Indeed, our results point to an elasticity of wages with respect to the regional unemployment rate ranging from −.127 to −.096, whereas Blanchflower and Oswald (1994) derived the “law” of an elasticity of −.1. The interesting addition to the literature on misperceptions concerns our estimation of the impact of unemployment misperception on wages (line 1 in either Tables 2 or 3). We find that each percentage point misperception of the unemployment rate is associ- ated with a wage decline of 0.7 % (see the results in columns 1 to 3)—a misperception of the labormarket situation by 10 p.p. would thus result in a wage penalty of 7 %, which is a noteworthy impact, in particular if one keeps in mind that the average mispercep- tion is 13 p.p. We further checked whether the impact could be different depending on whether the worker overshoots or undershoots when evaluating the country’s unemploy- ment rate. We would expect workers with a pessimistic view of the labormarket to lower their reservation wages and thus earn lower wages; instead, workers with an optimistic view of the labormarket could be overconfident and set too high reservation wages, though possibly having trouble finding a job. Consistent with that reasoning, we find that a pessimistic view of the labormarket leads to lower wages. An optimistic view, instead, has no significant impact on wages (as indicated by a formal test on the sum of the esti- mated coefficients in lines 1 and 2 in each model specification), possibly due to the fact that very few of these workers underestimate the unemployment rate. Column 4 further introduces controls for job and employer attributes. Interestingly, the magnitude of the impact of labormarket knowledge imperfections on wages is reduced—each percentage point misperception of the unemployment rate is now associated with a wage decline of 0.4 %—suggesting that part of the impact of misinformation on wages operates through worker matching to lower quality jobs.
Panel #2 (panel #4) of Table 1 shows the results for the CSM (EPS) equilibrium with a given (see last column in Table 1). When comparing this to our previous results in panel #1 (panel #3), we observe that, for both types of equilibria, changing the cost of staying low-skill through causes higher unemployment for low-skill workers— whereas the one for high-skill workers remains the same (or slightly decreases). Lower costs of skill upgrading lead to a higher share of high-skill workers (a lower q) while making the economy redirect its resources towards this sector. This, eventually, results in higher wages and a higher welfare level for high-skill workers and, consequently, a higher total welfare. Such a mechanism would also bolster the EPS equilibrium— notice that, unlike in panel #1, the CSM fails to exist for big productivity gaps (panel #2). 16 Appendix 2 shows how the model behaves for several other functional forms of the skill acquisition cost. For example, in the case of a linear cost function the cost of skill acquisition is rather high while welfare is low, but the adjustment to o¤shoring is very ‡exible and, as q decreases, the economy performs e¢ ciently with many highly productive workers. Higher non-linearity of the cost, instead, makes the adjustment to o¤shoring more rigid, but the skill procurement is cheap and results in better welfare outcomes.
assumptions (CIA) which we justify by exploiting not only cross-sectional but also longi- tudinal information from our rich household survey, the German Socio-Economic Panel (SOEP). In auxiliary analyses, we relax the CIA to identify effect bounds under weaker assumptions. Other sensitivity tests such as placebo regressions imply that remaining time-invariant unobservables are unlikely to lead to an upward bias of our estimates. Our main ﬁnding is that female caregivers reduce the probability to work full time by 4 percentage points (at a baseline probability of 35 percent). The effect is persistent over a period of eight years and seems to be mainly driven by switches to part-time work. High care intensities and longer episodes, however, also increase the long-run probability of leaving the labor force. When we move away from point identiﬁcation to effect bounds, the reduction in full-time work changes to an interval of 2.4 to 5.0 ppts. As another ﬁnd- ing, wages seem to be unaffected contemporaneously but are signiﬁcantly lower eight years after the start of a care episode.
indicate similarity in cultural identity (Falck et al., 2012; Suedekum, 2018). 3
Differences in dialect may affect social assimilation of migrants for serval reasons. Individuals may bear physiological costs when interacting with people speaking different dialects, or they may be discriminated. All this may hinder their social assimilation in the host society. 4 It is worth to mention that in addition to the cultural effect of dialect on social assimilation, there might be communication effects on their labormarketoutcomes as well. This should be less of a concern in our study setting, however. Given the popularization of Mandarin Chinese (i.e., putonghua), people can easily communicate with each other. 5 In addition, we use linguistic distance to capture the cultural effect of language, rather than language skills of the individuals. Lastly, we check the robustness of results by controlling for individual skills of local dialects. Thus, we identify variations in social assimilation caused by dialect differences reflecting persistent cultural differences across regions that developed over time, on the top of communication effects on labormarketoutcomes.
6 Since the beginning of this century Latvia has enjoyed solid economic growth and declining unemployment as a result of structural reforms undertaken in the country during its transition from a centrally planned to a market economy in the 1990s. Combined with relatively high emigration after the accession to the EU in 2004, which has contributed to, but was not the only factor behind declining unemployment, labor shortages emerged in 2005-2007 (Hazans and Philips, 2010; Rutkowski, 2007). The boom originated primarily from a rise in domestic demand and a housing boom, while the first signs of overheating came in 2006 with wages and prices, especially housing prices increasing rapidly (Blanchard et al., 2013). Most of Latvia’s banks were owned by Scandinavian banks and the interest rates were low while credit and mortgages were widespread. This “unhealthy boom” in combination with the world financial crisis caused a huge decrease in GDP which declined by 25% over just eight quarters with decline in domestic demand being 43% of GDP (Blanchard et al., 2013). Internal devaluation and fiscal consolidation policies in 2009 included cuts in public expenditures, wages and pensions, reductions in personal income tax allowances, resulting in a decline in unit labor costs and a substantial increase in unemployment. Declining unit labor costs and employment, withdrawal of low-productivity workers and increase in firms’ efficiency potentially led to an increase in productivity and improved firms' competitiveness (ibid).
rather different with respect to observed characteristics in 1989 (see Table 1, the rows “unmatched”), and potentially still with respect to unobserved characteristics. Given that, for example, people with different ages and education levels can be expected to have different developments of employment and labor income, this could help explain the diverging trends between the treatment and potential control groups before treatment assignment (as shown in Figure 2 and Figure 3). For the exogeneity assumption to hold, it thus seems important to control for these differences when estimating the ITT-effect (the effect of being offered improved commuting opportunities by public transit). To estimate the ITT we therefore use a difference-in-differences matching estimator (DIDM) (Heckman, Ichimura and Todd, 1997). 19 This type of estimator is analogous to the standard difference-in-differences (DID) regression estimator, but does not impose functional form restrictions in estimating the conditional expectations of the outcome variable, and reweights the observations according to the weighting function implied by the matching estimator. 20 The matching thus ensures that the treatment and control group are balanced in terms of observed characteristics, while the DID approach controls for unobserved but temporally invariant characteristics remaining after matching. 21
SUMMARY AND POLICY ADVICE
Results based on cross-section analysis suggest that overeducation indicates some form of market failure. Such studies find that many college graduates are employed in jobs that do not require a college degree and in which the skills they obtained in college are not fully put to use , , , . The same set of studies finds a wage penalty and reduced job satisfaction for overeducated workers. These findings need to be interpreted with caution, however, because they make no allowance for individual differences and preferences. Some workers may choose to work in jobs for which they are overeducated because they offer compensating non-financial advantages or better future job opportunities, or because it was the only job they could get due to low ability relative to their qualifications. These possibilities suggest that the market may at least in part be working efficiently. Why employers hire workers to positions in which they are mismatched is a question that requires answering, but adequately linked employer– employee information on this is lacking.
et al. ( 2008a ) and Almlund et al. ( 2011 ) provide comprehensive overviews over the empirical findings from labor economics and personality psychology.
In this paper, we study the importance of noncognitive skills in explaining diﬀerence in the labormarket performance of individuals by means of machine learning techniques. Empirical studies based on large-scale observational data usually contain a large number of measures of cognitive and noncognitive skills. Typically, some type of dimension reduction technique is applied in order to reduce the dimensionality problem and to obtain interpretable empirical results. Predominantly, this is done ex-ante via preprocessing the data by principal component analysis (PCA) and related factor modeling strategies or simply by index building. Moreover, some type of dimension reduction is implicitly accomplished by focusing on certain personality concepts (e.g. Big Five, locus of control) and disregarding covariates reflecting more closely alternative (complementary or competing) personality concepts. While conventional statistical approaches are mainly concerned with the within-sample explanatory power of noncognitive skills, the approach pursued in this paper is in the tradition of the machine or statistical learning literature to data analysis by focusing on the out-of-sample forecasting or classification qualities of noncognitive skills.