Health Insurance Costs and Employment Outcomes by Age. By Joanna N. Lahey. Texas A&M University

Transcription

1 Health Insurance Costs and Employment Outcomes by Age By Joanna N. Lahey Texas A&M University DRAFT WORKING PAPER May 2008 Bush School Working Paper # 601 No part of the Bush School transmission may be copied, downloaded, stored, further transmitted, transferred, distributed, altered, or otherwise used, in any form or by any means, except: (1) one stored copy for personal use, non-commercial use, or (2) prior written consent. No alterations of the transmission or removal of copyright notices is permitted.

2 Health Insurance Costs and Employment Outcomes by Age By Joanna N. Lahey Texas A&M University Abstract Health insurance costs are one reason that employers may be reluctant to hire older workers. Higher health insurance costs are often correlated with other factors of employment, such as firm size, which can also be correlated with employment outcomes. This paper uses state health insurance mandates, which are correlated with higher health care costs, as a source of exogenous variation in an instrumental variables (IV) strategy to identify the causal effects of health care costs on employment of older workers. Using this instrument, I find that increasing health care costs significantly lower men s employment. Thus it appears that rising health insurance costs for older workers are partly responsible for decreasing employment of older potential workers. Although people with higher health care costs are less likely to be employed, older workers do not seem to be singled out; employment and labor force rates for older potential workers are in fact less affected by higher health care costs than are employment outcomes for younger. This work was supported by a grant from Boston College s Center for Aging and Work from the Sloan Foundation. Thanks to Alejandro Bras, Hoai-luu Nguyen and Yiding Yu for excellent research assistance. Thanks also to Elizabeth O. Ananat, Lisa Bell, Mick Smyer, and Ebonya Washington for helpful comments and discussion. 1

3 I. Introduction Older workers who have been displaced from jobs often have a difficult time finding new jobs (Diamond and Hausman 1984). Recent experimental evidence (Lahey 2007a) suggests that employers prefer interviewing younger applicants to older. Surprisingly little work has explored the reasons for differential hiring for older workers. Employer surveys suggest many possible causes of this treatment, such as: worries about worker productivity, including shortened experience profiles and less worker flexibility; and higher worker costs in terms of salaries, benefits packages, and probability of a lawsuit (Rhine 1984). 1 One employer cost that is higher for older workers is health insurance. In most states, an older worker is more costly to insure than a younger worker. Workers who cost more may be willing to take a wage cut, in which case employment will be unaffected but wages will be lower for these more expensive workers (as in Gruber 1994). However, since older workers are often accustomed to getting higher salaries, it may be difficult for employers to cut their wages in order to offset health-care costs. These problems may be exacerbated in the future as health care costs are projected to increase over time. If older workers are unwilling to accept or firms are unable to offer lower wages based on health insurance costs, then firms will be reluctant to hire these workers. Additionally, if firms take health care costs into account for older workers but not for younger workers, then bias against older workers will result. 1 Lahey (2007b) finds some evidence that age discrimination laws decreased hiring of older men in states that had their own age discrimination laws compared to men in states that did not have these laws, but this finding cannot account for all differential hiring. 2

4 Firms may react to higher health insurance costs through other methods besides cutting back on wages or employment. For example, they could change the composition of their firms by reconfiguring their workforce to include fewer workers who would be covered by insurance. Along these lines, Cutler and Madrian (1996) suggest that rising insurance costs encourage companies that offer insurance to hire fewer workers for more hours. Part-time jobs do not have to offer insurance benefits, so firms may also increase their use of part-time workers compared to full-time workers. It is difficult to determine the extent to which health care and health insurance costs affect hiring and other employment decisions. Scott et al. (1995) find that older-age hiring is lower in firms that offer health insurance. However, firms that offer benefits such as health insurance are different than firms that do not. For example, firms that offer health insurance tend to be larger and have steeper earnings profiles (Idson and Oi 1999). Indeed, Hu (2003) suggests that large firms hire more younger workers than small firms do because they value long-term firm-specific human capital investments. Because of these other differences, it is not clear that hiring differences by age can be attributed to health insurance costs alone. Two recent papers look at whether there is a relationship between age differences in insurance costs and wages of older workers. Sheiner (1999) finds that in cities where health insurance costs are high, the age/wage profile is flatter, raising the possibility that older workers are in some sense paying for their insurance with reduced wages. However, cities with higher health insurance costs are different in many ways from those with lower costs; for example, they may differ in terms of HMO penetration, technological diffusion (Skinner and Staiger 2005), transportation costs, and comparable 3

5 wages of college graduates (Taylor and Fowler 2006), and these differences may also affect employment outcomes directly. An innovative study by Adams (2007) finds that when the state of New York enacted pure community rating, forcing insurers to charge a firm the same premium for older and younger workers, relative wages for older workers increased. It may, however, be easier for a firm to raise wages when benefit costs go down than to decrease wages when costs go up. Thus, while these two papers provide some evidence that wages may adjust based on health insurance costs, at least for currently employed workers, they cannot provide conclusive evidence on whether, as health costs for older workers increase, those workers accept lower wages dollar for dollar. If, in fact, wages do not adjust fully, then firm demand for older workers will decrease. I examine whether or not increased health insurance costs decrease employment for older workers by developing a new measure of average health care costs at the state/demographic-group level, using Medstat claims data. Because these costs may be directly related to wage and employment outcomes, I use state health insurance mandates as a source of exogenous variation. State health insurance mandates increase health care costs for some groups of workers but not others, because the incidence of the mandates often differs by demographic group. These mandates are plausibly unrelated to employment outcomes except through the channel of health insurance costs. Therefore I can use these state-level coverage mandates, which cause variation in the relative costs of covering different demographic groups, as an instrument for average health insurance costs by group. 2 2 To investigate possible labor-supply changes, which would interfere with my ability to measure labordemand changes, I looked at the effect of outside health insurance costs using Blue Cross/Blue Shield data 4

6 My results suggest that employers do react to increased health insurance costs for different demographic groups by becoming less willing to employ these groups. However, age is not the only factor that employers take into account when reacting to insurance costs; firms are no more sensitive to cost increases for older workers than they are for other groups. Thus I do not find evidence for employer bias against older workers in the weighing of benefit premia. Nonetheless, because older workers are typically more costly to insure than younger workers, elderly unemployment in the US can be partly accounted for by health care costs. Since the cost differential between older and younger workers may grow as overall health costs continue to rise, elderly unemployment may rise further in the future. II. Data I use Medstat data from 2001 to to create an index of how much a worker costs, in terms of actual average healthcare expenditure, by age/gender/state/year cell. Medstat data provide information on medical utilization and costs for enrollees in a variety of employer-provided health insurance programs. They include inpatient and outpatient health insurance claims filed by employees and their dependents in ninety-nine to two hundred (depending on the year of the survey) firms that self-insure. Neither employers nor health plans are identified by name in the database. The firms include a number of on the minimum cost for an independent BC/BS plan by age group in 22 states. I find that people with higher outside health care costs are not more likely to be employed. This finding suggests that costs for an outside insurance plan is not driving employment decisions of workers. However, these regressions are at best suggestive, because there may be an omitted variable that directly affects both outside health insurance costs and labor market outcomes, so I do not present them in the paper. I cannot use my state mandates instrument for these regressions because individual health insurance plans are exempt from state health insurance mandates, and as such do not usually cover the mandated services. 3 Years when the state identifiers are available unrestricted. 5

7 different industries, health care costs, plan provisions, and workforce characteristics. Each claim includes a patient identifier, a provider identifier, the date of the medical service, the claim amount, the co-payment and deductible amounts paid by the patient, the place of service hospital, physician office, intermediate care facility, etc. and ICD-9 and CPT-4 codes identifying the principal diagnoses and procedures performed. Although the Medstat data do not give exact age of patients, they provide information by age groupings 18-34, 35-44, 45-54, and To create the actual cost index, I estimate actual healthcare costs from these Medstat data using claims for state s in year t which had claims corresponding to state mandates. I look only at outpatient claims in which the employee s status is full time active, part time active or seasonal or COBRA continue. Within this restricted dataset, I calculate mean net pay, which I define as the average amount spent by employers on employee outpatient claims for each age-group/sex demographic group in state s in year t. I match this Medstat data for each year, state, and gender-age group to the Current Population Survey (CPS) to obtain outcomes on employment characteristics. 4 Figure 1 depicts actual normalized cost data from Medstat for different age groups that has been averaged for all states and plotted by age, together and separately by gender. As this figure demonstrates, actual health insurance costs increase with age. To obtain a source of variation in this cost index that is arguably exogenous to other employment outcomes, I use information on health insurance mandates. These are laws requiring insurance companies to include coverage for certain types of treatments in 4 Miriam King, Steven Ruggles, Trent Alexander, Donna Leicach, and Matthew Sobek. Integrated Public Use Microdata Series, Current Population Survey: Version 2.0. [Machine-readable database]. Minneapolis, MN: Minnesota Population Center [producer and distributor],

8 all of their insurance plans; these mandates are widely believed to raise the cost of coverage (e.g. Texas Department of Insurance 2005, US GAO 1996). Because many mandates are not very expensive, I include in my cost gradient only what are regarded as the five most expensive health insurance mandates: infertility treatment, mental health treatment, alcohol treatment, substance abuse treatment, and chiropractors (Milliman and Robertson 1997, Ralston et al. 1988, Texas Department of Insurance 2005, US GAO 1996). These different mandates have a different use incidence depending on the age, state, and gender of the potential worker. Therefore, by combining the Medstat data with information on state health insurance mandates, I can predict how much state legal requirements for coverage increase the cost of coverage for a worker in a given agegroup/gender/state/year cell. To create this mandated-cost index, I again restrict to outpatient claims for state s in year t where the employee s status is full time active, part time active or seasonal or COBRA continue. In addition, I restrict to procedures relevant to that subset of the five mandates that were present in state s in year t. Within this restricted dataset, mean mandated net pay is calculated as the average amount spent by employers on employee outpatient claims for each age/gender demographic group in state s in year t. This represents the average marginal mandated amount for state s in year t. Because this mandated-cost index is directly related to health care costs, but not necessarily related to other employment-related decisions, I can use it to instrument for the actual-cost index described above. I match this instrument by year, state, gender, and age-group to the CPS to predict the health care costs of individual potential workers. Table 1 presents summary statistics for these measures. Although actual costs 7

9 increase by age, it is important to note that the majority of the most expensive mandates primarily affect younger workers, and thus mandated costs decrease with age. Figure 2 shows normalized mandated cost data for the five most expensive mandates from Medstat averaged for all states and plotted by age. III. Methodology A. Main Regression I first examine the effect of actual health insurance costs on employment outcomes for older workers using both OLS and probit regression: Y (1) i = β 1netpaytotalasgt + X iβ2 + ϕ + σ + ε where Y includes outcomes such as Employed, Unemployed, Retired, ln(hourly wage), full-time and part-time. The variable netpaytotal is a variable indicating the actual average cost of insurance in a given age/state/gender/year cell. X is a vector of controls including gender, high school graduation and race. 5 Age dummies are given by ϕ and state dummies are given by σ. In some specifications, standard errors are clustered at the cell level. This equation assumes that health care costs are not related to other variables that affect employment outcomes and that health insurance costs drive employment outcomes rather than employment outcomes driving health care costs. The coefficient of interest, β 1, indicates the effect that health care costs have on employment outcomes under these assumptions. B. Instrumented Regression Because many things could influence both employment and wage levels and also increase health care costs, it is useful to use an instrumental variables (IV) procedure to 5 All regressions have also been estimated with a richer set of controls, including marital and veteran status. The results were unchanged. 8

10 instrument for health care costs with something that is less directly related to other costs. State insurance mandates requiring coverage of certain conditions or providers increase insurance costs for different demographic groups (Goodman et al. 1997, Jensen and Morrisey 1999, Price Waterhouse Coopers 2002). However, they are not necessarily directly related to employment outcomes except through the consequences of these healthcare costs. A good instrument will both be related to the outcome of interest, that is, state mandated costs will predict actual costs, and will be unrelated to other variables that predict actual costs; before using mandated costs as an instrument, it is necessary to check whether these two requirements hold. To check to see that estimated costs due to state mandates by demographic group predict actual costs, the first stage regression is given by the equation: netpaytotal astg = β 1 mandatednetpay astg + β 2 female i + ϕ + τ + σ + ε (2) where netpaytotal astg is the total actual costs found by age group/state/year/gender cell using the Medstat data, mandatednetpay astg is the predicted cost by age group/state/year/gender cell based on the five nationally most expensive mandates, female denotes female, ϕ is a set of age dummies, τ is a set of year dummies and σ is a set of state dummies. Because the regular netpay variables may be driven by states with higher costs, I also run this regression with normalized versions of netpaytotal astg and mandatednetpay astg, where each the value in each cell is divided by the value for the most expensive cell within that state, so that each index runs from zero to one. If 9

11 mandatednetpay astg is a valid instrument, β 1 would be expected to be positive and significant. The second stage instruments the netpaytotal astg variable with this mandatednetpay astg variable using two-stage least squares. Thus: netpaytotal α mandatednetpay + Xα + U u (3) astg = 1 astg 2 α3 + Y i = β 1netpaytotalastg + X iβ 2 + Uβ 3 + ε (4) where X is a set of controls including gender, high school graduation and race as well as dummies for age, state, and year. U includes variables of unobserved heterogeneity that may affect both mandated net pay and actual total net pay. Y includes labor market outcomes such as indicators for employed, unemployed, not in the labor force, full-time, and part-time, and continuous variables such as the natural log of hourly wage. To determine whether older workers are being singled out for their higher health insurance costs, I also examine age interactions with costs, and with costs instrumented with mandated costs: Y i = β 1 netpaytotal astg *over50 + β 2 netpaytotal astg + X i β 3 + ϕ + σ + ε (5) where the variables are defined as above, and over50 is an indicator for whether or not the individual is age 50 or older. IV. Results A. Medstat-CPS Results Without using an instrument for health care costs, OLS regressions using the actual cost index, called netpaytotal, show no significant effect of health care costs on employment in Table 2b, columns (1), (3), and (5). Higher health care costs are 10

12 associated with lower wages for women in column (11) with a coefficient of and a standard error of , although this could be because women with lower wages are likely to have higher health care costs because of some omitted variable. Normalized versions of the independent variable given in Table 3a show the same patterns, but the OLS result for the decrease in wage of women in column (11) is only marginally significant. OLS effects on full-time, part-time and not in the labor force (NILF) are all insignificant at conventional levels. However, there may be omitted variables related to both health insurance costs and employment outcomes that are driving these results in different directions. 6 To overcome this problem of potential omitted variables bias, I instrument for the actual cost of health insurance using state-level changes in health insurance mandates. As noted earlier, in order for mandated health care costs to be a good instrument for actual costs, it must be highly correlated with actual costs, as given by equation (4). The Medstat data show a very strong positive correlation between predicted costs from state mandates by age group/state/year/gender cell and actual costs at these cell levels in Table 2a. This positive correlation provides evidence for a strong first stage mandated costs positively predict actual costs. For the second stage of the instrumental variables technique, I estimate equations (3) and (4), instrumenting actual netpay (netpaytotal) with mandated netpay. The results of these regressions are presented in Tables 2b and 3. Both the actual and the normalized 6 I also ran similar regressions using the CPS Displaced Workers Survey (DWS), which follows people who lost their jobs due to a lay-off or plant closing, matched to the Medstat cost data. I looked at probability of re-employment by health-care cost for people who have been exogenously separated from their jobs. Unfortunately, there is only one year of the DWS that matches the Medstat data that I have available. The results I obtain are similar to those presented from the instrumented results described in the next section, but do not remain significant with state fixed effects added so I have not presented them in this paper. 11

13 costs show a negative relationship between employment and health care costs. The normalized results shown in Table 3a, column (4) show that a one percentage point increase in relative health care costs decreases the probability of employment by 1.39 percentage points for men. These results are negative but smaller in absolute value and not significant for women in column (6). 7 Since costs were instrumented by mandated health care costs, and many of the most expensive mandates are negatively correlated with age, this finding is not just the effect of employers being less likely to hire older potential workers. If workers with higher health care costs were willing to take a pay-cut to offset the costs of insurance, then we should not see an effect on employment for these groups. Employers would be indifferent between hiring a more expensive (in terms of health care costs) worker at a lower wage or a less expensive worker at a higher wage. The nonnormalized results of instrumented health care costs on wages presented in table 2(b), columns (8), (10) and (12), and the normalized results in table 3(a), columns (8), (10) and (12), show that there is no significant negative effect on wages for these groups. In general, the effect of health care costs on wages for this sample is positive and not significant. Thus it does not appear that workers with higher health care costs are willing to take wage cuts to offset higher health care costs. These instrumented results contrast with the OLS results for women reported above in Table 2b column (11); although women with lower wages have higher healthcare costs in general, this effect is not causal. Workers or employers may also adjust hours based on health care costs, since part-time workers do not have to qualify for health insurance by law. Table 3(b) provides 7 Significance diminishes when standard errors are clustered on cell; nevertheless, the results remain highly suggestive that as health care costs of a demographic cell increase, employers are less likely to hire people in that cell. 12

14 results on other outcomes by instrumented health insurance cost. I find that among men, those who cost more are less likely to be working full time (columns (2) and (4)). Men who cost more are also more likely to be working part-time, (columns (6) and (8)), and more likely to be not in the labor force (columns (10) and (12)). The above regressions demonstrated that higher health care costs lead to decreased employment overall. However, an additional question is whether or not employers are especially likely to discriminate against older workers based on their higher health care costs. Table 4 shows the results of an instrumented interaction between the netpay variable and an indicator for over age 50 (over50) given by equation (5). This interaction for the outcome of employed in column (2) is positive and significant for men (with a coefficient of and standard error.001), indicating that older men are more likely to be employed than younger men when each group s mandated health care costs increase. The interaction is negative and significant for men for the not in the labor force (NILF) outcome in column (6), with a coefficient of and a standard error of 0.001, suggesting that older men are less likely to be out of the labor force than younger men when their instrumented health insurance costs go up. Since increases in the instrument (the introduction of new mandates) primarily cause younger workers to become relatively more expensive, employers may realize that older workers are now relatively less expensive than younger and may discriminate less than they did prior to the mandate. Alternatively, employers may find it easier to fire or refrain from hiring younger workers based on health insurance costs because of age discrimination legislation, which would make the employment of younger workers more 13

15 sensitive to employment costs. Interacted results for other outcomes wages, part-time employment, and full-time employment are insignificant. V. Conclusions The question of why employers are less likely to hire older workers is an important one. This study suggests that health insurance costs are one of the reasons that employers are less likely to employ older workers. Demographic groups that have higher health insurance costs are less likely to be employed, and older workers, who are particularly likely to have high costs, are no exception to this rule. Four of the five most expensive health insurance mandates increase the relative cost of younger to older workers. In states where these mandates are present and older workers cost relatively less, employment rates for older workers are relatively higher than in states where older workers cost relatively more. These effects occur entirely through employment for men; older men are apparently unlikely to take wage cuts to compensate for their higher health insurance costs. However, for the most part employers treat all workers with higher health care costs similarly and do not specifically target older workers through taste-based discrimination. If anything, the relationship between health care costs and employment is weaker for older men than for younger men. If in the future more states mandate expensive provisions that primarily affect younger workers, thus bringing up the relative potential cost of a younger worker, employment may further increase for older workers. On the other hand, if further advances in health care at older ages cause the cost of older workers to rise faster than the cost of younger workers, then elderly unemployment may rise further in the future. In 14

16 addition, it is important to note that, health insurance costs may not be the only reason for lower employment opportunities for older job seekers and there is room for further research in this area. 15

17 Works Cited Adams Scott J. (2007)"Health Insurance Market Reform and Employee Compensation: The Case of Pure Community Rating in New York," Journal of Public Economics, 91(5-6), June 2007: Chesner, M. A. (1991). Group Comprehensive Major Medical Net Claim Cost Relationship by Area, Milliman and Robertson. Cutler, D. M. and B. C. Madrian (1996). Labor Market Responses to Rising Health Insurance Costs: Evidence on Hours Worked, National Bureau of Economic Research, Inc, NBER Working Papers: 5525, Diamond, Peter A. and Jerry A. Hausman "The Retirement and Unemployment Behavior of Older Men." Retirement and Economic Behavior. H. J. Aaron. Studies in Social Economics Series. Washington, D.C.: Brookings Institution. GAO (August 1996) United States General Accounting Office, Report to the Honorable James M Jeffords, US Senate Health Insurance Regulation Varying State Requirements affect cost of insurance. GAO-HEHS Goodman, J. C. and Matthews J. M. (1997). "The Cost of Health Insurance Mandates." National Center for Policy Analysis. Gruber, J. (1994). "The Incidence of Mandated Maternity Benefits," American Economic Review, American Economic Association, vol. 84(3), pages , June. Hu, L. (2003). "The Hiring Decisions and Compensation Structures of Large Firms." Industrial and Labor Relations Review 56(4): Idson, T. L. and W. Y. Oi (1999). "Workers Are More Productive in Large Firms." American Economic Review 89(2): Jensen, G. A. and M. A. Morrisey (1999). "Small Group Reform and Insurance Provision by Small Firms, " Inquiry 36(2): King, M. S. Ruggles, T. Alexander, D. Leicach, and M. Sobek. Integrated Public Use Microdata Series, Current Population Survey: Version 2.0. [Machine-readable database]. Minneapolis, MN: Minnesota Population Center [producer and distributor], Lahey, J. N. (2007a). Age, Women and Hiring: An Experimental Study. Journal of Human Resources (Forthcoming). Lahey, J. N. (2007b). Age Discrimination Laws and the Age Discrimination in Employment Act. Journal of Law and Economics 51 (Forthcoming). 16

18 Millman, R. (1997). "Estimated Additional Cost for Certain Benefits, 1997." National Center for Policy Analysis, from Price Waterhouse Coopers (April 2002), Factors Fueling Rising Healthcare Costs, Prepared for the American Association of Health Plans, Washington DC. Ralston, A., M. Power, et al. (1988). State Legislatively Mandated Life and Health Insurance Coverages. Legislative Extended Assistance Group, Iowa General Assembly. Rhine, S. H. (1984). Managing older workers: company policies and attitudes: a research report from the Conference Board. New York, N.Y., The Board. Scott, F. A., M. C. Berger, et al. (1995). "Do Health Insurance and Pension Costs Reduce the Job Opportunities of Older Workers?" Industrial and Labor Relations Review 48(4): Sheiner, L. (1999). Health care costs, wages, and aging, Board of Governors of the Federal Reserve System (U.S.), Finance and Economics Discussion Series: , Skinner, J. and D. Staiger (2005). Technology Adoption From Hybrid Corn to Beta Blockers. NBER Working Papers 11251: Taylor, L. L. and J. William J. Fowler (2006). "Comparable Wage Approach to Geographic. Cost Adjustment " NCES : Texas Department of Insurance (January 2005), Texas Mandated Benefit Cost and Utilization Summary for Calendar Year

19 Table 1 Summary Statistics for Medstat Index/CPS Matched Data All Male Female Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. (1) (2) (3) (4) (5) (6) employed unemployed lnweeklywage weeks worked over mandated netpay total netpay normalized mand. netpay normalized total netpay black married female age year Observations

20 Figure 1: Average Actual Netpay by Age 1 Average Actual Netpay Age netpaytotall netpaytotmen netpaytotwomen 19

21 Figure 2: Average Normalized Mandated Netpay by Age Average Normalized Mandated Netpay Age netpayall netpaymen netpaywomen 20

22 mandated netpay female Observations R-squared State dum? Age dum? Year dum? Table 2a 1st Stage Non-Normalized Netpay Variables All Men Women All (1) (2) (3) (4) (0.001)** (0.002)** (0.001)** (0.001)** (0.032)** (0.000)** Robust standard errors in parentheses. Controls for high school graduation and race are also included in the regressions. * significant at 5%; ** significant at 1% Normalized Netpay Variables Men (5) (0.001)** Women (6) (0.001)** Table 2b 2nd Stage Non-Normalized Netpay Variables Employed lnhourlywage All Men Women All Men Women OLS IV OLS IV OLS IV OLS IV OLS IV OLS IV (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) total netpay (0.0001) (0.002) (0.0001) (0.003)* (0.0001) (0.010) (0.0002)** (0.005)* (0.0002) (0.010) (0.0003)* (0.024) female (0.0016)** (0.018)** (0.0034)** (0.045)** Observations State dum? Age dum? Year dum? Robust standard errors in parentheses. Controls for high school graduation and race are also included in the regressions. * significant at 5%; ** significant at 1% 21

23 Employed Table3a 2nd Stage Normalized Netpay Variables lnhourlywage All Men Women All Men Women OLS IV OLS IV OLS IV OLS IV OLS IV OLS IV (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) normalnetpaytot (0.0087) (0.250)* (0.0112) (0.524)** (0.0145) (0.366) (0.0186)** (0.487) (0.0252) (1.101) (0.030) (0.553) female (0.0016)** (0.022)** (0.0035)** (0.042)** Observations State dum? Age dum? Year dum? Table3b 2nd Stage Netpay Variables Men Only Full-Time Part-Time NILF OLS IV OLS IV OLS IV OLS IV OLS IV OLS IV (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) total netpay (0.0001) (0.004)* (0.0001) (0.004)* (0.0001) (0.002)** normalized total netpay (0.0131) (0.592)** (0.0131) (0.592)** (0.0065) (0.473)** Observations State dum? Age dum? Year dum? Robust standard errors in parentheses. Controls for high school graduation and race are also included in the regressions. * significant at 5%; ** significant at 1% 22

24 Table 4 Instrumented Age Interactions Employed Log Hourly Wage NILF Part-time Full-time All Men All Men All Men All Men All Men (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) over50*netpay (0.001) (0.001)* (0.002) (0.003) (0.001) (0.001)** (0.001) (0.002) (0.001) (0.002) netpay (0.002) (0.005)* (0.005)** (0.010) (0.002) (0.005)** (0.003) (0.006)* (0.003) (0.006)* female (0.020)** (0.043)** (0.019)** (0.022)** (0.022)** Observations Age dummies? State dummies? Year dummies? Robust standard errors in parentheses. Controls for high school graduation and race are also included in the regressions significant at 10%; * significant at 5%; ** significant at 1% 23