Compensating wage differentials and the impact of health insurance in the public sector on wages and hours

Journal of Health Economics 38 (2014) 77–87

Contents lists available at ScienceDirect

Journal of Health Economics journal homepage: www.elsevier.com/locate/econbase

Compensating wage differentials and the impact of health insurance in the public sector on wages and hours Paige Qin a , Michael Chernew b,∗ a b

Harvard University, United States Harvard Medical School, 180A Longwood Avenue, Boston, MA 02115, United States

a r t i c l e

i n f o

Article history: Received 7 November 2013 Received in revised form 30 July 2014 Accepted 1 August 2014 JEL classification: H7 I1 J3 Keywords: Public Health insurance Wage Tradeoff Hours worked

a b s t r a c t This paper examines the trade-off between wages and employer spending on health insurance for public sector workers, and the relationship between coverage and hours worked. Our primary approach compares trends in wages and hours for public employees with and without state/local government provided health insurance using individual-level micro-data from the 1992–2011 CPS. To adjust for differences between insured and uninsured public sector employees, we create a matched sample based on an employee’s propensity to receive health insurance. We assess the relationship between state contribution to the health plan premium, state-level healthcare spending, and the wages and hours of state and local government employees. We find modest reductions in wages are associated with having employersponsored health insurance (ESHI), although this effect is not precisely measured. The reduction in wages associated with having ESHI is larger among non-unionized workers. Further, we find little evidence that provision of health insurance increases hours worked. © 2014 Elsevier B.V. All rights reserved.

In July 2012, the State Budget Crisis Task Force, led by former New York Lieutenant Governor Richard Ravitch and former Federal Reserve Board Chair Paul Volcker, released a report that examined the major threats to states’ fiscal sustainability in the aftermath of the 2008 financial collapse. State and local expenditures on Medicaid and health care compensation for current employees and retirees were identified as the leading causes of long-term fiscal imbalances for state and local governments (State Budget Crisis Task Force, 2012). In an estimate provided by the United States Government Accountability Office (GAO) in April 2012, health-related spending for state and local governments would be around 3.9% of national GDP in 2012 and 7.1% of GDP in 2060. In contrast, the sector’s non-health-related spending—such as the wages and salaries of state and local employees—was projected to decline as a percentage of national GDP, from about 10.4% of GDP in 2012, to 7.8% of GDP in 2060 (GAO, 2012).

∗ Corresponding author. Tel.: +1 617 432 0174. E-mail address: [email protected] (M. Chernew). http://dx.doi.org/10.1016/j.jhealeco.2014.08.001 0167-6296/© 2014 Elsevier B.V. All rights reserved.

This paper explores the relationship between insurance coverage and wages and hours of state and local government workers. Additionally, we explore the variation in state contribution to health insurance premium as well as state personal health care spending and examine their association with the wages and hours of public sector employees. In total, state and local governments reported an annual spending of $2.5 trillion in 2009 and employ over 19 million workers, or 15% of the national work force and 6 times as many employees as the federal government (State Budget Crisis Task Force, 2012). These workers include state and local government administrators, but also teachers, police officers and hospital employees. At the same time, almost all states have balanced operating budget requirements, which restrict borrowing across fiscal years.1 Furthermore, the structural imbalance in state budgets is exacerbated by the financial collapse of 2008; it took until the third quarter of

1 State and local governments often do not pre-fund liabilities such as pension and health insurance obligations to retirees. As these liabilities increase over time, the full costs of these benefits are not fully funded by state and local governments.

78

P. Qin, M. Chernew / Journal of Health Economics 38 (2014) 77–87

2011 for state and local government total tax receipts to return to pre-recession levels of 2007 (State Budget Crisis Task Force, 2012). Compared to their counterparts in the private sector, most public sector employees have employer-sponsored health insurance, enjoy lower deductibles and pay a much smaller share of the higher premium as a result of the lower deductibles (Clark et al., 2012). The theory of compensating wage differentials predicts that holding human capital and other variables influencing wages constant, individuals receiving higher fringe benefits are paid a lower wage than those receiving lower fringe benefits (Rosen, 1986). Therefore, as spending on health insurance rises, employers that provide workers with health insurance will lower wages, in order to keep total compensation the same (Summers, 1989). However, the situation in the public sector is complicated by state and local governments’ limited ability to adjust wages, as salaries and employee benefits are often determined by union contracts. The literature also examined the labor market’s response to rising employer spending on health insurance along other margins. Given health insurance’s status as a fixed cost per worker and wages as marginal cost per hour worked, an increase in fixed costs relative to marginal costs has led firms to substitute longer work weeks per employee for additional number of workers (Cutler and Madrian, 1998). The paper proceeds as follows. Section 1 presents previous evidence on labor market responses to increased health insurance spending. Section 2 describes the individual and state-level data we use. Section 3 contains the econometric methodology used to estimate the compensating wage differentials and changes in hours worked. Section 4 presents the results and concludes given the empirical findings.

1. Evidence on labor market responses to increasing spending on health insurance A standard compensating wage differential framework would involve regressing wages on the availability or cost of employersponsored health insurance, with an expected negative coefficient on the health insurance variable. However, as Currie and Madrian (1999) put it, most estimates of the average market value of employer-sponsored health insurance (ESHI) are either positive (wrong-signed), insignificant, or both. The challenge lies in eliminating omitted variable bias where unobserved human capital variables are often correlated with employer-sponsored health insurance status (i.e., more capable workers are more likely to receive employer-sponsored health insurance and higher wages). Of the recent papers that examine the relationship between wages and having health insurance, there is variation in the estimated value of health insurance as a percentage of wage compensation depending on sample selection and the estimation techniques used. For example, taking advantage of the rotating panel design of the Consumer Expenditure Survey (CEX) to track workers who changed health insurance status between the 2nd and 5th interviews, Miller (2004) used person fixed effects and found that having health insurance led to a 10–11% wage reduction among prime-aged male workers. Using husband’s firm size and union membership as instruments for wife’s health insurance coverage, Olson (2002) found that health insurance was valued at 20% of overall wages among employed married women. Studies that estimate the wage offset as a function of employer spending on health insurance report full or nearly full cost shifting to wages (Eberts and Stone, 1985; Gruber and Krueger, 1991; Lubotsky and Olson, 2010). There is also evidence of group-specific cost shifting—i.e., relatively slower wage increases for particularly

expensive groups such as older workers, workers with family insurance coverage and women of child-bearing age (Sheiner, 1995; Gruber, 1994). The implication that full cost shifting of employer health insurance payment to employee wages should have no effect on the equilibrium level of labor utilization is empirically confirmed by several studies (e.g., Gruber and Krueger, 1991; Gruber, 1994). However, despite the lack of overall change in labor input, the rise in health insurance spending has led to changes in the compositional mix of labor utilization—specifically, employers are responding to the rise in fixed employment costs by increasing hours per insured worker and decreasing employment (Cutler and Madrian, 1998; Gruber, 1994). There is somewhat mixed evidence on whether employers are expanding the share of the workforce that is ineligible for benefits (Montgomery and Cosgrove, 1993; Buchmueller, 1999). Several important issues remain unaddressed in the current literature. First, most of the findings on compensating wage differentials and hours do not distinguish between public and private sector workers, or are solely based on private sector industries. Cutler and Madrian (1998)’s finding that hours rose the most in industries with the fastest growth rates of health care spending excludes public sector employees. Very few papers to date have examined whether the competitive, private sector model of health benefits-wage trade-offs exists in the public sector, where wages are more rigid. Using data on school district finances in 1998 and 2007, Clemens and Cutler (2013) found that only a small fraction (around 15%) of the growth in benefit costs—including increases in both health care and pension costs—are offset through reduction in the wages of school district employees. In an earlier study of retirement system characteristics and wages of uniformed municipal employees, Ehrenberg (1980) found that increased employee pension contributions led to a compensating increase in their salaries and more generous retirement systems are associated with lower wages. To the extent that health care compensation for its employees is threatening the fiscal sustainability of state governments, it becomes important to gather evidence on the response of public sector wages and hours to rising health care spending. An important consideration when studying health insurance spending-wage trade-offs in the public sector are the roles played by public sector unions in negotiating employee benefits and wages. As noted by DiSalvo (2010), the dramatic growth of public sector unions during the second half of the 20th century has led unionized public sector employees (7.9 million) to overtake unionized private workers (7.4 million) in numbers for the first time in 2009. The aggressive political actions taken by public sector unions to both increase union members’ total compensation and the size of the public sector—using their unique position as the monopoly provider of government services—have occasionally been met with reformist moves, such as when New Jersey Governor Chris Christie issued an executive order banning state workers’ unions from making political contributions; however, rarely have these political counter-attacks against union power been successful. A recent paper (Anzia and Moe, 2012) found that unions increase the costs of government in the form of higher wages, better benefits and job protection for their workers, and these effects are substantively significant; specifically, municipal fire departments with collective bargaining during the 1990s spent 9% more per employee on salaries and wages, and 25% more on health and life insurance benefits, without decreasing fire protection employment levels. Another paper, Edwards (2010), quantified the increase in total compensation costs for state and local workforce due to public unionism to be 8.1%. Clemens and Cutler (2013) found that strong teachers unions mediate the relationship between benefit growth and increases in

P. Qin, M. Chernew / Journal of Health Economics 38 (2014) 77–87

total school district costs, and benefit costs tend to be shifted back onto workers where teachers’ unions are weak. Second, most estimates of health insurance-wage trade-off such as those in Miller (2004) and Olson (2002) use a single dummy variable for health insurance status, mostly due to data limitations. However, this obscures important distinctions, such as the employee’s eligibility for health insurance aside from his or her coverage status and the relative contributions to the health insurance premium made by the employer/employee. The CPS individual health insurance data we use partially solves this problem, in that we include a categorical variable on the extent of the employer’s contribution (none, partial or all). We also incorporate state-level data collected by the National Conference of State Legislators (NCSL) on total premium, state contribution to the premium and the required employee contribution for family coverage under the state health benefit plan. Lastly, the time period covered by our data, 1992–2011 is much more current than those used in previous estimates of labor market responses to rising health insurance spending, in which the most recent individual-level micro-data comes from 1993.

2. Data and summary statistics

79

Table 1 Sample means by employer-sponsored health insurance (ESHI) status in the public sector. Variable

With ESHI

Without ESHI

Hours per week

42.0 (6.9) $15.34 ($6.65) 41.1 (8.0) 0.72 (0.45) 0.028 (0.16) 0.23 (0.42) 0.18 (0.38) 0.57 (0.50) 0.47 (0.50)

38.9 (9.3) $12.92 ($6.55) 39.4 (8.6) 0.72 (0.45) 0.053 (0.22) 0.25 (0.43) 0.17 (0.38) 0.53 (0.50) 0.32 (0.47)

Hourly wage ($1992) Age Married Less than high school High school graduate Some college College graduate Union member

Data source: 1992–2011 March Current Population Survey (CPS) and Merged Outgoing Rotation Groups (MORG). Sample includes men aged 25–54, with a real ($1992) hourly wage between $4.05 and $42, who were not self-employed and who worked at least 40 weeks in the previous calendar year. Standard deviations are in parentheses.

2.1. Individual-level wages and hours data We linked person records from the 1992–2011 March Current Population Surveys (CPS), which contain demographic and health insurance data, with those in the Merged Outgoing Rotation Groups (MORG) which report wages and earnings of household members. The March CPSs distinguish the class of worker by reporting whether the individual is employed by the private sector, the public sector (federal, state or local government) or is self-employed. The data set also indicates whether an individual is the policy holder for employer-sponsored health insurance. Additional information on health insurance includes a categorical variable for whether an employer paid for all, part or none of the health insurance premium. Cross sections from the years 1992–2011 were then pooled together and following Cutler and Madrian (1998), analysis was restricted to prime-aged males between the ages of 25–54 employed in state or local governments. The focus on the labor market outcomes of men reflects the concern that the availability of public insurance such as Medicaid to women of child-bearing age and the potentially different labor market responses to health benefits by men and women could influence results. The age restrictions serve to eliminate changes in labor force participation due to school or retirement. The advantage of using earnings data from MORG, instead of the March CPSs is that the March earnings data covers the previous calendar year rather than the time of the survey. Following Cutler and Madrian (1998), we restrict the sample to employees who worked 40 or more weeks because employers are not required to provide health insurance to part-year workers, even if they work full-time when employed. This restriction eliminated 7.11% of the sample of state and local government workers or 3063 individuals, who worked less than 40 weeks per year. Hourly wages are measured by dividing earnings per week by the usual hours worked per week at the employee’s main job. Individuals with earnings below $4.05 per hour in 1992 dollars (approximately the minimum wage during the period; 391 individuals, or 0.98% of the public sector sample) or above $42 per hour in 1992 dollars (roughly the lowest top-coding level for earnings per week over the entire period; 2301 individuals, or 5.8% of the sample) are excluded. Table 1 presents the summary statistics for state and local government employees based on their employer-sponsored health

insurance (ESHI) status. On average, those with health insurance worked 42.0 h each week, while those without worked less, at 38.9 h per week. The average hourly wage for those with health insurance was $15.34 ($1992), and for those without ESHI was $12.92. Public sector workers with ESHI were older with an average age of 41.1 years (vs. 39.4 years for workers without ESHI), and more likely to hold a college or graduate degree (57% vs. 53% for those without ESHI). Forty-seven percent of employees with ESHI were union members, while only 32% of those without belonged to a union. 2.2. State contribution toward health insurance premium The National Conference of State Legislators (NCSL) reports the monthly premium, the state contribution to the premium and the required employee contribution for family coverage under 50 states’ employee health benefit plans. Data is available for years 1999–2006, 2009 and 2011. Up until 2009, the “standard benefit package” for family coverage was the lowest cost full-service HMO plan. In 2011, NCSL recorded premiums and state/employee contribution for a PPO and a lowest cost family plan. Following Clemens and Cutler (2013), we use figures from PPOs in 2011, as low cost plans likely understate state governments’ contribution to employee health plans. 2.3. State-level per capita personal health care spending Per capita personal health care spending data by state of residence is available from the National Health Expenditures Data provided by the Center for Medicaid & Medicare Services (CMS). Personal health care spending includes the total amount spent on the treatment of specific medical conditions, but excludes administrative costs, net spending on health insurance provision, government public health activity, non-commercial research, and investment in structures and equipment (Cuckler et al., 2011). To produce per capita personal health care spending by state of residence, the CMS Office of the Actuary adjusts health spending estimates by state of provider to account for individuals traveling across states for health care services.

80

P. Qin, M. Chernew / Journal of Health Economics 38 (2014) 77–87

3. Econometric methodology 3.1. Relationship between increasing health care spending, and wages and hours 3.1.1. Basic model The basic model follows the identification strategy as Cutler and Madrian (1998), estimating linear time trend models using hourly wages and hours worked per week as the outcome variables. In the following individual wage/hour determination equation: Ln(hourly wage) or hours worked = ˇ0 + ˇ1 ∗ HI + ˇ2 ∗ Year + ˇ3 ∗ (HI ∗ Time) + Z   + ε (1)

The dummy variable HI indicates whether the employee is the policy holder for employer (or union) sponsored health insurance.2 Year is a vector of year dummy variables. Time is a linear time trend. Z includes a set of demographic and job characteristics that influence individual wages or labor supply, such as age, age squared, marital status, education level (dummy variables for less than high school, high school degree, some college and college degree or beyond), state of residence and CPS industry and occupation codes. To reflect the increasing returns to education, we also include interaction variables of education dummies with a linear time trend (Katz and Murphy, 1992). The coefficient of interest, ˇ3 , captures the effect over time of having employer-sponsored health insurance on wages and hours worked relative to workers without employer-sponsored health insurance through state and local governments. According to the theory of compensating wage differentials, with hourly wage as the outcome variable, we would expect ˇ3 to be negative as the cost of fringe benefits rose. On the other hand, with hours worked as the outcome variable, based on findings of the use of overtime labor, we would expect ˇ3 to be positive to reflect lengthened work weeks for those with health insurance. It is worth noting that the estimation strategy will only be valid if state and local government employees with and without employersponsored health insurance would have similar trends in wages and hours conditional on demographic characteristics, in the absence of health insurance (Cutler and Madrian, 1998). In other words, the differences between the two groups are assumed to be the same over time. However, there are many reasons to believe that state and local government employees with different health insurance statuses will have different trends in wages and hours worked. Given that full-time public sector workers are much more likely than part-time employees to receive employer-sponsored health insurance, the estimate of the effect over time on wages of having health insurance may be biased due to different wage trends for full-time vs. part-time employees. The latest report on Employee Benefits in the United States released by the Bureau of Labor Statistics reveals that in March 2013, 99% of full-time state and local government workers had access to employer-sponsored health insurance, and 84% actually participated in these plans. In contrast, only 24% of part-time state and local government workers

2 About 42% of our sample are union members. Our data do not allow us to distinguish between union or employer sponsored coverage. For simplicity, we refer to coverage as employer sponsored coverage. However, the wage benefit tradeoff would apply only to employer sponsored coverage. We believe that most union members with health insurance coverage would get coverage through their employer, so we expect any bias would be small. Moreover, we control for union membership in the multivariate model.

had access to medical care benefits, and 17% actually participated in these plans (Bureau of Labor Statistics, 2013). While our data only records participation, and not access to employer-sponsored health insurance, it shows that during 1992–2011, 88% of full-time state and local government employees had employer-sponsored health insurance while 49% of part-time employees received such benefits. The relatively high rate of employer-sponsored health insurance among part-time workers recorded in our dataset may be due to the sample selection criteria for men ages 25–54 who worked 40 weeks or more. Also, the bias attributed to different wage trends for full-time and part-time workers in affecting our estimate of the wage trade-off is unlikely to be large, as part-time workers (N = 1107) consist of only 3% of our overall sample. Access to health insurance can also be contingent upon permanent vs. temporary worker status. While we restrict our sample to employees who worked 40 or more weeks each year, it may not be sufficient in distinguishing contractors from career employees; thus, to the extent that the wage premiums and working hours changed for permanent and temporary workers during 1992–2011, it will confound our estimation of the compensating wage differential due to health insurance. A source of compositional bias involves workers’ increasing self-selection into public sector jobs with generous medical care benefits as the cost of health insurance rose during this period. Thus, unobserved characteristics in the wages/hours determination equation such as risk-averseness that vary differently overtime between the two groups can also bias our estimation of the effect of rising health insurance costs on wages and hours. To address some of the concerns about the comparability of the two groups, a propensity score was developed to control for employee characteristics that influence whether public sector workers receive employer-sponsored health insurance. A logistic modeling process assigns each state and local government employee a probability (between 0 and 1) for receipt of employersponsored health insurance based on demographic characteristics such as age, education (highest grade obtained), marital and union status that also influence employee wages. To perform the match, we randomly sorted the sample of public sector employees with ESHI, before choosing the first worker and matching him with a worker from the randomly ordered sample of public sector employees without ESHI who has the most similar propensity score within a specified maximum distance. The process continued through all public sector workers with ESHI until no further matches could be found (Murray et al., 2003). We optimized the propensity score matching model by examining the balance—i.e., how similar employees with and without ESHI are on key variables—after matching. The model with the lowest standardized bias after matching was selected. We then conducted analysis in a sample of public sector employees matched on their propensity to receive employer-sponsored health insurance, by using a frequency weight that indicates the number of times the observation was used in the matching process. One source of bias not addressed by the propensity score model is the correlation between the receipt of health insurance and pension benefits in the public sector. Concurrent with the increased spending on employee health insurance by state and local governments are the rising costs of public sector pensions. As many states have raised the employee pension contribution rates as a percentage of wages in the wake of the Great Recession and potential state fiscal insolvency (NASRA Issue Brief, January 2014), the estimated health insurance spending-wage trade-off effect would be magnified. Unfortunately, the CPS individual-level data we use does not allow us to identify employee/employer contribution to public sector pension plans and control for employer spending on pensions.

P. Qin, M. Chernew / Journal of Health Economics 38 (2014) 77–87

81

5000

Further study is required to possibly isolate the effects of pension and health insurance spending growths on wages and salaries.

4500 4000 3500 1992 Dollars

3.1.2. Employer contribution toward health insurance We improve upon the basic model by comparing insured public sector employees with varying degrees of employer contribution toward health insurance. Aside from a dummy variable indicating whether an employee is the policy holder for employer-sponsored health insurance, the March CPS data also contains a categorical variable for whether the employer paid all, part or none of the health insurance premium. Therefore, we include dummy variables for zero, partial or full payment and interact them with a linear time trend, as in the slightly modified model below:

3000 2500 2000 1500 1000 500 0

Ln(hourly wage) or hours worked

Year

= ˇo + ˇ1 ∗ Paid Part + ˇ2 ∗ Paid All + ˇ3 ∗ Year

Fig. 1. Trend in per capita personal health care spending. Source: National Health Expenditure Accounts, Center for Medicaid & Medicare Services, 2011.



+ ˇ4 ∗ (Time ∗ Paid Part) + ˇ5 ∗ (Time ∗ Paid All) + Z  + ε (2a) The sample is now restricted to public sector employees with employer-sponsored health insurance, and we explore the variation in hourly wages and hours worked by differences in employer payment status. The omitted group includes employees for whom employers contributed zero dollars toward the health insurance premium. All control variables including demographic and job characteristics, as well as state fixed effects, remain the same as in Eq. (1). To incorporate NCSL’s data on average state government contribution to health plan premium for years 1999–2006, 2009 and 2011, we estimate the following among the sample of public sector employees with employer-sponsored health insurance: Annual earnings = ˇo + ˇ1 ∗ State Contribution + ˇ2 ∗ Year + Z   + ε

(2b)

or

We also include hours worked as the outcome variable, and estimate the trade-off between percentage increase in state-level health care spending and changes in usual hours worked per week. Hours worked = ˇo + ˇ1 ∗ HI + ˇ2 ∗ Log(PCPHCS) + ˇ3 ∗ (HI ∗ Log(PCPHCS)) + ˇ4 ∗ Year + Z   + ε (3b) The coefficient of interest ˇ3 captures the effect of a percentage increase in per capita personal health care spending on the usual hours worked per week of public sector employees with employersponsored health insurance. 4. Results 4.1. Trends in health care costs, wages, and hours of work

Hours worked = ˇo + ˇ1 ∗ Log(State Contribution) + ˇ2 ∗ Year + Z   + ε

(2c)

We substitute annual earnings, computed as weekly earnings times number of weeks worked per year, for log(hourly wage) as the outcome variable in Eq. (2b). State Contribution represents the average state contribution to the lowest cost full service HMO plan through 2009 or the PPO plan in 2011. 3.1.3. Incorporating state-level per capita personal health care spending We alternatively test the effect of state-level health care spending on state and local government employees’ wages and hours. We include annual earnings as the outcome variable, and alter Eq. (1) to include state-level per capita personal health care spending (PCPHCS) on the right hand side: Annual earnings = ˇo + ˇ1 ∗ HI + ˇ2 ∗ PCPHCS + ˇ3 ∗ (HI ∗ PCPHCS) + ˇ4 ∗ Year + Z   + ε

(3a)

The coefficient of interest ˇ3 captures the effect of a dollar increase in state health care spending on the annual earnings of employees with employer-sponsored health insurance coverage relative to uncovered employees. All demographic and job-specific controls, as well as state fixed effects, are the same as in Eq. (1).

Fig. 1 shows the trend in the national average of real ($1992) per capita personal health care spending from 1992 to 2009. Spending on health care steadily increased over this period, with a faster rate of growth since 2001. In 1992, the average per capita personal health care spending was about $2850; by 2009, the real health care spending had risen to about $4450, with a percentage increase of over 150%. Fig. 2 documents the trends in real hourly wages for public sector employees according to their employer-sponsored health insurance status. Real hourly wages increased by $1.40 between 1992 and 2011 for public sector employees without health insurance, and $0.58 for those with employer-sponsored coverage. While the steeper slopes seem to suggest faster rates of wage increase for employees without employer-sponsored health insurance relative to those who receive such benefits, it is uncertain whether the different growth rates are due to rising health insurance spending and significant cost-shifting to wages. The overall trends do not adjust for economic conditions, such as the recession in 2009 or demographic changes that could affect both health insurance coverage and wages. Fig. 3 presents the trends in hours worked for public sector employees by health insurance status. In contrast to the increase in weekly hours for employees with health insurance between 1979 and 1992 as documented in Cutler and Madrian (1998), usual hours worked per week fell for both groups in the public sector between 1992 and 2011.

82

P. Qin, M. Chernew / Journal of Health Economics 38 (2014) 77–87 17

Hourly Wage ($1992)

16 15

Public Sector with ESHI

14

Public Sector without ESHI

13 12 11 10

Year Fig. 2. Trends in hourly wages. Source: 1992–2011 March Current Population Survey (CPS) and Merged Outgoing Rotation Groups (MORG), available through the National Bureau of Economic Research.

4.2. Relationship between health insurance coverage and health care spending, and wages The results from estimating Eq. (1) are shown in the first column of Table 2. The relationship over time of having employersponsored health insurance on public sector wages is given by the coefficient on the key variable of interest, Time*Health Insurance, which is negative and statistically significant at the 5% level. Among state and local government employees, having employersponsored health insurance is associated with a 0.36% decline in wages per year, which, over the 19-year period between 1992 and 2011, equals to a 6.84% decrease. The demographic controls enter in the expected direction. Older, married, and better educated workers are more likely to have higher wages. Holding other variables constant, being in a public sector union is associated with a 17.0% increase in wages. The matching algorithm successfully matched 32,237 public sector employees who have employer-sponsored health insurance with 4,638 employees who do not, within 0.02 of the propensity score. Twenty individuals with ESHI were off common support—i.e., they had propensity scores that exceeded the maximum or fell below the minimum of the propensity scores of those without ESHI, and thus were excluded from the match. The average percentage absolute bias after matching was 1.032 and the percentage bias for each of the covariates including age, age squared, highest grade obtained, marital and union status was below 5%, indicating good balance. When we estimate Eq. (1) using a sample of public sector

Average Weekly Hours

45 44 43 Public Sector With ESHI

42 41 40 39

Public Sector Without ESHI

38 37

Year Fig. 3. Trends in hours worked. Source: 1992–2011 March Current Population Survey (CPS) and Merged Outgoing Rotation Groups (MORG), available through the National Bureau of Economic Research.

employees matched on their propensity to receive employersponsored health insurance, the coefficient on Time*Health Insurance is much less precise (and no longer statistically different from 0 at a .05 level). The magnitude of the trade-off is also smaller than that estimated using the unmatched sample of state and local government employees—having health insurance is associated with a 0.19% decline in wages per year. This estimate is consistent with an employer health insurance spending-wage trade-off of 78.0%. Now we look at only those with employer-sponsored health insurance, and examine the effect on wages based on variation in employer contribution to the premium. The results from estimating Eq. (2a) are presented in the third column of Table 2. The relationship over time on public sector wages of having fully paid employer-sponsored health insurance is −0.64% per year, and the coefficient is significant at the 10% level. Therefore, during 1992–2011, wages declined by 12.2% for public sector employees with fully paid health insurance relative to those whose employers did not contribute toward the cost of health insurance. Having partially paid health insurance leads to a smaller decrease in public sector wages per year, compared to having fully paid coverage, at −0.48% per year or 9.1% cumulatively. However, this effect is not precisely estimated. When we include NCSL’s data on state contribution to the premium, results from the state and time fixed effects regression in Eq. (2b) show that for one dollar increase in monthly state contribution, annual earnings decline by about 1.84 dollars. This effect is not precisely estimated, and suggests an employer health insurance spending-wage trade-off of only 15% (1.84/12). The last column of Table 2 presents the dollar-dollar tradeoff between per capita personal health care spending and annual wages. The coefficient on HI*PCPHCS in the public sector is −0.668, implying that a dollar increase in per capita health care spending leads to a 67 cents reduction in the wages of covered employees relative to the wages of those without employer-sponsored health insurance. However, this estimate of the health care spendingwage trade-off in the public sector is not statistically significant at the 10% level. The 95% confidence interval for the coefficient on HI*PCPHCS ranges from −1.63 to 0.11, and thus includes both −1 and 0. It is important to note that state per capita personal health care spending includes spending from both privately insured and publicly insured populations. Since Medicaid and Medicare enrollees report higher growth rates of health care spending than public and private sector employees receiving employer-sponsored health insurance, the estimated health care spending-wage tradeoff may be biased toward zero. However, to the extent that state and local governments’ contribution to employee health insurance premium grew faster than that for private employers during 1992–2009, the wage trade-off estimate may also be biased toward −1. 4.2.1. Estimating wage trade-offs excluding health and hospital employees The potential endogeneity between state-level health care spending and state wages can also bias our estimates of the wage offset. For example, increases in the compensation for medical personnel in state and local facilities can drive up personal health care spending as well as state contribution toward employee health premium. To eliminate this potential bias, we exclude state and local health and hospital workers from our sample and re-estimate the employer health spending-wage trade-off using the models in Table 2. The results including coefficients on key health insurance variables are presented in Table 3. The coefficients are similar to those obtained using a full sample of state and local government employees. Following Eq. (1), having health insurance is associated with a 0.36% wage reduction per year, or 6.84% during 1992–2011.

P. Qin, M. Chernew / Journal of Health Economics 38 (2014) 77–87

83

Table 2 The effect of employer-sponsored health insurance on public sector wages. Variables Model Health insurance Health insurance (HI) Time*Health Insurance

Log(hourly wage) (1)

Log(hourly wage) (1)a

0.174*** (0.0166) −0.00363** (0.00149)

0.160*** (0.0408) −0.00190 (0.00368)

Log(hourly wage) (2a)

Annual earnings (2b)

7612*** (1684)

0.128*** (0.0433) 0.171*** (0.0461) −0.00481 (0.00344) −0.00643* (0.00347)

Paid Part Paid All Time*Paid Part Time*Paid All

−1.841 (1.665)

State Contribution Per capita personal health care spending (PCPHCS)

0.585 (0.907) −0.668 (0.412)

HI*PCPHCS Demographics Age Age2 Married Union member Less than high school High school graduate Some college Time*Less than high school Time*High school graduate Time*Some college Constant Observations R2

Annual earnings (3a)

0.0539*** (0.00400) −0.000536*** (4.91e−05) 0.0665*** (0.00494) 0.170*** (0.0214) −0.432*** (0.0430) −0.262*** (0.0128) −0.137*** (0.0180) −0.00260 (0.00225) 7.34e−05 (0.000657) −0.000807 (0.00117) 1.323*** (0.104) 36,885 0.244

0.0476*** (0.0127) −0.000468*** (0.000162) 0.0793*** (0.0214) 0.160*** (0.0264) −0.366*** (0.0913) −0.285*** (0.0562) −0.196*** (0.0513) −0.00667 (0.00566) 0.00102 (0.00390) 0.00392 (0.00435) 1.463*** (0.291) 64,463 0.233

0.0515*** (0.00404) −0.000505*** (4.96e−05) 0.0625*** (0.00478) 0.162*** (0.0219) −0.454*** (0.0460) −0.255*** (0.0137) −0.120*** (0.0208) −0.00116 (0.00256) −0.000249 (0.000844) −0.00158 (0.00146) 1.411*** (0.0974) 32,054 0.226

1849*** (127.8) −18.37*** (1.520) 2476*** (276.7) 1668*** (333.5) −10,291*** (2422) −6962*** (1174) −3146*** (860.6) −307.3 (189.3) −84.59 (85.68) −90.42* (52.82) −10,605*** (2847) 17,564 0.261

1676*** (108.5) −16.33*** (1.304) 3131*** (223.9) 2097*** (289.0) −11,419*** (750.7) −7135*** (318.6) −4240*** (619.9) −115.5** (56.13) −46.29* (24.06) −2.212 (44.32) −16,128*** (4237) 33,501 0.276

Cluster standard errors, clustered at the state level are included in parentheses. Year and state fixed effects are included in all regressions. Sample is weighted to national totals. a Propensity score matched sample. *** p < 0.01. ** p < 0.05. * p < 0.1.

This estimate is statistically significant at the 5% level, and identical to that obtained using a full sample including health and hospital workers. The coefficient on Time*Health Insurance becomes −0.00119 in the propensity score matched sample, implying a much smaller (and not statistically significant) wage reduction, at 0.12% per year or 2.3% cumulatively. This translates into a wage-spending trade-off of 48.5%. Relative to those whose employers did not contribute toward their health insurance premium, those with partially funded health insurance experience a 0.68% wage reduction per year (or 12.9% during 1992–2011), and those with fully funded health insurance receive a 0.87% wage reduction per year (16.5% during 1992–2011). These estimates are significant at the 5% level, and indicate greater trade-offs than estimates in Table 2. In models of the monthly state contribution to health insurance premium excluding the health workers, annual earnings of insured

employees fell by 1.94 dollars, equivalent to a 16.2% trade-off; however, like the coefficient on State Contribution in Table 2, it is also imprecisely estimated and not statistically significant. Finally, for every dollar increase in state per capita personal health care spending, annual earnings declined by 78 cents, and this effect is significant at the 5% level. 4.2.2. Public sector unions and wage trade-offs Given the important roles public sector unions play in mediating benefit and wage growths for their members, we perform the regressions in Table 2 separately in samples of unionized and non-unionized public sector workers. The results are shown in Table 4. We find evidence consistent with a greater health insurance spending-wage trade-off among non-unionized public sector workers, with the exception of the coefficients on State Contribution. Having health insurance is associated with a 0.52% decline

84

P. Qin, M. Chernew / Journal of Health Economics 38 (2014) 77–87

Table 3 The effect of employer-sponsored health insurance on public sector wages excluding health and hospital employees. Variables Model

Log(hourly wage) (1)

Log(hourly wage) (1)a

Health insurance (HI)

0.176*** (0.0169) −0.00364** (0.00145)

0.150*** (0.0409) −0.00119 (0.00364)

Time*Health Insurance

Log(hourly wage) (2a)

Annual earnings (2b)

Annual earnings (3a) 8040*** (1488)

0.146*** (0.0417) 0.193*** (0.0458) −0.00680** (0.00327) −0.00873** (0.00333)

Paid Part Paid All Time*Paid Part Time*Paid All

−1.940 (1.656)

State Contribution Per capita personal health care spending (PCPHCS) HI*PCPHCS 35,510 0.243

Observations R2

62,444 0.233

30,878 0.225

16,992 0.265

0.511 (0.881) −0.778** (0.349) 32,218 0.278

Demographic controls are the same as those included in Table 2. Cluster standard errors, clustered at the state level are included in parentheses. Year and state fixed effects are included in all regressions. Sample is weighted to national totals. a Propensity score matched sample. *** p < 0.01. ** p < 0.05. * p < 0.1.

Table 4 The effect of employer-sponsored health insurance on public sector wages among unionized vs. non-unionized employees. Variables

Unionized

Non-unionized

Model (1) Health insurance (HI)

Log(hourly wage) 0.0910** (0.0354) 0.000456 (0.00257)

Log(hourly wage) 0.207*** (0.0177) −0.00521*** (0.00158)

Log(hourly wage) 0.150** (0.0640) −0.00260 (0.00441)

Log(hourly wage) 0.287*** (0.0468) −0.00972** (0.00399)

Log(hourly wage) 0.0336 (0.0742) 0.0768 (0.0851) −0.000678 (0.00521) −0.000241 (0.00517)

Log(hourly wage) 0.176*** (0.0595) 0.215*** (0.0731) −0.00677 (0.00457) −0.0101* (0.00552)

Model (2b) State Contribution

Annual earnings −3.629* (1.890)

Annual earnings −1.080 (2.011)

Model (3a) HI

Annual earnings 3472 (2218) 1.303

Annual earnings 7842*** (2240) −0.644

(1.015) −0.00692 (0.539)

(0.909) −0.548 (0.570)

Time*Health Insurance Model (1)a Health insurance (HI) Time*Health Insurance Model (2a) Paid Part Paid All Time*Paid Part Time*Paid All

Per capita personal health care spending (PCPHCS) HI*PCPHCS

Demographic controls are the same as those included in Table 2. Cluster standard errors, clustered at the state level are included in parentheses. Year and state fixed effects are included in all regressions. Sample is weighted to national totals. a Propensity score matched sample. *** p < 0.01. ** p < 0.05. * p < 0.1.

in wages per year for non-unionized employees, and this effect is statistically significant at the 1% level; the wage decline per year of having health insurance almost doubles in magnitude—at 0.97% annually when estimated in a propensity matched sample of non-unionized employees and remains statistically significant. The coefficients on Time*Paid Part and Time*Paid All also indicate greater reduction for non-unionized workers, at −0.00677 and −0.0101, respectively; the coefficients for unionized workers are much closer to zero and are imprecise. For every dollar increase in state per capita personal health care spending, non-unionized employees experience a 55 cents decline in wages, although this effect is imprecisely estimated; the corresponding estimate for unionized employees is essentially zero. 4.2.3. Wage trade-offs during the great recession Our time period 1992–2011 also covers the Great Recession starting in 2007, during which state government budgets were significantly constrained and may result in a different dynamic between benefit cost growth and wages as compared to prior periods of stable economic expansion. Thus, we divide our data into two time periods, 1992–2006 and 2007–2011, and estimate the wage regressions separately. The results are presented in Table 5. The coefficients indicate greater wage trade-offs during the Great Recession period unilaterally across all regressions, although some of the estimates are imprecise. Our results suggest that the recession and severely contracted budgets may have prompted state and local governments to restrict wage growth in face of high health insurance costs. 4.3. Relationship between health insurance coverage and health care spending, and hours The results from estimating Eq. (1) with hours worked as the outcome variable are presented in the first column of Table 6. The coefficient on Time*Health Insurance suggests that on average hours declined by 0.041 h per week per year during 1992–2011, or 0.78 h per week cumulatively over the 19-year period, for public

P. Qin, M. Chernew / Journal of Health Economics 38 (2014) 77–87

Variables

1992–2006

2007–2011

Model (1) Health insurance (HI)

Log(hourly wage) 0.177*** (0.0212) −0.00425* (0.00219)

Log(hourly wage) 0.239 (0.203) −0.00704 (0.0110)

Log(hourly wage) 0.233*** (0.0386) −0.00990** (0.00373)

Log(hourly wage) 0.493* (0.260) −0.0207 (0.0146)

Log(hourly wage) 0.0766* (0.0449) 0.124** (0.0466) 0.00155 (0.00474) −0.000625 (0.00469)

Log(hourly wage) 1.089** (0.447) 1.054** (0.486) −0.0588** (0.0250) −0.0561** (0.0270)

Model (2b) State Contribution

Annual earnings −0.530 (2.539)

Annual earnings −1.169 (5.805)

Model (3a) HI

Annual earnings 6385*** (1747) 0.967

Annual earnings 8674 (5972) −2.392

(1.051) −0.283 (0.432)

(4.326) −1.019 (1.309)

Time*Health Insurance Model (1)a Health insurance (HI) Time*Health Insurance Model (2a) Paid Part Paid All Time*Paid Part Time*Paid All

Per capita personal health care spending (PCPHCS) HI*PCPHCS

Demographic controls are the same as those included in Table 2. Cluster standard errors, clustered at the state level are included in parentheses. Year and state fixed effects are included in all regressions. Sample is weighted to national totals. a Propensity score matched sample. *** p < 0.01. ** p < 0.05. * p < 0.1.

sector employees with employer-sponsored health insurance relative to those without. The coefficient is not statistically significant at conventional levels. When we estimate Eq. (1) in a matched sample of public sector employees based on their propensity score to receive health insurance, the coefficient on Time*Health Insurance becomes 0.017—which equates to a 0.32 h increase in the length of the work week for covered employees over 1992–2011; similar to results from the unmatched sample, this estimate is also imprecise. The third and fourth columns of Table 6 present the results from estimating the effect of employer’s contribution toward health insurance premium on hours worked. Having partially paid coverage is associated with a 0.041 h decline in the length of the work week per year in the public sector, relative to the control group with zero employer contribution. The coefficient is not statistically significant and implies a shortened work week by 0.78 h over the time period 1992–2011. The corresponding estimate of the effect over time on hours of having fully paid health insurance is equal to −0.068 and sums to a 1.29 h decline in weekly hours worked during 1992–2011 relative to those with zero employer contribution to health insurance premium. In Eq. (2c), a 1% increase in monthly state contribution to the health plan premium is associated with a −0.0068 h decline in weekly hours worked and this effect is significant at the 10% level. As the average monthly state contribution to the lowest cost full-service HMO plan rose by around 189% between 1999 and 2009 as recorded by NCSL, our results suggest the

90

Public Sector Percent with Insurance

Table 5 The effect of employer-sponsored health insurance on public sector wages prior to & during the great recession.

85

85 80 75 70 65

Private Sector

60

Year Fig. 4. Share of workers with employer-sponsored health insurance. Source: 1992–2011 March Current Population Survey (CPS) and Merged Outgoing Rotation Groups (MORG), available through the National Bureau of Economic Research.

associated hours drop for insured state and local workers was around 1.3 h. The last column of Table 6 presents the estimates from Eq. (3b), on the trade-off between percentage increase in health care spending and changes in hours worked. The coefficient on HI*Log(PCPHCS) is −3.126 for public sector employees, implying a 0.03 h decrease in the length of the work week for covered employees for every percentage increase in per capita personal health care spending. The effect is statistically significant at the 1% level. Given that real per capita personal health care spending rose by 150% between 1992 and 2009 (Fig. 1), the length of the work week declined by about 4.5 h for covered employees relative to the hours of uncovered employees. This finding in the public sector is in contrast to the economic intuition and empirical findings presented by Cutler and Madrian (1998) that as fixed employment costs rise, the hours of covered employees should increase in substitution for additional workers hired. Whereas Cutler and Madrian (1998) found that the increase in hours for covered workers in the private sector was primarily a result of full-time workers shifting into over-time, public sector employers may be unable to increase the hours of full-time, insured employees due to the high price of over-time labor among unionized employees. Studies using CPS have shown that union coverage substantially increases the likelihood of receiving premium pay for over-time employment (Ehrenberg and Schumann, 1982; Trejo, 1993). Facing high rates of unionization among insured employees and rising costs of health insurance provision, public sector employers, therefore, may be enticed to expand the share of hours worked by employees not receiving health insurance, who were more likely to be part-time or non-unionized.3 Another important consideration is change in the composition of workers receiving employer-sponsored health insurance between 1992 and 2011. In our dataset, insurance coverage for state and local government employees reached a high of 89% in 1995, and dipped to a low of 84% in 2011 (Fig. 4). A difference of 5% of the overall state and local government workforce is large relative to the small share of noninsured employees. For instance, a decline in health insurance coverage over time for those with fewer hours of work relative to those who work more will bias the estimate of health insurance on hours worked upwards. As the pool of insured workers becomes increasingly populated by workers with longer work weeks, it will seem that hours of work are increasing

3 The average rate of union coverage and part-time status between 1992 and 2011 are 32.1% and 11.1% for public sector employees without employer-sponsored health insurance, and 46.6% and 1.6% for those with employer-sponsored health insurance.

86

P. Qin, M. Chernew / Journal of Health Economics 38 (2014) 77–87

Table 6 The effect of employer-sponsored health insurance on hours worked in the public sector. Variables Model Health insurance Health insurance (HI) Time*Health Insurance

Hours worked (1)

Hours worked (1)a

3.393*** (0.403) −0.0405 (0.0287)

1.736** (0.837) 0.0171 (0.0682)

Paid Part

Hours worked (2a)

Hours worked (2c)

28.48*** (7.190)

0.988 (0.727) 1.231* (0.704) −0.0408 (0.0566) −0.0676 (0.0560)

Paid All Time*Paid Part Time*Paid All

−0.679* (0.348)

Log(State Contribution) Log(per capita personal health care spending (PCPHCS))

1.796 (1.827) −3.126*** (0.872)

HI*Log(PCPHCS) Demographics Age Age2 Married Union member Less than high school High school graduate Some college Time*Less than high school Time*High school graduate Time*Some college Constant Observations R2

Hours worked (3b)

0.296*** (0.0666) −0.00359*** (0.000801) 1.114*** (0.108) 0.476** (0.188) −1.361*** (0.313) −0.615*** (0.197) −0.527* (0.282) −0.00794 (0.0256) −0.00250 (0.0146) 0.0264 (0.0208) 31.15*** (1.528) 36,903 0.044

0.298 (0.318) −0.00276 (0.00402) 1.953*** (0.540) 1.092** (0.454) −1.702** (0.738) −0.893 (0.686) −1.292 (1.496) −0.0123 (0.0712) −0.00108 (0.0791) 0.0832 (0.104) 31.09*** (6.181) 64,485 0.059

0.134* (0.0684) −0.00176** (0.000818) 0.777*** (0.102) 0.279 (0.199) −1.766*** (0.335) −0.884*** (0.225) −0.622** (0.249) 0.0219 (0.0259) 0.00491 (0.0148) 0.0331* (0.0194) 37.41*** (1.540) 32,065 0.026

0.351*** (0.111) −0.00443*** (0.00133) 0.635*** (0.146) 0.697** (0.284) −0.189 (0.588) −0.859 (0.615) 0.113 (0.561) −0.128** (0.0517) −0.00965 (0.0413) −0.0389 (0.0373) 38.41*** (2.569) 16,891 0.034

0.290*** (0.0743) −0.00349*** (0.000894) 1.056*** (0.113) 0.966*** (0.202) −1.561*** (0.384) −0.707*** (0.212) −0.629** (0.275) 0.0103 (0.0357) 0.00348 (0.0190) 0.0357 (0.0234) 17.62 (15.16) 32,365 0.052

Cluster standard errors, clustered at the state level are included in parentheses. Year and state fixed effects are included in all regressions. Sample is weighted to national totals. a Propensity score matched sample. *** p < 0.01. ** p < 0.05. * p < 0.1.

over time among those with health insurance (Cutler and Madrian, 1998).

2008. This yielded a total of 583 men in our sample. Our regression model was the following: Cognitive function = ˇo + ˇ1 ∗ HI + ˇ2 ∗ LATE + ˇ3 ∗ HI ∗ LATE

4.4. Trends in cognitive function Our identification assumption assumes no systematic differences in the trends of unobserved abilities of public workers with and without employer-sponsored insurance coverage. Our primary data do not allow rigorous investigation of that assumption because we observe only limited measures of ability (e.g. education). Therefore, we explore this concern using data from the University of Michigan Health and Retirement Study (HRS) and looking at men who are currently employed by state or municipal governments between 1996 and 2008. In order to achieve a sufficiently large sample size, we pooled enrollees from 1996 to 1998 and compared their cognitive function to enrollees pooled for years between 2004 and

+ Z + ε

(4)

HI is the dummy variable for having employer-sponsored health insurance. LATE is the dummy variable for whether or not the person-observation is in the 2004–2008 pool. Z is a set of demographic variables that include age, marital status, and education. The coefficient of interest, ˇ3 , captures the effect of having health insurance over time on cognitive function, measured in the HRS data as a sum of the total words recalled in a series of recall memory tests. The mean cognitive function score in the sample was 24.73 with a standard deviation of 3.87. The coefficient on HI*LATE was 0.33

P. Qin, M. Chernew / Journal of Health Economics 38 (2014) 77–87

with a p-value of 0.67, which suggests that trends in cognitive function were not systematically related to insurance status. In fact the point estimate suggests that those with coverage had systematically greater cognitive functioning over time than those without coverage, suggesting our estimate may in fact be conservative. 5. Conclusion This paper has analyzed the impact on the wages and hours of public sector employees due to the rise in health insurance spending. We find suggestive but not conclusive evidence of cost shifting of increasing employer spending on health insurance in the form of lower wages. Our estimates are reasonably consistent with those of Clemens and Cutler (2013), who found about a 15% tradeoff. Specifically, when we use the state’s contribution to premium to assess tradeoffs, we estimate a 15% tradeoff. When we propensity score match and exclude health sector employees, we estimate a tradeoff of about 48.5%, but it is not statistically different from the 15% estimate. In other models that are not propensity score adjusted, we find larger offsets, but the results are often very imprecise. More detailed analysis suggests that the tradeoff is larger among non-unionized workers and was bigger during the Great Recession. These finding suggest that institutional details (unionization) and economic environment matter. Public sector workers are not immune from the wage dynamics observed in the private sector, but the details and magnitudes of the effects differ. Appendix A. See Table A1. Table A1 Comparison of men with and without employer-sponsored health insurance (ESHI). Public sector Without ESHI

Education Less than high school High school graduate Some college College graduate Age group 25–34 35–44 45–54 Married Union member

With ESHI

1992

2011

Change

1992

2011

Change

8.7% 26.1% 14.9% 50.2%

3.8% 26.1% 14.6% 55.5%

−4.9% 0.0% −0.3% 5.3%

4.0% 25.8% 16.8% 53.4%

1.8% 19.5% 17.4% 61.3%

−2.2% −6.3% 0.6% 7.9%

40.6% 35.7% 23.7% 70.1% 27.0%

31.3% 31.3% 37.3% 71.1% 29.6%

−9.3% −4.4% 13.6% 1.0% 2.6%

27.2% 42.0% 30.8% 72.2% 49.3%

24.9% 36.1% 39.0% 71.1% 44.0%

−2.3% −5.9% 8.2% −1.1% −5.3%

Sample includes men ages 25–54 who are not self-employed.

References Anzia, S.F., Moe, T.M., (Unpublished draft) 2012, May. Public Sector Unions and the Costs of Government, Available at http://papers.ssrn.com/sol3/papers.cfm

87

Buchmueller, C.T., 1999. Fringe benefits and the demand for part-time workers. Applied Economics 31 (5), 551–563. Clark, R.L., Morrill, M.S., Hanson, E., Maki, J., 2012. State Health Plans During Times of Fiscal Austerity: The Challenge of Improving Benefits While Moderating Costs. Center for State & Local Government Excellence. Clemens, J., Cutler, D.M., 2013. Who Pays for Public Employee Health Costs? (No. w19574). National Bureau of Economic Research. Cuckler, G., Martin, A., Whittle, L., Heffler, S., Sisko, A., Lassman, D., Benson, J., 2011. Health spending by state of residence, 1991–2009. Medicare & Medicaid Research Review 1 (4). Currie, J., Madrian, B.C., 1999. Health, health insurance and the labor market. Handbook of Labor Economics 3, 3309–3416. Cutler, D.M., Madrian, B., 1998. Labor market responses to rising health insurance costs: evidence on hours worked. RAND Journal of Economics 29 (3), 509–530. DiSalvo, D., 2010. The trouble with public sector unions. National Affairs 5, 3–19. Eberts, R.W., Stone, J.A., 1985. Wages, fringe benefits, and working conditions: an analysis of compensating differentials. Southern Economic Journal, 274–280. Edwards, C., 2010. Public sector unions and the rising costs of employee compensation. Cato Journal 30, 87. Ehrenberg, R.G., 1980. Retirement system characteristics and compensating wage differentials in the public sector. Industrial and Labor Relations Review, 470–483. Ehrenberg, R.G., Schumann, P.L., 1982. Compliance with the Overtime Pay Provisions of the Fair Labor Standards Act. National Bureau of Economic Research. Employee Benefits in the United States, 2013, March. Bureau of Labor Statistics, http://www.bls.gov/news.release/pdf/ebs2.pdf Employee Contributions to Public Pension Plans, 2014, January. National Association of State Retirement Administrators (NASRA) Issue Brief, http://www.nasra.org/ files/Issue%20Briefs/NASRAContribBrief.pdf Government Accountability Office, 2012, April. State and Local Governments’ Fiscal Outlook, Washington, DC. Gruber, J., 1994. The incidence of mandated maternity benefits. The American Economic Review, 622–641. Gruber, J., Krueger, H.B., 1991. The incidence of mandated employer-provided insurance: evidence from worker’s compensation. In: Bradford, D. (Ed.), Tax Policy and the Economy, vol. 5. MIT Press, Cambridge, MA, pp. 111–143. Katz, L.F., Murphy, K.M., 1992. Changes in relative wages, 1963–1987: supply and demand factors. The Quarterly Journal of Economics 107 (1), 35–78. Lubotsky, D., Olson, C.A., 2010. The Wage-Employer Provided Health Insurance Trade off and Employee Premium Copayments. Miller, R.D., 2004. Estimating the compensating differential for employer-provided health insurance. International Journal of Health Care Finance and Economics 4 (1), 27–41. Montgomery, M., Cosgrove, J., 1993. The effect of employee benefits on the demand for part-time workers. Industrial and Labor Relations Realism 47, 87. Murray, P.K., Singer, M., Dawson, N.V., Thomas, C.L., Cebul, R.D., 2003. Outcomes of rehabilitation services for nursing home residents. Archives of Physical Medicine and Rehabilitation 84, 1129–1136. Olson, C.A., 2002. Do workers accept lower wages in exchange for health benefits? Journal of Labor Economics 20 (S2), S91–S114. Rosen, S., 1986. The theory of equalizing differences. Handbook of Labor Economics 1, 641–692. Sheiner, L., 1995. Health Care Costs, Wages and Aging: Assessing the Impact of Community Rating. Mimeo, Federal Reserve Board. State Budget Crisis Task Force, 2012. Report of the State Budget Crisis Task Force (revised 31.07.12). Summers, L.H., 1989. Some simple economics of mandated benefits. The American Economic Review 79 (2), 177–183. Trejo, S.J., 1993. Overtime pay, overtime hours, and labor unions. Journal of Labor Economics, 253–278.