How Do Medicaid Expansions Affect the Demand for Healthcare Workers? Evidence from Vacancy Postings

I. Introduction

The Affordable Care Act (ACA) has led to large increases in health insurance coverage in the United States. Much of the ACA’s increase in health insurance coverage can be accounted for by increased Medicaid enrollment, which rose by more than 16 million from 2010 to 2016. The rise in health insurance coverage from the ACA promises to increase demand for healthcare and to lead to greater competition for high-skilled healthcare workers. However, even before the ACA was passed, there was already widespread concern about the adequacy of the U.S. healthcare workforce. These concerns can be seen in a recent Senate hearing on shortages of healthcare workers (U.S. Senate 2018), reports of healthcare employers saying that they cannot find high-skilled workers (Society for Human Resource Management 2016), and research articles that predict even more severe shortages of healthcare workers on the horizon as the aging U.S. population’s healthcare needs continue to rise (Colwill, Cultice, and Kruse 2008; Cooper et al. 2002; Petterson et al. 2012; Sargen, Hooker, and Cooper 2011).¹ Despite the possibility of challenges arising from increasing demand for healthcare when there are already shortages of healthcare workers, little is known about how healthcare employers respond to increased competition for high-skilled workers.

Recent research finds that employers reduce their skill requirements for job vacancies when labor markets become tighter (Hershbein and Kahn 2018; Modestino, Shoag, and Ballance 2016, 2020), which suggests that one possible response to increased tightness in the healthcare labor market could be for healthcare employers to reduce their skill requirements to fill vacancies.² As higher-skilled healthcare workers deliver better and more cost-effective care, healthcare workers’ skill levels are an important measure of quality in the healthcare system.³ The fact that much of the increase in health insurance coverage from the ACA comes from Medicaid is also important. While theory predicts that healthcare employers will hire more workers in response to increased demand for their services, the quality of the marginal hires arising from Medicaid expansions has the potential to be low given Medicaid’s low reimbursement rates.

Understanding how employers’ demand for healthcare workers responds to increased Medicaid coverage is important for evaluating providers’ responsiveness to the needs of Medicaid patients and for considering potential spillover effects of Medicaid expansions, and it has the potential to provide insights into how healthcare employers respond to increased competition for healthcare workers more generally. Little is currently known about the impact of increased Medicaid eligibility on healthcare employers’ staffing decisions, despite concerns about the quality and amount of care that Medicaid patients receive, about the growing demand for healthcare paid for by public health insurance as baby boomers continue to reach Medicare eligibility, and about the capacity of the U.S. healthcare workforce to accommodate increases in healthcare demand.

The goal of this work is to examine the impact of Medicaid expansions on healthcare employers’ skill demand and to consider which types of employers hire more workers in response to Medicaid expansions. The main analysis draws on a database of more than 14 million job vacancies for healthcare workers collected from online job advertisements posted from 2010 to 2016. To obtain estimates of the impact of Medicaid eligibility on these vacancy postings, this study implements a simulated instrumental variables (IV) strategy that isolates variation in Medicaid eligibility that comes from the ACA’s changes in eligibility rules. The simulated IVapproach for studying the effect of Medicaid eligibility was pioneered by Currie and Gruber (1996a,b) and Cutler and Gruber (1996) and involves creating a measure that represents an area’s Medicaid generosity that is correlated with Medicaid eligibility and that is correlated with the outcome of interest only through its correlation with Medicaid eligibility. To create this simulated instrument and to calculate Medicaid eligibility, the study draws on data from the American Community Survey (ACS). The study also uses ACS data to estimate the impact of Medicaid eligibility on health insurance coverage, which is relevant for assessing the likely effects of Medicaid expansions on labor demand. To verify that the effect of Medicaid eligibility on realized hires and employment in the healthcare industry corroborates the findings from the vacancy data, the study also supplements the analysis with data from the Quarterly Workforce Indicators (QWI).

The health insurance analysis suggests that a ten percentage point increase in the share of an area that is eligible for Medicaid increases the share of the area with Medicaid coverage by 2.3 percentage points and the share with any health insurance coverage by 1.7 percentage points, which is consistent with previous research on the coverage effects of Medicaid eligibility. Consistent with healthcare employers seeking to hire more workers in response to increased Medicaid eligibility, a ten percentage point increase in the share of an area eligible for Medicaid leads to a 5.4 to 8.9 percent increase in the area’s number of vacancy postings. While an increase in job vacancies suggests that employers seek additional healthcare workers in response to Medicaid expansions, it does not guarantee that they will be able to hire additional workers, and an increase in vacancies could also arise from an increase in the number of people leaving their jobs rather than from employers growing their staffs. However, the QWI data verify that increased Medicaid eligibility is also associated with increased hiring and employment in the healthcare industry, which suggests that the vacancies represent new jobs and that employers are able to fill many of the additional vacancies.

Using the vacancy data’s information on jobs’ skill requirements, I show that an increase in Medicaid eligibility in an area leads to a reduction in mean skill requirements for new healthcare workers in that area within narrowly defined occupations. The baseline analysis uses the total number of specific skills required as the skill measure and draws on vacancies for all healthcare occupations, which provides a broad characterization of the effect of Medicaid on employers’ demand for healthcare workers. However, I also consider requirements for different types of specific skills, for certifications, and for experience and estimate separate specifications for vacancies for specific healthcare occupations. As with the main analysis, the results for the different skill measures and for specific occupations indicate that an increase in Medicaid eligibility is associated with a reduction in average skill requirements for new healthcare workers.

One possible explanation for this decrease in average skill requirements could be that Medicaid expansions increase competition for healthcare workers and in doing so lead healthcare employers to lower their skill requirements to fill vacancies. Alternatively, the decrease could also occur if healthcare employers that typically hire lower-skilled workers are the employers that post more vacancies in response to the ACA’s Medicaid expansion. Distinguishing between these two explanations is important because they have different spillover implications and suggest different policy interventions. Average skill requirements falling because increased Medicaid eligibility increases the tightness of healthcare labor markets would imply that Medicaid expansions have negative spillover effects on the previously insured. Addressing this issue might involve increasing the supply of high-quality healthcare workers or rationing care. In contrast, average skill requirements falling because increased Medicaid eligibility leads to providers that typically hire low-skilled workers increasing hiring is not evidence that Medicaid expansions have spillover effects on the previously insured. However, it does raise concerns about the quality of providers responsive to the needs of new Medicaid patients. Possible policy remedies would involve employing strategies to entice higher-quality providers to participate in Medicaid, such as increasing Medicaid’s higher reimbursement rates.

To understand why increased Medicaid eligibility leads to lower skill requirements for new healthcare workers, I take advantage of the fact that many of the vacancy postings contain the hiring employer’s name, which allows for estimating models with employer fixed effects. After accounting for employer heterogeneity, the effect of Medicaid expansions on skill requirements becomes statistically indistinguishable from zero, which suggests that Medicaid eligibility lowers the average skill requirements sought by causing lower-quality employers to hire more workers rather than by causing employers to lower their skill requirements. Using an approach that classifies employers based on their typical skill requirements prior to the ACA’s Medicaid expansion, I confirm that healthcare employers that typically hire lower-skilled workers post more job vacancies in response to increased Medicaid eligibility and that the posting behavior of employers that typically hire higher-skilled workers is unaffected by increased Medicaid eligibility.

The results presented here are at odds with fears that providers are not willing or able to increase capacity in response to increased Medicaid eligibility. The findings also do not indicate that increased competitive pressure from Medicaid expansions leads to employers lowering their skill requirements for new healthcare workers. However, the finding that providers that hire the least-skilled workers are the most responsive to Medicaid expansions suggests that Medicaid likely plays a role in various patterns that have been observed in the treatment that Medicaid patients receive. For example, previous research has found that Medicaid patients are often treated by lower-quality providers, likely in part because of Medicaid patients’ proximity to lower-quality providers (Bach et al. 2004; Chandra and Skinner 2003; Cutler, Lleras-Muney, and Vogl 2011; Goldman, Vittinghoff, and Dudley 2007; Jha, Orav, and Epstein 2011; Landon et al. 2007; Williams and Jackson 2005). While poorer areas having fewer resources likely factors into these patterns, the results from this study suggest that Medicaid plays a role in shaping local healthcare workforces and leads to areas with more Medicaid enrollees having lower-skilled healthcare workforces. Policies that induce higher-quality employers to accommodate demand from Medicaid could lead to poorer areas with more Medicaid patients having higher-skilled healthcare workforces.

Several facts are important to keep in mind when interpreting the results of the study. First, while an underlying assumption of this study is that receiving healthcare from higher-skilled workers is better than receiving healthcare from less-skilled workers, this study does not estimate the impact of the skills of healthcare workers on the quality of care they provide. Instead, this study relies on previous work that has established that higher-skilled workers provide better and more cost-effective care and are associated with improved health outcomes to support this assumption (for example, Aiken et al. 2003; Banki et al. 2013; Bartel et al. 2014; Boissy et al. 2016; Cortes and Pan 2015; Currie and Schnell 2018; Doyle, Ewer, and Wagner 2010). Second, detailed information on skill requirements is rare and is one of the main advantages of the vacancy data, which means that replicating the analysis using data on new hires to verify that the skills listed in vacancy postings match the skills of new hires is not possible for many of the skill measures. However, it should be noted that previous research that uses the same vacancy data and requires similar assumptions has shown that, for characteristics that are observable in both employment data and the vacancy data, areas that experience changes in skill requirements in the vacancy data experience corresponding changes in the skill levels of their workforces in the employment data (Hershbein and Kahn 2018).⁴ Finally, this analysis identifies the immediate effects of Medicaid expansions. As supply for some healthcare occupations may be slow to adjust to changes in demand, the long-term effects may differ from the short-term effects.

II. Background

C. Conceptual Framework

The mixed-economy model of Sloan, Mitchell, and Cromwell (1978) provides insights into the potential effects of Medicaid expansions on the healthcare workforce.⁷ In this model, providers, which can be hospitals, clinics, individual physicians’ offices, and any other suppliers of healthcare services, can treat both privately insured patients and government-insured patients.⁸ Providers face a downward sloping demand curve for treating private patients and receive a fixed price for treating Medicaid patients. Providers prefer to treat patients in the private market until the marginal revenue from the private market equals Medicaid’s reimbursement rate, at which point, they will prefer to treat Medicaid patients. If providers treat all the Medicaid patients who demand their services, they will then resume treating private patients. Providers face increasing marginal costs for each unit of healthcare they supply. The original model of Sloan, Mitchell, and Cromwell assumes that providers differ in their marginal costs for exogenous reasons. To consider the implications of Medicaid expansions on the skills of the healthcare workforce, I further assume that producing higher-quality services is costlier than producing lower-quality services, that providers producing higher-quality services employ higher-skilled workers, and that providers increase supply by hiring more workers.⁹

Figure 1 provides a graphical representation of the model. Prior to a Medicaid expansion, providers face a demand curve of ABCD. In this example, Provider 1 is the highest-quality and costliest provider, and Provider 3 is the lowest-quality and cheapest provider.¹⁰ Providers produce the quantity of medical services such that the marginal cost of providing additional services equals the marginal revenue generated from those services. Thus, a provider’s mix of public and private patients depends on its marginal cost curve. In this example, Provider 1 treats only private patients. Providers 2 and 3 see a mix of private and publicly insured patients, but because Provider 3’s marginal cost curve is lower than Provider 2’s, Provider 3’s marginal patient is private, while Provider 2’s is public.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

The Provider Response to a Medicaid Expansion

Notes: The graph illustrates providers’ supply response to Medicaid expansions as described in the mixed-economy model of Sloan, Mitchell, and Cromwell (1978).

A provider’s response to a Medicaid expansion depends on its marginal cost curve as well as on the degree of crowd-out from the Medicaid expansion. If there is no crowd-out, providers will face a marginal revenue curve of ABGH after a Medicaid expansion. While this expansion has no effect on Providers 1 and 2, it causes Provider 3 to move along its marginal cost curve and provide more care, which leads to the prediction that a Medicaid expansion with little or no crowd-out will lead to increased hiring if there exist providers with marginal costs that are lower than Medicaid’s reimbursement rate. Because Provider 3 hires less-skilled workers than the other providers, the model predicts that the Medicaid expansion will cause the average skill level of new healthcare workers employed in equilibrium to fall because the expansion leads to Provider 3’s share of overall employment rising. Note that the mixed-economy model does not guarantee the existence of providers with sufficiently low marginal costs. If no provider is producing at a marginal cost below the Medicaid reimbursement rate, the mixed-economy model predicts no changes in employment or skill requirements of healthcare workers.

If the increase in Medicaid coverage crowds out private health insurance, the marginal revenue curve can be represented by EFGH. In this case, Provider 1 loses privately insured patients, which leads to Provider 1 accepting Medicaid patients but supplying less care overall. This reduction in supply leads to the prediction that crowd-out from a Medicaid expansion will reduce employment and hiring by high-quality providers. Because the employment effect of crowd-out and the employment effect of the increased Medicaid coverage have opposite signs, the overall employment effect of a Medicaid expansion is unclear ex ante and depends on which effect dominates. As I show later, the crowd-out from the ACA’s Medicaid expansion is relatively modest, which suggests that the ACA’s Medicaid expansion would be likely to increase employment for healthcare workers. As increased Medicaid coverage and crowd-out both exert downward pressure on the average quality of healthcare workers employed in equilibrium, the mixed-economy model yields the prediction that Medicaid expansions reduce the average quality of healthcare workers regardless of which effect dominates. Since employment stocks can be slow to evolve and changes at the margin difficult to detect, the empirical analysis here studies the impact of increased Medicaid eligibility on the average skill requirements of new vacancies that employers post to understand the impact of increased Medicaid eligibility on the skills required for healthcare jobs, as posting a vacancy generally precedes hiring a new worker.

Medicaid’s reimbursement rates being lower than the reimbursement rates that providers can typically receive through private insurance is an important aspect of the mixed-economy model. While most states’ Medicaid reimbursement rates had been relatively flat leading up to the ACA, the ACA temporarily set national minimum Medicaid reimbursement rates that states could pay primary care providers equal to Medicare’s reimbursement rates for certain services. As Medicaid reimbursement rates were still below most private reimbursement rates even with the temporary rate increases, the predictions of the mixed economy model still hold even with the rate increases. Though these rate increases were only mandated in 2013 and 2014 and only applied to a subset of services, they could confound our empirical strategy if the size of the rate increases is correlated with the variation in Medicaid eligibility during the study period. In Online Appendix B, I show that the size of the rate increases is not correlated with changes in Medicaid eligibility from 2010 to 2016 and verify that the main skills results are robust to controlling for the rate increases.

In this mixed-economy model, the only benefit of treating Medicaid patients is the government reimbursement rate. In practice, however, providers may value treating Medicaid patients for other reasons as well. For example, some providers may view providing services to low-income patients as a social benefit to the community. Any utility that a provider gains from providing this community benefit could be incorporated into the mixed-economy model as an additional return to treating Medicaid patients over the government reimbursement rate. While the employment predictions of the model would still hold if providers value treating Medicaid patients for reasons in addition to the government reimbursement rate, the existence of nonmonetary returns to treating Medicaid patients means that a Medicaid expansion increasing employment no longer implies that there are providers who can treat Medicaid patients at a cost equal to the government reimbursement rate. Instead, an employment response indicates that there are providers whose marginal cost is less than the reimbursement rate plus the monetary value of the utility they gain from providing services to Medicaid patients.

While the mixed-economy model predicts changes to the average skill level of healthcare workers employed in equilibrium because of compositional changes in the share of care supplied by different providers, increased Medicaid eligibility also has the potential to affect average skill levels by causing providers to alter their skill requirements. Models that consider the implications of labor market tightness, such as the one in Modestino, Shoag, and Ballance (2016, 2020), suggest that increased Medicaid eligibility could lead to providers reducing their skill requirements if increased demand for healthcare from increased Medicaid eligibility leads to more competition for healthcare workers. In this framework, when workers are scarce relative to the number of open jobs, employers become less likely to encounter high-skilled workers, which results in employers lowering the minimum skill requirements for job vacancies. Several recent papers have examined the relationship between labor market tightness and skill requirements in the market as a whole and consistently find that employers lower their skill requirements when the job vacancies increase relative to the number of workers.¹¹

The labor-market-tightness model and mixed-economy model both predict that expanding Medicaid coverage will lead to falling average skill requirements of healthcare workers, but the two models provide different explanations for why this decrease occurs. The labor-market-tightness model predicts a decrease in the average skills of healthcare workers because Medicaid expansions lead to within-employer reductions in skill requirements, while the mixed-economy model predicts a decrease in the average skills of healthcare workers because Medicaid expansions lead to providers that typically hire lower-skilled workers providing a larger share of healthcare services. A key difference in the implications of these two mechanisms is that the within-employer decrease predicted by the labor-market-tightness model implies that Medicaid expansions have negative spillovers on the previously insured by causing their providers to employ less-skilled workers, while the mixed-economy model does not imply these spillovers. The two frameworks can be distinguished from each other empirically by estimating models with employer fixed effects. According to the labor-market-tightness model, the negative effect of a Medicaid expansion on skill requirements would be robust to including employer fixed effects, while the mixed-economy model would suggest that Medicaid expansions would have no effect on skill requirements after accounting for changes in the composition of providers hiring.¹²

While this study focuses on how a Medicaid expansion could affect the skills and quality of the workers providers hire, it is important to note that many factors affect the skills and quality of the workers that providers hire. Licensing, regulation, and the threat of lawsuits may all lead to the workers that providers hire being more homogeneous than they otherwise would be. However, despite these additional factors that may limit the scope for healthcare employers to hire workers with different characteristics, there is evidence of skill and quality heterogeneity within narrowly defined healthcare occupations. For examples of this evidence, refer to Aiken et al. (2003), Banki et al. (2013), Bartel et al. (2014), Boissy et al. (2016), Cortes and Pan (2015), Currie and Schnell (2018), and Doyle, Ewer, and Wagner (2010).

III. Data and Empirical Approach

IV. Effect of Medicaid Eligibility on Health Insurance Coverage, Job Vacancies, Hiring, and Employment

Medicaid eligibility’s effect on the healthcare labor market comes from Medicaid eligibility’s effect on health insurance coverage. Table 2 shows estimates of the effect of Medicaid eligibility on health insurance coverage. Each cell in Table 2 is the effect of Medicaid eligibility from separate regressions of Equation 1. The first column shows the OLS estimates, while the second column displays the 2SLS estimates.

The first row displays estimates of the effect of Medicaid eligibility on Medicaid coverage. The estimates imply that a ten percentage point increase in Medicaid eligibility leads to an additional 2.3 to 2.4 percent of the CZ having Medicaid coverage. The estimates for the OLS and 2SLS specifications are similar, which is consistent with much of the variation in Medicaid eligibility over the time studied coming from the ACA’s changes to Medicaid eligibility rules rather than from changes in economic conditions.

Row 2 displays estimates of the effect of Medicaid eligibility on the share of people in the CZ with employer-sponsored health insurance coverage. Neither the OLS estimate nor the 2SLS estimate is statistically significantly different from zero, suggesting that the ACA’s Medicaid expansion did not result in large amounts of crowd-out of employer-sponsored health insurance. Row 3 displays estimates of the effect of Medicaid eligibility on the share of people with privately purchased health insurance coverage. The estimates indicate that a ten percentage point increase in Medicaid eligibility decreases the share of the CZ with privately purchased health insurance by about 0.5 percentage points.

These results indicate that, unlike other Medicaid expansions that tended to crowd out employer-sponsored health insurance, the ACA’s Medicaid expansion crowded out privately purchased health insurance. A likely reason for the differential crowd-out from the ACA is that the ACA implemented several national changes to the individual market that increased the purchase of insurance directly from insurers. These changes included requiring insurers to accept all people who applied for coverage, limiting factors insurers could use for determining premiums, and mandating that insurers provide certain coverage. The ACA also created subsidies for purchasing health insurance on the individual market, and eligibility for these subsidies overlaps with Medicaid eligibility. All of these changes led to an increase in privately purchased coverage nationally. The negative effect of Medicaid eligibility on privately purchased insurance documented here reflects that the use of privately purchased insurance coverage increased less in areas that expanded Medicaid eligibility than it did in nonexpansion areas, rather than that privately purchased health insurance fell in Medicaid expansion areas in absolute terms.

As Row 4 indicates, this decrease in privately purchased health insurance led to a decrease in the share of the CZ having any private health insurance. The 2SLS estimates suggest that a ten percentage point increase in Medicaid eligibility reduces the share of people with private coverage by 0.5 percentage points. The 2SLS estimates suggest that 22 percent of the people gaining Medicaid coverage because of the Medicaid expansions would have had private coverage if not for becoming eligible for Medicaid.

The final row of Table 2 displays the estimated effect of an increase in Medicaid eligibility on the share of people in the CZ with any health insurance. Despite the evidence of crowd-out, Medicaid eligibility still leads to an increase in health insurance coverage. The 2SLS estimates imply that a ten percentage point increase in Medicaid eligibility increases health insurance coverage by 1.7 percentage points.

One potential concern with the analysis is that the ACA may noisily estimate CZ-level measures and thus may lead to classical measurement error, especially for small CZs. With the health insurance estimates, imprecise CZ-level estimates from the ACS would lead to the dependent variables suffering from classical measurement error, which could limit the precision of the estimates but will not lead to biased estimates. Small sample sizes in the ACS also have the potential to lead to classical measurement error in the share of the CZ that is eligible for Medicaid, which has implications for all the analysis since CZ-level Medicaid eligibility is the main regressor of interest. With OLS, classical measurement error in a regressor could lead to the estimate of the regressor’s effect suffering from attenuation bias. However, since classical measurement error in Medicaid eligibility is not systematically correlated with the simulated measure, the 2SLS estimator does not suffer from the same bias. As the OLS and 2SLS estimates tend to be similar to each other in Table 2 and in the results shown subsequently, classical measurement error arising from imprecise estimates of Medicaid eligibility from the ACS does not appear to result in major bias in the OLS estimates.

As explained in Section II.C, the mixed-economy model predicts that increased Medicaid eligibility will lead to an increase in the number of healthcare jobs if its positive effect on healthcare jobs from increasing Medicaid coverage dominates its negative effect on healthcare jobs from crowding out private coverage. The research that has found that the ACA’s Medicaid expansion has increased healthcare use coupled with the finding from this study that crowd-out from the ACA’s Medicaid expansion is small relative to the increases in health insurance coverage suggests that the positive employment effects may dominate. But even with theory predicting a likely increase in new hires and employment, concerns about shortages of healthcare workers and about Medicaid’s low reimbursement rates mean that it is still unclear whether providers would be willing or able to hire more workers. Despite these concerns, job growth in the healthcare sector has quickened since the passage of the ACA (Schoen 2016; Turner, Roehrig, and Hempstead 2017), and some early research has found that the increases in healthcare access for the ACA’s new Medicaid enrollees have not increased wait times or reduced the healthcare access of the previously insured (Benitez, Perez, and Tipirneni 2018; Carey, Miller, and Wherry 2018)—all of this points to the possibility that healthcare employers might be willing and able to hire more workers in response to increased Medicaid eligibility. The remainder of this section focuses on understanding the impact of the Medicaid expansion on vacancies, hiring, and employment of healthcare workers.

Panel A of Table 3 displays estimates of the effect of Medicaid eligibility on the inverse hyperbolic sine (IHS) of the number of vacancies posted.¹⁸ The first row shows estimates of the effect of Medicaid eligibility on the total number of vacancies posted, while the second row shows estimates of the effect on the number of vacancies posted with employer name. The 2SLS estimates indicate that a ten percentage point increase in the share of a CZ eligible for Medicaid increases the total number of vacancy postings for healthcare workers by 8.9 percent and the number of vacancy postings for healthcare workers with employer name by 5.4 percent.

These results indicate that providers seek to hire more workers in response to increased Medicaid eligibility, suggesting that providers at least try to increase capacity and supply in response to Medicaid expansions. But if there are not enough healthcare workers for hiring to increase, an increase in the number of vacancies may not lead to more hires. Instead, the additional vacancies could go unfilled. To consider if the increase in vacancy postings leads to more workers being hired, the first row of Panel B of Table 3 displays estimates of the effect of Medicaid eligibility on the IHS of healthcare hires from the QWI. The 2SLS estimate indicates that a ten percentage point increase in Medicaid eligibility increases hires by 4.2 percent. Thus, it appears that employers in the healthcare industry are able to increase the number of workers they hire in response to increased Medicaid eligibility.¹⁹

While not enough time has passed since the Medicaid expansions for a full study of employment dynamics, a discussion of the relationship between hiring and employment and of differences between expected short-run and long-run effects is still warranted. If adjustment were instantaneous, employment would rise to its new steady-state level immediately and remain relatively flat thereafter. Hiring would surge in the first quarter after Medicaid eligibility is expanded to allow employment to reach its new steady state and would subside thereafter. Even with immediate adjustment, hiring would still be higher than pre-expansion levels because more hires would be needed to maintain the higher steady-state employment. More likely, though, is that employment does not reach its steady state immediately and instead takes time to adjust. In this case, hiring would rise as the Medicaid expansion is being implemented and would then slowly fall to a new steady-state level as employment gradually rises to its new steady state.

Although a full examination of the employment dynamics is beyond the scope of this study, knowing the sign of the impact of Medicaid eligibility on employment is important for interpreting the results of this study. So far, the discussion has mostly centered on new vacancies and hires originating because of an increase in healthcare jobs, but vacancies and hires could also increase if increased Medicaid eligibility causes people to leave healthcare jobs. To consider if these new vacancies and hires are associated with employment growth, the final row of Table 3 examines the impact of Medicaid eligibility on employment in the healthcare sector. If the percent increase in employment were to match the percent increase in health insurance coverage, a ten percentage point increase in Medicaid eligibility would increase healthcare employment by 1.9 percent. The 2SLS estimate in Table 3, however, indicates that a ten percentage point increase in Medicaid eligibility leads to healthcare employment being 0.5 percent higher in the two years after the Medicaid expansion, which indicates that the immediate effect on healthcare employment is less than the immediate effect on overall health insurance coverage in percent terms. Possible explanations for why the effect on healthcare employment would be smaller than the effect on health insurance coverage in percent terms include that Medicaid may lead to smaller employment responses than private insurance and that some healthcare employment is not directly tied to health insurance or to the use of healthcare services. Again, though, it is important to note that this analysis does not necessarily indicate what the final impact on healthcare employment will be.

Columns 3 and 4 of Table 3 focus on observations that are in either the first or the last time period of the analysis data. While these estimates are still not the long-term effects, they exclude intermediate years of adjustment and can provide insights into the longer-term effects. The estimates rise across the specifications, as would be expected, though they are no longer statistically significant when employment is the dependent variable.

The estimates can be interpreted as the effect of a one percentage point change in Medicaid eligibility. The main 2SLS estimates imply that a one percentage point increase in Medicaid eligibility leads to a 0.23 percentage point increase in Medicaid coverage, a 0.17 percentage point increase in any health insurance coverage, a 0.89 percent increase in the overall number of vacancies posted, a 0.54 percent increase in the number of vacancies posted with employer name, and a 0.05 percent increase in healthcare employment. Several back-of-the-envelope calculations can help with interpreting these estimates. When evaluated relative to the CZ-level medians of their respective values, the estimates imply that a one percentage point increase in Medicaid eligibility leads to 1,117 more people becoming eligible for Medicaid, 268 more people having Medicaid coverage, 190 more people having any health insurance coverage, 5.4 additional vacancies, 1.7 additional vacancies with firm name, and 2.8 additional healthcare workers. Converting these estimates to represent the number of additional vacancy postings per 100 additional Medicaid enrollees suggests that an additional 100 Medicaid enrollees is associated with 0.6–2.0 additional vacancy postings and with 1.0 additional healthcare workers.

Estimates from the Oregon Health Insurance Experiment (Finkelstein et al. 2012) can provide a sense of the amount of increased healthcare services that would arise from providing Medicaid coverage to 100 additional people. The estimates from the Oregon Health Insurance Experiment imply that providing an additional 100 people with Medicaid coverage increases the annual number of inpatient hospital days by 10.8, the annual number of outpatient visits by 262, and annual spending on healthcare by around $110,000 in 2016 dollars. Because of the roughness of these calculations and because the new positions would not be expected to provide all of the services to the new Medicaid patients, I am hesitant to make strong statements about the link between the additional services and the additional employment. For instance, these calculations should not be interpreted to mean that an additional $110,000 pays for one additional worker. However, it is a reassuring plausibility check that the back-of-the-envelope calculations do not imply an outsized relationship between increased healthcare spending from Medicaid and healthcare employment.

V. Effect of Medicaid Eligibility on Skill Requirements

Section II.C discussed two conceptual models that predict that Medicaid expansions would lead to a lower average skill level of healthcare workers. Using the data on vacancy postings, I now examine the effect of Medicaid eligibility on skill requirements of new healthcare workers. As a simple way to examine the evolution of skill requirements, I first regress the number of skills in vacancy postings on a vector of occupation fixed effects and then plot annual mean residuals separately for CZs that experienced above-median increases in Medicaid eligibility during the study period and for all other CZs in Figure 2. Means of the residualized number of skills are similar for both sets of CZs until 2014, which is when the new eligibility rules went into effect for most of the states that expanded Medicaid coverage. Beginning in 2014, skill requirements began falling in CZs that experienced large increases in Medicaid eligibility, while skill requirements in other CZs remained steady. These trends are descriptive evidence consistent with the hypothesis that increased Medicaid eligibility leads to average skill requirements for new healthcare workers falling.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Means of the Residualized Number of Skills Required

Notes: The graph shows trends in the means of the residualized number of skills required in vacancies for healthcare jobs separately for commuting zones (CZs) with changes in Medicaid eligibility above the median and for all other CZs after controlling for the vacancy’s occupation.

To provide estimates of the impact of Medicaid eligibility on skill requirements after accounting for possible confounding factors, I estimate equations of the following form: (2) where i indexes the vacancy posting, o indexes the occupation at the six-digit SOC level, the γ are occupation-by-CZ fixed effects, the δ are occupation-by-year fixed effects, and all other variables are defined as before. Some specifications of Equation 2 will also include employer fixed effects to consider the role of employers in the effect of Medicaid eligibility.

Panel A of Table 4 displays estimates of the effect of Medicaid eligibility from estimating Equation 2 using the full sample of vacancies. As with the other tables of estimates, Column 1 displays OLS estimates, while Column 2 displays 2SLS estimates. As with the annual residualized means displayed in Figure 2, the estimates in Panel A suggest that increased Medicaid eligibility results in average skill requirements for job vacancies for healthcare workers falling within narrowly defined occupations. The 2SLS estimates indicate that a ten percentage point increase in Medicaid eligibility reduces the average skill requirements requested in job vacancies for healthcare workers by 0.24 skills per vacancy, or by 4.3 percent. Note that since these results control for six-digit SOC codes, they indicate that the negative impact of Medicaid eligibility on skill requirements occurs within occupation rather than because employers are hiring a different composition of occupations.

Panel B of Table 4 focuses on the sample of vacancy postings that contain the hiring employer’s name. The first two columns of Panel B replicate the analysis from Panel A with the restricted sample. Overall, the results are similar with the restricted sample, though the point estimates fall slightly and standard errors rise for both specifications. Neither of the estimates for the restricted sample is statistically significantly different from its corresponding estimate from Panel A, suggesting that the estimated effect of Medicaid eligibility on vacancies with employer names is not dramatically different from the estimated effect on the full sample. As explained in Section II.C, this reduction in average skill requirements for new vacancies could occur if employers that regularly hire lower-skilled workers increase hiring to accommodate the increased demand from Medicaid expansions, which would lead to employers that typically hire lower-skilled workers accounting for a higher share of total vacancies. Alternatively, average skill requirements for new vacancies could also fall if employers have to reduce their skill requirements in response to increased competition for healthcare workers. To distinguish between these two explanations, I supplement Equation 2 with employer fixed effects. While estimating models with employer fixed effects provides insights into whether average skill requirements fall within firms or because of compositional changes in the firms that hire, it is important to note that including employer fixed effects means that the effect of Medicaid eligibility is identified only from firms that hire before and after Medicaid is expanded. Firms that begin hiring in response to the Medicaid expansion will not contribute to identifying the estimates because these firms would have no variation in Medicaid eligibility across vacancy postings.

Columns 3–6 of Panel B of Table 4 display estimates from models that incorporate employer fixed effects. Columns 3 and 4 display estimates from models that include occupation-by-CZ fixed effects, occupation-by-year fixed effects, and employer fixed effects, which assumes a constant employer effect across occupations. Columns 5 and 6 display estimates from models that include employer-by-occupation fixed effects and occupation-by-year fixed effects, which allows each employer to have a different baseline skill requirement for each occupation. Regardless of whether the simpler employer fixed effects or the more flexible employer fixed effects are included, the point estimates of the effect of Medicaid eligibility on skill requirements change signs from the estimates in Columns 1 and 2 and become statistically indistinguishable from zero across all specifications. In short, these results do not provide evidence that increased Medicaid eligibility results in employers having to lower skill requirements. Instead, these results imply that average skill requirements fall in a CZ because employers that hire lower-quality workers post a greater share of total vacancies, which is consistent with the mixed-economy model.

The occupation fixed effects included in the results in Table 4 indicate that the negative impact of Medicaid eligibility estimated in the first two columns of Table 4 does not occur because employers are hiring lower-skilled occupations, but the results in Columns 1 and 2 of Table 4 do not rule out the possibility that the negative effect occurs because healthcare employers hire different kinds of workers within six-digit SOC codes in response to increased Medicaid eligibility. Even within six-digit SOC codes, there are different kinds of workers and jobs. For example, the six-digit SOC code for registered nurses includes vacancies for intensive care unit nurses, emergency room nurses, and dialysis nurses. In Table 5, I replicate the analysis in Table 4 but replace the occupation fixed effects with job-title fixed effects. The results changing dramatically when job-title fixed effects are included might suggest that the negative impact of Medicaid eligibility on average skill requirements occurs because firms are hiring different job titles within narrowly defined six-digit SOC codes in response to increased Medicaid eligibility. The results in Table 5, however, follow a similar pattern to the results in Table 4.

The inclusion of employer fixed effects leading to Medicaid eligibility having no effect on skill requirements could occur if employers that hire lower-skilled workers in the pre-period hire more workers in the post-period or if new employers that hire lower-skilled workers begin hiring in response to the Medicaid expansion. To verify that the impact of Medicaid eligibility on average skill requirements comes in part from firms that hired prior to the Medicaid expansion posting more vacancies in response to increased Medicaid eligibility, I use 2010–2012 vacancy data to classify employers’ typical skill requirements prior to the ACA’s Medicaid expansion. I then directly evaluate the type of employers that increase their vacancy postings in response to increased Medicaid eligibility. First, using 2010–2012 postings with employer name, I regress skill requirements on a vector of occupation fixed effects and year fixed effects. Next, I compute the mean residuals for each employer, which allows each employer to be ranked according to its typical skill requirements after accounting for the occupations and years it hires. For each CZ, I group employers into quartiles of this residual distribution and fix employers’ quartiles across all years of the data. Finally, I estimate the effect of Medicaid eligibility on the number of vacancy postings from employers in each quartile of the distribution using Equation 1.²⁰

Table 6 gives the estimated effects on vacancies for employers in each part of the skill distribution. The 2SLS estimate indicates that employers in the bottom quartile of the skills distribution respond to a ten percentage point increase in Medicaid eligibility by posting 22.7 percent more vacancies. The estimated effects on vacancies from all other employers are statistically indistinguishable from zero. The point estimates fall as employers’ skill residuals rise, though the estimates are not precise enough to distinguish statistically among the effects of Medicaid eligibility on vacancies of employers in the top three quartiles.

These results suggest that employers that hired lower-skilled workers in the pre-period hire more workers in the post-period in response to increased Medicaid eligibility, which provides support for the hypothesis that employers that hire lower-skilled workers are responsive to increased Medicaid eligibility. However, it is important to note that this analysis cannot rule out the possibility that some providers who had not posted vacancies prior to the Medicaid expansion post vacancies in response to the Medicaid expansion. Furthermore, since including firm fixed effects means that identification comes from firms that hire in both the pre- and post-periods, the analysis with firm fixed effects captures the effect of Medicaid eligibility on large firms that hire regularly rather than the effect on smaller firms that hire sporadically.

Table 7 considers the effect of Medicaid eligibility on different measures of skill requirements. To simplify the presentation and discourse, these results focus on three 2SLS specifications for each dependent variable. Specifications in Column 1 are for the sample of all vacancies and include occupation-by-year and occupation-by-CZ fixed effects. Specifications in Column 2 also include occupation-by-year and occupation-by-CZ fixed effects but are for the sample of vacancies with employer name. Specifications in Column 3 are for the sample of vacancies with employer name but replace the occupation-by-CZ fixed effects in Specification 2 with occupation-by-employer fixed effects. For reference, the first row of Table 7 reproduces the results with the number of skills as the dependent variable. Online Appendix B considers the relationship between the overall changes in Medicaid eligibility CZs experience from 2010 to 2016 and the annual changes in the outcomes and shows that the timing of the changes in outcomes aligns with the timing Medicaid expansion.

In Rows 2–5 of Table 7, the dependent variables are the number of cognitive skills, the number of health skills, the number of organizational skills, and the number of social skills. The results for all vacancies shown in Column 1 indicate that average skill requirements fall in response to increased Medicaid eligibility for all types of skills considered. A ten percentage point increase in Medicaid eligibility leads to a 6.8 percent decrease in the number of cognitive skills required, a 6.4 percent decrease in the number of health skills required, an 8.5 percent decrease in the number of organizational skills required, and a 7.2 percent decrease in the number of social skills required.

Row 6 of Table 7 considers the effect of Medicaid eligibility on the likelihood that a vacancy requires a certification. The most common certifications required include first aid certifications, cardiac life support certifications, and board certifications for physicians. Though there is likely a large variance in certifications’ rigor, some evidence points to certain certifications indicating higher quality and skill levels for providers. For example, Curtis et al. (2009), Reid et al. (2010), and Silber et al. (2002) all find that physicians being board certified is associated with higher quality of care and with improved health outcomes for patients. The estimate in Column 1 of Table 7 indicates that, as with the number of specific skills in vacancies, an increase in Medicaid coverage is associated with a reduction in average certification requirements.

The final row of Table 7 considers the effect of increased Medicaid eligibility on the required years of experience. While experience generally has positive returns in the labor market as a whole, evidence on the returns to experience for care quality in the healthcare profession is mixed. For example, while some evidence suggests that physicians having more experience is associated with better care (O’ Malley et al. 2007), other evidence indicates that experience is associated with lower-quality care (Choudhry, Fletcher, and Soumerai 2005), likely because more recently trained healthcare providers are more likely to know about recent norms and advances. About 42 percent of vacancies list experience requirements. For those vacancies that do not list experience, I code the vacancy as requiring zero years of experience. The point estimate on Medicaid eligibility when years of experience is the dependent variable in Column 1 of Table 7 is negative but statistically indistinguishable from zero.

Column 2 of Table 7 focuses on the sample of vacancies with nonmissing employer name. Though some of the estimates are no longer statistically significant, most point estimates are near their corresponding estimate from Column 1. For this subsample of vacancies, the effect of Medicaid eligibility on the experience requirements is now statistically significantly different from zero at the 5 percent level. Column 3 focuses on the same sample as Column 2 but includes employer-by-occupation fixed effects instead of occupation-by-CZ fixed effects. When employer heterogeneity is accounted for, all the point estimates of the effect of Medicaid eligibility fall in absolute value. Only one of the coefficients on Medicaid eligibility is statistically significantly different from zero at the 10 percent level, and none are statistically significant at higher significance levels.

Though there are exceptions, the patterns of results for most of the estimates in Table 7 follow the same pattern as when the dependent variable is the total number of skills. Increased Medicaid eligibility is associated with decreased skill requirements, but the negative impact of Medicaid eligibility largely goes away when employer fixed effects are included.

Table 8 shows 2SLS estimates from specifications with occupation-by-CZ and occupation-by-year fixed effects for the specific components of the cognitive skills, organization skills, and social skills, as well as for data entry and reporting for the full set of vacancies. Given that some of the specific skills have small baseline means and that the estimates for specific skills are noisy, interpreting differences across the specific skills is difficult. Fifteen out of 18 of the coefficients are negative, though only seven are statistically significant at at least the 10 percent level.

Table 9 focuses solely on vacancies for physicians and registered nurses. Together these occupations make up nearly one-half of all healthcare workers in SOC 29. In addition to these occupations being important healthcare providers, focusing on them also allows for considering the effect of Medicaid eligibility on specific requirements that have been shown to be associated with a higher quality of care. Panel A replicates the previous analysis with the sample of nurses and physicians. Note that a few of the coefficients on specific skill categories are only marginally statistically significant or are statistically insignificant for the sample of vacancies with firm name before accounting for employer heterogeneity. However, the estimated effect of Medicaid eligibility is generally similar to the estimated effect for the full sample. Increased Medicaid eligibility leads to lower skill requirements on average but does not appear to have major effects on skill requirements after including employer fixed effects.

Panel B of Table 9 focuses on vacancies for physicians and shows the effect of Medicaid eligibility on the likelihood that a vacancy lists board certification as a requirement. An increase in Medicaid eligibility is associated with a decrease in the likelihood that a vacancy requires physicians to be board certified for the full sample, but Medicaid eligibility has no effect on the likelihood that a vacancy requires board certification after accounting for employer heterogeneity. Note, however, that Medicaid eligibility also has no statistically significant effect when focusing solely on the vacancies with firm name. Thus, while the sample follows the same pattern of an effect in aggregate but not when firm fixed effects are included, these results do not display the same patterns for the full sample of vacancies and for the subsample with firm name.

Panel C of Table 9 focuses on the likelihood that a vacancy for a registered nurse lists a bachelor’s degree as a requirement. While a bachelor’s degree is not required for obtaining the registered nurse designation, having a bachelor’s degree has been found to be associated with better care for registered nurses. For example, Aiken et al. (2003) find that a 10 percent increase in a hospital’s proportion of nurses holding a bachelor’s degree is associated with a 5 percent decrease in the likelihood of patients dying within 30 days of admission. The results in Panel C display a similar pattern as the main estimates.

Table 10 displays estimates of the effect of Medicaid eligibility on the number of skills required for additional occupation groups. While the main analysis focused on all occupations with a two-digit SOC code of 29, Panel A of Table 10 focuses on the narrower set of occupations for healthcare practitioners that treat and diagnose patients, which are occupations with a three-digit SOC code of 29-1. This sample excludes the 26 percent of vacancies in the main sample for health technologists and technicians. The point estimates for the subsample of healthcare practitioners that treat and diagnose patients in Columns 1 and 2 are larger but not statistically different from the corresponding point estimates for the full sample. As with the main analysis, the point estimates fall and are typically no longer statistically significantly different from zero once employer fixed effects are included.

Panel B of Table 10 focuses on healthcare support occupations, which have a two-digit SOC code of 31. These occupations include many lower-skilled jobs in the healthcare labor market, such as aides and assistants. Though healthcare practitioners contain 5.5 times as many vacancies as the healthcare support occupations, healthcare support occupations are an important component of the healthcare sector. While the magnitudes of the estimates are not as large for SOC code 31 as they are for SOC code 29, the estimates remain negative and statistically significant at at least the 10 percent level, until employer fixed effects are included in the analysis. Once employer fixed effects are included in the analysis, the estimates become statistically indistinguishable from zero. Panel C of Table 10 shows estimates for SOC codes 29 and 31 combined. As would be expected since there are many more vacancies for SOC code 29 than for SOC code 31, the estimates are close to the baseline estimates.

Panel D of Table 10 shows a set of placebo estimates for farming, fishing, and forestry occupations (two-digit SOC of 45). While increased Medicaid eligibility could have broad effects on local labor markets, similarly sized effects on these occupations that would not be expected to experience first-order effects of expanded Medicaid eligibility would raise concerns about the possibility of unobserved economic trends related to the Medicaid expansion. The estimates shown in Panel D, however, are statistically indistinguishable from zero, regardless of the fixed effects included in the analysis.

Though the occupation fixed effects included in the baseline analysis mean that the negative impact of Medicaid eligibility does not occur because of compositional changes in the occupations being hired, they do not rule out the possibility that employers hire different types of workers in response to increased Medicaid eligibility, which could be a potential margin of adjustment. A possible response to increased Medicaid eligibility is for healthcare employers to hire more mid-level providers to treat patients. Table 11 shows estimated effects of Medicaid eligibility on the number of vacancies for registered nurses, nurse practitioners, and physician assistants for employers with firm name. For reference, the first row of Table 11 shows the results from the main specification. As with the overall number of vacancy postings for healthcare occupations, vacancies for registered nurses, nurse practitioners, and physician assistants all rise in response to increased Medicaid eligibility. The 2SLS estimate suggests that a ten percentage point increase in Medicaid eligibility increases the number of vacancies posted for registered nurses by 4.5 percent, which is similar to the estimate for all healthcare occupations. The 2SLS estimates for nurse practitioners and physician assistants imply that a ten percentage point increase in Medicaid eligibility increases the number of vacancies posted for nurse practitioners by 11.0 percent and the number of vacancies posted for physician assistants by 8.2 percent. While these estimates are consistent with healthcare employers trying to increase the use of nurse practitioners and physician assistants, neither estimate is statistically significantly larger than the estimate for all healthcare workers.

Table 12 considers the sensitivity of the estimates to various empirical choices. To keep the discussion concise, Table 12 focuses on three outcomes. Column 1 shows the 2SLS estimate of the effect of Medicaid eligibility on Medicaid coverage, Column 2 shows the 2SLS estimate of the effect on the IHS of the number of vacancies, and Column 3 shows the 2SLS estimate of the effect on the mean number of skills. Columns 1 and 2 include area and year fixed effects, while Column 3 includes occupation-by-area and occupation-by-year fixed effects. The first row of estimates shows the results from the baseline specification.

Row 2 of Table 12 no longer includes CZ-level characteristics as controls. As would be expected if the 2SLS estimate successfully corrects for any endogeneity in the Medicaid eligibility measure, the estimates do not change much when exogenous controls are removed from the estimation.

Row 3 shows estimates from specifications that do not weight based on population for the estimates of the effect on Medicaid coverage and on the IHS of the number of vacancies. As the specifications with the vacancy-level data are not weighted, I do not show an estimate in the third column for this robustness test. An advantage of weighting by population is that larger CZs with characteristics that are more precisely measured are given more weight. While the estimated effect on the IHS of vacancies falls relative to the baseline specification, the pattern of results remains similar to the baseline results.

Row 4 excludes all intermediate years and focuses only on observations for 2010–2016. Across all specifications, the estimated effect of Medicaid eligibility rises in magnitude, which is consistent with the full effect of the Medicaid expansion not occurring immediately after the expansion happens.

One channel through which the effects of Medicaid eligibility on labor demand could be operating is through an expansion in physical capacity at healthcare facilities. To consider the role of capacity in the estimates, I draw on data from the Area Health Resource File, which has information on the number of hospital beds and the number of beds in skilled nursing facilities. In Online Appendix B, I show estimates of the effect of Medicaid eligibility on the logs of these variables. The results are imprecisely estimated, but they provide no evidence that increased Medicaid eligibility has led to immediate changes in the number of hospital beds or the number of beds at skilled nursing facilities. While these results are consistent with healthcare employers not being able to expand capacity, they could also reflect that facility construction might take several years or that healthcare employers can expand production without additional capital construction. Another way to consider if changes in vacancy postings are associated with changes in capacity is to control for capacity measures. The results changing dramatically with the capacity measures included as controls would suggest that the changes in the outcomes are associated with changes in capacity. Row 5 of Table 12 shows results that include controls for the log of the number of hospital beds and the log of the number of beds in skilled nursing facilities. The estimated effects of Medicaid eligibility are similar with these additional controls. While this analysis does not imply that supply-side constraints are not important in healthcare, they suggest that changes in capacity as measured by the number of hospital beds and number of skilled nursing facility beds do not appear to be a key mechanism for the effect of Medicaid eligibility on vacancy postings.

Row 6 shows results from conducting the analysis at the state level rather than at the CZ level. Since the changes in Medicaid eligibility rules produce state-level variation in Medicaid eligibility, the results from conducting the analysis at the state level should be similar to the results obtained from conducting the analysis at the CZ level. To obtain the estimates in Row 6 of Table 12, I replace all the CZ-level measures with state-level measures and calculate the instrument using the same approach described in Section III but simulate Medicaid eligibility at the state level rather than at the CZ level. The results in Row 6 confirm that the state-level analysis produces estimates of the effect of Medicaid eligibility that are similar to the estimates obtained from the CZ-level analysis.

As explained in Section III, I construct the instrument for a CZ each year using all the observations in the ACS for that CZ. Within a CZ, the instrument changes values only when Medicaid eligibility rules change. Thus, after including CZ-specific fixed effects, the only remaining variation in the instrument comes from changes in the Medicaid eligibility rules. Similarly, to calculate the instrument for the specifications shown in Row 6 of Table 12, I use all observations for each state to construct the value of the instrument for each state and year combination. This approach for calculating the instrument is similar to but not exactly the same as the approaches used in Currie and Gruber (1996a,b). Currie and Gruber (1996a) use separate national random 10 percent samples each year to calculate simulated eligibility, while Currie and Gruber (1996b) use separate national random samples of 300 children from each age and each year to calculate their instrument. The sets of fixed effects included in each study’s 2SLS estimation mean that the remaining variation in the instruments comes solely from changes in the Medicaid eligibility rules for each of the studies. Therefore, the simulated eligibility measures calculated from each approach are valid instruments for Medicaid eligibility. To verify that I obtain similar results when calculating the instrument as in Currie and Gruber (1996a), Row 7 of Table 12 shows results from state-level specifications that use an instrument constructed using separate 10 percent national samples each year to calculate the instrument. As would be expected since both approaches summarize the same variation in Medicaid eligibility thresholds and estimate the same parameter, the results are similar with this alternative construction of the instrument.

I conclude this section by briefly summarizing several additional pieces of analysis. Online Appendix B displays tables of results and provides more comprehensive discussions when necessary.

As mentioned earlier, Online Appendix B discusses reimbursement rates and presents analysis that controls for changes in reimbursement rates for primary care providers during the study period to ensure that changes in reimbursement rates are not a threat to the identification strategy. Online Appendix B also displays results that test for heterogeneous effects of Medicaid eligibility for CZs with high and low reimbursement rates. The results are consistent with higher-quality providers posting additional vacancies in response to increased Medicaid eligibility when Medicaid’s reimbursement rates are higher, though the analysis cannot conclusively rule out that the effect of Medicaid eligibility is the same for places with high and low reimbursement rates for primary care.

Identification of the effect of Medicaid eligibility in this study comes from comparing how outcomes in CZs that experienced large changes in Medicaid eligibility due to the ACA’s Medicaid expansion change relative to how outcomes change in nonexpansion CZs. The analysis provides evidence that the healthcare labor market responds to changes in Medicaid eligibility. Whether these responses are associated with national increases in healthcare job vacancies, hiring, and employment or whether they reflect adjustments across CZs depends on assumptions about whether Medicaid expansions have spillover effects on nonexpansion CZs. If Medicaid expansions do not have spillover effects on nonexpansion CZs, the results presented here suggest that increased Medicaid eligibility leads to national increases in healthcare job vacancies, hiring, and employment. If, on the other hand, Medicaid expansions have spillover effects on nonexpansion CZs, the results presented here indicate that increased Medicaid eligibility leads to labor market adjustments across CZs. Online Appendix B discusses why labor market spillovers could occur and why they may be unlikely to occur in this setting. Online Appendix B also shows results from a difference-in-differences analysis that examines the impact of Medicaid expansions on nonexpansion CZs that are near expansion CZs by comparing how their outcomes change after the Medicaid expansion relative to how outcomes change for nonexpansion CZs that are farther from expansion CZs. The results shown in Online Appendix B provide no evidence that Medicaid expansions have effects on nonexpansion CZs that are near expansion CZs.

Finally, the main analysis focused on all vacancies for healthcare occupations, regardless of the type of employer that posted the vacancy. An issue with incorporating industry into the analysis is that detailed industry codes are missing for a high share of vacancies, which limits the ability to consider the possibility of heterogeneous effects across different provider types.

However, Online Appendix B displays results that focus on vacancies that are identified as being for the healthcare sector (two-digit North American Industry Classification System code of 62) and verifies that similar patterns hold for these vacancies. Online Appendix B also includes controls for industry when the industry of the vacancy is available and shows that the point estimates do not change dramatically when industry controls are included.

VI. Conclusion

An expansion of health insurance coverage and an aging population mean that certain areas have experienced large increases in healthcare demand in recent years and that other areas will experience large increases in the coming years. Much of the rise in health insurance in recent years has come from Medicaid. While increased Medicaid coverage would be expected to lead to new jobs in the healthcare sector, Medicaid’s low reimbursement rates lead to concerns about the quality of the newly hired workers, especially in light of concerns about shortages of high-skilled healthcare workers. By adding more stress to the healthcare workforce, the increased demand for healthcare workers accompanying increased Medicaid eligibility also has the potential to lead to employers reducing their skill requirements.

The goal of this study was to examine the impact of expanded Medicaid coverage on healthcare employers’ demand for healthcare workers. The analysis indicates that increased Medicaid coverage leads to healthcare employers posting more vacancies and hiring more workers. Average skill requirements for new vacancies fall in areas that experience increased Medicaid eligibility, as employers who typically hire lower-skilled workers post more vacancies in response to Medicaid expansions. Vacancy postings for employers that typically hire higher-skilled workers are unaffected by increased Medicaid eligibility. Despite dire warnings of exacerbated shortages from the ACA espoused by many and despite previous research that uses vacancy data and shows that downs-killing occurs when labor markets tighten more generally, this study did not find strong evidence of within-employer reductions in skill requirements as a result of the ACA’s Medicaid expansion.

These findings are relevant as policymakers try to ensure that Medicaid guarantees access to high-quality care. It is reassuring that there are employers that respond to increased Medicaid eligibility by posting more vacancies and hiring more workers, as the research that shows that Medicaid coverage leads to improved health suggests that care from less-skilled providers is still likely better than no care at all. But as other research suggests that care from high-skilled providers is better than care from lower-skilled providers, these results highlight the need to provide incentives to induce high-quality providers and healthcare employers to be responsive to Medicaid patients’ needs if policymakers want to improve the quality of care that Medicaid patients receive.