Census Bureau Issues Working Paper Entitled 'How Do Health Insurance Costs Affect Firm Labor Composition & Technology Investment?'

WASHINGTON, Sept. 28 -- The U.S. Census Bureau issued the following working paper (No. CES-23-47) entitled "How Do Health Insurance Costs Affect Firm Labor Composition and Technology Investment?"

The paper was written by Shan Ge, Lawrence D. W. Schmidt and Cristina Tello-Trillo.

Here are excerpts:

* * *

Introduction

While health insurance protects households from the financial repercussions of health shocks, its costs are large and rapidly rising. Health insurance is often sponsored and heavily subsidized by employers in the U.S.; as of 2021, over half of Americans are covered by employer-sponsored health insurance and U.S. employers contributed $16,253 on average in health insurance premiums per family plan./1

Unlike many other labor costs such as wages and payroll taxes, health insurance costs do not scale with individual productivity, so these costs are even larger as a fraction of overall compensation for low-income workers. Ceteris paribus, higher health insurance costs can act as a fixed cost or "head tax" and depress demand for labor, especially for the types of low-income workers that also face increasing displacement risk from other well-documented headwinds in the labor market./2

In part due to these potential labor market distortions, the role employers should play in health care provision is the subject of a heated policy debate. In this paper, we speak to these distortions by studying how plausibly exogenous variation in firms' health insurance costs affects employment, and how the effect varies across workers along the income spectrum.

Holding the health insurance take-up rate fixed, an increase in health insurance premiums would likely manifest as some combination of an increase in total labor costs per worker (earnings + firm contributions) paid by the firm and a decrease in take-home pay net of health insurance (earnings - worker contributions) received by the worker. The former case creates incentives for firms to lay off workers while the latter creates incentives for workers to leave. Thus, in theory, increases in health insurance costs should reduce firms' demand for labor. The answer becomes less clear when one considers labor market frictions. For instance, firms may be able to pass most of the cost increases onto workers if their labor supply is sufficiently inelastic, if workers can avoid paying higher premiums by obtaining comparable coverage elsewhere (e.g., through spouses or private plans), or if workers are inattentive to changes in their contribution to the benefit. Even without cost passthrough, labor adjustment costs might prevent firms from responding sufficiently to transitory cost increases. It is thus an empirical question whether and to what extent changes in health insurance costs affect firm employment./3

Moreover, the average employment effects may mask important heterogeneity across worker types, as there may be distributional implications of changes in health insurance costs. If employment does adjust in response to higher insurance premiums, one might expect such adjustments to be stronger for low-income workers since 1) as noted above, the same dollar increase in premiums leads to a larger proportional increase in firms' costs or employees' net-of-insurance pay for lower-paid workers and 2) many of these workers' tasks can be more easily performed by technology and/or benefits-ineligible (part-time or offshore) workers.

To test these hypotheses, we construct a unique employer-employee matched dataset spanning the period of 2012-2019. We combine Census administrative microdata on the universe of U.S. firms and their workers with information on firms' health insurance costs. Specifically, we collect this information from Form 5500, which is a mandatory filing for firms with benefit plans covering more than 100 participants. We also gather insurer regulatory filings to extract data on the losses of firms' insurance providers. We find that, following a plausibly exogenous increase in health insurance premiums, firms reduce employment. The decline in retention is particularly large among low-income workers, whose likelihood of unemployment also increases more than high-income workers. In conjunction with these patterns, the nature of firms' non-labor input usage changes. Specifically, we document an increase in investments in information technology following the shock, consistent with firms substituting labor with technology.

When studying how health insurance premiums affect firms and their workers, one must address potential endogeneity concerns. For example, firms that want to retain and attract workers may choose to offer more generous, but costlier health insurance plans. This suggests a correlation between premiums and other unobservable drivers of employment growth. We overcome this challenge by designing a novel identification strategy to isolate changes in health insurance premiums that are plausibly exogenous to firm-level conditions. Specifically, we use idiosyncratic variation in insurers' losses as an instrumental variable for premiums faced by their customers, i.e., employers.

We expect insurer losses to affect the premiums they charge for several reasons. First, existing evidence suggests that negative financial shocks often create incentives for firms to prioritize immediate cash flows over more distant ones./4

Second, ACA guidelines specifically cap insurers' profits in three consecutive years and therefore tie potential premium increases to recent losses. Higher recent losses provide room for insurers to raise prices under these guidelines. Third, past losses affect insurers' beliefs about future claims. We elaborate on these mechanisms in Section 2. Regardless of the mechanism, we argue that an insurer's decision to increase premiums in response to its prior losses reflects its own internal objectives and constraints, rather than its customers' labor and technology investment policies. This is our key exclusion restriction.

Using a two-stage-least-squares (2SLS) design, we study the effect of idiosyncratic shocks to firms' health insurance costs on subsequent firm and worker outcomes. Our instrument is highly relevant: the first-stage results suggest that a one-standard-deviation increase in insurers' losses predicts a 1.4% to 2% increase in premiums. Further, we provide direct evidence that past insurer losses do not predict firms having larger claims in the future but do predict higher insurer markups (proxied by the ratio of premiums to claims). This is consistent with loss-driven premium changes reflecting insurers' objectives rather than omitted employer or worker characteristics.

We find that an increase in the instrumented insurance premiums leads to a significant decline in firms' overall employment. Our estimates suggest that a 10% increase in premiums is associated with a 2-3% decline in firm-level employment. The employment changes are primarily driven by a reduction in the retention rate of existing workers rather than a decline in the number of new hires. These findings suggest that firms are responsive to these idiosyncratic shocks to health insurance costs.

These employment responses are robust across a wide battery of additional tests. Our benchmark specification includes firm fixed effects and year fixed effects to absorb average differences across firms and macroeconomic conditions. We find similar estimates in a more saturated specification with commuting zone-industry-year interactive fixed effects, which help to rule out confounding effects arising from local conditions or industry dynamics. To address the concern that some firms may be sufficiently large to impact insurers' reported losses, we show that results hold when we include only employers whose premium payment accounts for less than 1% of their insurers' total premium revenue. We further reduce the impact of potentially confounding firm and local conditions by constructing a "leave-one-out" instrument, where insurer losses are computed using only losses and premiums incurred in states outside of the employer's location. Our results remain robust to the alternative instrument design.

We also test the hypothesis that an increase in health insurance premiums should affect lower-income workers more than their more highly remunerated coworkers. For each worker, we compute her average earnings during the previous five years and then estimate how her retention likelihood is affected differently by health insurance premiums depending on her past earnings. Importantly, we use individual-level data and identify these differential effects by imposing firm-by-year interactive fixed effects in our 2SLS estimation. This specification helps us eliminate concerns related to employers' conditions or confounding dynamics related to industry and location. We find that an increase in healthcare premiums generates a negative effect on the retention likelihood of low-income, low-skill workers relative to their higher-earning coworkers. A 1% increase in health insurance premiums leads to a larger increase in job separation likelihood by 0.1 percentage points for low-income workers compared to workers whose past earnings are one standard deviation higher.

Since we do not observe reasons for separation, it is possible that our results are driven by larger increases in the probability of low-income workers voluntarily quitting for more desirable jobs elsewhere. To shed light on this issue, we examine the probability that a worker becomes unemployed following an increase in employer-sponsored health insurance premiums. We find that, following a 1% increase in premiums, low-income workers' probability of being unemployed increases by 0.05 percentage points more than high-income workers whose average past earnings are a one-standard-deviation higher. Therefore, our individual-level results on retention are unlikely driven by low-income workers quitting for more desirable jobs. In addition, we find that low-income workers also see a larger increase in the probability of being part-time. This is consistent with the idea that firms make some low-income workers ineligible for health insurance by converting them to part-time status, since firms are not required to provide health insurance to part-time workers.

A natural question is whether firms pass the increase in health insurance costs to workers. We find that the average participation rate (the number of plan participants as a ratio to the number of employees) drops significantly with our instrumented premiums. This could be a result of increased employee contributions to health plans./5

Another natural question is how workers' earnings change following an increase in health insurance premiums. Unfortunately, we are unable to fully answer this question since we only observe workers' net earnings, which equal gross earnings minus workers' health insurance and retirement plan contributions if they participate.

Taking stock, the results summarized so far are consistent with the hypothesis that rising health insurance premiums increase labor costs, especially for low-income workers.

As a result, employment sharply declines at firms facing positive premium shocks. Our findings are consistent with two (not mutually exclusive) mechanisms. First, higher premiums charged by insurers increase employer contributions, which incentivizes firms to lay off workers and/or cut hours for a subset of workers. Second, firms may elect to pass on part or all of the increased cost to workers, by reducing eligibility and/or pushing up employee contributions. This reduces workers' effective earnings and leads them to seek outside options. Regardless of the interpretation, low-income workers seem "worse off" as they are more likely to fall into unemployment.

From the firm's perspective, lower worker plan participation likely brings long-run consequences. The fact that, even absent formal mandates, firms contribute to workers' health insurance implies that they place some value on worker plan participation./6

When workers stop participating following a rise in premiums, they may be less productive (partly because they are more likely to leave, causing the firm to incur the cost of replacing them in the future)./7

Therefore, the marginal benefit of each worker for the firm declines, moving the firm's demand curve to the left. We thus cannot directly infer a labor demand elasticity from our estimates, because shocks to premiums likely induce shifts in both firm labor demand and labor supply curves.

What types of firms are more responsive to shocks to health insurance premiums? We explore several dimensions of heterogeneity. First, employment responses should be concentrated in cases where health insurance premiums account for a large fraction of a firm's total labor costs. Indeed, following instrumented increases in health insurance premiums, we only observe employment cuts among firms with high ex-ante premiums-to-total payroll ratios, but not those with low premiums-to-payroll ratios. Second, firms that have to compete in a tight labor market should be less able to pass the increased premiums to workers without losing workers to other employers. Consistent with this hypothesis, we find that firm-level employment responds more to changes in insurance premiums primarily in counties with lower unemployment rates. Third, firms with better future investment opportunities should be less responsive to a temporary increase in labor costs as their employment level is more likely to be below its desired level. Measuring investment options using industry-level Q based on public firms, we find firms in high-Q industries exhibit little sensitivity to premium shocks. Collectively, these results lend additional support to our main findings and shed light on the trade-offs faced by employers. Finally, we find that the employment-premium cost relation does not vary across firm size, suggesting that external financial constraints are unlikely to explain our findings. Instead, our findings are consistent with firms optimizing over their production function and responding to changes in labor costs.

We conclude by bringing additional data to test whether firms invest in automation technology in response to health insurance costs, potentially to replace low-income workers. To do so, we leverage the Aberdeen data that offer information on the budgets set by firms to purchase, install, and maintain information technology. The database covers individual establishments belonging to both public and private firms. We find that a 1% instrumented increase in health insurance premiums is associated with around a 1-2% increase in the IT spending per person in a firm. We find a similar estimate for the personal computer budget per person.

Our study contributes to two streams of research. First, we contribute to the growing literature on the effect of health insurance costs on firms and workers. In particular, our paper is related to studies examining the effect of health insurance costs on worker employment and labor force participation. The evidence is mixed. Using non-administrative data, Cutler and Madrian (1998) find that the rising cost of health insurance is associated with increasing work hours. Using similar data sources over a different sample period, Baicker and Chandra (2006) document that higher insurance premiums reduce the likelihood that a worker is employed full-time and the hours worked. Several other papers study the effect of the Affordable Care Act, which mandated many employers to offer health insurance plans to employees. A concurrent study by Almeida et al. (2021) finds that public firms do not change employment, but cut the number of covered workers. Mulligan (2020) argues that firms cut jobs to stay under 50 employees to avoid triggering the employer mandate. Dillender et al. (2022) find that part-time employment increases./8

Our paper is particularly related to Tong (2021), who finds that increased healthcare costs reduce capital expenditures and R&D among public firms, more so for financially constrained firms. We focus on the employment effect of mostly private, smaller firms and, importantly, document heterogeneous impacts of different workers within the firm.

To the best of our knowledge, all of the prior papers in this literature use aggregate shocks that often can confound with other macro or regional shocks, as well as affect many employers and employees' outside options for obtaining health insurance. For example, the Affordable Care Act affects a large number of employers in complex ways and also affects individuals' ability to obtain health insurance independent of employers. Our paper distinguishes itself by using idiosyncratic and exogenous shocks that generate variation within markets. With these shocks in hand, we can better tease out the exact mechanism, identify the causal effect of health insurance costs, and highlight the role of firms in responding to premium increases and adjusting their labor and technological inputs.

Our paper is related to contemporaneous work Ouimet and Tate (2022), who analyze all non-wage benefits, including health plans, leave, and retirement. The two studies not only examine different types of non-wage benefits, but also focus on different variations in benefits and document different worker-level outcomes. Consistent with our results, Ouimet and Tate (2022) find that higher benefits instrumented by peer firms' benefits lead to lower reliance on low-wage workers. We focus on idiosyncratic premium shocks arising from insurance providers and document the labor market impacts of increases in health insurance costs on low-income workers, who experience a greater increase in job separation and unemployment rates. The two studies complement each other and jointly advance the understanding of firms' responses to non-wage labor costs.

We are also related to Finkelstein et al. (2023), who calibrate a theoretical model about how the employer-sponsored health insurance regime contributes to higher equilibrium wages earned by high-skilled workers. We empirically examine how the premiums of employer-sponsored health plan premiums on the employment outcomes of workers across the skill spectrum.

Second, our instrument also highlights the role of insurers in transmitting shocks across geographical regions and firms. We thus complement existing work showing how financial and nonfinancial firms propagate shocks in the economy (e.g., Gilje et al. 2016, Cort'es and Strahan 2017, Giroud and Mueller 2019, Bena et al. 2022).

Finally, we add to the studies documenting the persistent decline in labor share as well as the demand for low- and middle-skill workers. Prior literature focuses on the impact of import competition (Bilal and Lhuillier 2021, David et al. 2013, Lu and Ng 2013, Pierce and Schott 2016), technological advancement (Doms et al. 1997, Acemoglu and Restrepo 2019, Acemoglu and Restrepo 2020), and distributional implications of policy changes (Tuzel and Zhang 2021, Engbom and Moser 2021). We add to this line of research by focusing on a less explored, yet important part of labor input cost, namely health insurance premiums. Our study highlights its heterogeneous effect on workers across income levels, potentially shedding light on another source of deteriorating labor demand for low-wage workers. Relatedly, rising health insurance costs can increase incentives to adopt substitute technologies, consistent with our evidence on technology investments in Section 8.

* * *

Conclusion

Employer-sponsored health insurance is a significant component of labor costs. We examine the causal effect of health insurance premiums on firms' employment, both in terms of quantity and composition, as well as technology investment decisions. To address endogeneity concerns, we design an instrument for insurance premiums using idiosyncratic variation in insurers' losses, that is plausibly exogenous to their customers, i.e., individual employers and their workers. Using Census microdata, we show that following an increase in increased premiums, firms reduce employment. Relative to higher-wage coworkers, lower-wage workers experience a larger increase in the probability of being laid off and remaining unemployed for two years following the shock. Firms also invest more in information technology, potentially to substitute for labor.

Our paper provides potential implications for policymakers. In particular, a downside of employer-sponsored health insurance is that it introduces dependencies between insurance costs and employment. Regulations might consider the value of subsidizing the costs of providing health insurance to low-income workers.

Our paper also speaks to the persistent decline in the labor share and a weakening demand for low- and middle-skill workers in the U.S. (e.g., Acemoglu and Autor (2011), Karabarbounis and Neiman (2014), Autor (2014), Autor et al. (2020), Kehrig and Vincent (2021)). A growing body of research investigates various determinants of this structural shift, including technological progress, trade exposure, and offshoring pressure, which are forces that have made substitutes increasingly competitive (e.g., Allen (2001), David et al. (2013), Pierce and Schott (2016), Dorn et al. (2017), Acemoglu and Restrepo (2019), Acemoglu and Restrepo (2020)). Health insurance costs, which have been rising faster than inflation and wages, could be an alternative and complementary channel that exacerbates the decline in firms' labor demand for these workers. Although our paper documents the partial equilibrium effects of idiosyncratic shocks to health insurance costs, we highlight that such costs could be distortionary to firms' employment and worker composition.

* * *

The paper is posted at: https://www2.census.gov/library/working-papers/2023/adrm/ces/CES-WP-23-47.pdf