Medicaid Expansion to Children Born after September 30, 1983

The cost of providing Medicaid to non-disabled children in 1991 was roughly $902 per year per enrolled child in 1991 USD (Wherry et al. (2018)). Hendren and Sprung-Keyser (2020) translate this to $1,484 in 2012 US dollars using the CPI-U-RS. For those obtaining insurance eligibility through the expansion, 44.6% take up Medicaid. This implies a cost per child-year of Medicaid eligibility of $663. Hendren and Sprung-Keyser (2020) assume this cost is spread uniformly over ages 8 through 14, and discount to back to age 8 using a 3% interest rate, which gives a discounted cost of $528.

Part of the cost paid by the government reflects compensation for care that otherwise would have gone uncompensated. Lo Sasso and Seamster (2007) use variation in childhood expansions of Medicaid to document that each additional child eligible for Medicaid reduces costs by $17 through both reductions in charity care and default on medical debt. The discounted value of these savings over the seven years of increased eligibility is $13.67. Assuming this falls to the government, it implies the net cost of eligibility to the government is $645 per year per child.

Wherry et al. (2018) revisit the birth discontinuity to look at health outcomes when children are 25 years old. They find significant reductions in chronic health conditions and healthcare utilization such as hospitalization and emergency room visits. Those born just after Sept 30, 1983 are 7.7 percentage points more likely to be covered (Wherry et al. (2018) Table 2). Correspondingly, they find significant reductions in hospitalization and prevalence of chronic conditions. Using the CCT bandwidth selector, Appendix Table 24 shows that being born after the Sept 30 cutoff leads to a 0.164 reduction in log costs of publicly insured hospitalizations and 0.049 reduction in log costs of publicly insured emergency department visits.

Translating these reductions into the dollars recouped by the government requires the base levels of public spending on these populations. Using the summary statistics from the paper, the authors’ calculations imply a total average cost of hospitalizations of $402 per year and $231 for emergency room expenses in 2012 USD. While the authors do not have a breakdown of the total costs paid by the government, Wherry et al. (2018) estimate that between 41-69% of total costs for this sample are paid by the government – Hendren and Sprung-Keyser (2020) take the midpoint of 55%. This suggests public expenditures of $221 on hospitalizations and $127 on emergency room costs per person per year. Multiplying by the 16.4% and 4.9% reductions, respectively, yields a total cost savings to the government of 36.25+6.24 = $42.49 (2012 USD) per person per year. Discounting these increased revenues back to the time when the initial outlays were made (when children were 8 at the earliest) yields a total cost savings of $26 using a 3% discount rate. These cost savings compare those born Oct 1 versus Sept 30, 1983 to those born just before. This discontinuity corresponds to a 0.87 increase in child-years of eligibility for Medicaid. This suggests that the discounted cost savings per year of Medicaid eligibility are $30 when individuals are 25 years old.

Because these reductions in expenditures are argued to be driven by reductions in chronic conditions that persist over time, their baseline specification assumes these effects persist as the individuals age. To account for this, Hendren and Sprung-Keyser (2020) assume that cost savings remain constant in levels, despite the growth in costs over the age distribution. Hendren and Sprung-Keyser (2020) then discount these back to age 8 using the real interest rate assumption of 3% for their baseline specification. For consistency with the earnings forecasts, Hendren and Sprung-Keyser (2020) assume the cost savings persist until age 65. But, their results suggest that the cost savings to government pays for itself by the time children are age 54.

In addition to impacts on future health costs, there could also be impacts on future earnings and tax revenue. Hendren and Sprung-Keyser (2020) make the arguably conservative assumption of no impact on earnings. They motivate this choice using estimates from Brown et al. (2015) who use administrative tax data to compute the mean tax revenue for children in their 20s and early 30s separately by birthday around the Sept 30 1983 threshold. Hendren and Sprung-Keyser (2020) note that Appendix Figure OA.3 reveals no significant changes in tax revenue around the discontinuity, although this test is potentially under-powered because it is not possible to zoom in on potentially eligible children.

Combining these estimates, their results suggest an infinite MVPF with a fairly wide 95% confidence interval of [0.26, \infty], so that one cannot reject an MVPF of 1, despite the point estimate suggesting the policy pays for itself.