In this fascinating paper, Barcellos and Jacobson examined the effect of turning 65 and gaining access to Medicare on use of health care services, out-of-pocket expenditures and measures of financial stress related to medical bills. The key to the analysis is the regression discontinuity research design.

A potential problem with this approach that is not discussed is the "look back" period to measure use of services and expenditures in MEPS. For those who turn 65, the period of measurement is between the ages of 64 and 65--not a full year at age 65 with access to Medicare. So those recorded as age 65 were age 64 for approximately half the time of measurement.

The "donut" RD in Appendix table 4 partly addresses this issue, but as can be seen in this and other tables, estimates are imprecise and small changes in specification result in some non-trivial change sin estimates. Consider the estimate of the effect of turning 65 on total medical spending. In the case where persons between ages 55 and 75 are used the estimate is -$2404. When those aged 50 to 80 are used but dropping those age 65, the estimate is -$1942–a 25% difference from a relatively small change. The analogous baseline estimate (in Table 3) is -$2168.

A second issue not addressed is the plausibility of the magnitudes of effect sizes and differences by race/ethnicity. Appendix Table 4 provides the relevant estimates. For non-Hispanic Black and Hispanic samples, almost all the RD estimates (10 out of 12) related to physician visits and inpatient care are positive and large, particularly when measured relative to number of people who gained insurance. Non-Hispanic Black and Hispanic persons were the most treated too. Therefore, the absence of a significant positive effect of turning 65 on physician visit and inpatient visit in Table 3 is likely due to the pooling of race/ethnicity samples. This raises an additional issue whether the decline in total spending observed in table 3 is a true effect.

Also, consider the fact that it well known that gaining insurance coverage (from uninsured) is associated with a 30% to 50% increase in utilization. Approximately 10% of the sample gained insurance coverage so utilization and total spending for the entire sample would go up by approximately 3% through this channel. However, in Table 3, spending went down by 34%. That implies that among the always insured group spending went down by a whopping 38%. This large negative effect seems implausible, particularly when many of the always insured keep their employer benefits and/or buy supplemental insurance (see Table 6 for evidence of significant increase in supplemental coverage)?

Table 6 also presents counter intuitive results. Out-of-pocket spending went down the most for the highest educated group who were least likely to gain insurance at age 65 (most likely to be insured) and most likely to have supplemental insurance that pays out-of-pocket expenses. Total spending also goes down most for this group.