The claim that Medicaid is better coverage than private insurance was all the rage a couple of weeks ago thanks to a press release from Southern Methodist University highlighting doctoral candidate and adjunct professor Manan Roy’s paper “How Well Does the U.S. Government Provide Health Insurance.”
The claim that Medicaid is better coverage than private insurance was all the rage a couple of weeks ago thanks to a press release from Southern Methodist University highlighting doctoral candidate and adjunct professor Manan Roy’s paper “How Well Does the U.S. Government Provide Health Insurance.”
Before you rush out to enroll yourself and your family in the cheaper, better, Medicaid system, it might be worth taking some time to evaluate whether the slender evidence in the paper supports such a sweeping conclusion.The paper compares Medicaid with private insurance. The comparison is based solely on data from birth certificates recorded at the time of birth. The measures of infant health include length, weight, weeks of gestation, and 5 minute Apgar score. Though these are numeric variables, the author turns them into dummy variables, reducing their already limited variation. Insurance coverage is based on individual recall 9 months after a child’s birth. Information on coverage before birth is not included.
Apgar scores are measured from 1 to 10. Infants with scores of 7 to 10 are considered clinically normal. The overwhelming majority of infants are clinically normal at birth. In a 2001 study of 151,891 births at Parkland Hospital in Dallas, Casey et al. reported that 131,581 full-term singleton live births had a 7-10 Apgar score, 561 had a 4-6 Apgar score, and 86 had a 0-3 Apgar score. The mean 5-minute Apgar score was 6.6±2.1 in infants born at 26 to 27 weeks of gestation, and 8.7±0.8 in infants born at 34 to 36 weeks.
The data used in Roy’s study are from the Early Childhood Longitudinal Survey, Birth Cohort (ECLS-B). The ECLS-B excluded children who died, thus ignoring an important health outcome. It also had only a 74.1 percent response rate. The existing ECLS-B sample has a mean Apgar score of 8.942 with a standard deviation of 0.682. If the Apgar scores were normally distributed, this would mean that more than 95 percent of all the children in the sample had Apgar’s suggesting clinical normality at birth. In fact, 98.6 percent of the children in the sample were normal at birth. The difference between the Apgar scores of the Medicaid sample and the private sample was -0.096.
It has long been known that people enrolled in Medicaid differ from those with private health insurance in important ways that affect health, and that those differences may not be captured by available data. The goal in this paper was to calculate how large the bias caused by unobservable variables would have had to have been in order to attribute the entire observed performance difference to selection bias. Pioneered by Altonji, the approach makes several assumptions. One is that the observed elements are chosen at random from the full set of factors that determine the outcome for the dependent variable. Another is that none of the included or omitted independent variables dominate the dependent variable.
Unfortunately, the data set includes no measure of maternal health, a variable that is likely to dominate outcomes, at least to the extent that variables with so little variability can be dominated. The observed independent variables are the typical grab bag of variables that show up in educational surveys–child’s gender, mother’s age, weight, and education, father’s age and education if available, the household’s socioeconomic quintile, parents’ marital status, race, geographic region, and urban or non-urban location.
Given that there is so little Apgar variation to begin with, it is not surprising that the author calculates that even a modest amount of selection on unobservables would erase the negative Apgar results for Medicaid. Slightly higher selection on the basis of unobservables would lead one to conclude that Medicaid has better outcomes if one assumes, as the author does, that the people covered by Medicaid are likely to have poorer birth outcomes than those covered by private insurance.
The problem is that we know little or nothing about how the distribution of the risk of poor live birth outcomes varies between the Medicaid and privately insured populations. Estimates of the number of births covered by Medicaid run as high as 40 percent, almost half of all births. State data suggest that mothers covered by Medicaid are likely to be younger, in the sample the average was 2.5 years younger than those who were privately insured, but this is not surprising given Medicaid means testing. Younger women tend to have higher birth weight children than older ones, unless they are very young, aged 17 or less, though this is subject to debate. Mothers insured by Medicaid are more likely to smoke, and smoking is associated with pre-term births and depressed Apgar scores, but older women are more likely to develop diabetes and gestational diabetes which increases risk. Whether socio-economic status affects Apgar scores is subject to debate.
Despite all of the zones of ignorance, the author asserts that “children on public HI [health insurance] appear to fare no worse, and possibly even better than their counterparts on private HI…” She argues that government-provided health insurance outperforms the private sector because CHIP “provides an alternative source of cheaper coverage coupled with a broader range of benefits than private HI.” And she writes that MEPS data show that “the average payments made by Medicaid (and/or CHIP) for medical services per enrollee are smaller than for those by private HI. Since payments constitute the bulk of the costs incurred by the health insurance provider, this simply corroborates the aforementioned evidence of public HI being a cheaper source of more benefits for infants.”
This is true only if one believes that the total cost of Medicaid is reflected by the payments it makes to providers. This is unlikely because MEPS explicitly excludes payments that are not directly linked to individual patients, payments such as grants for public and community health clinics, Medicaid disproportionate share payments, the deadweight loss from taxpayer financing. It also fails to account for the total costs of state and Congressional management and overhead.
