Although the dependent variables are binary, the large number of observations in
our sample ensures that regression via ordinary least squares is consistent.11 To account
for unobserved HMO-level effects (see Moulton (1990)), the observations are clustered
by HMO and location.12
Findings
Table 4-5 presents the regression estimates. The first two columns in Table 4-5
contain the estimates for the well-child visit measure for the two age cohorts. The last
two columns contain the estimates for the asthma medication measure for the two age
cohorts.
The first row of data in Table 4-5 presents the coefficient estimates for the variable
that indicates whether the HMO pays all of its doctors via a FFS arrangement. For both
age cohorts, the well-child visit success rate for enrollees in HMOs that pay all of their
doctors via FFS is six percentage points higher than for those enrolled in HMOs that pay
some of their doctors via capitation. Given the mean success rates, this difference implies
that the average probability that an enrollee receives a well-child visit is 10-20% higher
in an HMO that pays all of its doctors via FFS.
The opposite conclusion arises with regard to the asthma medication measure. For
both age cohorts, the success rate is lower for HMOs that pay all of their doctors via FFS.
The effect is statistically significant for the 5-9 year old cohort. The estimates imply that
1 As a specification test, the model was also estimated via probit. The results are largely
unchanged and are reported in Appendix Table B-1.
12 By clustering the observations, the estimates of the coefficient standard errors are adjusted to allow for
the possibility that the observations within each group are not independent. The enrollees are grouped
here by the HMO in which they are enrolled and the metropolitan area in which they reside.