William Greene 5330027886678
Frequency8864Histogram for Variable DocVis2 4 6 8 10 12 14 16 18 20
DocVis22 24 26 28 30 32 34 36 38 40Figure 11.1 Histogram of count variable DocVis
count variables. One of the issues considered in the study was whether the data
contained evidence of moral hazard, that is, whether health care utilization as
measured by these two outcomes was influenced by the subscription to health
insurance. The data contain indicators of two levels of insurance coverage:Public,
which is the main source of insurance, andAddon, which is a secondary optional
insurance. In the sample of 27,326 observations (family/years), 24,203 individuals
held the public insurance. (There is quite a lot of within group variation in this,
as individuals did not routinely obtain the insurance for all periods). Of these
24,203, 23,689 had only public insurance and 514 had both types. (One could not
have only the addon insurance.) To explore the issue, we have analyzed the DocVis
variable with the count data models described above. Figure 11.1 shows a histogram
for this count variable. (There is a very long tail of extreme observations in these
data, extending up to 121. The histogram omits the 91 observations with DocVis
greater than 40. All observations are included in the sample used to estimate the
models.) The exogenous variables in our model are:
xit=(1,Age, Education, Income, Kids, Public).(Variables are described in Table 11.2. Those listed are a small subset of those used
in the original study, chosen here only for a convenient example.)
Table 11.7 presents the estimates of several count models. In all specifications,
the coefficient onPublicis positive, large, and highly statistically significant, which
is consistent with the results in Riphahn, Wambach and Million (2003). The large
spike at zero in the histogram casts some doubt on the Poisson specification. As
a first step in extending the model, we estimated an alternative model that has a