Controlling for spending divided by income, there are only small
residual time-period effects on coverage levels (Exhibit 5). The
black line in Exhibit 5 shows the estimated effects of year dummy
variables from a logistic regression in which the probability of being
insured is modeled as a function of each respondent’s expenditureto-
income level.13 The estimated year effects—which serve as a proxy
for the effect of all factors other than the expenditure-to-income
variable included in the model—cluster around zero, indicating that
the eight-percentage-point rise in the percentage uninsured during
1979–1995 can be accounted for almost completely by changes in the
relative growth rates of per capita health spending and income.