We estimated three logistic regressions to quantify the extent to
which changes in the expenditure-to-income ratio account for
changes in the percentage uninsured. The dependent variable in
each model is a 0–1 indicator of whether the respondent has health
insurance. In the first model the independent variables are dummy
variables for each year, so that we could estimate the association
between year or time period and the percentage uninsured. The year
indicators serve as a proxy for all variables that might have affected
the percentage uninsured. In the second model we supplemented
the year indicators with the measure of per capita expenditures
divided by each individual’s personal income. In this model the parameter
estimates for the year indicators are a proxy for all variables
that might have affected the percentage uninsured other than the
expenditure-to-income ratio that is included in the model.