4.2. Sensitivity tests
The findings in this study are robust to alternative specifications and control variables. Specifically, results are insensitive to estimating equation (1) in log-linear expression. Conclusions are qualitatively unaltered when the regression variables expressed in levels are scaled by population. The TAX coefficients remain negative and significantly less than zero. As expected, the adjusted R2 in the scaled models are smaller, ranging from 18 percent to 42 percent. Results also hold when equation (1) is estimated using rank regressions.
In addition, because the theory is not rich enough to identify the ideal set of control variables, a battery of estimations was conducted using alternative control variables. The results from using these controls are not reported because when they were included with the set of control variables in equation (1), their estimated regression coefficients are not significantly different from zero and their inclusion does not affect the inferences drawn from the TAX coefficients. The unreported control variables are: (a) crime rates for burglary, robbery and larceny as reported in the Uniform Crime Reports 1993-1995; (b) the 1994 cost of living index in the CQ State Fact Finder; (c) personal income per capita per the Statistic Abstract of the United States; (d) number of new business starts in the state per the U.S. Bureau of the Census’ State and Metropolitan Area Data Book, (e) a Conning & Company index of state regulatory restrictions; and (f) a categorical variable indicating whether the state must approve changes in non-automobile property-casualty insurance rates (i.e., a rate regulated state), as reported by the National Association of Insurance Commissioners in a 1992 chart entitled, “Rate filing methods for property/casualty insurance, workers’ compensation, title.”
The failure to detect a relation between coverage and regulation is surprising, considering the extant regulatory literature. Harrington (1984) concludes that the evidence generally is consistent with insurance rates decreasing in the stringency of state regulatory oversight, implying that coverage should be greater in rate-regulated states. Consistent with Harrington (1984), Petroni and Shackelford (1999) find that premium-loss ratios are decreasing in regulation. In addition, Vines (1996) reports that customers in rate-regulated states bore less of the industry’s Federal income tax increase following the Tax Reform Act of 1986 than customers in other states, consistent with regulation introducing supply inelasticities. One reason this study may fail to detect regulatory effects is that its power (i.e., 50 states) is insufficient, given the imprecision in calibrating regulation and the aggregation of differentially regulated lines of business in a single regression model.
4.2. Sensitivity tests