**3.3. TAX**

The ideal tax measure would capture the relevant tax rate used by property-casualty insurers when they set insurance rates. Unfortunately, the data do not permit such fine calibrations. Instead we use a tax measure similar to the one in Petroni and Shackelford (1995, 1999). TAX is the sum of the total taxes collected from all insurers, both property-casualty and life and health, for the three years immediately preceding the investigated year as a percentage of the total property-casualty and life and health premiums written in each state for same three preceding years.

TAX is computed using data from the three preceding years to mitigate concerns about possible endogenous relations between current taxes and other regression variables. However, inferences are qualitatively unaltered when TAX is the current year’s taxes divided by the current year’s premiums, which is the measure used in Petroni and Shackelford (1995, 1999). On a separate note, TAX is a flawed measure to the extent life and health insurers are taxed differently from property-casualty insurers. Petroni and Shackelford (1995). however, contend that this effect should be immaterial because states tax the two sectors similarly.

**3.4. Control Variables**

Control variables are included in the regression to mitigate the potential that omitted correlated variables may lead to erroneous inferences about the effects of state taxes on selfinsurance. POP, the state’s total population, and WEALTH, gross state product per capita, are designed to capture cross-state differences in the demand for insurance. CAT, the state’s catastrophic losses, is designed to capture cross-state differences in states of nature. Insured losses are expected to be increasing in each control variable.

**3.5. Data**

The insured losses and premiums of each U.S. property-casualty insurer are collected for each year from the NAIC Property/Casualty Annual Statement Database and summed by state. Each state’s total taxes collected annually from insurance companies are gathered from the NAIC Insurance Department’s Resources Report. The population and gross state product of each state are collected from annual U.S. Statistical Abstracts.

Estimated catastrophic losses for each year are collected from The Fact Book, an annual publication by the Insurance Information Institute. A catastrophe is defined as an occurrence that causes more than $5 million in insured property damage. Catastrophic loss estimates are made by the Property Claim Services division of American Insurance Services Group Inc. using various sources of data, including claims adjusters and state and local officials. For those catastrophes that occurred in more than one state, estimated losses are allocated across the affected states based on their relative populations.

**3.6. Descriptive Statistics**

Table 1 presents descriptive statistics for the regression variables in 1993. Insured losses are segregated into automobile liability losses, automobile physical damage losses, and other property-casualty insured losses because each category is tested separately in the regression analysis below. Combining the three groups, the mean insured losses are $3.3 billion. The mean and median values for TAX are 1.6 percent, ranging from 0.6 percent (Wisconsin) to 2.7 percent (Mississippi). In 1993 the mean (median) state suffered $115 ($47) million in catastrophic damages or 3 percent of total insured losses. Descriptive statistics for 1994 and 1995 are similar to those reported in Table 1, except that 1994 had a major catastrophe, the Northridge, California earthquake, and 1995 was relatively unscathed by catastrophes with the median state registering none.