North Dakota and Wyoming are excluded from the analysis because they prohibit selfinsurance, and Texas is excluded because 1993 was the first year that it permitted self-insurance. For the 47 remaining states, the percentage of workers’ compensation not covered through selfinsurance ranges from 54 percent to 92 percent with a mean (median) of 77 (79) percent and a standard deviation of 9 percent. A categorical variable (NOPRIVATE) is added to the explanatory variables to identify the four states (Nevada, Ohio, Washington, and West Virginia) that restrict coverage to self-insurance or state funds.

Table 6 presents the results from testing workers’ compensation. The first column regresses the natural logarithm of incurred losses (as in the earlier tests) to establish that the workers’ compensation relation between insured losses and insurer taxes is similar to the relation found when all forms of property-casualty insurance are evaluated. Catastrophic losses are excluded from the explanatory variables because they account for few workers’ compensation claims. Consistent with the earlier findings, when the dependent variable is restricted to workers’ compensation insured losses, the coefficient on TAX is negative and significantly less than zero (t-statistic of -2.32).

The second column reports the results from the sensitivity test that employs the percentage of insured workers’ compensation payments. As predicted, the TAX coefficient is negative and significantly less than zero (t-statistic of -2.67). Unlike the earlier tests, the dependent variable in this test can be expressed as a percentage because total payments, including self-insured losses, are known. However, results hold when the dependent variable is expressed as the natural logarithm of total insured workers’ compensation payments (t-statistic of -1.92).

This analysis of the unique workers’ compensation data provides additional evidence that self-insurance is increasing in the state taxes levied on the property-casualty insurance industry. In particular, this result shows that the conclusions drawn in this paper are insensitive to whether the dependent variable is measured on an accrual basis, such as insured losses or premiums, or a cash basis, such as payments.

**Table 6 – OLS regression coefficient estimates (standard errors) [t-statistics] Dependent variables: natural logarithm of workers compensation losses incurred in 1993 for all insurance companies and workers’ compensation benefits paid by insurers as a percentage of all workers’ compensation benefits paid in 1993.**

In (LOSS) | % paid by insurers | ||

14.73 | 0.86 | ||

Intercept | (1.67) | (0.33) | |

[8.80] | [2.57] | ||

–0.43 | –0.10 | ||

In (TAX) | (-) | (0.19) | (0.04) |

[-2.32] | [-2.67] | ||

0.81 | –0.05 | ||

In (POP) | (0.08) | (0.01) | |

[10.35] | [-3.17] | ||

0.97 | 0.04 | ||

In (WEALTH) | (0.36) | (0.07) | |

[2.70] | [0.50] | ||

–3.26 | 0.05 | ||

NOPRIVATE | (0.20) | (0.04) | |

[-16.35] | [1.31] | ||

n | 44 | 45 | |

Adj. R^{2} |
0.92 | 0.17 |