One of the problems in the tort reform debate is the use of simplistic statistics. The identification of particular problems in assessing the existence of the need for tort reform or the effect of tort reform has generated a number of studies. Many of them conflicting. This is because they tend to be advocacy briefs rather then disinterested research. Robert Hunter, writing under the aegis of the Americans for Insurance Reform, is the latest into the foray. His report finds that doctors are being gouged by their insurers. He uses a pretty good data set of loss costs by state and he finds that the losses have not risen to reflect the crisis that is supposed to exist. Further, he finds that the growth in losses are not that much different in the last five years (so-called crisis years) from the previous five years. In fact the report claims that tort reform did not make a difference. While he does show statistics—he does no statistical tests. Not one.
Essentially what he does is look at states with a set of certain tort reform and states with fewer reforms. He then looks a the percentage loss cost growth and uses the eyeball (it looks close to me) test to say there are no differences. He also says that during the crisis years that loss costs grew less in states with relatively few reforms and more in states with the most reforms. Now this is pretty interesting as one would expect the reverse at first glance. However, one might actually expect states under more pressure (higher increases in loss costs, higher premiums, etc.) are more likely to put in place new reforms. States without loss cost growth are not likely to put additional reforms in place. This failure to control for causation or at least control for the fact that the choice of reform is an endogenous choice is quite important.
I looked at the NAIC’s page 14 (state page data) and did some simple statistical tests for the years 1994-2004. (I looked at firms that wrote more than $200,000 in premiums in any given state and year. This excludes companies that are not really in the medmal business. I also didn’t adjust for inflation as I was just doing this to avoid grading. I took the damage cap and non-economic damage cap information from the Appendix in Mr. Hunter's study.)
First, in Table 1 (click on table to enlarge) I looked at the med loss ratio by state and regressed it against whether the state had a punitive damage cap limitation or a non-economic damage cap limitation. (I used a state and year fixed effect model) and found that there appears to be of no statistical influence of the damage caps on the mean of the loss ratio. This result is not uncommon and it seems to lend support to the group complaining that tort reform is a sham. However, note that the standard deviation of the loss ratio is significant and positive. This implies that if in the previous year the loss ratio was more volatile, one saw a higher mean loss ratio this year. if we think about insurance pricing, price =expected losses + expense+ cost of risk. As cost of risk increase, prices go up. One can think of the standard deviation of the loss ratio as a proxy for the cost of risk. As the loss ratio becomes more “uncertain”, a prudent insurer will have to hold more capital to support the risk.
The question then becomes, does the cost of risk depend upon tort reforms? In Table 2, I estimate a similar regression looking at the standard deviation of the loss ratio as the dependent variable against the dummy variables for the damage caps and the lag of last year's loss ratio. Note that the punitive damage cap seems to have a significant negative effect on the standard deviation —thus reducing the cost of risk.
The only point I want to make is that the the tort reform story is more complicated than some simple averages might lead one to believe. My regression models are also simplistic and if I had more time I'd do a better job before running off the the New York Times to make an authoritative claim. The standards for analysis in this debate are too low. Every group has their favorite whipping boy and their statistics to back it up. However, just looking at point estimates and saying this estimate is bigger than another to make a conclusion ... is so 1930s. I don’t think these reports would even get an average grade as an undergraduate term paper. Shouldn't we expect more?