By Ted Frank
Dear Professors Black, Silver, Hyman, and Sage,
I was surprised at the counterintuitive preliminary results in your study covered in Reuters for which Professor Hyman was quoted. I read your paper. It seemed that 1990 figures were unusually high, so I downloaded and took a quick look at the 1990 data from the TDI website. Your paper mentions the 1988 and 1989 data collection problems, and it appears to my eye that these problems were resolved by filing closed claim forms in 1990 for a number of settlements that were actually made in 1989. See, e.g., the $14.7 million of settlements and defense costs in TDI #7800282, #7800283, #7900170, and #7900318. The payments were made in 1989, but the claims "closed" in May and June 1990. TDI #9700276 settled in August 1989, but wasn't closed until October 1990, adding $7.9 million in settlement and defense costs to the latter year. If I took the time to construct a spreadsheet macro, rather than just scrolling through, I'm sure I could find additional and similar errors. In contrast, I used 1998 as a control, and a spot-check of the "closed claim" dates for larger settlements seemed to take less time to report, and much more likely to be in the same year.
Did your paper sort data by the Q1G "closed claim" field, or by the Q1E "Date of settlement" field? Because it would appear that there is a strong possibility that the 1990 "closed claim" data, sorted by Q1G, is not directly comparable to that of later years, where data collection was more consistent. From the tables and graphs in your paper, it would appear that a 1991-2002 dataset would result in substantially different conclusions than a 1990-2002 dataset.
In addition, it seems to me that, given the variance involved in medical malpractice verdicts, there are problems of small sample size even in a state as large as Texas. An outlier of two $14 million settlements (as happened in 1990 TDI #7700251 and #8500332) can throw off the entire study. Was there a reason that a three-year moving average wasn't considered?
Finally, how much do malpractice rates increase when subjected to the same deflationary factors that your paper applied to malpractice expenses? I didn't see a chart on that subject, though the former was repeatedly described as a "spike." It seems unfair to compare the real rates of malpractice insurance expenses to the nominal malpractice insurance rates.
In a polite e-mail to me, Professor Silver has promised a reply.
With the 1990 data excluded, there's an increase of approximately 40% in real total cost per capita between 1991 to 2000 -- and that figure excludes the increase in costs of defending against claims found to be meritless. The study acknowledges this omission, but explains it away by saying "defense costs for large paid claims are only 15% as large as payouts on these claims." But these are the defense costs for less than a seventh of the total claims! Even if the average defense costs for inexpensive/no-payment claims is a third as costly as the average defense costs for large paid claims, it means the study is underestimating total costs by about a quarter and underestimating the increase in costs by much more than that.
Black et al. do identify another possible contribution to the insurance price increase: in the mid-1990s, Texas regulators put pressure on malpractice insurers to reduce prices; the resulting undercapitalization required a catch-up price increases. Though, as Walter has pointed out elsewhere, what are insurers catching up to if not the costs of defending and paying malpractice claims? Too, if undercapitalization forces insurers to charge too much, it doesn't explain why there isn't market entry from adequately capitalized insurers who don't need to overcharge to be properly capitalized. And as I've mentioned elsewhere, of course, there's no reason that those insurance companies can't be formed by the moneyed interests who claim that the insurance crisis is the result of insurance mismangement. One would think academic economists, of all people, when faced with analysis implying that economic actors are leaving profits on the table, would question the analysis more rigorously before reaching that result, or at least construct a model explaining the discrepancy.
Some of the factors the professors use to deflate insurance costs--"Texas economic growth", for example--are utterly irrelevant
, and could only be chosen to help manufacture a conclusion.
The press coverage is the expected level of shoddy. AP cites a number for "median" jury verdicts that (1) fails to identify that the figure is in 1988 dollars and (2) is two-thirds lower than the more relevant and unreported mean. The result is misleading by a factor of five. The Washington Post's Ceci Connolly concludes that the data reflect "amazing stability in the tort system"--an editorial comment that, due to the violently fluctuating graphs in the paper, is not even justified before one accounts for the potential flaws discussed above.
Update, April 1. See my discussion of the March 31 AEI conference on the subject.
At the conference, Professor Black took issue with me, protesting that I publicly accused him of manipulating the end-dates of the data. I don't believe my statements above can be fairly taken that way, but I will clarify: I am stating only that the data set is sufficiently ambiguous that it can be manipulated so that choosing different cut-off points creates different results, not that the authors deliberately chose their dates to do so. Professor Martin Grace made the same point. The authors report that their results are sufficiently robust that omitting the 1990 closed claims data entirely doesn't change the conclusion.
The latest version of the paper (not yet on SSRN as of March 29) includes real-dollar figures for insurance premium increases. It is also improved in that it includes citations to other studies about the magnitude of defense costs for claims without payment. The authors rationalize their use of Texas GSP as a deflator to show the magnitude of medical malpractice cost within the Texas economy as a whole. Update, April 2: I disagree that this is an effective means to do that. But my earlier statement that the GSP deflator "could only be chosen to help manufacture a conclusion" was based on the misunderstanding that the authors were claiming that the GSP deflator had applicability to questions about insurance rates. Because they aren't, the clause is unfair, and I've deleted it.