WPC& 2BJZ Courier#|T)x  @87X@HP LaserJet IIIHPLASIII.PRSx  @,\,A]X@26DVT ZCourier 10cpi#|TCG Times (Scalable)CG Times Italic (Scalable)CG Times Bold (Scalable)Courier 10cpiHP LaserJet IIIHPLASIII.PRSXv P7,\,A]XP2{ 6i04-12-96 03:42p sloan paper 95-07  *X01Í ÍX0Í Í"Sh ^;C]ddCCCdCCCCddddddddddCCȲY~~wCN~sk~CCCddCYdYdYCdd88d8ddddJN8ddddYYdYdHddddddCCCCYddddddddd8YYYYYY~Y~Y~Y~YC8C8C8C8ddddddddddYddddsddddddddd~d~d~d~ddddddddd8ddddHdoddd~d~d~d~8~8dvdddddddkdkNkdkd~d~dddddddddYCdddCC/NdddCYQQddddddFddddFCdhhd44ddzzdddwoodChF"Ȑdhd岲ddCCȐzȲCddodȐȅdCdYdsȐ`ȐȐȮzȐUwŐdȐYYCCCCŐz~ozoY~NYYYC8YooYdYzsdzdd~YYzozzz~CdzYzzzzCCdddddddzCzdYCd"Sh ^18MSS888S8888SSSSSSSSSS88Jxir{icx{8Aui{x`xoZi{xxxl888SS8JSJSJ8SS..S.SSSS>A.SSxSSJJSJS>q*"xxxxWWxxxWWkkxxx6uC;,/3Xu&_ x$&7XX>z-b81,b&_ x$&7X7zC;,/sXz_ p^7X2xxx,Hx  @87X@ 0.0316LLL0.608a!!! 444  0.068)(0.164)DDD(0.0169)(0.213) Income missing  #b P7P# < < < < < < & Tort Liability and Obstetricians' Care Levels< 5Frank A. Sloan* 3 Stephen S. Entman** 3EBridget A. Reilly** 4iCheryl A. Glass** 3Gerald B. Hickson** 3Harold H. Zhang***< This research was supported in part by Grant No. HS06499, "Birth Outcomes, Satisfaction with Care and Malpractice," from the U.S. Agency for Health Care Policy and Research. The Survey of Obstetrical Care and hospital chart abstraction, which was used in our empirical analysis, was supported by the same grant. We thank Betty McCall, Christoph Schenzler, and Beth Kulas for research assistance and Joni Hersch, Kathryn WhettenGoldstein, and anonymous reviewers for helpful comments. * Duke University ** Vanderbilt University ***Carnegie Mellon University!0*0*0*     @@5I. INTRODUCTION<In markets in which consumers misperceive risk, first party liability will result in nonoptimal output levels and nonoptimal levels of care per unit of output (Spence, 1977). A major goal of tort liability is to achieve an optimal level of injury prevention not achieved by imposing first party liability. Optimal prevention is obtained at a care level that minimizes the total cost associated with injuries. Total cost is the sum of the resource cost and utility loss attributable to injuries, prevention cost, and administrative cost of compensating injury victims (Calabresi, 1971). Many critics of tort liability as it exists in the United States maintain that such liability results in care  xP above the social optimum. That is, such liability overdeters.  Y ԍSee, e.g., Sugarman's (1985) review of tort liability and Huber's (1988) book on product liability. Calfee and Craswell (1984), Craswell and Calfee (1986), and Diamond (1974) showed that uncertainty about required levels of care resulted in potential injurers taking excess care. More recently, Kahan (1989) showed that uncertainty about the level of due care may result in injurers taking less than optimal care. Critics of tort liability as it relates to health care often argue that the threat of medical malpractice suits results in "defensive medicine." Although the term is only rarely used in relation to a socially optimal care level, the implication is that defensive medicine means  xP0 overdeterrence.x0 Y ԍAlthough overcompensation has been frequently asserted, there is no empirical evidence that medical malpractice claimants are overcompensated. If anything, the evidence points to undercompensation. See Sloan and van Wert (1991). Other reasons to expect under rather than overdeterrence is evidence that many potential meritorious claims are never brought  Y (California Medical Association, 1977; Weiler et al., 1993) and the lack of experience rating of medical malpractice premiums. There is a small body of literature that assesses the effects of tort liability on care levels of physicians  xPP (Reynolds, Rizzo, and Gonzalez, 1987; Rubin and Mendelson, 1993).5aP]  Y=" ԍIn the most widely cited study, Reynolds et al. (1987) used two approaches. In one, the authors queried physicians about specific changes in their practices that the respondents attributed to professional liability risk. The changes encompassed record keeping, tests or treatments, time spent with patients, and followup visits. Adding the projected cost of these changes to mean professional liability premiums, they estimated that the total cost of practice changes made to reduce the risk of medical malpractice claims premiums amounted to 14.1 percent of practice gross revenue in 1984. The authors also gauged utilization responses to'0*(( medical malpractice risk by regressing measures of quantity of service provided on medical malpractice premiums, their measure of physician malpractice risk, and on other factors. Cost calculations based on the regression results were similar to those based on the accounting approach. This convergence of findings gave the authors confidence in their results.5   Although inferences aboutP0*(( overdeterrence and its cost have often been made from these studies, they, like ours, are really about changes in physicians' care levels in response to changes in professional liability risk. Surveys of physicians' practices and the data used in past studies do not contain good patient health measures, and the dependent variables, such as the number of visits or tests performed, are highly aggregated. Such aggregate measures of responses to the threat imposed by tort liability are deficient in that they do not reflect the clinical decisionmaking process. When confronted with an increased probability of being sued, a physician does not simply order more diagnostic and therapeutic procedures. Rather s/he will probably respond in specific ways to specific clinical situations. It is therefore fruitful to study effects of changes in tort liability risk on practice changes made in particular circumstances that physicians frequently encounter. In our study, we assess practice changes made in response to the threat of tort liability in the field of obstetrics, which has one of the highest levels of premiums, claim frequency, and mean dollar value of paid  xP0 claims.w0 Y ԍSee American Medical Association (1992), and U.S. General Accounting Office (1987).w There is much "conventional wisdom" about effects of tort liability risk on obstetrical practice based  xP on obstetricians' perceptions of changes that have occurred.  Y ԍFor example, the Institute of Medicine (1989) undertook an informal letter survey of 132 heads of obstetrics departments. The most commonly reported response to the increased risk of medical malpractice litigation was increased frequency of cesarean sections, including liberalized criteria for cesarean sections, decreased frequency (or even elimination of) vaginal breech deliveries, and delivery of all twins by cesarean section. Other responses included: avoidance of midforcepts deliveries and decreased use of outlet forceps deliveries; increased use (or even universal use) of electronic fetal monitoring; increased antenatal testingultrasounds, alphafeto protein, and amniocentesis; increased referrals to tertiary centers for ultrasound examinations; increased documentation; and increased use of consultation. See Institute of Medicine (1989, p. 84) for the complete list of obstetricians' responses obtained from the survey.0*((ԌUsing data from Florida, our study evaluates the effect of the threat of tort liability on many of these utilization measures. In Section II, we present a conceptual framework for assessing the effect of exogenous changes, such as the threat of medical malpractice litigation, demand shifts, and factor price changes on the physician's care level. Section III describes our data. In Section IV, we discuss how we measured the threat of tort liability facing individual physicians. Section V provides our equation specification for assessing the extent of practice changes made in response to changes in professional liability risk. This section is followed by results in Section VI. Section VII discusses implications of our findings. % ԋII. LIABILITY RISK AND PHYSICIANS' CARE LEVELS<Determining the Physician's Care Level Assume that the market for physicians' services is competitive but price (p) depends positively on quality (Y). Assume also that patients' marginal valuation of quality is affected by such exogenous variables Z as their health, health insurance coverage, and income levels. Then the individual physician's demand curve is: ` ` p=p(Y;Z).*hh29@ppGN  U(1) Let the physician's total practice cost (C) depend on quantity (X), quality (Y), and exogenous factors W, such as input prices. ` ` C=C(X,Y;W)*hh29@ppGN  U(2) The total number of claims per time period that the physician anticipates (s) depends on the physician's  xP own quantity, quality, and exogenous determinants (L).1 Y ԍUnder a negligence rule, the injurer can avoid paying accident cost by setting care at the socially optimal care level, which is the standard set by the courts (see, e.g., Shavell, 1987). If courts always set the standard correctly, potential injurers could just set their care levels at the standard and thereby incur no accident cost. Care above the standard would be wasteful, both from the vantage points of society and the potential tortfeasor. Then there would be no demand for liability insurance or defendant losses at verdict, which is clearly counterfactual. If courts make errors in setting the due care standard, there is a market for liability insurance, and care may no longer be set at the socially optimal level (Cooter and Ulen, 1986; Danzon, 1985a). Further, it is then appropriate to view the relationship between careP'0*(( and potential losses imposed on tortfeasors as continuous. In the context of medical care, it seems likely that there would be some investment in accident prevention even if no tort liability were imposed on physicians, since patients probably place some value on prevention.K0*((Ԍ` ` s=s(Q,Y;L)*hh29@ppGN  U(3) We assume that the cost per claim that the physician bears is fixed at A. Included in A are the time, reputation, and psychic costs borne by the physician when s/he is sued, indemnity payments not covered by  xP insuranceK Y ԍVirtually all physicians have medical malpractice insurance. Such insurance is typically obtained on a group basis; thus, physicians do not typically select the amount of coverage. (See Danzon [1985b], chap. 7.) For these reasons and because our focus is on determining responses in quantity and quality to various exogenous changes, medical malpractice insurance can be considered to be exogenous., and any effect that adverse experience has on the premium the individual physician pays. Large claims should affect the reputation and psychic costs because such claims are much more likely to be publicized. The physician is assumed to be riskneutral; s/he maximizes profit by setting Q and Y: ` ` !=p(Y;Z)Q C(Q,Y;W) s(Q,Y;L)A.@ppGN(4) Using comparative statics analysis, we evaluated the effects of a positive shift in demand (Z), a factor price increase (W), an increase in claim frequency from an exogenous change in claim threat (L), and an increase in cost per claim (A) as anticipated by the physician.  xP Under a plausible assumption, a positive demand shift unambiguously raises quantity and quality.  Y ԍThe assumption is that increases in quality raise the marginal profitability of quantity  Y and, conversely, (!QY=pYĩCQYĩsQYA>0). Thus, we expect increased patient income and increased health insurance coverage to raise quality, as would clinical and demographic factors that put patients at higher risk of an adverse outcome. Increased wage rates should lower quality. Increases in L and A have ambiguous effects on quantity and quality. The source of the ambiguity is that we allow for the possibility that increased quantity raises claim frequency. If changes in quantity do not affect claim frequency, increases in L and A raise quality and, through their effects on demand, raise quantity: 0*((  xP as well.  YX ԍAlternatively, we could have defined claim frequency as the probability of a claim per unit of physician output. But this is just a special case of specification used here. Thus, in the context of our study, defining the number of obstetrical patients a physician treats as quantity and services per obstetrical patient treated as quality, one should expect that a greater threat of claims and higher awards lead to higher "quality"increased antenatal testing, more time spent with patients, and more "intensive" treatment at labor and delivery. 7>III. DATA <Mathematica Policy Research (MPR) conducted the Survey of Obstetrical Care in 1992 of 963 women who had given birth in 1987 in 31 counties in Florida for purposes of this study and related studies of medical malpractice and birth outcomes. Eightyseven percent of births/fetal deaths in Florida in 1987 occurred in these  xP counties.u b Y ԍThe excluded counties were in North Central Florida and in the Florida Panhandle.u The purpose of the survey was to obtain information about a respondent's pregnancy and delivery in 1987. The survey included questions about maternal risk factors, patterns of obstetrical care, choice of obstetrician, satisfaction with care received, birth outcomes at five years, and, conditional on an adverse outcome having occurred, whether a medical claim was filed against the obstetrician. Interviews were obtained by telephone; potential respondents who could not be contacted by telephone were interviewed in person. At the end of interview, respondents were asked for written permission to allow us to obtain the hospital charts for  xPp mother and child.i ap Y ԍThere were several reasons for conducting the interviews in Florida: availability of a public use file on all medical malpractice claims closed in the state since 1975 with identifiers of physician defendants; access to individual birth and death records in 1987 that identified the physician at the time the vital event occurred; and a populous state with many obstetricians, many malpractice claims, and appreciable variation in claim frequency within  Y" the state (Nye et al., 1988). Records with physician identities are not routinely released to the public.i Respondents were told that the survey dealt with satisfaction of care and other topics; medical malpractice was only mentioned when respondents were asked whether they had filed a medical malpractice claim.< 0*((Ԍ The survey contains an oversample of fetal and infant deaths, births of which the Apgar score at five minutes was less than six, and of obstetricians who had four or more claims filed against them arising from care they rendered during 197783. Patients of obstetricians who did not practice in Florida for at least three years during 197783 and/or did not practice in Florida in 1987 were excluded, as were patients who used nonobstetricians (such as family practitioners or nursemidwives) for their obstetrical care or initiated prenatal care in the third trimester of their 1987 pregnancy. With the one exception noted below, we used sample weights in our regression analysis. Before we began the survey, a pilot study of maternal recall was conducted at Vanderbilt University Hospital to ascertain whether women could accurately recall details of a pregnancy and labordelivery that  xP occurred five years before the survey. Recall was generally found to be quite accurate.=  Yh ԍSee Githens et al. (1993). Other articles based on these data are Entman et al. (1994),  YS Hickson et al. (1994), Hoerger and Howard (1995), Sloan et al. (1995), and Sloan and Hsieh  Y> (1995).#b P7P#= ' IV. MEASURING THE THREAT OF TORT LIABILITY<We used several alternative measures of the threat of tort liability. Each measure refers to claims or payments by medical malpractice insurers made on behalf of obstetricians arising from care rendered during 197783. Data on medical malpractice claims and payments came from a public use file of all closed medical malpractice claims in Florida maintained by the Florida Department of Insurance. The public use data identifies the defendant, amount paid on behalf of the defendant to claimants, administrative expenses allocated to the claim, as well as other pertinent information about the claim. The threat measures were: (1) number of claims per exposure year incurred by the individual obstetrician; (2) a set of binary variables classifying the individual obstetrician according to his/her claims experience; (3) number of claims per exposure year incurred by all obstetricians in the physician's county; and (4) total payments, indemnity payments plus administrative expenses, incurred per exposure year in the obstetrician's county. Several aspects of our definitions are noteworthy. First, we analyzed obstetrician behavior in 1987. By`"O 0*(( 1987, the obstetricians would have known about most of the claims arising from care rendered during 197783. Many of these claims had not been resolved by 1987, although obstetricians, their insurers, and their attorneys plausibly had formed expectations about likely outcomes. If we had selected an earlier period than 197783, the parties would have known more about the outcome. However, the legal environment prevailing in 1987 would have been less close to the period from which the claims/losses had been generated. Given the long claims tail  xP in obstetrics, use of a period closer to 1987 would have resulted in a serious undercount of claims and losses.Y  Y( ԍSee Sloan et al. (1989) for further discussion.Y Second, even if claims were measured perfectly, from the vantage point of decisionmakers, there is likely to be noise in the claims and loss data. Actual experience from the recent past may provide the best estimate of future experience. But since actual experience partly reflects unique circumstances, no decisionmaker plausibly expects history to precisely repeat itself. For this reason, in an alternative specification of the threat, we classified obstetricians into four categories according to the number of claims they had per exposure year from 197783: zero claim; low claim (>0 to 0.5); high claim (>0.5 to 0.74); and very high claim (>0.74). Given our oversampling of obstetricians with adverse claims' experience, there was appreciable variation among the respondents' obstetricians. Since the sample was selected in large part based on the claims' experience of the obstetrician, when we used this specification, we did not use the sample weights. Third, although claim size is likely to be exogenous to physicians, the physician's claim frequency may be endogenous. In our model, a physician's claim frequency (s) depends on the physician's care level (Y). Further, it is not clear that a physician's claim frequency from a prior period can be viewed as exogenous to the current care decision. We tested whether the threat was exogenous in specifications using the threat variable  xP based on the continuous individual obstetrician claims' experience measure.c{ Y! ԍThe test we used is equivalent to a Hausman specification test.c When the null hypothesis of exogeneity was rejected, we substituted a threat variable for obstetrician claims' frequency in the obstetrician's county. We again tested for exogeneity. In no case was the county measure endogenous. With the exception of one variable, the obstetrician's gender, all of the instrumental variables used for the exogeneity test were`",0*((  xP defined for the county in which the respondent's obstetrician practiced.x  YX ԍThe explanatory variables used in our exogeneity test were: by educational attainment, the percentage of women in the county who gave birth16+ years, 1315, with less than 12 years the omitted reference group; percentage of women with Medicaid the principal source of payment expected at discharge who gave birth; percentage of women with other government insurance or selfpay (no insurance) who gave birth, with privately insured women the omitted reference group; percentage of nonwhite population; percentage of population in poverty; per capita income; population density; lawyers per square mile; and obstetrician gender. Information on educational attainment of women who gave birth was computed from data on all births occurring in Florida in 1987. The payer mix variables were computed from unpublished Florida hospital discharge data for 1988, the first year discharge data became available for Florida hospitals. Data on the other county variables came from  Y[ Florida Statistical Abstracts. Information on the obstetrician's gender came from the American Medical Association.x Any available measure of the individual obstetrician's quality was logically a candidate for inclusion as an explanatory variable in the main equations. Since all but one of the instrumental variables were countyspecific, it made sense to use a threat measure defined at the county level when exogeneity of the individual claims' measure was rejected rather than an instrumental variables approach. /V. EQUATION SPECIFICATIONÐ<Measures of Effects of Tort Liability on Practice Behavior We assessed effects of the threat of medical malpractice litigation on the method of delivery, antenatal testing, and on various measures of maternal satisfaction with care received. Method of Delivery. We analyzed the choice between cesarean section and vaginal delivery. Forty percent of  xP the deliveries in our sample were by cesarean section.g  Y ԍThis is a high percentage compared to the national mean of 24%, and even for Florida overall28% in 1987. Several hospitals in South Florida had appreciably higher rates (Silver and Wolfe [1989]).  xP0 In addition to the tort liability variables, we included the following groups of explanatory variables:0 Y# ԍThere is a large literature on determinants of cesarean section rates. See, for example, Anderson and Lomas (1985), Danforth (1985), Gould, Davey, and Stafford (1989), de Regt  Y|% et al. (1986), Goyert, Bottoms, Treadwell, and Nehra (1989), Localio et al. (1993), Myers and Gleicher (1988), Sachs (1989), Shiono, McNellis, and Rhoads (1987), and Stafford (1990a, 1990b, 1991).0 A0*(( previous cesarean section; maternal clinical risk factors; family income and insurance; demographic characteristics of the mother; characteristics of the hospital at which the birth occurred; and characteristics of the obstetrician who delivered the infant. Maternal clinical risk factors included were (for target pregnancy unless otherwise indicated): firstborn child; previous stillbirth; diabetes during this or previous pregnancy; hypertension during this or previous pregnancy; placenta previa or abruption; toxemia; previous history of toxemia; premature rupture of membranes; birthweight less than 1,500 grams; multiple gestation; breech presentation; dystocia; and fetal distress (bradycardia, tachycardia, heartbeat slowing down too much after each contraction, meconium staining). Placenta previa or abruption, premature rupture of membranes, and fetal distress were only included in the regression for cesarean section versus vaginal delivery. Income was measured as the family's annual income in 1987. If income was missing, we set income to zero and included a separate variable identifying cases with income missing. We included binary variables for private conventional health insurance, HMO, Medicaid, and other health insurance coveragewith uninsured persons the omitted reference groupall defined as of the date of the target delivery.  xP Demographic variables of the mother were: binaries for age in 1987 of 19 or lessm Y8 ԍThere were too few observations to allow us to use a lower age threshold.m and 35 or more; nonwhite (almost all were black); Hispanic; married, including living with a partner; and binaries for high school dropout, high school graduate, some college, with college graduates the omitted reference group. We included variables describing the level of obstetrical care at the hospital at which the delivery  xP  occurred: service level II, service level III, with service level I omitted. y Y) ԍThe hospital data came from the 1987 American Hospital Association (AHA) Annual Hospital Survey. In a few cases, we did not have a hospital identifier for the place where the woman delivered, either because the delivery did not occur in a hospital or because we could not match the hospital listed on the vital statistics file with a hospital on the American Hospital Association's Annual Survey of Hospitals for 1987. Rather than lose these cases, we included a variable for hospital missing.`" 0*((ԌFinally, we included variables describing the obstetrician who delivered the baby: (1) board certification status; (2) year of graduation from medical school195573, 197377, 1978+, with before 1955 the omitted reference group; and (3) whether the obstetrician graduated from a medical school in a less industrialized country.  xP@ Antenatal Testing. We examined the effect of tort liability and other factors on the number of ultrasounds[_@ Y ԍUltrasounds are performed to detect unsuspected fetal disorders that may become clinically important, such as placenta previa, identification of the number of fetuses, documentation of presence/absence of fetal heart motion. In addition, ultrasounds are used for evaluation of the uterus, gauging the amount of amniotic fluid, gauging gestational age, and evaluation of fetal anatomic structures (see, e.g., Eden and Boehm [1990, pp. 24750]; D'Alton and DeCherney [1993]). Since the purposes differ, ultrasounds are often repeated during the course of the pregnancy.[  xP and whether the following tests were performed or not: amniocentesis; YO ԍAmniocentesis is used to diagnose chromosomal and Mendelian disorders and for detecting elevated levels of alphafeto protein (Eden and Boehm [1990, p. 285]). alphafeto protein;  Y ԍThe alphafeto protein test is designed to identify neural tube defects and for detecting Down's syndrome. For a description of the test and indications for performing it, see Eden and Boehm (1990, pp. 269, 275). In 1987, this test was not yet used to detect Down's. On the relationship of alphafeto protein testing to professional liability, see Holtzman (1989). and glucose  xP` tolerance test (threehour).}`  Y ԍThis threehour test is performed on persons with positive findings on the onehour test.} We obtained information on the number of ultrasounds from hospital charts. We only were able to obtain hospital charts for a subsample of persons surveyed. Thus, the number of observations in our analysis of ultrasounds was appreciably smaller than in our analysis with other dependent variables. Women were asked in our survey whether an ultrasound was performed but not about the number of ultrasounds performed. Virtually all women answered affirmatively, a pattern confirmed by the charts. For the other tests, we obtained information for the dependent variables from the survey. In all equations for diagnostic testing, we included a common set of variables for family income and health insurance and demographic characteristics of the motherage 35+, race, and ethnicity. Thus, the following discussion of equation specification focuses on variables unique to each test. 0*((ԌFor maternal risk factors in the analysis of ultrasounds, we included variables for hypertension during previous or target pregnancy and genetic disorder of mother or father (actual or suspected). The maternal risk factors in our analysis of amniocentesis were mother or father born with genetic disorder and mother born with birth defect (e.g., spina bifida, club foot, cleft lip). The analysis of alphafeto protein testing contained an additional maternal risk factor, previous stillbirth. Maternal risk factors in our analysis of the threehour test for diabetes were obesity (respondent weighed 200 pounds or more when she became pregnant with the target infant) and parity. Patient Satisfaction with Care During Labor and Delivery. Interviewers read respondents several statements and were asked whether they thought the statements were descriptive of the physician who delivered the baby they had in 1987. Respondents were given a choice along a fivepoint scale (agree to disagree). We selected four of these statements for analysis. "You feel that (DELIVERY DOCTOR) was genuinely interested in you and your baby  dduring labor and delivery." "You feel that (DELIVERY DOCTOR) fully explained the reason for each test and xx\procedure during your labor and delivery." "You feel that (DELIVERY DOCTOR) ignored what you told (him/her) during your xx\labor and delivery." "After the delivery, if you had questions, you felt you could call (DELIVERY xx\DOCTOR)." We used ordered probit analysis to determine the effect of liability risk and other factors on patient satisfaction with care during labor and delivery. The highest degree of agreement with these positive statements about physician behavior took the lowest value on the scale. Thus, if the liability threat made obstetricians try to please their patients more, the coefficients on the threat variable should have a negative sign. Explanatory variables were tort liability threat, income, insurance, age, raceethnicity, and education. Other Utilization During Pregnancy. In preliminary empirical analysis, we gauged the influence of tort liability on the decision to refer the patient to a specialist during pregnancy, mean length of visit, and waiting time to see the prenatal physician. Very few parameter estimates, including those on our liability risk variables, were' 0*(( statistically significant at conventional levels. Therefore, analysis of these decisions was dropped. 7V. RESULTS<Method of Delivery: Cesarean Section versus Vaginal Delivery Weighted and unweighted means and standard deviations of the explanatory variables are shown in Table 1. Comparing weighted with unweighted means, it is evident that we oversampled obstetricians with high claims' frequency, women with toxemia during the index pregnancy, and low birthweight infants. We slightly undersampled cases of dystocia. In general, weighted and unweighted means are quite similar. The specification test for exogeneity failed to reject the null hypothesis that individual obstetrician claim frequency was exogenous. Thus the explanatory variable for the threat of tort liability was defined for the physician's own claims' experience. Claim frequency has a negative coefficient which would not be expected if obstetricians performed cesarean sections as a defensive practice. The coefficient, however, is not statistically significant. By contrast, other factors do have statistically significant impacts: for the motherprevious cesarean section (+), first live birth (+), diabetes (+); for the birthplacenta previa or abruption (+), birthweight less than 1,500 grams (), multigestation (+), breech presentation (+), dystocia (+), and fetal distress (+); for demographic and financial characteristics of the mother and her householdincome (+), private conventional insurance (+), age (+), Hispanic (+), married (+), and low educational attainment (+); and for provider characteristicsgraduated from medical school since 1955 (+). With one exception, alternative specifications of the threat of tort liability were not statistically significant either (Table 2). High claim frequency raised the probability that a cesearean section was performed when a countyspecific measure of claim frequency was used. The associated parameter estimate was statistically significant at the 10 percent level. The specification including a set of binary variables to depict the individual obstetrician's prior claims' experience did not approach statistical significance at conventional levels. Although the coefficient on mean liability payment per exposure year was positive, the coefficient was not statistically significant at conventional levels. Antenatal Testing' 0*((ԌThe regressions used actual claim frequency of the individual obstetrician as a measure of the threat of tort liability, in none of these equations did we reject the null hypothesis of exogeneity (Table 3). For two of the four tests, the coefficient on the threat variable was positive and statistically significant at the 10 percent level or better, implying that such testing occurs defensively. However, signs on the threat coefficients in the ultrasound and amniocentesis analysis were negative, suggesting the opposite.  More affluent women were more likely to have had an amniocentesis and alphafeto protein test. Also, some insurance coefficients attained statistical significance at conventional levels, suggesting, in general, that the uninsured receive less antenatal testing. Satisfaction with Care The null hypothesis of exogeneity of the individual claims threat variable was consistently rejected in our analysis of mother's satisfaction with care received. Thus, the threat variable included in the analysis shown (Table 4) was defined as claim frequency in the obstetrician's county. None of the threat coefficients attained statistical significance at conventional levels. Moreover, signs varied. Since higher satisfaction took on low values, if the threat of claims had a deterrent effect, the coefficient should have been negative. In some specifications, relatively affluent and insured women were more satisfied with the care they received than were nonwhite women. However, patterns differed among equations, suggesting that the dependent variables capture different dimensions of the care process. , VII. DISCUSSION AND CONCLUSIONS<Our results suggest that some antenatal testing is responsive to variation in the threat of being sued. But for most measures included in our study, half of the antenatal testing variables, the decision to perform a cesarean section, and various dimensions of maternal satisfaction with care, our empirical analysis failed to reveal that obstetricians practice more "defensively" in areas with relatively high suit rates. Three recent studies have examined the relationship between liability threat and obstetrician defensive practices.  zP% Localio et al. (1993) examined the relationship between malpractice claims and cesarean delivery. Using data on all childbirths in sampled hospitals in New York State in 1984, the authors used five measures'0*(( with appropriate denominators to represent liability risk: relative premium levels; perceived risk by geographic area as reported in a physician survey; claims against the hospital's obstetric staff; claims opened against hospital; and claims against individual physicians. Claim frequency was measured alternatively as claims open and claims paid. Positive and statistically significant effects were obtained for all but the variable representing  xP@ individual physicians' actual claims' experience. @ Y ԍOther studies, based on much smaller samples and many fewer controls, found no effect of tort liability on cesarean section rates. Silver and Wolfe (1989) regressed cesarean section rates for 41 states on medical malpractice premiums, also defined for the state, and other explanatory variables. The coefficient on the premium variable, although positive, had an  Y< associated tvalue well below 1.0. Goyert et al. (1989) included several alternative variables for tort liability, including measures of the physician's own claims experience and perception of the influence of the current medicallegal climate on obstetrical decisionmaking in a multivariate analysis of cesarean section, with the delivery as the observational unit. None of the liability variables had a statistically significant effect on the probability of having a cesarean section. However, there were only 11 physicians in the sample of 1,533 deliveries.  The authors did not consider the possibility that their tort liability threat variables may have been endogenous. Even the claims' experience of a moderately sized hospital may be influenced by the experience of a very small number of members of the medical staff (Sloan and Hassan, 1990).  zP Baldwin et al. (1995) tested the hypothesis that physicians with greater malpractice claims exposure, either through personal experience or in their practice environment (their county) use more prenatal resourcesultrasound use, referral and consultation, and a weighted index of service useand have a higher cesarean delivery rate than do other physicians. Their data set was for a random sample of obstetricians and family physicians in Washington State. Data on outcome variables were obtained by medical abstractors from office records. In contrast to our study, Baldwin and coauthors limited inclusion of abstracted records to patients defined as "low risk" at the time they entered care. The authors found no indication that physicians with prior suit experience or those located in areas with frequent medical malpractice claims practiced any differently from other physicians. Using the same data base as the present study, we assessed the relationship between obstetrician claims'~ 0*((  zP experience and maternal satisfaction with care (Hickson et al., 1994). We found that women who obtained care from obstetricians with high previous claim frequency were less satisfied with the care they had received. As in the current analysis, the claims were from care delivered during 197783 and the index pregnancy occurred in 1987. The association we detected plausibly reflected the endogeneity of individual obstetrician claim frequency in the satisfaction with care analysis. It is noteworthy that physicians might prevent claims with good "bedsize manner," but this aspect of "defensiveness" rarely enters the public debate about medical malpractice policy. It could be said that the previous studies and we understated the effect of tort liability on practice behavior. If, given the threat of medical malpractice, all physicians exercise due care, analysis of a single cross section would not reveal that tort liability affects practice patterns, even if the threat of being sued led all physicians to adopt a higher care standard. We found, for example, that over three quarters of the patients did receive electronic fetal monitoring. In a much higher percentage of cases, the nurses who abstracted the hospital charts found that pertinent aspects of the case had been documented. The cesarean section rate was elevated in our sample. However, it is known that mistakes do occur, and while not completely a "straw man", the argument that all physicians exercise due care, or care exceeding any due care standard that the courts may impose, should not be pushed too far. We only assessed practice patterns in Florida. With data from a single state, it is generally not possible  xP to assess variations in liability rules on behavior. Y\ ԍA decision of a state court below the level of its Supreme Court would not be applicable statewide. With one state, the natural variation in threat largely occurs from geographic differences in litigiousness and in the propensity of patients/consumers to suffer adverse outcomes. Possibly, an interstate analysis would have revealed more effects. Very preliminary interstate analysis of cesarean section rates using medical malpractice premiums to measure professional liability risk did not show an effect on choice of delivery method (Silver and Wolfe, 1989). Also, the analysis based on Washington State data reached essentially the same conclusion as our study, and results from the New York study using claim frequency of the individual physician are consistent with ours. Finally, an important issue not investigated here is whether use of the types of diagnostic and#b0*(( therapeutic procedures included in our analysis really reduces injury cost. Recent evidence suggest that cesarean  xP section rates can be reduced without having an adverse impact on perinatal mortality.m Y ԍSee, e.g., O'Driscoll and Foley (1983), Porreco (1985), and Sachs (1989).m The evidence on effectiveness of ultrasound in pregnancy is mixed (Thacker, 1985). Many procedures have beneficial health  xP effects,y Y ԍThere is some evidence that screening for gestational diabetes reduces perinatal mortality (Everett, 1989) or that use of amniocentesis is helpful for detecting Down's  Y syndrome (Haddow et al., 1992). but again the marginal benefit depends on the circumstances under which the procedures are used.\ Y> ԍSee, e.g., Luthy et al. (1987) and Thacker (1989).\ Probably the strongest case against the use of tort liability in this context is that optimal care cannot be operationally defined because so many causation issues remain unresolved. In view of uncertainty about what the care standard should be, it may be at least reassuring that the effects of tort liability are as limited as this study's evidence implies they are.  0*((  Y 7l#b P7P##Xv P7[hXP#REFERENCES<  Y X` hp x (#%'0*,.8135@8: 34` `  #* 0.31c (0.17)9 @ 0.13 (0.34)N  U0.12 (0.33)  Y- Nonwhite` `  #* 0.51b (0.22)9 @ 0.11 (0.31)N  U0.12 (0.33) ` `   Y Hispanic` `  #* 0.73a (0.17)9 @ 0.15 (0.35)N  U0.11 (0.32)  Y Married` `  #* 0.64a (0.20)9 @ 0.88 (0.33)N  U0.84 (0.37)  Y < High school # * 1.01a (0.29)9 @ 0.07 (0.26)N  U0.08 (0.27)  Yu High school grad # * 0.60a (0.17)9 @ 0.39 (0.49)N  U0.41 (0.49)  YG Some college` `  #* 0.46a (0.16)9@ 0.32 (0.47)N  U0.30 (0.46) Obstetrical unit level II` `  # * 0.23 (0.16)9 @ 0.57 (0.50)N  U0.58 (0.49) Obstetrical unit level III` `  # * 0.016 (0.20)9 @ 0.15 (0.35)N  U0.19 (0.40) Obstetrical unit missing` `  #*0.091 (0.26)9@ 0.08 (0.28)N  U0.07 (0.25)  YJ& OB grad 195573 #* 0.63a (0.24)9 @ 0.59 (0.49)N  U0.60 (0.49) ` `  #  Y( OB grad 197377 # * 0.66a (0.26)9 @ 0.26 (0.44)N  U0.28 (0.45) )0*0*0* Table 1 cont. ` `  # Probit Regression @ Means and Standard Deviations ` `  #*hh29@ WeightedN Unweighted Explanatory Variables  _ yx dddy  YX OB grad 1978+ #* 0.58c (0.31)9 @ 0.08 (0.28)N  U0.07 (0.25) Foreign M.D. grad #* 0.026 (0.16)9 @ 0.20 (0.40)N  U0.18 (0.38) ` `  #* N = 955 _____________________________________   Y9 1Cesarean Section = 1 if performed and 0 otherwise.  Y aStatistically significant at the one percent level (twotail test)  Y: bStatistically significant at the 5 percent level (twotail test)  Y cStatistically significant at the 10 percent level (twotail test) ;0*0*0* Table 2. Effects of Tort Liability on Cesarean Section v. Vaginal Delivery:   Alternative Specifications ` `  #*hh29Individual OBN Coefficient Tort Variable` `  #*hh29 or CountyppGN(Standard Error)  H yx dddy 1. Claim frequency* #* hh29Individual OBN0.089 (0.24)  Y 2. Claim frequency* #* hh29 CountyppGN 0.82c (0.48) 3. Mean liability payment**hh29@ CountyN  U 0.31 (0.21) _____________________________________ *Per exposure year **Per exposure year ('00000$)  Y cStatistically significant at the 10 percent level (twotail test)  0*0*0*    X  X   Y Table 3. Diagnostic Procedures1 Explanatory44 Ultrasound<<.5Amniocentesis Alphafeto Protein`Diabetes Test Variables44 ( # )<<.5 (0 1)DDCK (0 1)LLY`3 hour (0 1)  _ y! dddy  YX Intercept44 0.94a '(0.21)<<.52.89a<(0.49)DDCK1.34aR(0.19)LLY`0.59a!!g(0.17)T$T$n  Y* Claim frequency 0.48c '(0.26)<<. 51.30<(1.05)*K 0.48bR(0.22)LLY` 0.38c!!g(0.21)  Y Hypertension44 0.58a '(0.19)<<.5 < (  )DDCK R (  )LLY` !!g (  ) Motherfather genetic disorder 0.34 '(0.59)<<.5 0.60<(1.40)DDCK 0.008 (0.78)` !!g (  ) Mother had birth defect44  ' (  )<<.50.72<(0.80)DDCK0.21R(0.34)LLY` !!g (  ) Previous stillborn  ' (  )<<.5 < (  )DDCK1.36R(1.17)LLY` !!g (  ) Mother obese44  ' (  )<<.5 < (  )DDCK R (  )LLY` 0.01!!g(0.43)  Y Parity44  ' (  )<<.5 0.27<(0.74)DDCK R (  )LLY` 1.08b!!g(0.54)  Y Income ('00000$) 0.20 '(0.18)<<.5 0.98a<(0.27)DDCK 0.97aR(0.19)LLY` 0.061(0.17)  Yu Income missing 0.58 '(0.49)<<.5 0.29<(0.70)DDCK 0.66cR(0.36)LLY`0.89b!!g(0.38) Private conventional  Y0 insurance44 0.27 '(0.20)<<.5 0.86b<(0.42)DDCK 0.50aR(0.18)LLY` 0.28c!!g(0.17)T$T$nT$T$n  Y HMO44 0.63b '(0.25)<<.5 0.62<(0.55)DDCK 0.71aR(0.23)LLY` 0.53b!!g(0.21)  Y Medicaid44  '(  )<<.5 1.04<(0.83)DDCK 1.07aR(0.35)LLY` 0.18!!g(0.32)  Y" Other insurance  '(  )<<.5 1.28a<(0.50)DDCK 0.28R(0.29)LLY` 0.17!!g(0.26) Medicaid and other insurance 0.24 '(0.35)<<. 5 <(  )DDCK R(  ) LLY` !!g (  )  Y3' Age > 3444 0.41b '(0.17)<<.5 1.86a<(0.19)DDCK 0.087 (0.16)LLY `0.027 (0.15)  Y) Nonwhite44 0.61 '(0.27)<<.50.72b<(0.36)DDCK0.91aR(0.22)LLY`0.010 (0.18)  Y* Hispanic44 0.046 (0.18)<<.5 0.062 (0.25)DDCK 0.032 (0.14)LLY` 0.47a!!g (0.14) 44 N = 268<<. 5 N = 747DDCK N = 708LLY` N = 716,--- Table 3 cont. _____________________________________  Y R2 = 0.11  Y R2 (C) = 0.08 F (11,256) = 3.0  Y_ aStatistically significant at the one percent level (twotail test)  Y bStatistically significant at the 5 percent level (twotail test)  Y cStatistically significant at the 10 percent level (twotail test) *County level specification  Y0 1The first regression was estimated with ordinary least squares. The other regressions are probits. #x6X@X@#---  Y #Xv P7[hXP#Table 4. Satisfaction with Care from M.D.: Ordered Probit  44 Seemed Explained Ignored WhatR Could` 44 Interested5 Fully You Said Telephone   y!(dddy  Y Constant44  '0.99a<<.50.98a<DDC1.60aKR0.90a 44  '(0.22)<<.5(0.20)<DDC(0.23)KR(0.23) Claim44  ' 0.63<<.50.08<DDC 0.22KR0.19LLY`!!gT$T$n 44  '(0.49)<<.5(0.45)<DDC(0.49)KR(0.50)  Y Age < 2044  ' 0.10<<.50.46b<DDC 0.05KR 0.30LLY` 44  '(0.22)<<.5(0.21)<DDC(0.21)KR(0.21)  Y Age > 3444  '0.59a<<.5 0.05<DDC 0.05KR0.16 44  '(0.15)<<.5(0.12)<DDC(0.14)KR(0.14)  Y < High school  '0.18<<.5 0.25<DDC0.49bKR0.21 44  '(0.22)<<.5(0.19)<DDC(0.20)KR(0.21)  YD High school grad  '0.30b<<.50.31a<DDC 0.08KR0.37a 44  '(0.13)<<.5(0.12)<DDC(0.13)KR(0.13) Some college44  '0.11<<.50.12<DDC 0.18KR0.18 44  '(0.12)<<.5(0.11)<DDC(0.13)KR(0.12)  Y Income ('00000$)  '0.06650.310cDDC 0.0316R0.608a`!!g 44  '(0.068)5(0.164)DDC(0.0169)R(0.213)  Yu Income missing  '1.12a<<.51.62a<DDC1.06aKR1.24a 44  '(0.42)<<.5(0.46)<DDC(0.44)KR(0.41) Private conventional  '  Y insurance44  '0.56a<<.50.13<DDC0.36aKR 0.02 44  '(0.14)<<.5(0.13)<DDC(0.15)KR(0.15)  Y HMO44  '0.54a<<.50.12<DDC0.07KR 0.31c 44  '(0.19)<<.5(0.17)<DDC(0.18)KR(0.19)  Y# Medicaid44  '1.10a<<.50.31<DDC 0.24KR0.64b 44  '(0.323)5(0.26)<DDC(0.26)KR(0.32)  YJ& Other insurance  '0.54a<<.50.37b<DDC0.28KR0.29LLY 44  '(0.21)<<.5(0.19)<DDC(0.20)KR(0.23)  Y) Nonwhite44  '0.14<<.50.38a<DDC0.21KR0.27c 44  '(0.16)<<.5(0.15)<DDC(0.15)KR(0.16))--- Table 4 cont.  Satisfaction with Care from M.D.: Ordered Probit 444 Seemed Explained Ignored What Could 444 Interested Fully You Said Telephone y Tdddy Hispanic444  0.18<<<0.08DDD 0.110.25b 444  (0.13)<<<(0.50)DDD(0.12)(0.13) Inter 2444  0.33a<<< 0.37aDDD 0.28a 0.19a 444  (0.05)<<<(0.05)DDD(0.05)(0.03) Inter 3444  0.34a<<< 0.64aDDD 0.50a 0.21a 444  (0.05)<<<(0.05)DDD(0.06)(0.04) Inter 4444  0.98a<<< 1.13aDDD 1.01a 0.74a 444  (0.07)<<<(0.06)DDD(0.07)(0.05)  444 N = 885 N = 879 N = 881 N = 881  _____________________________________  aStatistically significant at the one percent level (twotail test) bStatistically significant at the 5 percent level (twotail test) cStatistically significant at the 10 percent level (twotail test)