Essays in the Economics of Medical Malpractice Law

by

Ity Shmuel Shurtz

A dissertation submitted in partial satisfaction of the

requirements for the degree of

Doctor of Philosophy

in

Economics

in the

Graduate Division

of the

University of California, Berkeley

Committee in charge:

Professor Emmanuel Saez, ChairProfessor David Card

Professor Justin McCrary

Spring 2011

Essays in the Economics of Medical Malpractice Law

Copyright 2011by

Ity Shmuel Shurtz

1

Abstract

Essays in the Economics of Medical Malpractice Law

by

Ity Shmuel Shurtz

Doctor of Philosophy in Economics

University of California, Berkeley

Professor Emmanuel Saez, Chair

This dissertation explores the interaction between medical malpractice law and medicaltreatment. The first chapter addresses the question: How do malpractice lawsuits affectphysician behavior? In this chapter, I study the impact of malpractice claims against ob-stetricians, a specialty that is regarded as particularly subject to malpractice concerns, ontheir choice of whether to perform C-sections, a common procedure that is thought to besensitive to physician incentives. I find that immediately after an adverse event (definedas an obstetrical procedure that ultimately leads to a malpractice claim), C-section ratesjump discontinuously by 4%. The increase in C-section rates persists even 4.5 years afterthe adverse event. Several other findings provide support to the view that fear of litigationand damage to reputation explain the results, rather than a mere response to the negativeoutcome that brought about the malpractice claim. First, unsuccessful claims, which, atthe time of the adverse event, are perceived as less harmful to physicians’ reputation, donot lead to an increase in C-section rates. Second, the impact on C-section rates is largerfor patients insured by a commercial insurance provider, for which reputational concerns arelikely to be stronger, since they are less constrained in their choice of physicians. In addition,the impact is smaller for experienced physicians, but not for those with a prior history oflitigation claims. I also find evidence of peer effects: following an adverse event, a physician’scolleagues also have higher C-section rates. Overall, this chapter shows that following anadverse event physicians adopt more conservative and costly treatment strategies and thattheir response is likely to be related to fear of litigation and damage to reputation.

The impacts of malpractice regulations and financial incentives for providers are typicallystudied independently. In the second chapter of this dissertation, I show that in order tomake both positive and normative statements about medical malpractice liability, one mustconsider the legal and financial incentives faced by healthcare providers jointly. I develop asimple model of physician behavior to show that the effect of tort reforms on treatment deci-sions depends critically on physicians’ financial incentives. When treatment is not profitableat the margin, liability reduction leads to a decrease in treatment levels; conversely when

2

treatment is profitable, liability reduction leads to an increase in treatment levels. Motivatedby this simple theoretical framework, I analyze the impact of a tort reform in Texas thatreduced malpractice liability on C-section rates and common pediatric surgical procedures.Consistent with the theory, the data show that the rate of C-sections for commercially in-sured mothers, which are thought to be profitable, increase by about 2% relative to the rateof C-sections for mothers on Medicaid, which are considered to be unprofitable. Similarly,the reform increases the incidence of profitable pediatric procedures relative to unprofitableones. These findings help explain why the existing literature on optimal medical malpracticelaw is inconclusive and underscore the importance of understanding the economic incentivesat play when designing legal regulations.

i

To Adi

ii

Contents

List of Figures iv

List of Tables vi

1 The Impact of Malpractice Litigation on Physician Behavior: The Case ofChildbirth 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Background on Medical Malpractice and Related Work . . . . . . . . . . . . 31.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.4 The Effect of an Adverse Event . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.4.1 Selection Around an Adverse Event . . . . . . . . . . . . . . . . . . . 51.4.2 Short-Run Effect on an Adverse Event . . . . . . . . . . . . . . . . . 61.4.3 Long-Run Effect of an Adverse Event . . . . . . . . . . . . . . . . . . 71.4.4 The Causal Role of Litigation . . . . . . . . . . . . . . . . . . . . . . 9

1.5 Response to News About a Lawsuit . . . . . . . . . . . . . . . . . . . . . . . 121.5.1 Selection Around a First Contact . . . . . . . . . . . . . . . . . . . . 121.5.2 Response to a First Contact . . . . . . . . . . . . . . . . . . . . . . . 13

1.6 Response of Peers to an Adverse Event . . . . . . . . . . . . . . . . . . . . . 141.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15Tables & Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2 The Interaction between Optimal Medical Malpractice Law and Physi-cians’ Financial Incentives 522.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542.3 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2.3.1 Patients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552.3.2 Providers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552.3.3 Providers’ behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

2.4 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 582.4.1 Background - Texas Tort Reform . . . . . . . . . . . . . . . . . . . . 59

iii

2.4.2 Childbirth Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 602.4.3 Common Pediatric Surgery Analysis . . . . . . . . . . . . . . . . . . 62

2.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66Appendix B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67Tables & Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

iv

List of Figures

1.1 C-section rates 1992-2008, Florida. . . . . . . . . . . . . . . . . . . . . . . . 321.2 Distribution of Physicians’ Prior Claims. . . . . . . . . . . . . . . . . . . . . 331.3 Distribution of Claim Payments. . . . . . . . . . . . . . . . . . . . . . . . . . 341.4 Selection around Adverse Event. . . . . . . . . . . . . . . . . . . . . . . . . . 351.5 Selection on Observables, Adverse Event Panel. . . . . . . . . . . . . . . . . 361.6 Short-Run Effect of an Adverse Event. . . . . . . . . . . . . . . . . . . . . . 371.7 Long-Run Effect of an Adverse Event, Matching Approach. . . . . . . . . . . 381.8 Long-Run Effect of an Adverse Event, Event Study Approach. . . . . . . . . 391.9 Short-Run Effect of an Adverse Event by claim Success. . . . . . . . . . . . . 401.10 Short-Run Effect of an Adverse Event by Insurance Type. . . . . . . . . . . 411.11 Distribution of Physicians’ Experience, Adverse Event Panel. . . . . . . . . . 421.12 Short-Run Effect of an Adverse Event, by Experiences. . . . . . . . . . . . . 431.13 Short-Run Effect of an Adverse Event, by Prior Claims. . . . . . . . . . . . . 441.14 Distribution of Claims Report Timing. . . . . . . . . . . . . . . . . . . . . . 451.15 Selection around First Contact. . . . . . . . . . . . . . . . . . . . . . . . . . 461.16 Selection on Observables, First Contact Panel. . . . . . . . . . . . . . . . . . 471.17 Short-Run Effect of a First Contact. . . . . . . . . . . . . . . . . . . . . . . 481.18 Short-Run Effect of a First Contact, by claim success. . . . . . . . . . . . . . 491.19 Short-Run Effect of an Adverse Event on Same-Hospital Peers. . . . . . . . . 501.20 Short-Run Effect of an Adverse Event on Same-Hospital Peers. . . . . . . . . 51

2.1 The Effect of a Liability Decreasing Reform On Treatment. . . . . . . . . . . 752.2 C-section Rate: TX, FL and CA. . . . . . . . . . . . . . . . . . . . . . . . . 762.3 C-section Rate Medicaid vs. Commercial, TX. . . . . . . . . . . . . . . . . . 772.4 C-section Rate Medicaid vs. Commercial, FL. . . . . . . . . . . . . . . . . . 782.5 C-section Rate Medicaid vs. Commercial, CA. . . . . . . . . . . . . . . . . . 792.6 C-section Rate Medicaid vs. Commercial. . . . . . . . . . . . . . . . . . . . . 802.7 Ratio of Commercial to Medicaid Surgeries High and Low Tercile Attractiveness. 812.8 Change in Ratio of Commercial to Medicaid Surgeries Pre-Post Reform by

Attractiveness, TX. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

v

2.9 Change in Ratio of Commercial to Medicaid Surgeries Pre-Post Reform byAttractiveness, TX 2001. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

2.10 Change in Ratio of Commercial to Medicaid Surgeries Pre-Post Reform byAttractiveness, CA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

vi

List of Tables

1.1 Summary Statistics - Inpatient Data: All Sample, Adverse Event and FirstContact Panels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.2 Selection Around an Adverse Event - High-Risk, Age and Insurance Type . . 181.3 Selection Around an Adverse Event - High-Risk, Age and Insurance Type . . 191.4 Short-Run Effect of an Adverse Event . . . . . . . . . . . . . . . . . . . . . . 201.5 Matching Physician to Same-County-Colleagues, 7 Year Adverse Event Panel 211.6 Long-Run Effect of an Adverse Event, Event Study Approach . . . . . . . . 221.7 Short-Run Effect of an Adverse Event, Successful and Unsuccessful Claims . 231.8 Short-Run Effect of an Adverse Event, Private Insurance Mothers and Medi-

caid Mothers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241.9 Short-Run Effect of an Adverse Event, High-Experience and Low-Experience

Physicians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251.10 Short-Run Effect of an Adverse Event, Physicians With and Without Prior

Claim History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261.11 Selection Around the a First Contact - High-Risk, Age and Insurance Type 271.12 Selection Around a First Contact - Predicted C-section Rates . . . . . . . . 281.13 Short-Run Effect of a First Contact . . . . . . . . . . . . . . . . . . . . . . . 291.14 Short-Run Effect of a First Contact, Successful and Unsuccessful Claims . . 301.15 Short-Run Effect of an Adverse Event on Peers from the Same Hospital . . . 31

2.1 Summary Statistics Childbirth Sample . . . . . . . . . . . . . . . . . . . . . 692.2 The Effect of a Liability Decreasing Reform, Diff in Diff Estimates: TX, CA

and FL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 702.3 The Effect of a Liability Decreasing Reform, Diff in Diff Estimates by Risk

Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 712.4 Summary Statistics Pediatric Surgeries Sample . . . . . . . . . . . . . . . . . 712.5 Effect of Reform on Pediatric Surgery by Attractiveness Group . . . . . . . . 722.6 Effect of Reform on Pediatric Surgery . . . . . . . . . . . . . . . . . . . . . . 732.7 Procedure Incidence by Attractiveness Tercile . . . . . . . . . . . . . . . . . 74

vii

Acknowledgments

I am very grateful to Professors David Card, Raj Chetty and Emmanuel Saez for theirguidance and support in my research.

1

Chapter 1

The Impact of Malpractice Litigationon Physician Behavior: The Case ofChildbirth

1.1 Introduction

Medical malpractice liability law has attracted much attention in the past decades. It isoften argued that fear of lawsuits might encourage the provision of high-cost, low-benefitmedical treatment, i.e. defensive medicine. This view is supported by evidence from self-reported data: studies that survey physicians, consistently find that physicians report prac-ticing defensive medicine both by accepting fewer high-risk patients, and by choosing moreconservative procedures and more diagnostic tests (Studdert et al. [2005], Kessler and Mc-Clellan [1997], Reyes [2010b]). Nevertheless, despite the high stakes attributed to the designof optimal medical malpractice law, very little is known about the extent to which medicalmalpractice law affects healthcare providers’ behavior, and the mechanisms underlying thiseffect.

In particular, it is not well-understood why physicians’ choice of procedures would bestrongly affected by medical malpractice law. Physicians are typically insured against mal-practice litigation, and furthermore, physicians’ premiums are not experience-rated and aretypically set at the specialty level (Sloan [1990], Fournier and McInnes [2001]). Thus, sincephysicians do not bear the financial costs of medical malpractice litigation it is a puzzle:why do physicians care so deeply about medical malpractice litigation?

One hypothesis, linking medical malpractice litigation and medical treatment is thatmedical malpractice litigation might harm a physician’s reputation. The various stages ofthe litigation process, from the initial investigation by the plaintiff, through the lawsuitand to the final payment, are visible to hospitals, colleagues, patients and lawyers, therebymagnifying the reputational consequences of incidents which result in litigation.

2

Motivated by that hypothesis, this study concentrates on the effect of physicians’ personalexposure to malpractice lawsuits. It examines the effect of an adverse event, defined as anobstetrical procedure that ultimately leads to a malpractice claim, on treatment patternsof obstetricians, that are regarded as particularly sensitive to malpractice concerns (Reyes[2010b]). Particular attention is paid to an adverse event, rather than later stages in thelitigation process, since physicians are likely to worry about the damage to their reputationfrom an impending law suit, and alter their behavior accordingly - immediately after anadverse event.

Physician’s responses are studied by examining their decision to perform C-sections, acommon procedure that is thought to be sensitive to physician incentives (see Currie andMacLeod [2008]), and that is often argued to be associated with the practice of defensivemedicine. Inpatient data from Florida, matched with data on physicians’ educational back-ground and malpractice claim history is used to conduct the analysis.

This study extends the existing literature in two ways. First, it stresses physicians’responses immediately following an adverse event, which has not been rigorously studiedbefore. Second, it investigates the mechanisms in the impact of an adverse event on treat-ment, by analyzing the heterogeneity in the response across physicians, patients and typesof claims, and examining the causal role of fear of litigation in the response.

To evaluate the short-run response of physicians, I use a regression discontinuity design,estimating the “break” in C-section rates immediately after an adverse event. In order toestimate the long-run effect of an adverse event, I take two approaches, each based on adifferent identification assumption. The first is a matching method, pairing each affectedphysician with an individually-tailored control group. The second is an event study regres-sion, estimating the response to an adverse event by controlling for physician and time fixedeffects, as well as other covariates.

The empirical analysis yields the following main findings. First, following an adverseevent there is a clear dicontinuous jump in C-section rates of about 1 percentage point,which reflects an increase of roughly 4% in C-section rates. Second, the increase in C-sectionrates is persistent, lasting at least 4.5 years after the adverse event. Finally, I find evidencethat physicians’ close peers are also more likely to perform C-sections following the adverseevent.

While fear of litigation and damage to reputation can explain an immediate responseto the adverse event, physicians may also respond to the negative outcome that broughtabout the malpractice claim. Therefore, I investigate the causal role of fear of litigation anddamage to reputation in the effect of an adverse event, by analyzing heterogeneity in theresponse across patients, physicians and claim-types. I find that the response to an adverseevent is concentrated among successful claims, which, at the time of the adverse event, aremore likely to be perceived as harmful to physicians’ reputation than unsuccessful claims.Since the severity of outcomes in the two types of claims is similar, this result supports thecausal role of malpractice litigation. Moreover, privately insured patients who are likely to beboth less constrained in their ability to choose a prenatal physician and in their access to the

3

legal system, are subject to a greater increase in C-sections following an adverse event thanpatients insured by Medicaid, providing additional support to the causal role of malpracticelitigation.

The observed increase in C-section rates following adverse events is concentrated amongless experienced physicians. However, physicians’ responses are not significantly differentbetween physicians with and without prior malpractice claim histories. Since a response tonegative outcomes is expected to wane with prior adverse event exposure, whereas fear ofharm to reputation is likely to increase with prior adverse event exposure, the latter resultfurther supports the role of litigation in physicians’ response to the adverse event. Overall,the results suggest that reputation and fear of litigation are important in physicians’ responseto the adverse event. It is important to note that the results do not rule out the effect ofphysicians’ response to the negative outcome that is associated with an adverse event.

A major concern about the interpretation of the results as reflecting a change in practicepatterns is that there might be a change in patient composition following the adverse event.I address this concern by testing for observed differences in the number of births, risk level ofthe mothers pool, mothers’ mean age and the share of mothers insured by a private carrieraround the adverse event. I find no evidence of a change in the characteristics of mothersfollowing the event, which alleviates those concerns.

The remainder of the paper is organized as follows. Section 2 describes previous work onmedical malpractice litigation and healthcare provision. Section 3 presents the data, section4 presents evidence on physicians’ response to an adverse event, section 5 presents evidenceon physicians’ response to the first contact regarding a claim, section 6 presents evidence onpeer effects and section 7 offers concluding remarks.

1.2 Background on Medical Malpractice and Related

Work

A growing body of empirical work attempts to evaluate the relationship between the threat ofa lawsuit (“malpractice pressure”), and delivery of healthcare using a variety of identificationapproaches. In the case of elderly patients, in a seminal paper, Kessler and McClellan [1996],find that while tort reforms have no significant effect on health outcomes they significantlyreduce medical costs (Sloan and Shadle [2009], reassessed those results and found that tortreforms do not significantly affect medical decisions). Baicker et al. [2007] find that highermalpractice awards and premiums are associated with higher Medicare spending. In thearea of childbirth, Currie and MacLeod [2008] show that Joint and Several Liability reformsreduce complications of labor and procedure use, whereas caps on noneconomic damagesincrease them. Kim [2007] on the other hand, using variation in claim numbers in otherspecialties, finds that obstetricians’ procedure choice is insensitive to malpractice pressure.In a recent working paper, Lakdawalla and Seabury [2009] exploit variation in the generosity

4

of local juries to identify the causal impact of medical malpractice litigation on medical costsand mortality, and find that Liability Pressure is associated with improved outcomes, namelyreductions in patient mortality.

This paper builds on and is directly related to several recent papers which study theassociation between healthcare and personal experience of malpractice litigation, using in-patient data from Florida, matched with physician history of malpractice claim data. Grantand McInnes [2004] relate the change in Florida obstetricians’ propensity to perform C-sections between 1992 and 1995 to their malpractice experience in 1993 and 1994. They findthat claims which ultimately resulted in a large indemnity payment were associated with anincrease in C-section rates and by contrast claims which ultimately resulted in a small indem-nity payment were associated with a decrease in C-section rates, with overall small effect onC-section rates. Gimm [2010] uses, inpatient data from 1992-2000 aggregated in physician-year cells and does not find statistically significant evidence that physicians changed theirpractice patterns by increasing C-section rates in response to malpractice claims. Recently,Dranove and Watanabe [2010] examine whether physicians change C-section rates after thefirst contact regarding a lawsuit. They find that during the quarter of contact there is ahospital wide increase in C-section rates of 0.5 percentage points. In addition, they find thatthree quarters after the first contact physicians increase C-section rates by 1.3 percentagepoints for a period of 1 quarter. Overall their results imply small, short-lived increases inC-section rates after a physician contacted about a malpractice claim for the first time.

1.3 Data

I use the universe of all births recorded in the Florida Hospital Inpatient Discharge Data(“inpatient data”) spanning the years 1992-2008. Physicians who performed less than 25deliveries throughout the entire period are excluded from the analysis, leaving about threemillion births performed by 2,307 physicians, which comprise 99.8% of all births. The inpa-tient data was merged with the Practitioner Profile Data File (“profile data”). The profiledata contains detailed information about physicians’ education history, allowing to create ameasure of physicians’ experience. Next, Medical Professional Liability Files (“claims data”)for the years 1979-2008 were matched to the data. The claims data was matched using bothlicense number and physician name. The claims data contains a history of closed medicalmalpractice claims, payments made if any, severity of injury, as well as the dates of injury,report and closing of malpractice claims.

I then create an adverse event panel : a five year balanced panel, including physicianswho appear in the data 10 quarters pre and post the adverse event. The first adverse eventthat is covered by the inpatient data period is chosen for each physician. A panel for thetime of the first contact regarding a claim, a first contact panel, was created in an analogousway.

Figure 1.1 plots C-section rates in Florida for the years 1992-2008. The figure shows,

5

consistent with the national trend (MacDorman et al. [2008]), that C-section rates have in-creased substantially from roughly 23% in 1996 to 38% in 2008. Table 1.1 provides summarystatistics for three groups: the full sample, the adverse event panel and the first contactpanel. Patient characteristics in the full sample are very similar to those in the two panels.Notably, in the panel samples there are lower rates of mothers under Medicaid, lower ratesof Afro-American and Hispanic mothers, and lower incidence of risk factors.

Figure 1.2a presents the distribution of physicians’ prior claims history, indicating thatroughly 55% of the 459 physicians in the adverse event panel, did not have prior history ofmalpractice litigation at the time of the adverse event, and approximately 90% of the physi-cians experienced no more than four claims. Figure 1.2b shows a very similar distributionfor the first contact panel. Figure 1.3 summarize the distribution of nominal payments perclaim rounded to the closest multiple of $50,000, in the adverse event and first contact panels,respectively. While roughly 31% of the claims, in the adverse event panel, are unsuccessfuland result in a payment of zero (The first bar in Figure 1.3a shows a frequency of about 37%since it includes claims which resulted in low payments), there are claims with payments ofabout $1,000,000 or more. Interestingly, Payment amounts tend to ”bunch” around $250,000and $500,000, which are standard per claim ceilings of physicians’ premiums, suggesting thatparties tend to reach a settlement based on the physicians’ coverage.

1.4 The Effect of an Adverse Event

In order to analyze the effect of an adverse event, the timing of an adverse event was normal-ized to zero for all physicians, and other quarters were defined relative to this base period.I study the short-run effect of an adverse event by estimating the “break” in C-section ratesimmediately before and after the adverse event using a regression discontinuity design typeestimation. The identification of the short-run effect is based on assuming that the differencebetween physician’s treatment behavior immediately before and after the adverse event isthe result of the adverse event. However, using this approach one can not evaluate physi-cians’ response away from the discontinuity around the adverse event. In order to study thelong-run effect of an adverse event, two approaches are used, each based on a different iden-tification assumption. The first is a matching method, pairing each affected physician withan individually-tailored control group. The second is an event study approach, controllingfor physician and time fixed effects, as well as other covariates.

1.4.1 Selection Around an Adverse Event

One might be concerned that the estimates of the effect of the adverse event reflect a changein the composition of the sample following the adverse event. Hence, it is important to checkwhether following the adverse event there was a change in the composition of the sample’sobservable characteristics which are likely to be associated with a change in C-section rates.

6

To address this matter I first plot the number of births around the adverse event to examineif it indicates a change in the per period sample size. Figure 1.4a plots per period birth num-bers, showing no apparent change in the sample size around the adverse event. Next, I definehigh-risk deliveries: deliveries with one of a set of risk-factors and delivery complications,markedly breech position, previous C-section, hemorrhage, hypertension, multiple gestationand oligohydramnios1 (see a similar classification in MacDorman et al. [2008]). Figure 1.4bplots the per period share of high-risk mothers in the adverse event Panel. While there isa downward trend in the share of high-risk mothers is the sample, there is no visible dis-crete change in the share of high-risk mothers following the adverse event, indicating thatrisk levels are similar for mothers just before and after the adverse event. As an additionalcheck, the per-period average mothers’ age in the sample is plotted in Figure 1.4c. Thisfigure shows an upward trend in the average age of mothers over time, but there is no evi-dence of a discontinuity in mother’s average age in the period following the adverse event,offering additional evidence that no change in the risk composition of the sample took placefollowing the adverse event. Figure 1.4d plots per period rate of privately insured mothers,showing a small and statistically insignificant increase after the adverse event. To quantifythe graphical evidence, columns (1)-(6) of Table 1.2 report, the estimates of a model analo-gous to (1.1), replacing C-section rates by high-risk, age and rate of mothers under privateinsurance, respectively. The estimates are consistent with the figures with very small andstatistically insignificant coefficients.

Finally, to combine the three measures of selection, a linear probability model, estimatingthe effect of age, each of the high-risk factors and insurance type, on the probability ofundergoing a C-section in the pre-event period is used. The average predicted C-sectionrates using the model’s estimates for the entire panel are plotted in Figure 1.5. The Figureshows that predicted C-section rates decrease over time. There is an increase in C-sectionrates of roughly 0.3 percentage points following the adverse event, but as Table 1.3 shows,it is not statistically significant.

1.4.2 Short-Run Effect on an Adverse Event

The discontinuity in C-section rates is estimated using a regression discontinuity design typemodel. The estimation equation has the form:

C-sectionit = α + τD + β1time+ β2time2 + β3time ·D + β4time

2 ·D + εit (1.1)

where time ∈ {−10, ...− 1, 0, ...9} is the number of elapsed quarters since the adverse eventand D ∈ {0, 1} is a dummy variable that indicates the post event periods so that D = 1if time ≥ 0, and D = 0 if time < 0. τ is the coefficient of interest in this specificationas it captures the effect of the adverse event on physician behavior. To test the robustness

1High-risk includes the following diagnoses: Previous C-section, breech position, multiple gestation, hy-pertension, early onset, hemorrage, obesity, diabetes, polyhydramnios, oligohydramnios & distress.

7

of the results, I control for confounding factors by estimating two additional specifications:basic controls, adds a set of patient characteristics - age, race and risk factors (includingmother’s previous C-section history, breech position and hypertension2). Full controls, adds,in addition to the basic controls, physician and quarter fixed effects.

Results. Figure 1.6 plots average per period C-section rates 10 periods prior to, and 10periods following the adverse event. For visual reference, a quadratic regression model fitis added separately to the periods before and after the adverse event. Figure 1.6 shows anapparent jump of about 1 percentage point immediately following the adverse event, implyingan increase of roughly 4% in C-section rates. Note that mean C-section rates away from theadverse event are quite smooth, confirming that absent the adverse event, C-section ratesare not likely to change discotinuously. Column (1) of Table 1.4, displays the estimationresults of equation (1.1). The estimates support the graphical evidence conveyed in Figure1.6, indicating a statistically significant increase of 1.1 percentage points in C-section ratesfollowing the adverse event. columns (2) and (3) of Table 1.4 report the estimates of thebasic and full controls models, showing a statistically significant increase in C-section ratesof 0.9 and 0.76 percentage points respectively.

1.4.3 Long-Run Effect of an Adverse Event

A Matching Approach

In order to evaluate the long-run effects of malpractice litigation I employ a complementaryapproach based on a comparison between the treated physician group and a control groups.The time span of the panel is expanded to seven years, 10 quarters prior the adverse eventand 18 quarters after the adverse event, leaving a smaller sample of 338 physicians. I create acontrol group by matching each physician with colleagues who appear in the data throughoutthe relevant 7 years. In order to generate a control group which best controls for the factorsaffecting C-section rates of the treated physician, one would prefer to choose physicians thatare as close as possible to the treated physician. On the other hand, there is a concernthat the adverse event affects the physician’s close peers (an issue which is studied below).With this trade-off in mind, the control group that was chosen for the analysis is the setof physicians who treat patients from the same county as the treated physicians, excludingphysicians who work at the same hospital as the treated physician3.

Using the control group, I construct an estimator for the difference in C-section ratesbetween the treatment and control groups. To make the visual representation clearer, theresults are summarized in six-months periods.

Formally, for individual physician i, i = 1, ..., N , in six-month period t, t ∈ {−5, ... −2The full list of coefficients includes: previous C-section, breech position, multiple gestation, hypertension,

early onset, hemorrage, obesity, diabetes, polyhydramnios, oligohydramnios, anemia, distress and feto.3Same county is defined as the county in which most of a physicians’ patients reside and same hospital is

defined as the hospital in which most of a physician’s deliveries are performed

8

1, 0, ...8}

τt =1

N

N∑i=1

{C-sectionit −J

1

J

∑1

C-sectionjt} (1.2)

where , j ∈ {1...J} is the set of physicians in the control group.Results. Figure 1.7a plots average C-section rates five six-months periods before and

nine six-months periods after the adverse event for all 338 physicians who are included inthe seven years panel, and for the control group. For visual reference, a quadratic regressionmodel fit is added separately to the periods before and after the adverse event for the controland treatment groups. The figure shows that prior to the adverse event, physicians in thetreatment group tend to perform less C-sections relative to the average C-section rate in theircounty. This result is not surprising because, as Table 1.1 shows, the mother population inthe panel tends to be of a higher socioeconomic status with lower incidence of risk factors.Following the event, there is a jump in C-section rates of the treated physicians and the gapbetween the groups narrows from about 1% to less than 0.5%. The dashed black verticalline indicates that two year have past since the adverse event, representing the approximateend of the Statute of Limitation period, after which, about 95% of the treated physicianswere contacted and notified that they face a medical malpractice lawsuit. After the end ofthe first two years following the adverse event, the average rate of the treated physiciansis growing closer the county average and four years after the adverse event, they are aboutequal.

In order to estimate the long-run effects of the medical malpractice litigation process,I estimate the average difference between the treatment and the control group for the fivesix-months periods prior to the adverse event, and for the five six-months periods startingafter the end of two years following an adverse event. The results are summarized in column(1) of Table 1.5. In the period prior to the adverse event there is a statistically significant gapof 1 percentage point between average C-section rate of the treated physicians and averagecounty C-section rate. In the 5 six-months period starting two years after the adverse eventthere is a very small and statistically insignificant gap between average C-section rate of thetreated physicians and average county C-section rate.

In order to make inference based on the estimation result using this method, one mustassume that the treatment and control group would have followed a similar trend in C-sectionrates if it weren’t for the adverse event. However, it is apparent from Figure 1.7a that thepre treatment C-section rates of the physicians who experience a lawsuit are lower thanthose of the control group, suggesting that this assumption may not hold. To address thisconcern, I restrict the control group to include only physicians who have similar experience,defined as a maximum 3 year gap from the treated physician, and physicians who use similarpractice patterns, defined as less than 10 percentage point gap in average C-section ratein the pre adverse event period. This restriction leaves 251 physicians for whom there is acontrol group. The results of this refinement are summarized in Figure 1.7b and column (2)

9

of Table 1.5. Figure 1.7b shows that prior to the adverse event, C-section rates for physiciansin the treatment and the control group are quite similar. Following the adverse event, thereis a jump in C-section rates of the treated physicians, and 4 years after the adverse event,C-section rates of the treatment group are roughly 2 percentage points higher than thoseof the control group. Column (2) of Table 1.5 confirms the visual impression, showing verysmall insignificant difference in C-section rates prior to the adverse event and a statisticallysignificant difference of 1.7 percentage points between the treatment and control group, in the5 six-months period after the end of two years following an adverse event. These estimatessupport the preceding analysis suggesting a long-run increase in C-section rates of roughly1.5 percentage point.

An Event Study Approach

The second strategy I am taking in order to evaluate the long-run effect of an adverse eventis an event study approach. The main estimation equation is

C-sectionjit = α + δktimekit + β1physiciani + β2quartert + β3Charjit + εjit (1.3)

In (1.3) , physician is a vector of physician dummies, quarter is a vector of quarter dummiesand Char is a vector of mother personal characteristics. The variables of interest are theevent time indicator variables, time, a vector of dummies for the number of elapsed quarterssince the adverse event, k ∈ {−10, ... − 1, 0, ...9}. The indicator variable timekit = 1 if aphysician i experienced an adverse event in quarter t− k. For example, time05,1995Q3 = 1, ifphysician 5 experienced an adverse event in the third quarter of 1995. Using this specification,δk is the effect of an adverse event k period following its occurrence.

Results. Figure 1.8 plots the estimates of δk in equation (1.3). The thin vertical linesreport the 95% confidence interval of the estimates. Columns (1) and (2) of Table 1.6 reportthe estimates and standard errors of the baseline specification, including physician quarterfixed effect, and of the full controls specification, adding mother characteristics, respectively.Figure 1.8 shows that before the adverse event the estimates of δk are not statisticallydifferent from zero. Immediately following the adverse event there is a jump of roughly 1percentage point in C-section rates. C-section rates continue to increase and the estimate ofδ9, which estimates the effect of the adverse event on C-section rates 2.5 years following theadverse event, is roughly 2.5 percentage points.

1.4.4 The Causal Role of Litigation

While fear of damage to reputation, amplified by an impending law suit may explain physi-cians’ response to the adverse event, another interpretation of the results might be that theincrease in C-section rates after the adverse event is a response to the negative outcome thatbrought about the malpractice claim (e.g. an emotional response which changes treatmentpatterns, see Redinbaugh et al. [2003]). I use the heterogeneity in the data to compare the

10

short-run response to the adverse event in subsets of claim types, patients and physicians,in order to examine if the response is caused, at least in part, by fear of harm to reputation,amplified by the impending lawsuit.

Which type of claims do physicians respond more to? I exploit the fact that while 69% ofthe claims in the adverse event panel were successful and resulted in a positive payment (paidclaims), 31% of the claims were unsuccessful, and ultimately resulted in zero payment (non-paid claims). Under the assumption that immediately after the adverse event, physicianscan predict whether a lawsuit is expected or not, claims that are unsuccessful ex-post, areless likely to be perceived as harmful for reputation ex-ante. Consequently, if the fear oflitigation is responsible for the increase in C-section rates, a weaker response is expectedamong physicians who experienced an unsuccessful claim. A comparison between the severityof adverse outcomes of successful and unsuccessful claims shows that they are quite similar(e.g. in both groups the proportion of deaths is roughly 24%), therefore, if physiciansrespond only to the negative outcome that brought about the claim, one would expect asimilar response in successful and unsuccessful claims.

Figures 1.9a and 1.9b display average per period C-section rates, among paid and non-paid claims respectively. Figure 1.9a shows that for paid claims, following the adverse event,there is a discernible jump of roughly 1.5 percentage points, implying a 6% jump in averageC-section rates. Yet, in the case of non-paid claims there is no apparent discontinuity inC-section rates. Consistent with the graphical evidence, the estimates in column (1) ofTable 1.7 show, for paid claims, a statistically significant coefficient of about 1.4 percentagepoints that reflects an increase of 6% in C-section rates. Adding additional covariates incolumns (2) and (3) of Table 1.7, resulted in statistically significant coefficients of 1 and0.9 percentage points, respectively. In non-paid claims, the coefficient estimates in columns(4)-(6) of 1.7 are small and statistically insignificant, supporting the interpretation that thefear of litigation plays a role in physicians’ response.

The Role of Patients’ Socioeconomic Status. Another way to learn about the causalrelationship between the threat of a lawsuit and treatment patterns is by analyzing physi-cians’ response by patients’ socioeconomic status. There are two main reasons to think thatphysicians are more responsive to a threat of a lawsuit in patients of a high socioeconomicstatus then in patients of a low socioeconomic status. First, patients of a high socioeconomicstatus are thought to have better access to the legal system (Burstin et al. [1993]), makingthem more likely to sue in the case of subsequent adverse events. Second, patients of ahigh socioeconomic status are considered to be less constrained in their choice of prenatalphysicians (Hoerger and Howard [1995]), magnifying the reputational aspects of a lawsuit.I test the hypotheses that physicians are more responsive to an adverse event when treatinghigher socioeconomic status patients, by using the mothers’ type of insurance carrier. SinceMedicaid is a means based program, the population of mothers under Medicaid is likely tobe of a lower socioeconomic status relative to the population of mothers insured by a privatecarrier.

Figures 1.10a and 1.10b plot average C-section rates for mothers under private insurance

11

and Medicaid respectively, and Table 1.8 presents the estimation results similar to the speci-fications in Table 1.4, using one regression with separate coefficient for the two patient types.Figure 1.10 shows a clearly discernibly jump in average C-section rates in mothers under pri-vate insurance of about 1.5 percentage points, which implies an increase in C-section ratesof about 5% . For mothers under Medicaid there appears to be a much smaller increasein C-section rates. The estimates in column (1) of Table 1.8 show, for privately insuredmothers, a statistically significant increase of 1.6 percentage points. Column (2) shows astatistically significant coefficients of 0.91 and Column (3) show an insignificant coefficientof 0.76 percentage points. The estimation results for mothers under Medicaid show smallerand insignificant coefficients.

The Role of Experience. Little is known about the interaction between physicians’ ex-perience and the effect of litigation on practice patterns. Highly experienced physicians arelikely to have a more established reputation and thus, it is expected that they would be lesssensitive to the reputational implications of an adverse event. According to an interpretationof the effect of the adverse event as a response to patients’ negative outcomes, other thingsbeing equal, experienced physicians are more likely to be exposed to prior adverse events,making them less sensitive to subsequent adverse events. I study if as the two interpreta-tions predict, less experienced physicians respond more to the adverse event, by exploitinginformation on physicians’ educational history. Experience is defined as the time betweenthe beginning of residency and the adverse event. This measure of experience was chosenover other options for two reasons: (1) it is available for the largest number of physiciansin the sample, thereby leading to the lowest loss of estimation power; (2) other measures,like the year a physician began to practice, leave more room for interpretation and thusinclude some outliers. This experience measure is available for 448 of the 459 physicians inthe adverse event panel. As Figure 1.11 shows, the distribution of experience has a medianof 16 years, thus I split the physician sample into two groups: high-experience, with morethan 16 years of experience at the time of the adverse event, and low-experience, with 16years of experience or less at the time of the adverse event.

Figures 1.12a and 1.12b show average per period C-section rates over the five years of theadverse event panel for the low-experience and high-experience physicians respectively, andTable 1.9 displays the estimation results. Figure 1.12a shows a jump of roughly 2 percentagepoints in average C-section rates for low-experience physicians. In Figure 1.12b there is noapparent change in C-section rates following the adverse event, for high-experience physi-cians. The estimation results in Column (1) of Table 1.9 show that average C-section ratesfor low-experience physicians jumped by 2.3 percentage points, reflecting a jump of roughly9% in C-section rates. The estimates in columns (2) and (3) show a statistically significantincrease of 1.8 and 1.9 percentage points respectively. For high-experience physicians, theestimation results in column (4) of Table 1.9 show, consistent with the graphical evidence,a very small and statistically insignificant increase in C-sections.

The results show that, as both interpretations of the effect of the adverse event predict,the response is concentrated among low-experience physicians. In order to try to distinguish

12

between the two interpretations it would be interesting to test if low-experience physiciansresponded more to the adverse event because they have less exposure to prior adverse events.To do this, I divide physicians into two groups, those with prior litigation experience andthose with no prior litigation experience. A stronger response in the group with no priorlitigation history would indicate that physicians respond to the negative outcome, and priorexposure to adverse events generates a weaker response. It is important to note that physi-cians may have been exposed to bad outcomes which did not result in litigation and hence,are not recorded in the data, and therefore this test should not be viewed as conclusive aboutthe nature of the interaction of the response with experience.

Average C-section rates for the two groups are plotted in Figures 1.13aand 1.13 and Table1.10 shows the estimation results. Examining Figure 1.13 , it appears that in both groupsthere is a jump in C-section rates of roughly one percentage point. Column (1) of Table1.10 shows a 1 percentage point increase, but the estimates are not statistically significant.The estimates in Columns (2) and (3) of Table 1.10 are 0.9 and 0.7 respectively and bothare not statistically significant. Columns (4) of Table 1.10 shows that for physicians withprior claim history there is a statistically significant increase of 1.3 percentage point. Theestimates in columns (5) and (6) of Table 1.10 both show a statistically insignificant ofabout 0.8 percentage points. The evidence suggest no interaction between the response tothe adverse event and prior claim history.

1.5 Response to News About a Lawsuit

Medical malpractice insurance policies typically require to report any incident in which apatient may be considering filing a claim, promptly following the incident. Figure 1.14 showsa histogram of the frequencies of the time between the adverse event and the report to theinsurer. 95% of the claims are reported less than two years following the adverse event,consistent with the Statute of Limitation which typically places a time limit of two yearsfrom the adverse event on pursuing a legal remedy. Assuming that physicians are likely tocomply with this requirement, the time of report of a claim to the insurer coincides with thetime physicians are first contacted regarding a claims and learn that they are likely to befacing a lawsuit. Hence, I use a similar analysis, using the time of first contact as a proxy fornews about a malpractice claim, in order to study the effect of fear of litigation separatelyfrom the response to an adverse event. I concentrate on malpractice claims reported morethan a year after the adverse event, for which the time of report is far enough from theadverse event and thus likely to be a better proxy for news about a lawsuit.

1.5.1 Selection Around a First Contact

As in the case of the adverse event, it is important to check whether there was a change inthe patient sample composition around the time of the first contact. Figure 1.15a plots per

13

period birth numbers, showing no apparent change in the sample size around the adverseevent. Figure 1.15b plots the per period rate of high-risk mothers in the first contact panel.It shows that the rate of high risk mothers is quite smooth around the report of the adverseevent. Figure 1.15c Plots the mean age of mothers in the first contact panel. It shows anupward trend in the mean age of mothers over time, but there is no sharp discontinuity inthe period following the adverse event. Finally, Figure 1.15d plots the per period share ofprivately insured mothers, showing no sharp change around the first contact. Table 1.11confirms these observations, indicating no finding of a change is the risk level of mothersaround the time of first contact.

Here too, I fit a linear probability model, estimating the effect of age, each of the high-risk factors and insurance type, on the probability of undergoing a C-section in the pre-eventperiod. I plot the average predicted C-section rates for the entire period of the panel in Figure1.16. The Figure shows no apparent jump in predicted C-section rates, as the estimates inTable 1.12 confirm.

1.5.2 Response to a First Contact

Figure 1.17 plots average per period C-section rates for all claims in the first contact panelthat were reported more than a year after the adverse event. Similarly to the analysisabove, a quadratic regression model fit is added separately to the periods before and afterthe time of report. The figure shows a modest increase in C-section rates following thefirst contact. The corresponding estimates are displayed in Table 1.13. Column (1) reportsthe baseline regression estimations. The estimation results show a statistically insignificantincrease of about 0.5 percentage points in C-section rates following the first contact regardinga malpractice claim. In columns (2) and (3), additional covariates are added analogous tothe analysis of the adverse event panel, showing similar results.

In the previous section I found that physicians respond by considerably increasing C-section rates following an adverse event in cases for which the adverse event is likely tobe followed by a lawsuit. That is, when a claim is anticipated, the response preemptsthe claim itself. It may not surprising then, that when physicians are contacted about ananticipated claim, there is no evidence of a change in practice patterns. In order to measurethe effect of news regarding a lawsuit on treatment, one needs to isolate claims which werenot anticipated by physicians. Assuming that successful claims are more anticipated thanunsuccessful claims, the report of unsuccessful claims is more likely to reflect “bad news”.Thus, I study the response of physician to “bad news” about a lawsuit by splitting thesample, once more, to paid and non-paid claims.

Figures 1.18a and 1.18b plot average C-section rates for paid and non-paid claims, re-spectively. Figures 1.18a and 1.18b indicate that for non-paid claims there is a discontinuityis C-section rates after the time of the first contact and for paid claims there no paralleljump. The estimation results in columns (1)-(6) of Table 1.14 show an increase in C-sectionrates of about 2 percentage points for non-paid claims which are boarder line insignificant

14

(P-values of 0.143-0.064 ) and very small and insignificant estimates for paid claims. Theseresults suggest that physicians respond to new information about a lawsuit. However, onecan not rule out the hypotheses that physicians do not respond to news about a lawsuit.

1.6 Response of Peers to an Adverse Event

A further question that arises in this context is whether physicians’ lawsuits affect their peers.The entire department or hospital may change treatment patterns in response to a lawsuit,or, possibly close colleagues who work together with the treated physician may change theirpractice patterns. I address this question by using the same method that was used to studythe short-run response to the adverse event in the previous sections. In this section, todecrease computation time, inpatient discharge data is aggregated and transformed from apatient-discharge unit of observation to a physician-quarter unit of observation. I generatea five year balanced panel similar to the adverse event Panel, but instead of including thetreated physicians, all the physicians who are affiliated with the hospital of the treatedphysician (like in the previous section, affiliation to a hospital is defined as the hospital inwhich most of a physician’s deliveries are performed) and appear throughout the relevant 5year time period are included (“peer panel”). This generates a data set of 45,440 observationswhich include 558 physicians, some of whom appear more than once in the sample.

I first examine if, there a hospital-wide response to the adverse event. Figure 1.19 depictsaverage hospital C-section rates 10 periods before and 10 periods after the adverse event. Thefigure shows that there is an upward trend in C-section rates in the peer panel but it appearsto be smooth around the adverse event, implying no evidence of hospital-wide response to theadverse event. The estimates in column (1) of Table 1.15 confirm this impression, showinga statistically insignificant coefficient of 0.001.

Next I test whether the adverse event affects treated physicians’ close peers. In order tostudy this question, I generate a measure of “closeness” between physicians. The measureis created based on three parameters: physician experience, patient geographical proximityand patient socioeconomic status. Geographic location is calculated using the most frequentzip-code among each physician’s patients. This location is used in order to calculate thedistance between the main practice area of the treated physician and each of his colleagues.The rate of mothers under Medicaid that were treated by the physician prior to the time ofadverse event is used as a measure of patients’ socioeconomic status. Close peers are definedas peers who work at the same hospital with similar experience (a difference of no more than3 years), who treat patients in adjacent neighborhoods (up to a 10 miles distance), and withsimilar socioeconomic background (difference in the rate of mother under medicaid is lessthan 0.4). Figure 1.20 plot average C-section rates, for close and remote peers respectively.In Figure 1.20a there is a discernible jump of about 1 percentage point in C-section rateswhile in Figure 1.20b, the figure appears to be smooth around zero. Columns (2)-(3) ofTable 1.15, show a statistically significant coefficient of 1.3%, for close peers, and, a small

15

and insignificant coefficient for remote peers. These results suggest that while there is nohospital-wide response to the adverse event, close peers - possibly close colleagues who worktogether with the treated physician tend to perform higher rates of C-section following theadverse event.

1.7 Conclusions

Despite its central importance to the design of malpractice law, very little is known about theeffect of malpractice law on medical treatment, let alone about its underlying mechanisms.This study investigated the role of physicians’ exposure to malpractice litigation on medicaltreatment patterns. In addition to rigorously measuring the effect of an adverse event onC-section rates, this study sheds light on the mechanisms that underlie the response.

Following an adverse event physicians increase C-section rates by roughly 4% and thiseffect persists for at least 4.5 years following an adverse event. These findings are importantbecause they show that medical malpractice litigation encourages physicians to adopt moreconservative and costly treatment strategies. The role played by fear of litigation in elicitingphysicians’ response is supported by several additional findings. First, adverse events whichultimately result in an unsuccessful claims, and are perceived at the time of the adverseevent as less likely to harm physicians’ reputation, do not lead to an increase in C-sectionrates. Second, physicians increase C-section rates more in privately insured patients, whoare likely to have better access to the legal system as well as more freedom to choose theirphysician. In addition, the observed increase in C-section rates following an adverse eventis concentrated among less experienced physicians. However, responses of physicians withand without prior malpractice claim histories are not significantly different. These resultsare inconsistent with a theory of response to the negative outcome which brought about theadverse event.

The evidence in this paper suggest that, potential reputation loss following a malpracticeclaim, leads to a change in physician treatment patterns, possibly resulting in excessivelyconservative behavior. In future work it would be interesting to analyze optimal medicalmalpractice law in the light of these finding. In addition, it would be interesting to extendthe analysis to other areas of health care provision and test whether these results apply toother aspects of healthcare provision.

16

17

Table 1.1: Summary Statistics - Inpatient Data: All Sample, Adverse Event and FirstContact Panels

Full Sample Adverse Event Panel Reported Panel(1) (2) (3)

Age (median) 27 27 27Mother Hispanic 18.8% 12.4% 13.2%Mother African American 21.3% 18.5% 19.2%Mother Other Race 59.9% 69.2% 67.7%Anemia 8.4% 7.6% 7.5%Breech Position 3.5% 3.6% 3.5%Diabetes 0.7% 0.6% 0.6%Early Onset 7.5% 6.7% 6.8%Hemmorage 1.9% 1.8% 1.8%Hypertension 4.8% 4.3% 4.4%Multiple Gestation 1.1% 1.1% 1.1%Obesity 0.3% 0.2% 0.5%Oligohydramnios 2.4% 1.8% 1.9%Distress 3.3% 3.9% 3.3%Polyhydramnios 0.6% 0.5% 0.6%Previous C-section 14.1% 13.1% 13.4%Medicaid 41.7% 34.7% 35.5%Commercial 48.4% 56.9% 56.2%Physician # 2,307 459 494Observations 2,981,742 403,336 434,771

NOTE. Table entries are means unless otherwise noted. Column (1) includes all the deliveriesin the Florida Inpatient Data in the years 1992-2008. Column (2) includes all the deliveriesin the adverse event panel. Column (3) includes all the deliveries in the first contact Panel.

18

Tab

le1.

2:Sel

ecti

onA

round

anA

dve

rse

Eve

nt

-H

igh-R

isk,

Age

and

Insu

rance

Typ

e

Age

Hig

h-R

isk

Pri

vate

Insu

rance

Bas

elin

eC

ontr

ols

Bas

elin

eC

ontr

ols

Bas

elin

eC

ontr

ols

(1)

(2)

(3)

(4)

(5)

(6)

Eve

nt

Dum

my

-0.0

018

-0.0

013

0.01

600.

0107

0.00

810.

0086

(0.0

050)

(0.0

050)

(0.0

200)

(0.0

725)

(0.0

084)

(0.0

082)

2nd

Ord

erY

ear

Quar

ter

Pol

ynom

ial

no

yes

no

yes

no

yes

Num

ber

ofP

hysi

cian

s45

945

945

945

945

945

9O

bse

rvat

ions

403,

336

403,

336

403,

336

403,

336

403,

336

403,

336

NO

TE

.A

llco

lum

ns

rep

ort

esti

mat

esof

model

sak

into

the

bas

elin

em

odel

spec

ified

ineq

uat

ion

(1),

repla

cing

C-s

ecti

onw

ith

hig

h-r

isk,

age

and

shar

eof

pri

vate

lyin

sure

dm

other

s.H

igh-r

isk

incl

udes

the

follow

ing

condit

ions:

Pre

vio

us

C-s

ecti

on,

bre

ech

pos

itio

n,

mult

iple

gest

atio

n,

hyp

erte

nsi

on,

earl

yon

set,

hem

orrh

age,

obes

ity,

dia

bet

es,

pol

yhydra

mnio

s,ol

igoh

ydra

mnio

san

ddis

tres

s.C

olum

ns

(2),

(4)

and

(6)

add

aquad

rati

cp

olynom

ialfo

rth

ere

leva

nt

under

lyin

gquar

ter.

Sta

ndar

der

rors

clust

ered

by

physi

cian

show

nin

par

enth

esis

.

19

Table 1.3: Selection Around an Adverse Event - High-Risk, Age and Insurance Type

predicated C-section

Event Dummy 0.0030(0.0026)

Number of Physicians 459Observations 403,336

NOTE. The table reports estimates of a models akin to the baseline model specified inequation (1), replacing C-section with predicted C-section. Standard errors clustered byphysician shown in parenthesis.

20

Tab

le1.

4:Shor

t-R

un

Eff

ect

ofan

Adve

rse

Eve

nt

Bas

elin

eB

asic

Con

trol

sF

ull

Con

trol

s(1

)(2

)(3

)

Eve

nt

Dum

my

0.01

150.

0090

0.00

74(0

.004

7)(0

.003

9)(0

.003

6)P

atie

nt

Char

acte

rist

ics

no

yes

yes

Physi

cian

,Q

uar

ter

FE

no

no

yes

Num

ber

ofP

hysi

cian

s45

945

945

9O

bse

rvat

ions

403,

336

403,

336

403,

336

NO

TE

.A

llco

lum

ns

rep

ort

esti

mat

esof

model

sak

into

the

bas

elin

em

odel

spec

ified

ineq

uat

ion

(1).

Col

um

n(2

)in

cludes

,in

addit

ion

toth

ebas

elin

esp

ecifi

cati

on,

aquad

rati

cp

olynom

ial

for

age,

dum

my

vari

able

sfo

rra

cean

dfo

rpat

ients

’co

ndit

ions

asfo

llow

s:pre

vio

us

C-s

ecti

on,

bre

ech

pos

itio

n,

mult

iple

gest

atio

n,

hyp

erte

nsi

on,

earl

yon

set,

hem

orrh

age,

obes

ity,

dia

bet

es,

pol

yhydra

mnio

s,ol

igoh

ydra

mnio

s,an

emia

,dis

tres

san

dfe

to.

Col

um

n(3

)ad

ds

physi

cian

and

quar

ter

fixed

effec

ts.

Sta

ndar

der

rors

clust

ered

by

physi

cian

show

nin

par

enth

esis

.

21

Table 1.5: Matching Physician to Same-County-Colleagues, 7 Year Adverse Event Panel

τt τtAll Physicians Close Match

Time of Event (Years) (1) (2)

Pre event period -0.0099 -0.0011(0.0024) (0.0023)

2.5-4.5 years after adverse event -0.0030 0.0170(0.0028) (0.0030)

F-test for equality of Refrom Effect (Prob¿F) 0.0006 0.0000# of Physicians 338 251

NOTE. Column 1 includes all the colleagues from the same county excluding colleaguesfrom the same hospital. Column 2 includes all colleagues from the same county with lessthan a 3 year gap in experience and less than 10 percentage points gap in C-section rate inthe 10 quarters pre-event period. Pre-event period is defined as the 5 six-months periodsprior to the adverse event. 2.5-4.5 years after adverse event is the 5 six-months periodsstarting 2 years following the adverse event. Standard errors are calculated usingbootstrapping, analytic asymptotic variance estimator (Abadie and Imbens (2006)) showvery similar results.

22

Table 1.6: Long-Run Effect of an Adverse Event, Event Study Approach

Baseline Full Controls

coeffiecient Standard Errors coeffiecient Standard Errors

period -10 -0.0001 (0.0057) 0.0060 (0.0060)period -9 0.0008 (0.0059) 0.0021 (0.0059)period -8 -0.0003 (0.0055) -0.0038 (0.0053)period -7 -0.0028 (0.0053) -0.0028 (0.0053)period -6 -0.0018 (0.0051) 0.0000 (0.0046)period -5 0.0061 (0.0047) 0.0043 (0.0043)period -4 -0.0020 (0.0045) -0.0052 (0.0038)period -3 0.0025 (0.0044) -0.0002 (0.0038)period -2 0.0009 (0.0042) 0.0014 (0.0034)period -1period 0 0.0111 (0.0043) 0.0067 (0.0036)period 1 0.0127 (0.0045) 0.0083 (0.0037)period 2 0.0139 (0.0045) 0.0109 (0.0038)period 3 0.0132 (0.0048) 0.0115 (0.0043)period 4 0.0149 (0.0050) 0.0131 (0.0045)period 5 0.0120 (0.0056) 0.0089 (0.0053)period 6 0.0220 (0.0056) 0.0158 (0.0054)period 7 0.0144 (0.0056) 0.0106 (0.0056)period 8 0.0256 (0.0057) 0.0211 (0.0060)period 9 0.0248 (0.0060) 0.0206 (0.0061)Quarter & Physician FE Yes Yes Yes YesNumber of Physicians 459 459 459 459Observations 403,336 403,336 403,336 403,336

NOTE. Full controls includes, in addition to the baseline specification, a quadratic polynomial forage, dummy variables for race and for patients conditions as follows: previous C-section, breechposition, hypertension, early onset, hemorrhage, oligohydramnios. Standard errors clustered byphysician.

23

Tab

le1.

7:Shor

t-R

un

Eff

ect

ofan

Adve

rse

Eve

nt,

Succ

essf

ul

and

Unsu

cces

sful

Cla

ims

Succ

essf

ul

Unsu

cces

sful

Bas

elin

eB

asic

Con

trol

sF

ull

Con

trol

sB

asel

ine

Bas

icC

ontr

ols

Full

Con

trol

s(1

)(2

)(3

)(4

)(5

)(6

)

Eve

nt

Dum

my

0.01

450.

0103

0.00

870.

0049

0.00

560.

0042

(0.0

053)

(0.0

044)

(0.0

042)

(0.0

091)

(0.0

076)

(0.0

068)

Pat

ient

Char

acte

rist

ics

no

yes

yes

no

yes

yes

Physi

cian

&Q

uar

ter

FE

no

no

yes

no

no

yes

Num

ber

ofP

hysi

cian

s31

631

631

614

314

314

3O

bse

rvat

ions

283,

198

283,

198

283,

198

120,

138

120,

138

120,

138

NO

TE

.Succ

essf

ul

clai

ms

are

clai

ms

whic

hre

sult

edin

apay

men

tla

rger

than

zero

.U

nsu

cces

sful

clai

ms

resu

lted

ina

zero

pay

men

t.A

llco

lum

sre

por

tes

tim

ates

ofm

odel

sak

into

the

bas

elin

em

odel

spec

ified

ineq

uat

ion

(1).

Col

um

ns

(2)

and

(4)

incl

ude

inad

dit

ion

toth

ebas

elin

esp

ecifi

cati

ona

quad

rati

cp

olynom

ial

for

age,

dum

my

vari

able

sfo

rra

cean

dfo

rpat

ients

’co

ndit

ions

-se

enot

esfo

rta

ble

(2).

Col

um

ns

(3)

and

(6)

add

quar

ter

and

physi

cian

fixed

effec

ts.

Sta

ndar

der

rors

clust

ered

by

physi

cian

show

nin

par

enth

esis

.

24

Tab

le1.

8:Shor

t-R

un

Eff

ect

ofan

Adve

rse

Eve

nt,

Pri

vate

Insu

rance

Mot

her

san

dM

edic

aid

Mot

her

s

Pri

vate

(Pan

elA

)M

edic

aid

(Pan

elB

)

Bas

elin

eB

asic

Con

trol

sF

ull

Con

trol

sB

asel

ine

Bas

icC

ontr

ols

Full

Con

trol

s(1

)(2

)(3

)(1

)(2

)(3

)

Eve

nt

Dum

my

0.01

630.

0091

0.00

760.

0042

0.00

720.

0057

(0.0

057)

(0.0

049)

(0.0

048)

(0.0

075)

(0.0

060)

(0.0

054)

Pat

ient

Char

acte

rist

ics

no

yes

yes

no

yes

yes

Physi

cian

Quar

ter

FE

no

no

yes

no

no

yes

Num

ber

ofP

hysi

cian

s45

945

945

945

945

945

9O

bse

rvat

ions

369,

739

369,

739

369,

739

369,

739

369,

739

369,

739

NO

TE

.P

anel

Are

por

tses

tim

ates

for

the

even

tdum

my

inte

ract

edw

ith

apri

vate

insu

rance

dum

my.

Pan

elB

rep

orts

esti

mat

esof

the

even

tdum

my

inte

ract

edw

ith

aM

edic

aid

dum

my.

All

colu

mns

rep

ort

esti

mat

esof

model

sak

into

the

bas

elin

em

odel

spec

ified

ineq

uat

ion

(1),

incl

udin

gti

me

and

tim

esq

uar

edin

tera

cted

wit

hin

sura

nce

typ

e.C

olum

n(2

)in

cludes

inad

dit

ion

toth

ebas

elin

esp

ecifi

cati

ona

quad

rati

cp

olynom

ial

for

age,

dum

my

vari

able

sfo

rra

cean

dfo

rpat

ients

condit

ions

-se

enot

esfo

rta

ble

(2).

Col

um

n(3

)ad

ds

physi

cian

and

quar

ter

fixed

effec

ts.

Sta

ndar

der

rors

clust

ered

by

physi

cian

show

nin

par

enth

esis

.

25

Tab

le1.

9:Shor

t-R

un

Eff

ect

ofan

Adve

rse

Eve

nt,

Hig

h-E

xp

erie

nce

and

Low

-Exp

erie

nce

Physi

cian

s

Low

-Exp

erie

nce

Hig

h-E

xp

erie

nce

Bas

elin

eB

asic

Con

trol

sF

ull

Con

trol

sB

asel

ine

Bas

icC

ontr

ols

Full

Con

trol

s(1

)(2

)(3

)(4

)(5

)(6

)

Eve

nt

Dum

my

0.02

320.

0185

0.01

880.

0025

0.00

04-0

.002

7(0

.006

6)(0

.005

4)(0

.005

1)(0

.006

7)(0

.005

5)(0

.004

9)P

atie

nt

Char

acte

rist

ics

no

yes

yes

no

yes

yes

Physi

cian

Quar

ter

FE

no

no

yes

no

no

yes

Num

ber

ofP

hysi

cian

s24

624

624

620

220

220

2O

bse

rvat

ions

203,

255

203,

255

203,

255

189,

767

189,

767

189,

767

NO

TE

.E

xp

erie

nce

counte

das

ofth

eb

egin

nin

gof

resi

den

cy.

Hig

h-E

xp

erie

nce

:m

ore

than

16ye

ars

ofex

per

ience

,L

ow-E

xp

erie

nce

:16

year

sor

less

ofex

per

ience

.A

llco

lum

ns

rep

ort

esti

mat

esof

model

sak

into

the

bas

elin

em

odel

spec

ified

ineq

uat

ion

(1).

Col

um

ns

(2)

and

(4)

incl

ude,

inad

dit

ion

toth

ebas

elin

esp

ecifi

cati

on,

aquad

rati

cp

olynom

ial

for

age,

dum

my

vari

able

sfo

rra

cean

dfo

rpat

ients

’co

ndit

ions

-se

enot

esfo

rta

ble

(2).

Col

um

ns

(3)

and

(6)

add

aphysi

cian

and

quar

ter

fixed

effec

t.E

xp

erie

nce

isav

aila

ble

for

448

out

ofth

e45

9physi

cian

sin

the

adve

rse

even

tpan

el.

Sta

ndar

der

rors

clust

ered

by

physi

cian

show

nin

par

enth

esis

.

26

Tab

le1.

10:

Shor

t-R

un

Eff

ect

ofan

Adve

rse

Eve

nt,

Physi

cian

sW

ith

and

Wit

hou

tP

rior

Cla

imH

isto

ry

Fir

stC

laim

≥Sec

ond

Cla

im

Bas

elin

eB

asic

Con

trol

sF

ull

Con

trol

sB

asel

ine

Bas

icC

ontr

ols

Full

Con

trol

s(1

)(2

)(3

)(4

)(5

)(6

)

Eve

nt

Dum

my

0.01

040.

0094

0.00

700.

0131

0.00

840.

0088

(0.0

065)

(0.0

054)

(0.0

049)

(0.0

066)

(0.0

055)

(0.0

052)

Pat

ient

Char

acte

rist

ics

no

yes

yes

no

yes

yes

Physi

cian

Quar

ter

FE

no

no

yes

no

no

yes

Num

ber

ofP

hysi

cian

s25

325

325

320

620

620

6O

bse

rvat

ions

215,

614

215,

614

215,

614

187,

722

187,

722

187,

722

NO

TE

.A

llco

lum

ns

rep

ort

esti

mat

esof

model

sak

into

the

bas

elin

em

odel

spec

ified

ineq

uat

ion

(1).

Col

um

ns

(2)

and

(4)

incl

ude,

inad

dit

ion

toth

ebas

elin

esp

ecifi

cati

on,

aquad

rati

cp

olynom

ial

for

age

dum

my

vari

able

sfo

rra

cean

dfo

rpat

ients

’co

ndit

ions

-se

enot

esfo

rta

ble

(2).

Col

um

n(3

)an

d(6

)ad

dphysi

cian

and

quar

ter

fixed

effec

t.Sta

ndar

der

rors

clust

ered

by

physi

cian

show

nin

par

enth

esis

..

27

Tab

le1.

11:

Sel

ecti

onA

round

the

aF

irst

Con

tact

-H

igh-R

isk,

Age

and

Insu

rance

Typ

e

Age

Hig

h-R

isk

Pri

vate

Insu

rance

Bas

elin

eC

ontr

ols

Bas

elin

eC

ontr

ols

Bas

elin

eC

ontr

ols

(1)

(2)

(3)

(4)

(5)

(6)

Eve

nt

Dum

my

0.02

90-0

.097

5-0

.002

4-0

.002

7-0

.006

5-0

.006

0(0

.025

6)(0

.092

1)(0

.008

0)(0

.008

0)(0

.010

5)(0

.010

8)2n

dO

rder

Yea

rQ

uar

ter

Pol

ynom

ial

no

yes

no

yes

no

yes

Num

ber

ofP

hysi

cian

s23

223

223

223

223

223

2O

bse

rvat

ions

211,

951

211,

951

211,

951

211,

951

211,

951

211,

951

NO

TE

.F

irst

conta

ctsa

mple

incl

udes

clai

ms

rep

orte

dm

ore

than

aye

araf

ter

the

adve

rse

even

t.A

llco

lum

ns

rep

ort

esti

mat

esof

model

sak

into

the

bas

elin

em

odel

spec

ified

ineq

uat

ion

(1),

repla

cing

C-s

ecti

onw

ith

hig

h-r

isk,

age

and

shar

eof

pri

vate

lyin

sure

dm

other

s.H

igh-r

isk

incl

udes

the

follow

ing

condit

ions:

Pre

vio

us

C-s

ecti

on,

bre

ech

pos

itio

n,

mult

iple

gest

atio

n,

hyp

erte

nsi

on,

earl

yon

set,

hem

orrh

age,

obes

ity,

dia

bet

es,

pol

yhydra

mnio

s,ol

igoh

ydra

mnio

san

ddis

tres

s.C

olum

ns

(2),

(4)

and

(6)

add

aquad

rati

cp

olynom

ial

for

the

rele

vant

under

lyin

gquar

ter.

Sta

ndar

der

rors

clust

ered

by

physi

cian

show

nin

par

enth

esis

.

28

Table 1.12: Selection Around a First Contact - Predicted C-section Rates

predicated C-section

Event Dummy -0.0024(0.0042)

Number of Physicians 211,951Observations 232

NOTE. First contact sample includes claims reported more than a year after the adverseevent. The table report estimates of a models akin to the baseline model specified inequation (1), replacing C-section by predicted C-section. Standard errors clustered byphysician shown in parenthesis.

29

Tab

le1.

13:

Shor

t-R

un

Eff

ect

ofa

Fir

stC

onta

ct

Bas

elin

eB

asic

Con

trol

sF

ull

Con

trol

s(1

)(2

)(3

)

Eve

nt

Dum

my

0.00

500.

0081

0.00

52(0

.007

2)(0

.006

0)(0

.005

7)P

atie

nt

Char

acte

rist

ics

no

yes

yes

Physi

cian

Quar

ter

Fix

edE

ffec

tno

no

yes

Obse

rvat

ions

211,

951

211,

951

211,

951

Num

ber

ofP

hysi

cian

s23

223

223

2

NO

TE

.F

irst

conta

ctsa

mple

incl

udes

clai

ms

rep

orte

dm

ore

than

aye

araf

ter

the

adve

rse

even

t.A

llco

lum

ns

rep

ort

esti

mat

esof

model

sak

into

the

bas

elin

em

odel

spec

ified

ineq

uat

ion

(1).

Col

um

n(2

)in

cludes

inad

dit

ion

toth

ebas

elin

esp

ecifi

cati

ona

quad

rati

cp

olynom

ial

for

age

dum

my

vari

able

sfo

rra

cean

dfo

rpat

ients

’co

ndit

ions

asfo

llow

s:P

revio

us

C-s

ecti

on,

bre

ech

pos

itio

n,

mult

iple

gest

atio

n,

hyp

erte

nsi

on,

earl

yon

set,

hem

orrh

age,

obes

ity,

dia

bet

es,

pol

yhydra

mnio

s,ol

igoh

ydra

mnio

s,an

emia

,dis

tres

san

dfe

to.

Col

um

n(3

)ad

ds

physi

cian

and

quar

ter

fixed

effec

ts.

Sta

ndar

der

rors

clust

ered

by

physi

cian

show

nin

par

enth

esis

.

30

Tab

le1.

14:

Shor

t-R

un

Eff

ect

ofa

Fir

stC

onta

ct,

Succ

essf

ul

and

Unsu

cces

sful

Cla

ims

Succ

essf

ul

Unsu

cces

sful

Bas

elin

eB

asic

Con

trol

sF

ull

Con

trol

sB

asel

ine

Bas

icC

ontr

ols

Full

Con

trol

s(1

)(2

)(3

)(4

)(5

)(6

)

Eve

nt

Dum

my

-0.0

011

0.00

240.

0000

0.01

940.

0205

0.01

61(0

.008

6)(0

.007

2)(0

.007

0)(0

.013

1)(0

.010

9)(0

.009

4)P

atie

nt

Char

acte

rist

ics

no

yes

yes

no

yes

yes

Physi

cian

Quar

ter

FE

no

no

yes

no

no

yes

Num

ber

ofP

hysi

cian

s15

315

315

379

7979

Obse

rvat

ions

143,

329

143,

329

143,

329

68,6

2268

,622

68,6

22

NO

TE

.F

irst

conta

ctsa

mple

incl

udes

clai

ms

rep

orte

dm

ore

than

aye

araf

ter

the

adve

rse

even

t.Succ

essf

ul

clai

ms

are

clai

ms

whic

hre

sult

edin

apay

men

tla

rger

than

zero

.U

nsu

cces

sful

clai

ms

resu

lted

ina

zero

pay

men

t.A

llco

lum

ns

rep

ort

esti

mat

esof

model

sak

into

the

bas

elin

em

odel

spec

ified

ineq

uat

ion

(1).

Col

um

ns

(2)

and

(4)

incl

ude

inad

dit

ion

toth

ebas

elin

esp

ecifi

cati

ona

quad

rati

cp

olynom

ial

for

age,

dum

my

vari

able

sfo

rra

cean

dfo

rpat

ients

condit

ions

-se

enot

esfo

rta

ble

(2).

Col

um

ns

(3)

and

(6)

add

quar

ter

and

physi

cian

fixed

effec

ts.

Sta

ndar

der

rors

clust

ered

by

physi

cian

show

nin

par

enth

esis

.

31

Tab

le1.

15:

Shor

t-R

un

Eff

ect

ofan

Adve

rse

Eve

nt

onP

eers

from

the

Sam

eH

ospit

al

All

Hos

p/P

hys

Rem

ote

Pee

rsC

lose

Pee

rs(1

)(2

)(3

)

Eve

nt

Dum

my

0.00

10-0

.002

80.

0129

(0.0

036)

(0.0

024)

(0.0

060)

Num

ber

ofP

hysi

cian

s55

837

320

8O

bse

rvat

ions

45,4

4021

,160

6,96

0

NO

TE

.T

he

pee

rsa

mple

incl

udes

all

the

physi

cian

sfr

omth

esa

me

hos

pit

alw

ho

app

ear

thro

ugh

the

whol

e5

year

ssa

mple

per

iod.

Clo

seP

eers

are

pee

rsfr

omth

esa

me

hos

pit

al,

wit

hat

mos

ta

3ye

ars

exp

erie

nce

gap,

10m

ile

dis

tance

(usi

ng

mos

tco

mm

onpat

ient

zip-c

ode)

,an

d0.

4p

erce

nta

gep

oints

gap

inra

teof

Med

icai

dpat

ients

,in

the

pre

-eve

nt

per

iod.

rem

ote

pee

rsar

eth

eco

llea

gues

from

the

sam

ehos

pit

alw

ho

are

not

Clo

seto

trea

ted

Physi

cian

.Sta

ndar

der

rors

clust

ered

by

physi

cian

show

nin

par

enth

esis

.

32

Figure 1.1: C-section rates 1992-2008, Florida.

.2.2

5.3

.35

.4C

-se

ctio

n R

ate

1992q1 1996q1 2000q1 2004q1 2008q1Time (Quarters)

NOTE: The figure depicts C-section rates in Florida from 1992-Q1 to 2008-Q4. The sampleconsists of all deliveries in the Florida Hospital Inpatient Discharge Data in the relevanttime period.

33

Figure 1.2: Distribution of Physicians’ Prior Claims.

(a) Adverse Event Panel.

0.2

.4.6

Fre

que

ncy o

f P

hysic

ians

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15Prior Claim Number

(b) First Contact Panel.

0.1

.2.3

.4.5

0 1 2 3 4 5 6 7 8 9 10 11

Fre

que

ncy o

f P

hysic

ians

Prior Claim Number

NOTE: Panel A and B of this figure depict the frequency of physicians by the number ofclaims they experienced prior to the adverse event and the first contact respectively. Forexample, in Panel A, 253 of the 459 physicians in the sample did not experience prior claims.The Florida Medical Professional Liability Files were used to calculate the number of priorclaims for each physician.

34

Figure 1.3: Distribution of Claim Payments.

(a) Adverse Event Panel.

0.1

.2.3

.4

0 200000 400000 600000 800000 1000000

Fre

que

ncy o

f C

laim

s

Indemnity Paid

(b) First Contact Panel.

0.1

.2.3

.4

0 200000 400000 600000 800000 1000000

Fre

que

ncy o

f C

laim

s

Indemnity Paid

NOTE: Panels A and B of this figure show the frequency of claim payments rounded to theclosest multiple of $50K, in the adverse event and the first contact panels, respectively. TheFlorida Medical Professional Liability Files are used to generate the figures. Payments arein nominal terms.

35

Figure 1.4: Selection around Adverse Event.

(a) Per Period Number of Births.

17000

19000

21000

23000

Num

ber

of

Birth

s

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9Time from Adverse Event (Quarters)

(b) Per Period Share of High-Risk Mothers.

.28

.29

.3.3

1

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

Share

of H

igh

-Ris

k M

oth

er

Time from Adverse Event (Quarters)

(c) Per Period Average Age of Mothers.

27

27

.22

7.4

27

.6

Avera

ge M

oth

ers

A

ge

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9Time from Adverse Event (Quarters)

(d) Per Period Share of Privately Insured Mothers.

.54

.55

.56

.57

.58

.59

.6

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

Sha

re o

f P

rivate

ly insure

d M

oth

ers

Time from Adverse Event (Quarters)

NOTE: These figures show how observable characteristics evolve around the adverse event. Thevertical line denotes the time of the adverse event. Panels A-D plot the per-period number of births,Share of high-risk mothers, average mother age and share of privately insured mothers, respectively.

36

Figure 1.5: Selection on Observables, Adverse Event Panel.

.25

.255

.26

.265

.27

Pre

dic

ted C

-section r

ate

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9Time from Adverse Event (Quarters)

NOTE: This figure shows predicted C-section rates in the adverse event panel. The verticalline denotes the time of the adverse event. The prediction was done by regressing, usingOLS, C-section dummy on the high-risk covariates as well as age and insurance typedummies in the pre-reform period. The figure plots the average per period predictedC-section rate.

37

Figure 1.6: Short-Run Effect of an Adverse Event.

.24

.25

.26

.27

.28

.29

C-s

ectio

n r

ate

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

Time from Adverse Event (Quarters)

NOTE: The figure plots per period C-section rates in the adverse event panel. The verticalline denotes the time of the adverse event.

38

Figure 1.7: Long-Run Effect of an Adverse Event, Matching Approach.

(a) All Same County Colleagues.

.25

.27

.29

.31

.33

.35

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 4

Statute of Limitation About Here

C-s

ection r

ate

Time from Adverse Event (Years)

(b) Similar Same County Colleagues.

.25

.27

.29

.31

.33

.35

-2.5 -2 -1.5 -1 -.5 0 .5 1 1.5 2 2.5 3 3.5 4

C-s

ection r

ate

Time from Adverse Event (Years)

Statute of Limitation About Here

NOTE: Panels A and B of this figure depict C-section rate in six-months periods 2.5 years beforeand 4.5 years after the adverse event. The control group in panel A is comprised by physiciansfrom the same county excluding physicians from the same hospital. The Control group in panel Bis a subgroup of the Control group in panel A, including only physicians with similar experienceand similar pre-event C-section rates. The vertical red line denotes the time of the adverse eventand the vertical dashed black line denotes the approximate end of the Statute of Limitation.

39

Figure 1.8: Long-Run Effect of an Adverse Event, Event Study Approach.

-.02

-.01

0.0

1.0

2.0

3.0

4

C-S

ection R

ate

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

Time from Adverse Event (Quarters)

NOTE: This figure plots the coefficients of dummies for time from injury, obtained from anOLS regression with controls for physician and quarter fixed effects (equation 2). The thinvertical red lines report the 95% confidence interval of the coefficients. Table 1.6 reportsthe coefficients and standard errors shown in the figure.

40

Figure 1.9: Short-Run Effect of an Adverse Event by claim Success.

(a) Successful Claims.

.24

.25

.26

.27

.28

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ection r

ate

Time from Adverse Event (Quarters)

(b) Unsuccessful Claims.

.26

.27

.28

.29

.3.3

1

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ection r

ate

Time from Adverse Event (Quarters)

NOTE: Panels A and B of this figure depict per-period C-section rates for Paid and Non-Paid claims respectively, in the adverse event panel. The vertical red line denotes the timeof the adverse event.

41

Figure 1.10: Short-Run Effect of an Adverse Event by Insurance Type.

(a) Private Insurance Mothers.

.26

.27

.28

.29

.3.3

1

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ection r

ate

Time from Adverse Event (Quarters)

(b) Mediciad Mothers.

.23

.24

.25

.26

.27

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ection r

ate

Time from Adverse Event (Quarters)

NOTE: Panels A and B of this figure depict C-section rates for privately insured mothersand Medicaid mothers, respectively, in the adverse event panel. The vertical red line denotesthe time of the adverse event.

42

Figure 1.11: Distribution of Physicians’ Experience, Adverse Event Panel.

0.0

5.1

.15

Fre

qu

en

cy o

f P

hysic

ian

s

0 4 8 12 16 20 24 28 32 36 40 44 48Years of Experience

NOTE: This figure depicts frequency of physicians by years of experience. Experience isdefined as time from the beginning of residency. The figure uses the Profile Data and theFlorida Medical Professional Liability Files to calculate the level of experience at the timeof the adverse event.

43

Figure 1.12: Short-Run Effect of an Adverse Event, by Experiences.

(a) High-Experience Physicians.

.23

.24

.25

.26

.27

.28

.29

.3

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ectio

n r

ate

Time from Adverse Event (Quarters)

(b) Low-Experience Physicians.

.23

.24

.25

.26

.27

.28

.29

.3

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

Time from Adverse Event (Quarters)

C-s

ectio

n r

ate

NOTE: Panels A and B of this figure depict C-section rates for high-experience and low-experience physicians, respectively. High-experience physicians are physicians with morethan 16 years of experience and low-experience physicians are physicians with 16 years ofexperience or less. The vertical red line denotes the time of the adverse event.

44

Figure 1.13: Short-Run Effect of an Adverse Event, by Prior Claims.

(a) Physicians with No Prior Claims.

.25

.26

.27

.28

.29

.3.3

1

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ection r

ate

Time from Adverse Event (Quarters)

(b) Physicians with 1 or More Prior Claims.

.22

.23

.24

.25

.26

.27

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ection r

ate

Time from Adverse Event (Quarters)

NOTE: Panels A and B of this figure depict C-section rates for physicians without and withprior claim history, respectively. The vertical red line denotes the time of the adverse event.

45

Figure 1.14: Distribution of Claims Report Timing.

0.0

5.1

.15

Re

po

rt P

erio

d D

en

sity

0 5 10 15 20

Statute of Limitation About Here

Time from Adverse Event (Quarters)

NOTE: This figure depicts frequency of claims by the timing of report relative to theadverse event.

46

Figure 1.15: Selection around First Contact.

(a) Per Period Number of Births.

9000

10000

11

000

12000

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

Time from First Contact (Quarters)

Num

ber

of

Birth

s

(b) Per Period Share of High-Risk Mothers.

.27

.28

.29

.3.3

1.3

2

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

Time from First Contact (Quarters)

Share

of H

igh

-Ris

k M

oth

er

(c) Per Period Average Age of Mothers.

27

27.2

27.4

27.6

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

Time from First Contact (Quarters)

Avera

ge M

oth

ers

A

ge

(d) Per Period Share of Privately Insured Mothers.

.54

.56

.58

.6

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

Time from First Contact (Quarters)

Share

of P

rivate

ly insure

d M

oth

ers

NOTE: These figures show how observable characteristics evolve around the time of the first contact.The vertical line denotes the time of the first contact. Panels A-D plot the per-period number ofbirths, share of high-risk mother, average mother age and share of privately insured mothers,respectively.

47

Figure 1.16: Selection on Observables, First Contact Panel.

.26

.265

.27

.275

.28

.285

.29

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

Time from First Contact (Quarters)

Pre

dic

ted

C-s

ectio

n r

ate

NOTE: This figure shows predicted C-section rates in the first contact panel, for claimswith a first contact a year or more after the adverse event. The vertical line denotes thetime of the first contact. The prediction was done by regressing, using OLS, C-sectiondummy on the high-risk covariates as well as age and insurance type dummies. The figureplots the average per period predicted C-section rate.

48

Figure 1.17: Short-Run Effect of a First Contact.

.24

.26

.28

.3.3

2

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ectio

n r

ate

Time from First Contact (Quarters)

NOTE: The figure plots per period C-section rates in the first contact panel, for claimsreported more than a year after the adverse event. The vertical line denotes the time of theadverse event.

49

Figure 1.18: Short-Run Effect of a First Contact, by claim success.

(a) Successful Claims.

.24

.25

.26

.27

.28

.29

.3

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ection r

ate

Time from First Contact (Quarters)

(b) Unsuccessful Claims.

.26

.27

.28

.29

.3.3

1.3

2

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ection r

ate

Time from First Contact (Quarters)

NOTE: Panels A and B of this figure depict per-period C-section rates for Paid and Non-Paid claims respectively, in the first contact panel for claims reported more than a year afterthe adverse event. The vertical red line denotes the time of the first contact.

50

Figure 1.19: Short-Run Effect of an Adverse Event on Same-Hospital Peers.

.26

.28

.3.3

2.3

4

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ectio

n r

ate

Time from Adverse Event (Quarters)

NOTE: The figure plots per period C-section rates in the peer panel, including allphysicians who work at the same hospital as the treated physician and appear through the5 year sample period. The vertical line denotes the time of the adverse event.

51

Figure 1.20: Short-Run Effect of an Adverse Event on Same-Hospital Peers.

(a) Close Peers.

.26

.28

.3.3

2.3

4

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ection r

ate

Time from Adverse Event (Quarters)

(b) Remote Peers.

.26

.28

.3.3

2.3

4

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

C-s

ection r

ate

Time from Adverse Event (Quarters)

NOTE: Panels A and B of this figure depict C-section rates for close and remote peers, respectively.Close peers are defined as peers from the same hospital who have at most 3 years gap in experiencefrom the treated physician, who treat patients in nearby neighborhoods - at most 10 miles apart,and who treat patients of a similar socioeconomic status - at most 0.4 percentage points in shareof pre-event Medicaid patients. The vertical red line denotes the time of the adverse event.

52

Chapter 2

The Interaction between OptimalMedical Malpractice Law andPhysicians’ Financial Incentives

2.1 Introduction

It is often argued that there is too much medical malpractice liability, and that healthcareproviders deliver excessive, low-benefit treatment because of fear of malpractice lawsuits.This view is consistent with evidence from self-reported data according to which, defensivemedicine is very common (Studdert et al. [2005], Reyes [2010a]). This view suggests thatoptimal medical malpractice law is very important, and a growing body of literature istrying to study the relationship between medical malpractice law and healthcare provision.Nevertheless, the mechanisms driving the response of medical treatment to malpractice laware not well-understood, and the empirical evidence about the magnitude of the response ismixed, providing little normative direction.

A separate body of literature studies the response of healthcare providers to financialincentives. The theory and evidence in this area show that when healthcare providers bearthe cost of performing procedures, they tend to perform fewer procedures. The treatment ofMedicare patients, under the Prospective Payment System is a well-known example of sucha situation (see Ellis and McGuire [1986]). In contrast, when procedures are profitable, likein the case of a Fee For Service reimbursement system , providers tend to perform more ofthem (see Ellis and McGuire [1993]).

This paper investigates the interaction between healthcare providers’ financial incentivesand their response to malpractice law. Its key result is that the role medical malpracticeliability plays in health care providers’ decisions varies according to the financial incentivesthey face. When providers bear the cost of the marginal treatment, they tend to provide toolittle care; malpractice liability offsets this tendency and increases the amount of treatment

53

provided. When providers’ marginal procedure is profitable, they tend to provide unneces-sary, low-benefit care; malpractice liability offsets this tendency and decreases the amountof care provided.

Using a simple model of providers’ behavior I show that the effect of a tort reformon treatment depends on healthcare providers’ financial incentives. In particular, liabilityreduction leads to a decrease in treatment level when treatment is not profitable. Conversely,liability reduction leads to an increase in treatment level when treatment is profitable. Theintuition behind this result is that when treatment is costly, it is provided excessively dueto fear of legal liability. Consequently a decrease in liability leads to a decrease in treatmentlevel. On the other hand, when treatment is profitable, providers tend to shower patientswith unnecessary low-benefit care, and a liability-decreasing reform would only exacerbatethis tendency.

The distinction between the role of medical malpractice law under different providers’financial incentives is important because it emphasizes that optimal malpractice law dependscrucially on the financial incentives of healthcare providers. If high healthcare expendituresare driven by fear of litigation, reducing liability may help curb excessive treatment. Onthe other hand, if high healthcare expenditures are driven by the profitability of healthcare,reducing liability will only make things worse.

I test the predictions of the model using a large tort reform which took place in Texasin June 2003 (“the reform”). The reform is attractive because it limited damage amountsfor which providers may be liable, which substantially lowered the liability risk faced byhealthcare providers (Hyman et al. [2009]).

The financial incentives faced by healthcare providers are not typically observed directly.To address this problem, I follow previous studies which use the distinction between com-mercially insured patients and those insured by Medicaid, which is known to pay low ratescompared to commercial insurers. I use the heterogeneity in physicians’ financial incentivesthat is created by the differences in generosity in reimbursement rates between commercialinsurance carriers and Medicaid as a proxy for providers’ financial incentives. Using thisapproach, I cannot separately identify the effect of the reform on patients under each insur-ance type. The main testable prediction offered by the model is that following a liabilitydecreasing reform, the treatment levels of the two insurance type groups would diverge.

I study two medical procedure classes. First, I estimate the effect of the reform on thelikelihood of undergoing a C-section. While in the aggregate it appears as if the reform didnot have an effect on the overall incidence of C-sections, analyzing the effect of the reform byinsurance type reveals that C-section rates increase by 2% for commercially insured mothersrelative to mothers insured by Medicaid. This effect comes mostly from low-risk births, forwhich providers are more likely to be sensitive to financial incentives, supporting the causalrole of the reform.

Second, I examine the effect of the reform on the incidence of common pediatric surgeries.I cannot directly observe which procedures are more attractive to providers, so I analyze theeffect of the reform in two stages. The first, I generate a measure of the attractiveness of each

54

procedure based on revealed preferences. Using the notion that during the weekend there isa tendency to perform only necessary procedures, I proxy attractiveness by higher weekdayproportion of procedures relative to the weekend. The second, I match the procedure’smeasured attractiveness with the procedure’s incidence, following the reform. Consistentwith the model, I find that, following the reform, attractive procedures are associated witha larger increase in their incidence.

This paper helps to reconcile the gap between the modest effect of malpractice law, typi-cally measured in the literature and the evidence from self-reported data, which suggest thatdefensive medicine is very common. In particular, this paper shows that a small aggregateresponse to medical malpractice law may be the sum of offsetting responses associated withdifferent financial incentives. This paper makes it clear that the effect of malpractice lawon medical treatment is best analyzed by separately considering situations according to thetype of financial incentives at play.

The remainder of the paper is organized as follows. Section two provides a literaturereview. Section three presents a simple model of providers’ decision making. Section fourdescribes the data that is used for the study, the empirical design and results and sectionfive concludes.

2.2 Literature Review

In early empirical work on the effect of liability pressure on treatment, Brennan et al. [2004]find that total costs per discharge are larger in hospitals that face higher claim rates (seeKessler and Rubinfeld [2007] for more examples from early literature). In a seminal paper,Kessler and McClellan [1996] examine the effect of malpractice pressure by studying theimpact of two broad classes of tort reforms on medical costs and on outcomes for a popu-lation of elderly heart patients. They find that while tort reforms have no significant effecton health outcomes, they significantly reduce medical costs, which they interpret as evi-dence for defensive medicine. Kessler and McClellan [2002b] extend their work and find thatboth managed care and liability reduce procedure use without affecting health outcomes. Inanother follow up paper, Kessler and McClellan [2002a] find that increases in malpracticepressure lead to significant increases in diagnostic expenditures but not in therapeutic ex-penditures. Recently Sloan and Shadle [2009] revisited the results of Kessler and McClellan[1996] finding no significant affect of tort reforms on medical decisions at all.

Baicker et al. [2007] find that greater liability pressure is associated with greater expendi-tures on diagnostic tests. Lakdawalla and Seabury [2009] exploit variation in the generosityof local juries to identify the causal impact of liability pressure on medical costs and mortal-ity, finding that liability pressure is associated with improved outcomes (namely, reductionsin patient mortality).

In the context of childbirth, Dubay et al. [1999] find that when malpractice pressure ishigher, physicians perform more C-sections, especially on mothers of lower socioeconomic

55

status with no evidence for better outcomes. Kim [2007] finds that, the performance ofC-sections and the use of ultrasound, forceps and vacuum, is insensitive to measures ofliability pressure such as the number of obstetrician claims per 1,000 births. Using tortreforms, Currie and MacLeod [2008] find that malpractice pressure affects both treatmentdecisions and patient outcomes. They show that Joint and Several Liability reforms reducecomplications of labor and procedure use, whereas caps on noneconomic damages increasethem.

2.3 Theory

I study a model of a representative healthcare provider’s behavior closely related to Ellisand McGuire [1986]. Providers’ utility function is assumed to include patients’ benefits,providers’ financial incentives and expected medical malpractice liability. I demonstratethat two types of over-treatment may arise and analyze how they are affected by malpracticeliability.

2.3.1 Patients

LetB(q) be patients’ benefits from health treatment, in dollar terms. As in Ellis and McGuire[1986] it is assumed that there a single input to healthcare, q, and that patients accept anytreatment prescribed by the provider. The benefit function is assumed to satisfy

Condition 1 B(q) is concave, and reaches a maximum at some quantity q′, after whichtotal benefits fall.

Benefits to the patients, are assumed to be equal to the full social benefit from treatment.Note that, because of the social costs that are associated with treatment, the socially optimaltreatment level, q∗ , is lower than the maximum benefits treatment level q∗ < q′.

2.3.2 Providers

Profits, π(q), capture the providers’ financial incentives which are associated with treatment.Two broad cases are considered. The first is profitable treatment (“Fee For Service (FFS)type incentives”)

π = D(q) (2.1)

where marginal profit in this case, d(q), satisfies d(q) ≥ 0, d′(q) ≤ 0. More broadly this profitstructure captures situations where additional treatment is profitable. The second case isnon profitable treatment (“Prospective Payment System (PPS) type incentive”)

π = −C(q) (2.2)

56

where marginal cost in this case, c(q), satisfies c(q) ≥ 0, c′(q) ≥ 0. This profit structure mayfit any treatment decision for which providers bear costs.

P (q) is the probability of facing medical malpractice liability. It is assumed that thefunction P (q) satisfies the following condition:

Condition 2 P (q) is convex, reaches a minimum at q′, and increases after that point

Intuitively, medical malpractice liability system aligns the patient’s benefits with theprovider’s incentives. When the provider’s decision increases the patient’s benefits it alsodecreases the likelihood of malpractice liability. H is the expected cost of facing medicalmalpractice liability. Note that in practice providers are typically insured against malpracticeclaims payments, so the costs associated with litigation which enter H include time loss,damage to reputation, promotion, etc. (see discussion in Currie and MacLeod [2008]). Forsimplicity, providers are assumed to be risk neutral with respect to medical malpracticeliability. Providers’ utility is given by

U(q,H) = V (π(q), B(q))− L(P (q)H) (2.3)

where V is a concave function of π(·) and B(·) and L is a convex function of P (q)H.

2.3.3 Providers’ behavior

The provider problem ismaxqV (π(q), B(q))− L(P (q)H)

and the solution is given by the first order condition

αdπ

dq+ b(q)− βp(q)H = 0 (2.4)

where b(q) is marginal patient benefits, p(q) is the marginal probability of facing liability,

α =∂V∂π∂V∂B

= MRSBπ, and β =∂L

∂(P ·H)∂V∂B

. Intuitively α reflects the rate at which a provider

is willing to trade a dollar of profit for a dollar of patient benefits. Analogously β reflectsthe rate at which a provider is willing to trade a dollar of patient benefits with a dollar ofexpected malpractice liability. One reason to think that α 6= β is that the provider’s financialincentives reflect a combination of physician incentives and hospital incentives while the fearof liability might reflect only the physician’s liability costs.

Non Profitable Treatment

When providers’ profits are given by (2.2), the first order condition in (2.4) becomes

−αc(q) + b(q)− βp(q)H = 0. (2.5)

57

Note that under PPS type incentives providers set q < q′. Intuitively, suppose that q′ isprovided, a very small decrease in q would not affect patients benefits and the likelihood offacing malpractice liability because b(q′) and p(q′) both equal zero, but it would decreasetreatment costs.

It is easy to show that, under PPS type incentives, there exists a unique H∗ such thatthe social optimum is attained, that is characterized by

−αc(q∗) + b(q∗)− βp(q∗)H∗ = 0. (2.6)

This implies that under PPS type incentives, two regions of provision exist. The first, whenH < H∗, is the under-provision region, 0 < q < q∗, where too little care is provided due tolow liability pressure and high cost of treatment. The second, when H > H∗, is the defensivemedicine region, q∗ < q < q′, a region in which low-benefit healthcare is provided due tohigh liability pressure as illustrated in Figure 2.1.

Profitable Treatment

When profits are given by (2.1), the first order condition in (2.4) becomes

αd(q) + b(q)− βp(q)H = 0 (2.7)

When providers face Fee For Service type incentives, they provide care in the region q > q′,the profit-seeking region. Intuitively, so long as q ≤ q′, there is no trade-off in treatment provi-sion. Increasing q increases patient benefits and profit and decreases likelihood of malpracticeliability. Only when q > q′, a trade-off emerges. While additional treatment is profitable, itdeceases patient benefits and increases the likelihood of facing medical malpractice liability.

The Effect of a Tort Reform

Consider now the effect of a tort reform on providers’ behavior for the two types of financialincentives. Intuitively, Under PPS type incentives, financial considerations tend to lowertreatment intensity and malpractice law offsets this tendency. An increase in malpracticeliability pressure would lead to an increase in treatment intensity. In contrast, under FeeFor Service type incentives, financial consideration tend to raise treatment intensity andmalpractice law decreases this tendency. In this case, an increase in liability leads to adecrease in treatment intensity. The intuition is summarized by the following proposition.

Proposition 1. Under conditions 1 and 2, The effect of an increase in expected costs offacing malpractice liability depends on providers’ financial incentives. When providers facePPS (Fee For Service) type incentives, an increase in expected cost of facing malpracticeliability decreases (increases) healthcare provision.

58

Proof of Proposition 1. The logic of the proposition is straightforward. For PPS typeincentives, providers bear the cost of treatment and they trade off this cost with their bene-fits from additional treatment, which include both patient benefits and lower expected costsof malpractice liability. As the cost of medical malpractice goes up, providers tend to pro-vide more medical treatment. The logic for Fee For Service type incentives is similar. SeeAppendix A for the detailed proof.

The analysis above distinguishes between three regions of treatment: under-provision,defensive medicine and profit seeking medicine. It is important to distinguish betweenthe different types of treatment for optimal medical malpractice policy. Under defensivemedicine, physicians provide an excessive level of care due to fear of lawsuits, so it is desir-able to reduce malpractice liability. In the profit seeking region, low-benefit care is providedbecause it is profitable and despite the exposure to malpractice liability. In the region ofunder provision, low levels of care are provided because of its costs and despite the lossof benefits to patients and the exposure to malpractice liability. In both of these cases anincrease in liability pressure is desirable.

The corollary below captures this intuition, showing that when there is over-provisionof health care, it is important to distinguish between defensive medicine and profit seekingmedicine.

Corollary 1. The welfare effect of a liability decreasing tort reform when q > q∗ depends onproviders’ financial incentives. When providers face PPS (Fee For Service) type incentives,a decrease in expected costs of facing medical malpractice liability increases (decreases) socialwelfare.

Proof of corollary 1. A liability decreasing tort reform moves society towards the socialoptimum q∗, when providers face PPS type incentives and away from the social optimum q∗

when providers face Fee For Service type incentives. QED.

2.4 Empirical Analysis

The objective of the empirical analysis is to study the relationship between treatment inten-sity and tort reforms under PPS and under Fee For Service type incentives. Unfortunately,financial incentives faced by providers are typically unobservable, making it challenging toaddress the question without making additional assumptions.

I contrast two important cases for which π(q) is likely to be of the PPS type and Fee ForService type, respectively. In particular, I adopt an approach often used in the literature onproviders’ response to financial incentives. Medicaid reimbursement to providers is much lessgenerous than commercial insurance reimbursement (see Zuckerman et al. [2009] and for thecase of Texas see e.g. Davila et al. [2002]). Holding other things equal, the differential payoff

59

(i.e. reimbursement net of costs) to providers who treat Medicaid patients is more likelyto be negative whereas the differential payoff for providers who treat patients insured by acommercial carrier is likely to be positive. Using this notion, Currie and Gruber [2001] showthat the lower fee differentials between C-section and vaginal childbirth under Medicaid thanunder commercial insurance are associated with lower C-section rates for Medicaid mothers.Gruber et al. [1999] find that the lower fee differentials can explain between one half andthree-quarters of the difference between Medicaid and commercial C-section rates.

Relying on this approach, I study the change in treatment intensity following the reformfor patients insured by a commercial carrier and for patients under Medicaid. With thisapproach I cannot separately identify the effect of the reform on each insurance type. Themodel shows that following a liability decreasing tort reform, there is a differential responseto the reform for patients under different insurance types. The incidence of procedures forpatients insured by a commercial carrier would increase while the incidence of proceduresfor patients under Medicaid would decrease. Hence, the main testable prediction that thetheoretical analysis offers is that there will be a divergence in treatment levels between thetwo insurance types.

I study areas for which both Medicaid and commercial insurance are common, and taketwo approaches to test the model’s hypothesis. The first approach focuses on childbirth,one of the most common medical procedures that is thought to have a high sensitivity tomedical malpractice liability pressure (Smarr [1997]). The second approach is to analyzecommon childhood surgical procedures, another area that is regarded to be highly sensitiveto malpractice liability pressure (McAbee et al. [2008], Fanaroff [2010]).

2.4.1 Background - Texas Tort Reform

In June 1 2003, Texas passed a tort reform reducing malpractice liability by capping noneco-nomic damages in medical malpractice claims. Texas law requires a constitutional amend-ment authorizing the legislature to determine limits for noneconomic damages in healthcareliability claims. This necessary constitutional amendment was adopted by the voters onSeptember 13, 2003 and was preceded by months of a fierce public debate on which opposingsides spent approximately twenty million dollars (Roberson and Torbenson [2007]).

The reform imposed a cap on noneconomic damages1 in medical malpractice cases filedafter September 1, 2003. The cap limits noneconomic damages against physicians and otherindividuals who are licensed healthcare providers to $250,0002. Furthermore, the reformmade a number of changes in the procedures regarding medical malpractice claims. An

1“Noneconomic damages” means damages for physical pain and suffering, mental or emotional pain oranguish, loss of consortium, disfigurement, physical impairment, loss of companionship and society, inconve-nience, loss of enjoyment of life, injury to reputation, and all other nonpecuniary losses other than exemplarydamages.

2All the cap amounts mentioned here are nominal and not adjusted for inflation. A separate $250,000 capapplies to each hospital, with total noneconomic damages capped at $500,000 for all health care facilities.

60

important change is that a claimant must now file, within 120 days after the claim is filed,an expert report, a written report by an expert regarding the expert’s opinion concerning howthe standard of care rendered by the physician or healthcare provider failed to meet applicablestandards and the causal relationship to the harm suffered by the claimant. Generally, suchexperts must be physicians or persons in the same occupation as the healthcare provider.

A recent study by Silver et al. [2008] uses claim level data to estimate the effect of thereform on jury verdicts, post-verdict payouts, and settlements. Using simulations based onmedical malpractice cases closed during 1988–2004, they find that the cap affects 47-percentof verdicts and substantially reduces the mean of noneconomic damages and verdicts. Theyalso find that in cases settled without trial, the noneconomic cap reduces predicted meantotal payout by 18-percent. Their evidence confirms that the reform has a large impact onmedical malpractice pressure in Texas. More generally, Avraham [2007] finds that liabilitydecreasing reforms tend to reduce the number of cases and average awards.

2.4.2 Childbirth Analysis

As mentioned above, there is a body of evidence showing that C-sections are sensitive toproviders’ financial incentives that are generated by the difference in reimbursement rates be-tween commercial insurances and Medicaid (Gruber et al. [1999], Currie and Gruber [2001]).This makes the decision to perform a C-section a potentially good candidate to study theresponse of providers to tort reforms when they face different financial incentives. I studythe effect of the reform on the decision to perform a C-section rather than a vaginal birth.

I use a standard differences in differences methodology:

C-sectionist=a+Quartert+ Insurances+ b1Charist+ b2Insurance*Reform+ b3Hosp+εist(2.8)

where the estimates of the coefficient Insurance ∗Reform capture the relative effect of theTexas reform on the probability of preforming a C-section by type of insurance. Char isa vector of the mother’s personal characteristics, Quarter is a vector of dummy variablefor each quarter in the relevant time period, Hosp is a vector of dummy variables for eachhospital.

Data

I use the deliveries in the periods 2000Q1-2007Q4 for mothers aged of 25-343 from the TexasPublic Use Data File (PUDF). California Patient Discharge Data and Florida Inpatient Dataare used as control groups. The data contain information on patient demographics, length ofstay, discharge status (alive or dead), diagnosis (including primary and secondary ICD-9CM

3The reason to choose this age group is that in Texas, about 80% of mothers under 25 are insured byMedicaid and about 80% of mothers above 34 are commercially insured. Focusing on ages 25-34 generatesa sample that is more comparable.

61

and diagnosis related group (DRG) codes), source of payment, and procedure codes. Thedata also include discharge quarter and hospital identification number.

Table 2.1 shows descriptive statistics of the Texas, Florida, and the California populationsfor mothers insured commercially and for mothers insured by Medicaid. In all three states,The rate of Afro-American and Hispanic mothers is higher among Medicaid mothers andMedicaid mothers are also more likely to have undergone a previous C-section.

Figure 2.2 plots C-section rates for Texas, Florida and California for the periods 2000Q1-2007Q4. The figure shows that C-section rates in Texas, Florida and California are increasingsteadily during the sample period. C-section rates in California are lower than C-section ratesin Texas and Florida by roughly 4% and C-section rates in Florida appear to be growingslightly faster than C-section rates in Texas. Based on the raw data it does not appear thatthere is a striking effect of the Texas reform on the aggregate level of C-section rates.

Results

Figures 2.3, 2.4 and 2.5 plot C-section rates for commercially insured mothers and for mothersinsured by Medicaid in Texas, Florida and California. Figure 2.3 suggests that prior to thereform, C-section rates for the two insurance types in Texas are similar and they both followa similar time trend. Following the reform, the increase in C-section rates for Medicaidappears to slow down while commercial insurance C-section rates appear to maintain thesame trend. C-section levels for the two types of insurances diverge. In contrast, as Figure2.4 and Figure 2.5 show, in Florida and California Medicaid and commercial C-section ratesappear to follow a similar trend.

The baseline regression estimation results are summarized in columns (1)-(3) of Table2.2. Consistent with the graphic illustration, the estimates reveal that following the reform,the overall incidence of C-sections for mothers insured by a commercial insurance carrierrelative to mothers insured by Medicaid, increased by approximately 2%. Columns (4)-(9)of Table 2.2, show the estimates from placebo regressions for Florida and California. There isno statistically significant evidence of an increase in the incidence of C-sections following thereform in those two states. In Florida, there is a small positive coefficient that is marginallyinsignificant. This result is consistent with Figure 2.4 which shows a small increase in thegap between C-section rates of commercially insured mothers and mothers under Medicaidin Florida. The pattern may be explained by a “softer” liability decreasing reform in Floridaat the end of 2003. The Florida reform set a cap of noneconomic damages at $500,000, andmay have had a similar but smaller effect on C-section rates in that state (see Avraham[2006]).

The model predicts that the reform would have a stronger effect on treatment when ∂b(q)∂q

is smaller. Intuitively, other things being equal, when patients’ benefits are more sensitiveto treatment decisions, the financial incentives have a smaller effect on treatment. In order

62

to examine the validity of the results, I divide the sample into high-risk and low-risk groups4

and compare the relative effect of the reform on mothers insured by a commercial carrierand mothers under Medicaid, for the high-risk and the low-risk groups (i.e. test whether∂q∂H |high risk

< ∂q∂H |low risk

).

Figure 2.6a and Figure 2.6b plot C-section rates of commercially insured mothers andmothers insured by Medicaid, for the low-risk and high-risk groups, respectively. The figuresshow that, consistent with the model, the divergence in C-section rates between commerciallyinsured mothers and mothers under Medicaid, appears to be larger for low-risk births.

Table 2.3 summarizes the regression estimation results for this specification. As Figure2.6a and Figure 2.6b suggest, There is a statistically significant difference between the effectof the reform on high-risk and low-risk mothers. For low-risk mothers, following a liabilitydecreasing reform, the overall incidence of C-sections for commercially insured mothers rel-ative to mothers insured by Medicaid increased by approximately 2.3% while for high-riskmothers it increased by about 1.3%, supporting the causal role of the reform.

In summary, analysis of the response of C-section rates to the reform shows a divergencein C-section rates between the commercial and Medicaid groups. The causal effect of thereform in this result is supported by the fact that the response is stronger in low-risk mothersthan it is in high-risk mothers. The results suggest that the response to the tort reform indeeddepends on providers’ financial incentives.

2.4.3 Common Pediatric Surgery Analysis

In this section, I augment the empirical analysis with a second approach, studying the effectof the reform on the incidence of common pediatric surgical procedures. As in the case ofchildbirth, I cannot directly observe providers’ financial incentives, and furthermore, otherconsiderations in the providers’ decisions to perform surgery such as patient benefits are alsounobserved, making it impossible to know how attractive a given procedure is in the eyesof the healthcare provider. In order to overcome this problem I use revealed preferences torecover the relative attractiveness of pediatric surgeries. Like in the previous section, theanalysis exploits the fact that it is more profitable to treat a commercially insured patientthan a patient under Medicaid. I create a proxy for a procedure’s attractiveness based onthe notion that in the weekend there is a tendency to perform only necessary surgeries.That is, compared to weekday surgeries, weekend surgeries are less likely to depend on thefinancial incentives a procedure offers (see Card et al. [2009]). Therefore, a higher weekdayproportion of commercial insurance patients relative to Medicaid patients, suggests thatcommercial insurance provides stronger financial incentive to perform the procedure.

Using the attractiveness proxy, I test the hypothesis that following the reform procedures

4High-risk is defined as one of the following diagnoses: Previous C-section, breech position, early on-set, polyhydramnios, oligohydramnios, obesity, diabetes, multiple gestation, distress, hypertension andhemorrhage.

63

with high attractiveness would be associated with an increase in the share of patients insuredby a commercial provider relative to procedures with low attractiveness. Intuitively, highattractiveness procedures are prevalent during the weekdays, revealing that they are preferredby providers. The liability decreasing reform is expected to magnify the tendency to performthese procedures. Conversely, unattractive procedures have a similar proportion on weekdaysreflecting the fact that providers are not interested or are not able to perform a higher numberof those procedures. Therefore following the reform these procedures are not expected todemonstrate a change in the proportion of commercially insured patients.

The empirical analysis therefore has two stages. In the first stage, a proxy for per-procedure relative attractiveness is estimated. In the second stage, the differential associationbetween a procedure’s attractiveness and the effect of the reform on the proportion of patientsinsured by a commercial carrier is tested.

First stage: To generate the proxy, I first calculate the ratio of weekend procedures tototal procedures for patients insured by a commercial carrier and for patients under Medicaid,during a period of 4 years before the reform. I then calculate the difference in the log of thecommercial and the Medicaid ratios. Namely,

Attractiveness(procedurei) = log(weekend#

total#)aid − log(

weekend#

total#)comm (2.9)

As explained above, a high proxy implies that providers tend to perform a large number ofprocedures for commercially insured patients during the week, suggesting that the procedureis attractive.

Second stage: In the second stage, the association between attractiveness and the effectof the reform on the proportion of patients under a commercial insurance is tested usingtwo estimation equations. In the first equation, I pool procedures into three groups: high,medium and low, by their attractiveness. I then regress:

log(commijt

aidijt) = Hospj + Y eart + Procedurei + γ · index gr ∗ Post reform+ εijt (2.10)

where log(commijtaidijt

) is the ratio of commercially insured patients to patients insured by Medi-

caid in a hospital-procedure-year cell, Hospj is a vector of hospital dummies, Y eart is a vectorof year dummies, Procedurei is a vector of procedure indicators and index gr∗Post reformis the attractiveness group multiplied by an indicator for the reform. The second equationis

log(commijt

aidijt) = Hospj+Y eart+Procedurei+γ ·Attractiveness∗Post reform+εijt (2.11)

where Attractiveness ∗ post reform is the attractiveness proxy multiplied by an indicatorfor the reform. The model predicts that γ > 0, reflecting a differential response to the reformbased on physicians’ incentives that are proxied by the Attractiveness measure.

64

Data

Sample generation. For this part too, the Texas Public Use Data File (PUDF) is used. TheCalifornia Patient Discharge Data is used to create the placebo tests. I extract dischargesfor which, the ICD9 code of the main procedure corresponds to one of the common pediatricsurgical procedures in the US5. Only patients aged nineteen or less are included in the data.Table 4 shows summary statistic for the data. Note that, the median age in the Medicaidsample is smaller and has a higher rate of Hispanic and African-American patients in bothstates.

The Attractiveness index. I generate the proxy by applying the formula (2.9) to a sixteenquarter period prior to the reform. In order to use discharges for which the timing ofadmission is related to the main procedure performed, I include only discharges for whichthe procedure was done in the first two days of admission and I eliminate newborns. Ieliminate procedures with attractiveness proxies that are larger than 0.5 or smaller than-0.5, which are outliers caused by very low weekend incidence. Table 2.7 includes the list ofthe procedures by their attractiveness group.

Results

Figure 2.7a plots log(commijtaidijt

) of high attractiveness procedures and log(commijtaidijt

) of low at-

tractiveness procedures against time, controlling for hospital fixed effects. To generate thefigure, I regressed log(

commijtaidijt

) on a vector yearXindex gr dummies and hospital fixed effects,

and plotted the regression coefficients for the period 1999-2007 for the high attractivenessand low attractiveness groups, omitting the medium attractiveness group. The figure showsthat relative to the omitted medium attractiveness group the proportion of commerciallyinsured patients in high attractiveness procedures increased following the reform while theproportion of low attractiveness procedures remained unchanged.

The estimates of equation (2.10) are displayed in column (1) of Table 2.5. In columns(2) of Table 2.5 a hospital year interaction is added to control for local trends in Medicaidenrollment. The estimates in columns (1) - (2) of Table 2.5 show that there is a statisticallysignificant difference between the effect of the reform on the high attractiveness and the lowattractiveness groups: while the reform had a positive and significant effect on the proportionof commercially insured patients for the case of high attractiveness group, it had no effecton the proportion of commercially insured patients low attractiveness group.

Columns (3)-(4) of Table 2.5 report results from a placebo regressions which replicatethe regressions in columns (1)-(2) for a fictitious reform in the first quarter of 2001 and forthe time period 1999-2002. The results of the placebo reform are statistically insignificant.Columns (5)-(6) of Table 2.5 report results from a placebo regression using the Californiadata. Each procedure in this specification was assigned the Texas attractiveness proxy

5See appendix B for a detailed description of the generation of the common surgical procedures in theUnited States list.

65

and a fictitious 2003 reform in California was analyzed. The results of this regression arestatistically insignificant as well, showing that the observed effect is not driven by a trendin treatment patterns of particular procedures.

Figure 2.8 plots the percentage change in the ratio of patients insured by a commercialcarrier to patients insured by Medicaid following the reform, against the attractiveness proxy.In order to generate the figure, I first ran a regression with hospital and year fixed effectssimilar to (2.11). I then pooled procedures’ attractiveness proxies into 0.05 wide bins andfor each bin computed the mean difference in log(

commijtaidijt

) following the reform (between

2002 and 2004). The mean difference was plotted against median bin value and a linear fitline was added. Figure 2.9 is a placebo test. It is constructed in an analogous way using afictitious reform which took place in the beginning of 2001. Figure 2.10 is another placebotest that is using the California data as a placebo test analogous to the one done to generateFigure 2.7b.

Figure 2.8 shows that, consistent with the prediction of the model, the reform has adifferential effect on log(

commijtaidijt

). Higher attractiveness is associated with a larger increase

in the proportion of commercially insured patients. Figure 2.9 shows that following thefictitious reform in 2001 there doesn’t seem to be a differential change in log(

commijtaidijt

). Figure

2.10 shows that there doesn’t seem to be a parallel differential change in log(commijtaidijt

) in

California, implying that the effect in Figure 2.8 is not driven by a trend in treatmentpatterns of particular procedures.

The estimates of equation (2.11) are displayed in column (1) of Table 2.6. In columns(2) of Table 2.6, a hospital year interaction is added to control for local trends in Medicaidenrollment. The estimates in columns (1)-(2) of Table 2.6 show a positive and statisticallysignificant coefficient for γ. Columns (3) and (4) of Table 2.6 show the estimates of afictitious reform in 2001. The estimates are small and insignificant. Columns (5)-(6) ofTable 2.6 show the estimates of another placebo test that is using the California data as aplacebo test analogous to the one done in Table 2.5.

This section shows that in the case of pediatric surgery the response to the reform is con-centrated among procedures that are revealed to be attractive from the provider perspectivebased on the pre-reform period. The results in this section provide additional support tothe idea that the response of treatment to malpractice law depends on providers’ financialincentives.

2.5 Conclusions

Using a simple model I have shown in this paper that the relationship between medicalmalpractice law and medical treatment depends crucially on the financial incentives facedby healthcare providers. Medical malpractice law plays a different role in regulating theex-ante behavior of healthcare providers, under different financial incentives. When high

66

healthcare expenditures are driven by fear of litigation, reducing liability may help curbexcessive treatment. On the other hand, when high healthcare expenditures are the result ofthe profitability of healthcare provision, a liability reduction would only make things worse.

Using a large liability reducing tort reform in Texas, I examine the effect of medicalmalpractice liability-reduction on medical treatment for the case of childbirth and for thecase of pediatric surgeries. In the case of childbirth, I find that C-sections in commerciallyinsured mothers become more prevalent. In common pediatric surgeries, the incidence ofprocedures that are likely to be attractive in commercially insured patients, increases forthat group.

Interestingly, Currie and MacLeod [2008] find an increase in C-section rates followingsimilar reforms which put caps on noneconomic damages. Their result is surprising, sinceC-sections are regarded as a conservative and more expensive treatment relative to vaginalbirth. The finding in this paper may be viewed as providing the “micro foundations” fortheir result by showing that when C-sections are likely to be profitable, C-section rates areexpected to increase following a reduction in malpractice liability. This response may offsetother effects of the reform resulting in an overall increase in C-section rates.

Finally, The results in this paper may help reconcile the gap between the modest effectsthat are typically measured in the literature on the relationship between malpractice law andtreatment and the evidence from self-reported data, according to which defensive medicineis very common. In particular, it shows that a small aggregate response to malpractice lawmay be the sum of offsetting responses arising from different types of financial incentives.

Appendix A

Proof of Proposition 1. To see the first part of the proposition, one can find the effectof a change in H on q under PPS by totally differentiating (2.5) w.r.t H

−α∂c(q)∂q

∂q

∂H+∂b(q)

∂q

∂q

∂H− β[

∂p(q)

∂q

∂q

∂HH + p(q)] = 0

∂q

∂H=

βp(q)

[−α∂c(q)∂q

+ ∂b(q)∂q− β ∂p(q)

∂qH]

(2.12)

∂q

∂H> 0

67

similarly, to see the second half totally differentiating (2.7) w.r.t H

α∂d(q)

∂q

∂q

∂H+∂b(q)

∂q

∂q

∂H+ β[

∂p(q)

∂q

∂q

∂HH + p(q)] = 0

∂q

∂H=

βp(q)

[α∂d(q)∂q

+ ∂b(q)∂q− β ∂p(q)

∂qH]

(2.13)

∂q

∂H< 0.

QED.

Appendix B

Generation of the common procedure list. Using the Healthcare Cost and Utilization Projectdatabase year 2000 sample, I extracted a list of pediatric procedures with incidence of morethan 500 cases. Of the hundred most common pediatric procedures in the US by age group, Ikept only procedures with ICD9 codes which represent an inpatient surgery (i.e. I eliminateddiagnostic tests and non surgical procedures). The sixty most common surgeries in thesample were included in the final list. The full list of procedures by their attractivenessproxy group appears in Table 2.7.

68

69

Tab

le2.

1:Sum

mar

ySta

tist

ics

Childbir

thSam

ple

Tex

asC

alif

orn

iaF

lori

da

Med

icai

dC

omm

erci

alM

edic

aid

Com

mer

cial

Med

icaid

Com

mer

cial

Age

(mea

n)

30-3

430

-34

25-2

930

-34

25-2

930-

34

Moth

erH

isp

an

ic64

.1%

23.7

%64

.8%

23.1

%30

.6%

16.9

%M

oth

erA

fric

an

Am

eric

an

10.5

%7.

8%4.

4%2.

3%26

.7%

11.

9%

Moth

erO

ther

Rac

e25

.4%

68.5

%30

.8%

74.6

%42

.7%

71.2

%P

revio

us

C-s

ecti

on

22.3

%16

.3%

20.0

%13

.3%

19.9

%15

.2%

Bre

ech

2.8%

3.4%

2.7%

3.3%

3.2%

3.8

%E

arly

On

set

7.3%

6.5%

6.2%

5.8%

8.3%

6.8

%H

emorr

age

1.6%

1.6%

1.7%

1.6%

2.2%

1.8

%H

yp

erte

nsi

on

4.5

%4.

7%3.

4%3.

5%5.

1%4.9

%D

istr

ess

0.2%

0.4%

0.6%

0.6%

0.3%

0.4

%M

ult

iple

Ges

tati

on0.9

%1.

2%0.

8%1.

1%1.

1%1.3

%D

iab

etes

1.2

%0.

8%0.

9%0.

7%1.

1%0.7

%O

bes

ity

0.5%

0.3%

0.5%

0.4%

0.6%

0.3

%O

ligo

hyd

ram

nio

s3.

1%2.

2%2.

5%2.

4%2.

8%2.5

%P

oly

hyd

ram

nio

s0.

5%0.

7%0.

5%0.

4%0.

8%0.8

%O

bse

rvat

ion

s46

5,2

6160

0,75

974

2,32

41,

140,

892

242,3

33

444

,476

NO

TE

:S

amp

lein

clu

des

moth

ers

bet

wee

nag

es25

-34

insu

red

by

aco

mm

erci

alca

rrie

ror

by

Med

icaid

inth

ep

erio

d199

9-Q

1to

2007

q4

usi

ng

the

Tex

as

Cal

ifor

nia

and

Flo

rid

aIn

pat

ient

Dat

are

spec

tive

ly.

70

Tab

le2.

2:T

he

Eff

ect

ofa

Lia

bilit

yD

ecre

asin

gR

efor

m,

Diff

inD

iffE

stim

ates

:T

X,

CA

and

FL

Tex

asC

alif

ornia

Flo

rida

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Ref

XIn

s0.

0209

0.01

940.

0175

-0.0

032

-0.0

032

-0.0

031

0.00

490.

0069

0.00

68(0

.005

1)(0

.004

6)(0

.002

9)(0

.002

7)(0

.002

4)(0

.002

5)(0

.004

2)(0

.003

7)(0

.003

6)In

s’ty

pe

yes

yes

yes

yes

yes

yes

yes

yes

yes

Ris

kG

roup

No

No

yes

No

No

yes

No

No

yes

Quar

ter

yes

yes

yes

yes

yes

yes

yes

yes

yes

Age

,R

ace

No

No

yes

No

No

yes

No

No

yes

Hos

pit

alF

EN

oye

sye

sN

oye

sye

sN

oye

sye

sO

bs’

1,05

9,81

81,

059,

818

1,05

6,54

11,

883,

216

1,88

3,21

61,

597,

030

686,

809

686,

809

686,

809

NO

TE

:Sta

ndar

der

rors

clust

ered

atth

ephysi

cian

leve

lfo

rT

exas

and

Flo

rida

and

atth

eH

ospit

alle

velfo

rC

alif

ornia

,ar

ere

por

ted

inpar

enth

eses

.E

ach

regr

essi

onin

cludes

aco

nst

ant.

Dep

enden

tva

riab

lein

all

model

sis

the

indic

ator

for

C-s

ecti

on.

71

Table 2.3: The Effect of a Liability Decreasing Reform, Diff in Diff Estimates by Risk Group

(1) (2) (3)

ReformXInsuranceXHigh 0.0145 0.0128 0.0134(0.0055) (0.0048) (0.0049)

ReformXInsurance Low 0.0234 0.0223 0.0226(0.0031) (0.0029) (0.0029)

Insurance type yes yes yesRisk Group yes yes yesRisk GroupXInsurnance yes yes yesQuarter yes yes yesAge, Race No No yesHospital FE No yes yesObservations 1,059,818 1,059,818 1,056,541

F-test for equality of Refrom Effect (Prob > F ) 0.058 0.044 0.054

NOTE: High-risk group is defined as a discharge with at least one of the following diagnoses:Previous C-section, breech position, early onset, polyhydramnios, oligohydramnios, obesity,diabetes, multiple gestation, distress, hypertension and hemorrhage. Low-risk is defined asany discharge that is not high risk. Standard errors clustered at the physician level, arereported in parentheses. Each regression includes a constant. Dependent variable in allmodels is the indicator for C-section.

Table 2.4: Summary Statistics Pediatric Surgeries Sample

Texas California

Medicaid Commercial Medicaid Commercial

Age (median) 5-9 10-14 5-9 10-14Hispanic 58.2% 25.0% 45.1% 19.0%African American 11.5% 7.5% 3.6% 1.8%Other Race 30.2% 67.5% 51.4% 79.3%Observations 79,105 85,144 102,437 123,341

NOTE: A sample of all discharges of patients aged 19 or less who are insured by Medicaid ora commercial insurance carrier for which main procedure in the discharge record is includedin the most common pediatric procedures list (See appendix B for details on the generationof the list).

72

Tab

le2.

5:E

ffec

tof

Ref

orm

onP

edia

tric

Surg

ery

by

Att

ract

iven

ess

Gro

up

Ref

orm

pla

ceb

oT

exas

2001

pla

ceb

oC

A20

03

(1)

(2)

(3)

(4)

(5)

(6)

hig

hat

trac

tXR

efor

m0.

1833

0.19

55-0

.024

5-0

.024

50.

0676

-0.0

047

(0.0

351)

(0.0

351)

(0.0

528)

(0.0

535)

(0.0

346)

(0.0

366)

low

attr

actX

Ref

orm

0.06

430.

0578

0.01

620.

0347

0.07

30-0

.023

5(0

.033

6)(0

.033

6)(0

.050

1)(0

.052

1)(0

.030

8)(0

.033

4)A

ttra

ctiv

enes

sin

dex

yes

yes

yes

yes

yes

yes

Por

cedure

yes

yes

yes

yes

yes

yes

Ref

orm

yes

yes

yes

yes

yes

yes

Hos

pit

alye

sye

sye

sye

sye

sye

sY

ear

yes

yes

yes

yes

yes

yes

Hos

pit

alX

Yea

rN

oye

sN

oye

sN

oye

sF

-tes

tfo

req

ual

ity

ofR

efro

mE

ffec

t(Prob>F

)0.

0005

0.00

000.

496

0.32

90.

878

0.58

1O

bse

rvat

ions

10,2

0110

,201

4,01

64,

016

11,9

8611

,986

NO

TE

:P

anel

Ash

ows

esti

mat

ion

resu

lts

ofre

gres

sion

(10)

.P

anel

Bpre

sents

resu

lts

from

anal

ysi

sof

afict

itio

us

refo

rmin

the

beg

innin

gof

2001

.P

anel

Cpre

sents

resu

lts

from

are

gres

sion

anal

ysi

sof

CA

assi

gnin

gea

chpro

cedure

the

attr

acti

venes

spro

xy

from

Tex

as.

Sta

ndar

der

rors

are

rep

orte

din

par

enth

eses

.E

ach

regr

essi

onin

cludes

aco

nst

ant.

Dep

enden

tva

riab

lein

all

model

sis

log

rati

oof

com

mer

cial

pro

cedure

sto

Med

icai

dpro

cedure

s.

73

Tab

le2.

6:E

ffec

tof

Ref

orm

onP

edia

tric

Surg

ery

Ref

orm

pla

ceb

oT

exas

2001

pla

ceb

oC

A20

03

(1)

(2)

(3)

(4)

(5)

(6)

Att

ract

index

XP

ost

Ref

0.19

080.

2391

0.00

260.

0112

-0.1

048

-0.0

313

(0.0

704)

(0.0

682)

(0.1

142)

(0.1

156)

(0.0

744)

(0.0

712)

Pro

cedure

FE

yes

yes

yes

yes

yes

yes

Ref

orm

yes

yes

yes

yes

yes

yes

Hos

pit

alye

sye

sye

sye

sye

sye

sY

ear

yes

yes

yes

yes

yes

yes

Hos

pit

alX

Yea

rN

oye

sN

oye

sN

oye

sO

bse

rvat

ions

10,2

0110

,201

4,01

64,

016

11,9

8611

,986

NO

TE

:P

anel

Ash

ows

esti

mat

ion

resu

lts

ofre

gres

sion

(11)

.P

anel

Bpre

sents

resu

lts

from

anal

ysi

sof

afict

itio

us

refo

rmin

the

beg

innin

gof

2001

.P

anel

Cpre

sents

resu

lts

from

are

gres

sion

anal

ysi

sof

CA

assi

gnin

gea

chpro

cedure

the

attr

acti

venes

spro

xy

from

Tex

as.

Sta

ndar

der

rors

are

rep

orte

din

par

enth

eses

.E

ach

regr

essi

onin

cludes

aco

nst

ant.

Dep

enden

tva

riab

lein

all

model

sis

log

rati

oof

com

mer

cial

pro

cedure

sto

Med

icai

dpro

cedure

s.

74

Tab

le2.

7:P

roce

dure

Inci

den

ceby

Att

ract

iven

ess

Ter

cile

Low

Med

ium

hig

h

Orc

hio

pex

y57

5F

emor

al

Div

isio

nN

ec500

Dors

al/

Dors

olu

mF

us

Ant

534

Oth

erL

oca

lD

estr

uc

Skin

694

Ton

sill

ecto

my

591

Cl

Red

-Int

Fix

Tib

/F

ibu

673

Des

troy

Loc

Lu

ng

Les

Nec

793

Mu

sc/T

end

Ln

gC

han

ge

Nec

651

Bil

at

Ind

Ing

Her

nR

ep780

Tot

Rep

air

Tet

ral

Fal

lot

892

Pro

stR

epair

Ven

tric

Def

735

Per

cuE

nd

osc

Gast

rost

om

y899

Cru

ciat

eL

igR

epai

rN

ec97

7R

evis

Cle

ftP

ala

tR

epair

883

Op

enR

edu

ctM

an

dib

leF

x1,0

30

Sp

inal

Str

uct

Rep

air

Nec

990

Loca

lD

estr

Ova

Les

Nec

947

Fre

eS

kin

Gra

ftN

ec1,1

82

Tot

alS

ple

nec

tom

y1,

066

Rep

air

Of

Cle

ftL

ip955

Nep

hro

ure

tere

ctom

y1,2

12

Rep

air

Of

Gas

tros

chis

is1,

090

Pec

tus

Def

orm

ity

Rep

air

1,0

27

Oth

erG

ast

rost

om

y1,3

42

Non

exci

sD

ebri

dem

ent

Wn

d1,

143

Oth

Un

iS

alp

ingo-O

op

hor

1,0

38

Oth

Per

iton

Ad

hes

ioly

sis

1,4

91

Oth

erC

ran

ioto

my

1,26

3S

egO

steo

pla

sty

Maxil

la1,3

40

Cra

nia

lO

steo

pla

sty

Nec

1,6

59

Inte

rnal

Fix

atio

n-F

emu

r1,

389

Per

iton

eal

Inci

sion

1,4

61

Dors

al/

Dors

olu

mF

us

Post

2,2

89

Syst

emic

-Pu

lmA

rtS

hu

nt

1,39

9P

erit

on

sill

ar

I&

D1,5

08

Part

Sm

Bow

elR

esec

tN

ec2,3

14

Dec

orti

cati

onO

fL

un

g1,

546

Ven

tric

ulo

stom

y1,5

22

Oth

erB

rain

Exci

sion

2,4

03

Cor

rect

Ure

tero

pel

vJu

nc

1,70

6C

left

Pala

teC

orr

ecti

on

1,6

08

Myri

ngoto

my

WIn

tub

ati

on

3,0

33

Op

enR

ed-I

nt

Fix

Hu

mer

us

2,26

2C

lose

dR

ed-I

nt

Fix

Fem

ur

2,0

56

Ton

sill

ecto

my/A

den

oid

ec3,7

33

Op

Red

-Int

Fix

Tib

/Fib

ul

3,52

6O

pen

Red

uc-

Int

Fix

Fem

ur

2,3

99

Occ

lud

eT

hora

cic

Ves

Nec

4,0

62

Clo

sR

ed-I

nt

Fix

Hu

mer

us

4,22

9O

pR

ed-I

nt

Fix

Rad

/U

lna

2,6

89

Lap

aro

scop

icC

hole

cyst

ec7,7

58

Exc

Wou

nd

Deb

rid

emen

t6,

511

Ven

tric

lS

hu

nt-

Ab

dom

en2,7

85

Pylo

rom

yoto

my

8,7

52

Lap

Ap

pen

dec

tom

y19

,790

Ure

tero

neo

cyst

ost

om

y3,5

36

Rep

lace

Ven

tric

leS

hu

nt

4,0

79

Cre

atE

sop

hagast

rS

ph

inc

4,9

85

Oth

erA

pp

end

ecto

my

29,9

67

Tot

al51

,841

67,2

62

45,1

46

NO

TE

:P

roce

du

ren

ames

app

ear

inth

esh

ort

han

dfo

rmth

at

isu

sed

inth

eH

CU

PN

ET

data

base

.

75

Figure 2.1: The Effect of a Liability Decreasing Reform On Treatment.

q'q*

Patient

Benefits

B(q)

Quantity of

Treatment q

Defensive medicine H>H*

Profit seekingmedicine

1

Perfect agent No liability

PPS Fee For Service

H↓ → q ↓ H↓ → q ↑

PPS Type Incentives FFS Type Incentives

H↓ → q ↓

Underprovision H<H*

NOTE: This figure illustrates the response of a healthcare provider to a liability decreasingtort reform. There are two types of financial incentives: (1) Fee For Service, in whichhealthcare is provided in the region q > q′. (2) Prospective Payment System, in whichhealthcare is provided in the region q < q′.

76

Figure 2.2: C-section Rate: TX, FL and CA.

.2.2

5.3

.35

.4.4

5

2000q1 2002q1 2004q1 2006q1 2007q4

C-s

ectio

n R

ate

Time (Quarters)

Texas

NOTE: This figure plots quarterly C-section rates in Texas Florida and California formothers aged 25-34. The solid vertical line (separating quarters 2003-Q3 and 2003-Q4)denotes the time at which the Texas reform was enacted. The figure was constructed usingInpatient data from those three states.

77

Figure 2.3: C-section Rate Medicaid vs. Commercial, TX.

.25

.3.3

5.4

2000q1 2002q1 2004q1 2006q1

Medicaid

Commercial

C-s

ectio

n R

ate

Time(Quarters)

2007q4

NOTE: This figure plots quarterly C-section rates in Texas for mothers aged 25-34 insuredby a commercial carrier and by Medicaid. The solid vertical line (separating quarters2003-Q3 and 2003-Q4) denotes the time at which the Texas reform was enacted. The figurewas constructed using the Texas Inpatient Data.

78

Figure 2.4: C-section Rate Medicaid vs. Commercial, FL.

.25

.3.3

5.4

2000q1 2002q1 2004q1 2006q1

Time(Quarters)

C-s

ectio

n R

ate

Commercial

Medicaid

2007q4

NOTE: This figure plots quarterly C-section rates in Florida for mothers aged 25-34insured by a commercial carrier and by Medicaid. The solid vertical line (separatingquarters 2003-Q3 and 2003-Q4) denotes the time at which the Texas reform was enacted.The figure was constructed using the Florida Inpatient Data.

79

Figure 2.5: C-section Rate Medicaid vs. Commercial, CA.

.25

.3.3

5.4

2000q1 2002q1 2004q1 2006q1

Time(Quarters)

C-s

ectio

n R

ate

Commercial

Medicaid

2007q4

NOTE: This figure plots quarterly C-section rates in California for mothers aged 25-34insured by a commercial carrier and by Medicaid. The solid vertical line (separatingquarters 2003-Q3 and 2003-Q4) denotes the time at which the Texas reform was enacted.The figure was constructed using the California Inpatient Data.

80

Figure 2.6: C-section Rate Medicaid vs. Commercial.

(a) Low-Risk

.05

.1.1

5.2

.25

2000q1 2002q1 2004q1 2006q1

Time(Quarters)

C-s

ection R

ate

Commercial

Medicaid

2007q4

(b) High-Risk

.6.6

5.7

.75

.8

2000q1 2002q1 2004q1 2006q1

Time(Quarters)

C-s

ection R

ate

Commercial

Medicaid

2007q4

NOTE: These figures plot quarterly C-section rates in Texas for mothers aged 25-34 insuredby a commercial carrier and by Medicaid for high-risk and low-risk mothers. The solid ver-tical line (separating quarters 2003-Q3 and 2003-Q4) denotes the time at which the Texasreform was enacted. High-risk is defined as a discharge with one of the following diagnoses:Previous C-section, breech position, early onset, polyhydramnios, oligohydramnios, obesity,diabetes, multiple gestation, distress, hypertension and hemorrhage. The figures were con-structed using the Texas Inpatient Data.

81

Figure 2.7: Ratio of Commercial to Medicaid Surgeries High and Low Tercile Attractiveness.

(a) Texas

-.2

-.1

0.1

.2.3

1999 2000 2001 2002 2003 2004 2005 2006 2007

time(Years)

High Attractiveness

Low Attractiveness

Ra

tio

of com

merc

ial to

Medic

aid

Pro

cedu

res (

in L

og

s)

(b) California

-.1

0.1

.2.3

1999 2000 2001 2002 2003 2004 2005 2006 2007

time(Years)

Ratio

of com

merc

ial to

Med

icaid

Pro

ced

ure

s (

in L

ogs)

Low Attractiveness

High Attractiveness

NOTE: Panel A plots annual ratios of pediatric surgeries in commercially insured patients topediatric surgeries in Medicaid patients for the period 1999-2007, with controls. The commonpediatric surgeries data (described in text) was used. The curve reports dummy coefficients ofthe interaction between year and high and low attractiveness proxies from an OLS regressionwith hospital fixed effects. Panel B is done in an analogous way using the California commonpediatric surgery data and assigning each procedure the Texas attractiveness index.

82

Figure 2.8: Change in Ratio of Commercial to Medicaid Surgeries Pre-Post Reform byAttractiveness, TX.

10

-.4

-.2

0.2

.4

-.4 -.2 0 .2 .4Attractiveness

ΔR

atio

of C

om

me

rcia

l to

Me

dic

aid

Pro

ce

du

res

NOTE: This figure plots percentage change in the ratio of patients insured by acommercial carrier to patients insured by Medicaid following the reform (between 2002 and2004), against the attractiveness proxy. To construct the figure I pooled attractivenessproxies in 0.05 bins. For each bin I calculated the mean percentage change of the ratio ofcommercial to Medicaid surgeries, controlling for year and hospital fixed effects. Thesample and data are described in the text.

83

Figure 2.9: Change in Ratio of Commercial to Medicaid Surgeries Pre-Post Reform byAttractiveness, TX 2001.

11

-.4

-.2

0.2

.4

-.4 -.2 0 .2 .4Attractiveness

ΔR

atio

of C

om

me

rcia

l to

Me

dic

aid

Pro

ce

du

res

NOTE: This figure plots percentage change in the ratio of patients insured by acommercial carrier to patients insured by Medicaid following a fictitious reform in 2001(between 2000 and 2002), against the attractiveness proxy. The figure was constructed inan analogous way to Figure 2.8

84

Figure 2.10: Change in Ratio of Commercial to Medicaid Surgeries Pre-Post Reform byAttractiveness, CA.

12

-.4

-.2

0.2

.4

-.4 -.2 0 .2 .4

Attractiveness

ΔR

atio

of C

om

me

rcia

l to

Me

dic

aid

Pro

ce

du

res

NOTE: This figure plots percentage change in the ratio of patients insured by acommercial carrier to patients insured by Medicaid following a fictitious reform in 2003 inCalifornia (between 2002 and 2004), against the attractiveness proxy. The figure wasconstructed in an analogous way to Figure 2.8, assigning the Attractiveness proxies fromTexas to the California data

85

Bibliography

R. Avraham. An Empirical Study of the Impact of Tort Reforms on Medical MalpracticeSettlement Payments. The Journal of Legal Studies, 36(S2):183–229, 2007.

Ronen Avraham. Database of State Tort Law Reforms (DSTLR 2nd), 2006.

Katherine Baicker, Elliott S. Fisher, and Amitabh Chandra. Malpractice Liability Costs AndThe Practice Of Medicine In The Medicare Program. Health Aff, 26(3):841–852, 2007.

TA Brennan, LL Leape, NM Laird, L. Hebert, AR Localio, AG Lawthers, JP Newhouse,PC Weiler, and HH Hiatt. Incidence of adverse events and negligence in hospitalizedpatients: results of the Harvard Medical Practice Study I. Quality and Safety in HealthCare, 13(2):145, 2004.

H.R. Burstin, W.G. Johnson, S.R. Lipsitz, and T.A. Brennan. Do the poor sue more? Acase-control study of malpractice claims and socioeconomic status. JAMA, 270(14):1697,1993.

David Card, Carlos Dobkin, and Nicole Maestas. Does medicare save lives?*. QuarterlyJournal of Economics, 124(2):597–636, 2009.

J. Currie and W.B. MacLeod. First Do No Harm? Tort Reform and Birth Outcomes*.Quarterly Journal of Economics, 123(2):795–830, 2008.

Janet Currie and Jonathan Gruber. Public health insurance and medical treatment: theequalizing impact of the medicaid expansions. Journal of Public Economics, 82(1):63 –89, 2001. ISSN 0047-2727.

Yolanda Davila, Elizabeth Kaufman, and Stout Sarah. Assessing the impact of medicaidand schip reimbursement rates on access to care in texas,. Technical report, The LyndonB. Johnson School of Public Affairs The University of Texas at Austin, May 2002.

David Dranove and Yasutora Watanabe. Influence and Deterrence: How Obstetricians Re-spond to Litigation against Themselves and Their Colleagues. American Law and Eco-nomics Review, 12(1):69–94, 2010.

86

L. Dubay, R. Kaestner, and T. Waidmann. The impact of malpractice fears on cesareansection rates. Journal of Health Economics, 18(4):491–522, 1999.

RP Ellis and TG McGuire. Provider behavior under prospective reimbursement. Cost sharingand supply. Journal of health economics, 5(2):129, 1986.

R.P. Ellis and T.G. McGuire. Supply-side and demand-side cost sharing in health care. TheJournal of Economic Perspectives, 7(4):135–151, 1993.

J.M. Fanaroff. Medical Malpractice/Expert Testimony/Disclosure of Errors. Pediatrics inReview, 31(3):e24, 2010.

G.M. Fournier and M.M. McInnes. The case for experience rating in medical malpracticeinsurance: an empirical evaluation. Journal of Risk and Insurance, 68(2):255–276, 2001.

Gilbert W. Gimm. The Impact of Malpractice Liability Claims on Obstetrical PracticePatterns. Health Services Research, 45:195211, 2010.

Darren Grant and Melayne Morgan McInnes. Malpractice experience and the incidence ofcesarean delivery: A physician-level longitudinal analysis. Inquiry, 41(2):170–188, 2004.

Jon Gruber, John Kim, and Dina Mayzlin. Physician fees and procedure intensity: the caseof cesarean delivery. Journal of Health Economics, 18(4):473 – 490, 1999. ISSN 0167-6296.

T.J. Hoerger and L.Z. Howard. Search behavior and choice of physician in the market forprenatal care. Medical Care, 33(4):332–349, 1995.

D.A. Hyman, B.S. Black, C. Silver, and W.M. Sage. Estimating the effect of damages caps inmedical malpractice cases: evidence from Texas. Journal of Legal Analysis, 1(1):355–409,2009.

D. Kessler and M. McClellan. Do doctors practice defensive medicine? The QuarterlyJournal of Economics, pages 353–390, 1996.

D. Kessler and M. McClellan. How liability law affects medical productivity. Journal ofHealth Economics, 21(6):931–955, 2002a.

D. Kessler and M. McClellan. Malpractice law and health care reform: optimal liabilitypolicy in an era of managed care. Journal of Public Economics, 84(2):175–197, 2002b.

Daniel P. Kessler and Daniel L. Rubinfeld. Chapter 5 Empirical Study of the Civil JusticeSystem. In A.M. Polinsky and S. Shavell, editors, Handbook of Law and Economics,volume 1 of Handbook of Law and Economics, pages 343 – 402. Elsevier, 2007.

87

D.P. Kessler and M.B. McClellan. The effects of malpractice pressure and liability reforms onphysicians’ perceptions of medical care. Law and Contemporary Problems, pages 81–106,1997.

B. Kim. The Impact of Malpractice Risk on the Use of Obstetrics Procedures. The Journalof Legal Studies, 36(S2):79–119, 2007.

Darius Lakdawalla and Seth A. Seabury. The Welfare Effects of Medical Malpractice Liabil-ity. SSRN eLibrary, 2009.

M.F. MacDorman, F. Menacker, and E. Declercq. Cesarean birth in the United States:epidemiology, trends, and outcomes. Clinics in perinatology, 35(2):293–307, 2008.

G.N. McAbee, S.M. Donn, R.A. Mendelson, W.M. McDonnell, J.L. Gonzalez, and J.K. Ake.Medical diagnoses commonly associated with pediatric malpractice lawsuits in the UnitedStates. Pediatrics, 122(6):e1282, 2008.

E.M. Redinbaugh, A.M. Sullivan, S.D. Block, N.M. Gadmer, M. Lakoma, A.M. Mitchell,D. Seltzer, J. Wolford, and R.M. Arnold. Doctors’ emotional reactions to recent death ofa patient: cross sectional study of hospital doctors. British Medical Journal, 327(7408):185, 2003.

Jessica Wolpaw Reyes. The effect of malpractice liability on the specialty of obstetrics andgynecology. Working Paper 15841, National Bureau of Economic Research, March 2010a.

J.W. Reyes. The Effect of Malpractice Liability on the Specialty of Obstetrics and Gynecol-ogy. NBER Working Paper, 2010b.

Jason Roberson and Eric Torbenson. Tort reform: Debate still thrives over limit on damagesin texas malpractice suits. Dallas Morning News, The (TX), 2007.

Charles M. Silver, David A. Hyman, and Bernard S. Black. The Impact of the 2003 TexasMedical Malpractice Damages Cap on Physician Supply and Insurer Payouts: SeparatingFacts from Rhetoric. as published in Texas Advocate, pp. 25-34, Fall 2008, 2008.

F.A. Sloan. Experience rating: Does it make sense for medical malpractice insurance? TheAmerican Economic Review, pages 128–133, 1990.

F.A. Sloan and J.H. Shadle. Is there empirical evidence for Defensive Medicine? A reassess-ment. Journal of health economics, 28(2):481–491, 2009.

L.E. Smarr. Comparative Assessment of the PIAA Data Sharing Project and the NationalPractitioner Data Bank: Policy, Purpose, and Application. Law AND ContemporaryProblems, 60:59, 1997.

88

D.M. Studdert, M.M. Mello, W.M. Sage, C.M. DesRoches, J. Peugh, K. Zapert, and T.A.Brennan. Defensive medicine among high-risk specialist physicians in a volatile malpracticeenvironment. Jama, 293(21):2609, 2005.

Stephen Zuckerman, Aimee F. Williams, and Karen E. Stockley. Trends In Medicaid Physi-cian Fees, 2003-2008. Health Aff, 28(3):w510–519, 2009.