<%BANNER%>

Three Essays on Health Economics


PAGE 1

THREE ESSAYS ON HEALTH ECONOMICS By SCOTT HANKINS A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2006

PAGE 2

Copyright 2006 by Scott Hankins

PAGE 3

TABLE OF CONTENTS page LIST OF TABLES...............................................................................................................v LIST OF FIGURES..........................................................................................................vii ABSTRACT....................................................................................................................... ix CHAPTER 1 INTRODUCTION........................................................................................................1 2 CERTIFICATE OF NEED REGULAT IONS AND HEALTH OUTCOMES............5 Introduction................................................................................................................... 5 Previous Literature........................................................................................................8 Certificate of Need and Costs................................................................................8 Certificate of Need and Quality.............................................................................9 Quality of Neonatal Health.........................................................................................11 Description of Data.....................................................................................................13 Model Specification....................................................................................................13 Results.........................................................................................................................16 Conclusion..................................................................................................................23 3 HOSPITAL COMPETITION AND PRICES.............................................................29 History of Certificate of Need....................................................................................30 Previous Literature......................................................................................................32 Description of data......................................................................................................35 The Repeal of Certificate of Need and Hospital Competition....................................41 Hospital Competition and Prices................................................................................43 Conclusion..................................................................................................................46 4 MALPRACTICE LAWSUITS AND MEDICAL PROCEDURE USE.....................64 Previous Literature......................................................................................................65 Description of the Data...............................................................................................70 Physicians and Medical Malpractice Lawsuits...........................................................73 Responses to Lawsuits................................................................................................75 Obstetrical Procedures................................................................................................75 iii

PAGE 4

Caesarean Sections in Philadelphia............................................................................78 Caesarean Sections in Allegheny................................................................................80 Labor Inductions in Philadelphia................................................................................81 Labor Inductions in Allegheny...................................................................................82 Conclusions.................................................................................................................82 5 CONCLUSION...........................................................................................................98 LIST OF REFERENCES...................................................................................................99 BIOGRAPHICAL SKETCH...........................................................................................103 iv

PAGE 5

LIST OF TABLES Table page 2-1 States without Certificate of Need (CON)...............................................................25 2-2 Apgar scores.............................................................................................................25 2-3 Summary statistics (standard deviations in parentheses).........................................26 2-4 Results of the first model.........................................................................................26 2-5 Health outcome if premature....................................................................................27 2-6 Results of the second model.....................................................................................27 2-7 Results of the third model........................................................................................28 3-1 Percentage of patient s crossing county borders.......................................................47 3-2 Number of hospitals in the state...............................................................................47 3-3 Percentiles of length of stay (by county type)..........................................................47 3-4 Effect of CON repeal on the HHI (standard errors in parentheses).........................48 3-5 Regression results using OLS (with hospital fixed-effects).....................................49 3-6 Quantile regression results for m odel 1 (with Hospital Fixed-Effects)....................49 3-7 Quantile regression results for m odel 2 (with hospital fixed-effects)......................50 3-8 Quantile regression results for m odel 3 (with hospital fixed-effects)......................50 4-1 Number of doctors by type and county....................................................................84 4-2 Number of patients by doctor type and county........................................................84 4-3 Number of licenses issued and number of doctors sued by county and year the license was issued.....................................................................................................85 4-4 Philadelphia results with csections as the dependent va riable (standard errors in parentheses)..............................................................................................................86 v

PAGE 6

4-5 Philadelphia segmented results with c-sections as the dependent variable (standard errors in parentheses)................................................................................87 4-6 Allegheny results with c-sections as th e dependent variable (standard errors in parentheses)..............................................................................................................88 4-7 Philadelphia results with inductions as th e dependent variable (standard errors in parentheses)..............................................................................................................89 4-8 Philadelphia segmented results with inductions as the dependent variable(standard errors in parentheses)...................................................................90 4-9 Allegheny results with inductions as the dependent variable (standard errors in parentheses)..............................................................................................................91 vi

PAGE 7

LIST OF FIGURES Figure page 3-1 Average price of a C-Secti on (by Payer Type and Year).........................................51 3-2 Average price of a premature birth (by payer type and year)..................................51 3-3 County-level HHI for 1994 (num ber of hospitals in county)...................................52 3-4 County-level HHI for 2004 (num ber of hospitals in county)...................................53 3-5 Number of hospitals per county (1994 vs. 2004)......................................................54 3-6 Number of patients per hospital (1994 vs. 2004).....................................................54 3-7 Hospital-specific HHIs (1994 vs. 2004)...................................................................55 3-8 Hospital-specific HHI averaged at the county-level for 1994..................................56 3-9 Hospital-specific HHI averaged at the county-level for 2004..................................57 3-10 Map of MSAs and major cities in Pennsylvania......................................................58 3-11 Average hospital-specific HHI (by type of county).................................................59 3-12 Average hospital charge (by type of county)...........................................................59 3-13 Distribution of charges for metro counties...............................................................60 3-14 Distribution of charges for non-metro counties.......................................................60 3-15 Averagelength of stay (by type of county)...............................................................61 3-16 Percentage of births delivered by caesarean section (by type of county)................61 3-17 Percentage of births delivere d prematurely (by type of county)..............................62 3-18 Percentage of payers who are HMOs (by type of county).......................................62 3-19 Percentage of payers who ar e Medicaid (by type of county)...................................63 4-1 Percentage of births that ar e caesarean sections (by county)...................................92 vii

PAGE 8

4-2 Percentage of births that are labor inductions (by county).......................................92 4-3 Percentage of births that are vaginal births af ter c-sections (by county).................93 4-4 Percentage of births that are prolonged pregnancies (by county)............................93 4-5 Percentage of births that e xperience fetal distress (by county)................................94 4-6 Percentage of births that l ead to neonatal death (by county)...................................94 4-7 Percentage of births that are premature (by county)................................................95 4-8 Percentage of births that are breech (by county)......................................................95 4-9 Percentage of births to mothers over 35 (by county)...............................................96 4-10 Percentage of births to medicaid mothers (by county).............................................96 4-11 Cumulative probability of a firs t lawsuit in allegheny county.................................97 4-12 Cumulative probability of a firs t lawsuit in Philadelphia county.............................97 viii

PAGE 9

Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THREE ESSAYS ON HEALTH ECONOMICS By Scott Hankins August 2006 Chair: David Figlio Major Department: Economics There are approximately 4 million babies born every year in the United States. My research investigates two distinct issues that may affect both health outcomes of newborns and mothers as well as the costs of their care. Th e first chapter introduces the issues of hospital competition. In the second chapter, I find that increased levels of hospital competition lead to better outcomes for some premature babies. The third chapter investigates the relationship betw een competition and prices. I find that increased competition leads to lower prices and this effect is greater for the lower cost, i.e., least complicated patients. The fourth chapter investigates physician decision making in response to medical malpractice la wsuits. I show that doctors respond to lawsuits by increasing the number of caesarean sections performed, a result consistent with defensive medicine. ix

PAGE 10

CHAPTER 1 INTRODUCTION With approximately 4 million babies bor n every year in the United States, Caesarean sections, and othe r obstetrics procedures, are some of the most common medical procedures performed. At the same time, obstetrics patients are rarely the focus of health economics research. Little is known about the effects of competition on procedure use or health outcomes in this patient population. Hea lth care costs are a growing concern in America. Much academic research has investigated the causes of growing medical costs and, more recently, the benefits and determinants of procedure use. Frequently, health economists rely on Me dicare data as it is r eadily available for all states. By construction, Medica re patients are older and this data ignores a large section of hospital admissions. My research uses data on obstetrics admissions, both mothers and babies. The vast majority of these pa tients are healthy and young which leads to a considerably different population to study. W ith this data, I study the effects of hospital regulation on health outcomes and prices. I also use this data to investigate the effects of malpractice lawsuits on phys ician procedure use. Researchers have documented vast differe nces in procedure use among different geographic areas that lead to higher hea lth care costs without corresponding health improvements. It is unclear whether the tendency to use Medicare data skews these conclusions. It is reasonab le to assume that the underl ying distribution of health outcomes differs between Medicare Patients a nd new mothers and babies. The cause of procedure use variation is stil l an open question. While it is possible that some of these 1

PAGE 11

2 differences are due to patient and physician preferences, th e wide range of legal and competitive environments may be responsible. Obstetrics data sheds new light on these issues. In my first chapter, I use the repeal of Certificate of Need (CON) regulations as a proxy for hospital competition and study the effects of this competition on newborn health outcomes. These regu lations started in New York in 1965 with the intent of controlling health care costs. At the time, hos pitals were paid on a cost-plus basis. That is, their costs were covered plus a given percenta ge as profit. Because of this regulatory structure, hospitals had no incentives to cont rol costs since they we re all but guaranteed of covering them. There was also a fear th at as insurance dulled the price sensitivity of consumers, hospitals in competitive markets w ould engage in a medical arms race and supply a socially excessive amount of medical care. Federal regulation in 1972 re quired hospitals to obtain st ate approval of capital improvements in order to receive Medicare/M edicaid payment for health care that used these improvements. In 1974, federal regulati on required all states to implement a CON program by 1980. In the early 1980s the federal requirement for CON programs was repealed. In the following year, 14 states eliminated their CON la ws. This framework allows me to study the relationship between changes in hospital competition and health outcomes. I find that most babies benefit from bei ng born in a state without Certificate of Need. It appears that the repeal of these laws does not have an immediate effect on newborn health, but outcomes shift over a period of time. In the short run, very

PAGE 12

3 premature babies may be harmed by the repeal. However, it is possible that this effect is due to small sample issues. The second chapter uses de tailed hospital discharge da ta from Pennsylvania for all new mothers and babies in the years surr ounding Pennsylvanias repeal of Certificate of Need regulations. I exam ine hospital competition as the regulatory regime changed and determine the type of hospital most likely to be affected by the repeal. To measure hospital competition, I calculate a hospital-specific HirschmanHerfindahl Index (HHI). Instead of using county borders as the definition of the hospitals market, the hospital-specific HHI us es a patients zip code as the geographic market and constructs a weighted average of these smaller geographic areas. I argue that the hospital-specific HHI is superior to the standard HHI as, it measures each hospitals competitive environment at a much finer geograph ic level. I find that the repeal of CON did not have a measurable effect on hosp ital competition although, it appears that small hospitals in more competitive areas are more lik ely to close. I use the changing levels of hospital competition over these years to investigate the effects of competition on the prices charged by hospitals. I also find that increased levels of competition lead to lower prices. The third chapter uses the same Pennsylvani a discharge data to explore the effect of medical malpractice lawsuits on physicia n behavior. While th ere are a number of papers that investigate the i ssue of litigation induced medical procedures or defensive medicine, this chapter directly measures how doctors respond when they are sued for malpractice. This is distinct from previ ous research which was not able to link an

PAGE 13

4 individual doctors lawsuit experience with their procedur e use and therefore looked at the how changes in the legal environment might be related to changing procedure use. Using the data from Philadelphia and A llegheny Counties, I show that doctors increase the number of caesarean sections they perform after being sued for malpractice. This is consistent with the hypothesis of defens ive medicine. I also find that doctors in Allegheny County respond to a laws uit in a more dramatic fash ion relative to doctors in Philadelphia County. It is pos sible that the larger number of lawsuits relative to Allegheny County causes doctors in Philadelphia to be hyper-sen sitive to the threat of lawsuits given.

PAGE 14

CHAPTER 2 CERTIFICATE OF NEED REGULATIONS AND HEALTH OUTCOMES Introduction Certificate of need regulations affect th e number of hospitals in a community as well as the number and type of services that a hospital can provide. Initially designed to control costs, these regulations may limit co mpetition and potentia lly impact the quality of healthcare. While there has been much res earch on the impact of Certificate of Need laws (CON) on health care costs, few have l ooked at the effect of these laws on health outcomes. In this paper, I document a negati ve correlation between Certificate of Need laws and health outcomes of newborn babies. Specifically, I use data from the U.S. Department of Health and Human Services Linked Birth/Infant Death File (which includes information from almost every birt h certificate matched to the corresponding death certificate), and a differe nce-in-difference approach to show that babies in states with CON laws are less likely to be healt hy (as measured by the 5-minute Apgar score). By comparing health outcomes in states that re pealed their laws at different times, I also show that the positive effects, i.e., greater probabilities of a hea lthy birth, of removing CON restrictions is due to those states which re pealed their laws at least 10 years earlier. In addition, these positiv e effects, of increased hospital competition, are greater for most premature births. Certificate-of-Need regulations started in New York in 1965 with the intent of controlling health care costs. At the time, hospitals were paid on a cost-plus basis. That is, their costs were covered plus a given percentage as profit. Because of this regulatory 5

PAGE 15

6 structure, hospitals had no incentives to cont rol costs since they were all but guaranteed of covering them. There was also a fear that as insurance dulled th e price sensitivity of consumers, hospitals in competitive markets wo uld engage in a medical arms race and supply a socially excessive am ount of medical care. Certificate of need requires state approva l of a hospitals large capital projects, including either the expansion of existing facilities or the in troduction of new services. For example, the purpose of Missouris CON st atute is cost containment through health cost management, assurance of community need and the prevention of unnecessary duplication of health care services. CON is based on a goal of public accountability through public review of proposed health care services, value promotion and negotiation among competing interests (Mei er, July 2001). States differ on the level of investment or changes in the level of service (regardless of changes in capital) that requires prior approval. Often a decrease in the level of se rvices as well as an increase must go through approval process. Federal regulation in 1972 required hospita ls to obtain state approval of capital improvements in order to rece ive Medicare/Medicaid payment for health care that used these improvements. In 1974, federal regulati on required all states to implement a CON program by 1980. The Reagan revolution in the early 1980s removed the federal requirement for CON programs and, subseque ntly, some states eliminated their CON laws. However, most did not, and some e xpanded their programs. Currently, 14 states do not have a CON program ( Table 2-1 ). There is a perception within the health care industry that CON programs protect incumbents (i.e., Stiglers capture theory of regulation). The director of government

PAGE 16

7 relations for the Missouri State Medical Asso ciation said, The Certificate of Need program has outlived its usefulness. It doesn t do anything but stifle competition and innovation. Its extremely bureau cratic, and no one relishes ha ving to go through it. Its the people who have existing projects who want it to continue. It helps them keep competition out of their bac kyard (Meier, July 2001). Economic theory suggests that certificate of need regulation will lead to inefficient results in competitive markets. However, the hospital market is far from competitive. There is a mix of non-profit and profit hospitals in most geogr aphical markets. There are also areas with only one hospital. While there has been research on competition and medical care, there is not a generally accepted theory of how competition affects prices/costs and quality. As will be discu ssed, others have looked at the effects of competition and prices/costs. It is hypothesized that either prices/costs will be higher than they would be in the absence of pure comp etition or that quality will be lower. This can occur because incumbents are protect ed from competition because competitors (either new entrants or exis ting hospitals that wish to expand) must go through the certificate of need process. This process is costly and, in many states, an existing hospital is allowed to testify against a potential entr ant. As with any market barrier, it is reasonable to expect less strenuous competition and possibly less innovation. If these barriers limit competition along both the price as well as the quality dimension, then it is reasonable to expect sub-optimal health outco mes in the states with CON requirements. Previous studies have shown that prices/costs tend to be higher in states with CON requirements; this paper shows that there is al so a decrease in health outcomes in those same states.

PAGE 17

8 Previous Literature The existing literature on certificate of need and hospital competition covers two distinct areas. The earliest research looked at the impact of both CON regulation and its subsequent repeal in some st ates, as well as general competition within the health care market, on costs. More recently, researcher s have begun to investigate the effect of competition, and specifically CON programs, on the quality of health care. Certificate of Need and Costs Sloan (1988) provides a detailed review of the early literatu re. In general, the early studies find that CON has no stat istically significant effect on health care costs and most agree that if there were to be an effect, it would increase costs rather than lower costs. CON proponents argued that these early studies were flawed because of their limited time horizon immediately following the introduction of these programs It was argued that the programs needed time to get up and running be fore they could be expected to control costs. More recent studies have l ooked at longer periods of tim e and evaluated the effects of repeal of the programs (a good survey is Morrisey, 2000). These studies suggest that, if anything, CON programs have tended to incr ease costs (Sloan, Morrisey and Valvona, 1988) and found that the repeal of CON ha d no effect on hospital costs per capita (Sherman, 1988). Conover and Sloan (1998) found that mature CON programs result in a slight reduction in bed supply bu t higher costs and higher profits. Antel, Ohsfeldt, and Becker (1995) analy zed hospital costs allowing for interaction between regulation programs other than CON. Using state data on hospitals costs per day, per admission, and per capita, they found that CON had no statistically significant effect in any of their empirical specifications Health care costs a nd prices charged both

PAGE 18

9 increase with the duration of the CON program Using Medicare price and cost data from 1977, Noether (1988) found prices were higher relative to costs and therefore profits were higher in states with a CON program. CON proponents also argued that hospitals were different than other industries and therefore competition did not, and could not, work. However, the empirical evidence doe s not support this. Melnick, et al. (1992) looked at transaction prices that a large Ca lifornia preferred provider organization (PPO) negotiated with hospitals in 1987. While controlling for many factors, they found that the PPO paid more in the less competitive mark ets. This suggests that competition does work to lower prices in hospital markets. Certificate of Need and Quality While much research has been done on the de terminants of health care quality (see Tancredi, 1988 for a review), very little work has been devoted to examining the effects of health care regulation or competition on quality. In one of the few papers, Ho and Hamilton (2000) looked at the effects of hosp ital consolidation on h ealth care quality. Analyzing California hospital care before and after mergers and acquisitions between 1992 and 1995, they looked at several proxies for quality of care. They find 90-day readmission rates for heart attack patients and discharges within 48 hours for normal newborn babies increased in some cases. While mortality and readmission rates are reasonable, though imperfect, metrics for health ca re quality, it is unclear that an increase in early discharges should be considered as such. The authors equate early discharge with cost cutting measures of hospitals, but the connection with quali ty is not discussed. Furthermore, they find no measurable effect on inpatient mortality for heart attack and stroke patients.

PAGE 19

10 A recent paper by Kessler and McClellan ( 2000) looks at hospital competition and Medicare beneficiaries heart attack car e from 1985 to 1994. They find the welfare effects, i.e., expenditures and treatment and pa tient health outcomes, of competition to be ambiguous in the 1980s. However, in the 1990s, competition unambiguously improves social welfare. In addition to these papers on competition and the quality of h ealthcare, two papers have focused on the effect of CON regulations on Coronary Artery Bypass Graft (CABG) quality. Robinson, et al. (2001) looked at outcomes in Pennsylvania for the years 1994 1999. The elimination of the states CON pr ogram is found to increase the number of open-heart surgery programs by 25% without a significant increase in the number of surgeries performed. With this limited sample, quality (as measured by the mortality rate) was not impacted by this reallocation of volume. Va ughan-Sarrazin, et al. (2002) consider a larger dataset of Medicare beneficiaries who underwent CABG surgery between 1994 and 1999. The authors find highe r mortality rates fo r CABG patients in states without CON programs (5.1% compared to 4.4%). There is one potential problem with this paper, it compares cross-sectiona l differences across states. The authors cite several papers that point to a negative c onnection between hospita l volume and mortality for CABG surgery. While they do not state su ch a hypothesis, it woul d appear that there is a type of learning by doing in open-heart surgery. This is a r easonable hypothesis as open-heart surgery is a complicat ed and lengthy process in whic h the skills of the surgeon and the other medical professionals could have a great impact on patient outcomes. While the previous literature looke d at both CON regulation and hospital competition as determinants of health car e costs, the question how CON regulations

PAGE 20

11 affect health care quality ha s only begun to be addressed. This paper will attempt to evaluate the issue of certificate of need regulation, hospital competition and the quality of care. Specifically, the effect of CON regulation on the qua lity of neonatal care will be investigated. Quality of Neonatal Health Birth certificates in the United States r ecord many different kinds of information about the parents as well as the baby. In formation about both parents socioeconomic background, as well as factors affecting the mo ther and babys health are collected. One variable recorded is the -minute Apgar score for the newborn baby. I use the Apgar score as a proxy for the qua lity of neonatal health. The Apgar score is a subjective measure of the infants condition based of heart rate, respiratory effort, muscle tone, reflex irritability, and color. Each of these factors is given a score of 0, 1, or 2; the sum of thes e 5 values is the Apgar score, which ranges from 0 to 10. A score of 10 is optimal, and a very low score raises a flag about the subsequent health and the survival of the in fant. The Apgar score was designed to be a useful measure of the need fo r resuscitation and a predictor of the infants chances of surviving the first year of life. A recent paper by Casey, et al. (2001) shows that for premature infants (26 to 36 weeks of gesta tion), the neonatal mortality rate was 315 per 1000 births for an infant with an Apgar scor e of 0, as compared with 5 per 1000 births for an infant with an Apgar score of 7. Si milar results are demonstrated for full term babies. Despite the emphasis placed on low birth we ight and poor health outcomes, in the popular press, there is some ev idence that low birth weight is not itself the sole predictor of infant mortality. One paper concludes that the threshold weight below which mortality

PAGE 21

12 is significantly greater is the 3rd percentile (McIntire, et al (1999)). A recent working paper by Almond, et al. (2002) uses twins to compare the correlation between birth weight and various health outcomes. The au thors find that the heavier twin is no more likely to survive past the first year than the lighter twin. They also find that the Apgar score is more highly associated with infant mortality but is not correlated with birth weight. This is because a low Apgar score may be caused by a birth trauma not related to prematurity. For these reasons, low birth wei ght is not used as a health outcome. After consultation with a labor and deliver y nurse, I chose Apgar scores of 8 and above to be healthy. 1 The reasoning for this is as follows, while the Apgar score is a subjective measure based on the health care professionals opinion, there appears to be some agreement on what score denotes a healthy baby. For example, one nurse may give a baby an Apgar score of 9 while another may gi ve a score of 8, the score is different but both professionals would agree that the baby was healthy. It is highly unlikely that one professional would score a baby a 6 and anot her score the same child as an 8. In addition, while the percentage of Apgar scores either 9 or 10 have changed from 1983 to 1999, the percentage of Apgar scores greater than or equal to 8 has stayed relatively stable. 2 While the break between score of 8 and 9 appears to be the natural break (see Table 2-2 ), I have chosen to be conservative and defined the break point for health to be between 7 and 8. ____________ 1 As a robustness check, the same models were estimated with a cutoff of 7 or 9. The conclusions do not change with the different cutoff points. 2 One potential reason for the shift from Apgar scores of 10 to Apgar scores of 9 is the increasing use of pain medication over this time period, specifically epidurals. These medications tend to dull the babys responses and therefore affect the Apgar score.

PAGE 22

13 Description of Data The data set and the descript ion of the variables are from the National Center for Health Statistics Matched Birt h/Death Certificates file for 1983 and 1999. The beginning of the sample was chosen to be 1983 becau se in years preceding that time the Matched Birth/Death was incomplete for many of the states for earlier years. The data for 1983 are based on the total number of births in 46 states. In 1983, California, Delaware, Oklahoma and Texas did not report Apgar scores on their birth certific ate therefore these four states were excluded from the sample. Unfortunately, California and Texas are two of the 15 states without a CON program. New Mexico was also dropped from the sample because it did not record gestational age in 1983. Louisi ana was excluded from the analysis because it was the only state to not implement a CON regulation while it was required by the federal government 3 Arizona, California, Dela ware, and Georgia as well as the District of Columbia only reported a 50 percent random sample of their births. The data set is limited to those births for which all variables of interest, described in the model specification secti on, are available. In order to control for differences in prenatal care in other countries, an observ ation was also dropped if the mother was a resident of another country 4 We are left with 2,60 2,155 observations in 1983 and 2,920,950 observations in 1999, for a total of 5,523,105 observations. Model Specification The basic model estimated is, ju year noCON noCON X outcome3 2 199 99 89 ____________ 3 Louisiana actually started its CON regulations in 1991, after 11 states had repealed their CON laws. 4 Most of the discarded observations are lacking an Apgar score or gestational age.

PAGE 23

14 where outcome is an indicator of either a healthy birth or a ne onatal death. For the models that estimate the probability of a h ealthy birth, the dependent variable, is a dummy variable that takes the value of one if a baby has a 5-minute Apgar score greater than or equal to 8 depending on the model. Because the 5-minute Apgar score is a measure of health at the 5-mi nute mark, it is an imperfect measure of hospital quality at best. It is likely that many quality related problems would occur af ter the 5-minute mark and would not be picked up in this data. Noneth eless, it is reasonable to believe that some quality related problems would occur before th e Apgar score. Because the vast majority of problems will not be captured in the Apgar score, the estimates can be considered a lower bound. The fact that we find any resu lts with the healthy dependent variable, suggests that these laws have an effect. Because the Apgar score may not pick up a ll health problems, the same model is estimated with the dependent variable an indicat or of neonatal death. That is, the variable is equal to one if the baby died in its first ye ar of life and is zero otherwise. If CON laws have an effect, we would expect the results to have opposite signs b ecause of the nature of the indicators. For example, if CON laws are beneficial then the estimated coefficients on the noCON indicators will be negative if the dependent variable is healthy (indicating that dropping a states CON program redu ces the probability of a healthy baby). Similarly, the estimated coefficients will be positive for the noCON indicators if the dependent variable is death (indicating that dropping a states CON program increases the probability of a baby dying in its first year). The variables of interest are the noCON variables; noCON89 is a dummy variable that takes the value of one if the state has dropped its Certificate of Need program by

PAGE 24

15 1989 and noCON99 is an indicator that is e qual to one if the state dropped its CON program between 1990 and 1999. The noCON variab les are the difference in differences estimates of the effect of dropping a states certificate of need pr ogram either in the 1980s or the 1990s relative to the states that still maintain these regulations. That is, noCON89 is the difference between 1983 and 1999 in the difference between states with a CON program and those that dropped their CON program in the 1980s (a similar interpretation holds for noCON99). As previ ously mentioned, of the 45 states in the sample and Washington D.C., 14 had droppe d their CON programs by 1999 (11 in the 1980s and 3 in the 1990s). The obvious question is why did only some of the states repeal their regulations in the 1980s. And then why did the second wave of repeals occur in the 1990s. On the surface, endogeneity appears to be a problem However, it is likely to be less of a concern than is initially apparent. Omitted variables are the most common form of endogeneity; to the extent to which the underlying factors do not change much over time, the omitted variables problem is dealt with using state level fixe d effects. If there are time-varying omitted variables, we may have a reverse causality problem. That is to say, the dependent variable influences the variable of interest. In this case, health outcomes, or underlying health trends, would have to affect the repeal of a states certificate of need regulations. This is highly unlikely, in order for this to be true one would have to argue that politicians or bureaucrats obser ve a downward trend in health outcomes, relative to other states, and conclude that the way to fix this problem is through the repeal of the states CON regulation. One could argue that obvious solution from a planners perspective

PAGE 25

16 would be to argue for more regulation in this case, rather than less. If this is the case, then the potential bias in this case works against finding any effect from the repeal of CON. The mother and child characteristics contai ned in X are: the mothers age at the time of birth and the babys birth we ight rounded to the nearest 100 grams 5 (See Table 2-3 ). Year99 is an indicator that is e qual to one if the year is 1999; while, uj represents state fixed effects and is a random error. All models are estimated as a linear probability model with state fixed effects to control for unobserved state specific variation. As the certificate of need laws vary at the state level, all of the standard errors are corr ected for clustering of the errors at the state level as described in Moulton (1990). State fixed effects and state level laws imply that we are estimating probabilities at the state level, i.e., the av erage probability of a healthy baby in a state, conditional on observed characte ristics. Because the vast majority of births in the United States are healthy ones, we are trying to explain a rare event, that of an unhealthy birth or a neonatal death. The ra rity of this event, combined with the limited information recorded on the birth cer tificate, makes one suspect that a goodness of fit measure such as R2 is going to be poor. Indeed, this is what is found for all of these models. Results As mentioned in the previ ous section, the dependent variable (Healthy) is an indicator equal to one if the A pgar score is greater than or equal to 8, and zero otherwise. ____________ 5 While birth certificates record other useful information, i.e mothers education, number of prenatal visits, etc. not all states report these variables. The model was limited to mothers age and birth weight to maximize the number of states included in the data. Gestational age is used in later models.

PAGE 26

17 The coefficients all have the expected sign (see Table 2-4 column 1). The variable year99 is positive and significant, which sugge sts that the likelihood of a healthy baby increased from 1983 to 1999, given the advances in medical technology (such as fetal monitoring), this intuitively make s sense. The coefficient on birth weight is positive and statistically significant. Babies with higher birth weight are more likely to have an Apgar score of 8 or higher, an increase of one st andard deviation (600 gr ams or 21 ounces) leads to an increase in the probabil ity of a healthy birth by 4 %. The variables of interest in this paper are the noCON indicators. This variable takes the value of one if a state does not have a Certificate of Need program and zero otherwise. The noCON89 coefficient is both st atistically and economi cally significant. A baby born in a state that removed its CON program in the 1980s is .59 % more likely to be healthy than a baby born in a state with a CON program. While this may appear economically insignificant, the estimated increa se in the probability of a healthy birth only increased 1.2% from 1983 to 1999. As the vast majority of babies are born healthy, it is impossible for the effect of CON repeal to lead to a large change in the likelihood of a healthy birth. The noCON99 coefficient is negative but not statis tically significant. This implies that the states which dropped th eir restrictions on hos pital competition in the 1990s did not see an increase in the likelihood of a healthy birth. This result is not surprising, since the entire 2nd wave of CON removals occurr ed in the later part of the 1990s. 6 The differential impact between the ea rly and later groups implies a time lag between the removal of competition barriers a nd an increase in hospital competition or the increased competitions effect on health outcomes. Given the time required to ____________ 6 The three states that dropped their CON regulations in the 1990s did it in either 1995 or 1996

PAGE 27

18 finance and construct major capital improvement s, a delayed effect on health outcomes is not surprising. The lack of an effect implies that it takes time for there to be an increase in hospital competition and for this increase to have an effect on health outcomes. Given the length of time it takes to build new a hospital, a lag in th e effect of removing constraints on competition is to be expected. Because the Apgar score is an imperfect measure of a newborn babys health, the same basic model was estimated with an i ndicator of neonatal death as the dependent variable. Again, the coefficients have the expected signs. An increase in the babys weight of 600 grams (one standard deviation) leads to a 2% reducti on in the probability of death. While the coefficient on mothers age is positive, indicating increasing age increases the chance of death, it is difficult to argue that the coefficient is economically significant. The coefficient on the year99 indicator is negative, which again makes sense given the improvements in medical technology. The coefficients on both of the noCON variables are statistically a nd economically insignificant. This is not completely unexpected since very few babies are so sick that they are in danger of dying. However, it raises the question of the effects of CON re gulation on high-risk, i.e., premature, births. Because premature births are more likely to have health problems and are more likely to die, there is the possibility of more aggre ssive medical interventions and hence, more quality related issues. The Interactions of Certificate of Need regulation and Premature Births: Because CON programs may affect neonatal health by limiting Neonatal Intensive Care Units (NICUs), the same model is estimated wi th the inclusion of different indicators of premature birth interacted with the certificat e of need indicators as before. These high

PAGE 28

19 risk births are more likely than a normal birt h to require the services of a NICU. While not all premature births need a NICU, the pr esence of a NICU indicates that the hospital staff is more prepared to ha ndle these high-risk births. The average length of gestation is 39 w eeks, while the definition of full term pregnancy is 40 weeks. Anything less than 37 weeks is considered premature and increases the risk of medical complications for the infant and the mother. Premature births are relatively rare in the data (approximately 10% are born at less than 36 weeks in 1983 and 12% in 1999). Of these premature birt hs, the majority of premature births are born between 36 and 32 weeks (variable P36), th is represents 7% of the total number of births in 1983 and 9% in 1999. As can be seen in Table 2-5 all types of premature births have increased relative to 1983. Babies born premature are less likely to be healthy and more likely to die. Although these babies are le ss healthy, their prospects have improved over the time period. For those births with a gestational age between 32 and 36 weeks, the probability of being healthy has increased from 92% to 95%, while the mortality has decreased from 2% to 1%. To test the effect of CON regulations on premature births, an indicator of prematurity is created that e quals one if the baby had a ge stational age of 36 weeks or less. The first of these models estimated is 7 ju year noCON PRE noCON PRE PRE noCON noCON X outcome3 3 2 1 2 199 99 89 99 89 ____________ 7 The secondary interactions are not repo rted but are available upon request.

PAGE 29

20 where PRE is an indicator of premature birth (i.e., PRE equals 1 if gestation is less than or equal to 36 weeks). The coefficient on the interacted term can be interpreted as a difference in difference in differences estimate. For example, the interaction of PRE with one of the noCON indicators gives us the difference between these high-risk births and other births in the difference between states with a certificate of need program and those without a program. Give n that noCON is a difference in difference estimate, the coefficient on the interaction term is the difference in difference in differences estimate. First, the model was estimated with health as the dependent vari able. As expected, being born prematurely significantly reduces the probability of being healthy (-7.7%). The coefficient on the interaction of the noCON indicators with the premature indicator is positive for both the states that dropped thei r CON in the 1980s and in the 1990s. Again, we see that only the interacti on with noCON89 is statistically significant. The estimated impact on premature babies is 1.98% (Column 1, Table 2-6 ) in those states that repealed their CON programs in the 1980s. These posi tive coefficients indicate that removing barriers to competition results in a higher like lihood of a healthy birth. As before, the higher probability of a healthy birth in the states that rem oved their CON restrictions in the 1980s versus those that rem oved the restrictions in the 19 90s is reasonable given the time needed for increased competition to have an effect on health care. Again, this may appear to be a small effect. However, wh en compared to the average change in the probability of being healthy over this time peri od, the effect of CON repeal leads to a large increase.

PAGE 30

21 Next, the same model was estimated with death as the dependent variable. As would be expected, being born prematurely in creases the chance of death (2.96%). The interactions of the prematur e indicator and the noCON indi cators are not as easy to interpret as with the health regression. Premat ure babies born in states that removed their CON restrictions in the 1990s are slightly mo re likely (.23%) to di e relative to those states that still maintain CON restrictions. There is no statistically significant effect on those babies born in the states that removed their CON restrictions in the 1980s. This seems to imply that there are short run costs to the removal of CON, although this effect is small from an economic significance point of view. Because it is possible that the effect of cer tificate of need regulation varies with the gestational age of the baby, i ndicators of prematurity by sp ecific weeks are created. The variables P28, P32, P36 are indicators of prem ature births (i.e., P28 indicates gestation less than or equal to 28 weeks while P32 indicates gestation greater than 28 and less than or equal to 32 weeks). These indicators of premature birth are interacted with the noCON indicators and the following model estimated is 8 : ju noCON P noCON P noCON P noCON P noCON P noCON P PPP year noCON noCON X outcome3 2 1 3 2 1 3 2 1 3 2 199 3699 3299 28 89 3689 3289 28 363228 99 99 89 ____________ 8 As before, the secondary interactions are not reported.

PAGE 31

22 Not surprisingly, the probability of a hea lthy birth is monotonically increasing in the prematurity indicators and the probability of death is monotonically decreasing ( Table 2-7 ). The extremely premature births (P28) ar e much less likely to be born healthy (47%) and more likely to die (34%) than a full term birth. The births with a gestational age between 28 and 32 weeks (P32) have a better chance of a healthy birth (17%) and are less likely to die (4%) than the extremely premat ure births. The babies born between 32 and 36 weeks are only slightly more likely to be unhealthy (3%) or to die (.25%) than a full term birth. For all but the oldest premature births, th e removal of CON restrictions does not appear to have an effect. For the P36 births the interactions of the noCON indicators are positive for the health outcomes regardless of when the CON restrictions were removed. As before, the effect of removing the CON re strictions is larger for the states that repealed their laws in the 1980s versus those that repealed in the 1990s. These results are also economically significant given the re duced probability of being healthy if born prematurely ( Table 2-5 ). Again, we see an effect in the short run wh en the dependent variable is death. The interaction of noCON99 and P28 is positive ( 3.07%); indicating the repeal of Certificate of Need in the 1990s has led to an increas e in mortality for the extremely premature births. We also see a decrease in mortality for the babies born between 28 and 32 weeks. The interaction of noCON99 and p32 is -.82%. It appears that thes e babies benefit from the repeal. There is no effect from the repeal of CON in the 1980s. It is possible that these contradictory re sults are due to the small number of very premature babies, especially in those states that dropped their laws in the 1990s, in this

PAGE 32

23 data set. However there is a possible economic explanation for these results. The certificate of need process was not established simply to hinder competition but also to maintain acceptable levels of quality in the hospital industry. One argument in favor of the certificate of need process is to maintain a critical volume at area hospitals, especially in services like NICUs. It is conceivable that, in the short-term, the removal of CON leads to worse outcomes for very high-risk infants by expanding the number of places where a premature baby can be born within a geographical area. Newly opened hospital facilities may not have high enough volumes of th ese special needs of births to allow the medical staff to maintain their skills of a more established hospital. A baby born with less than 28 weeks of gesta tion is extremely small and is likely to have health problems. A baby of this size is di fficult to intubate or administer an I.V. to when such steps are necessary and problems in these processes could impact the 5-minute Apgar score. These are also tasks for which there may be learning by doing present. If experience leads to greater medical proficie ncy, learning by doing could result in better health outcomes in those hospitals that perf orm a large number of these procedures. This hypothesis would lead us to e xpect that very high-risk in fants would be harmed by the repeal of CON regulations, as seen in these em pirical findings. In the long-run, the repeal of CON regulations allows co mpetitor hospitals to establ ish NICUs and to achieve high enough volume levels to improve outcomes. It is important to remember that the vast majority of births benefit immediately from the repeal of CON regulations and given time extremely premature births benefit as well. Conclusion Restricting competition within the hospital market appears to negatively affect the quality of care as measured by health outcome s. Complementing the earlier findings that

PAGE 33

24 certificate of need regulation doe s not control health care co sts, this paper shows that CON regulations are associated with lower A pgar scores and a slightly higher incidence of neonatal deaths. However, a small subsection of high-risk births may benefit from a CON program. More research to determin e the cause of this is necessary. As this data does not include information on costs, it is impossible to say whether or not the certificate of need process is socially efficient. Yet, given the negative impact of these laws on the overwhelming majority of births, Certificate of Need regulations may not be the optimal social policy.

PAGE 34

25 Table 2-1. States without Ce rtificate of Need (CON) State Year Dropped Arizona 1985 California* 1987 Colorado 1987 Idaho 1983 Kansas 1985 Minnesota 1984 New Mexico 1983 North Dakota 1995 Ohio 1995 Pennsylvania 1996 South Dakota 1988 Texas* 1985 Utah 1984 Wyoming 1985 does not record Apgar scores Table 2-2. Apgar scores 1983 1999 Apgar Percent of Births Total Number Percent Dead Total Died Percent of Births Total Number Percent Dead Total Died 0 0.06 1,547 0.61 950 0.07 2,164 0.55 1180 1 0.21 5,543 0.77 4284 0.18 5,263 0.82 4303 2 0.14 3,629 0.55 1989 0.08 2,415 0.56 1356 3 0.14 3,752 0.36 1333 0.09 2,607 0.32 826 4 0.20 5,285 0.23 1225 0.14 4,079 0.18 739 5 0.38 9,889 0.15 1468 0.25 7,357 0.12 872 6 0.79 20,680 0.08 1754 0.59 17,184 0.08 1354 7 1.99 51,743 0.04 1821 1.51 43,961 0.04 1596 8 8.72 226,889 0.01 2618 7.06 206,303 0.01 2140 9 62.62 1,629,369 0.00 6699 82.05 2,396,692 0.00 5492 10 24.74 643,829 0.00 2102 7.97 232,925 0.00 439 Total 2,602,155 1.01 26,243 2,920,950 0.69 20,297

PAGE 35

26 Table 2-3. Summary statistics (standard deviations in parentheses) Variable 1983 1999 Total Mom's Age 25.554 27.209 26.429 (5.307) (6.156) (5.831) Birth Weight (100s grams) 33.514 33.144 33.319 (5.944) (6.152) (6.058) Gestation (weeks) 39.405 38.772 39.070 (2.825) (2.620) (2.737) Healthy 0.961 0.971 0.966 (0.194) (0.168) (0.181) Died 0.010 0.007 0.008 (0.100) (0.083) (0.091) (P28) Gestation <= 28 weeks 0.009 0.010 0.009 (0.093) (0.098) (0.096) (P32) 28 weeks < Gestation <=32 weeks 0.014 0.016 0.015 (0.119) (0.127) (0.123) (P36) 32 weeks < Gestation <=36 weeks 0.071 0.092 0.082 (0.256) (0.289) (0.274) Table 2-4. Results of the first model (1) (2) Healthy Died Coefficient Std Error Coefficient Std Error noCON89 0.5808 (0.2040) -0.0197 (0.0613) noCON99 -0.0807 (0.2612) 0.0795 (0.0324) Weight 0.6696 (0.0132) -0.3350 (0.0076) Age -0.0226 (0.0028) 0.0091 (0.0012) Year99 1.2182 (0.0690) -0.4593 (0.0246) Constant 74.2231 (0.4208) 12.0051 (0.2452) Observations 5523105 5523105 Number of state 45 45 R-squared 0.05 0.05

PAGE 36

27 Table 2-5. Health outcome if premature 1983 Prematurity Percentage of Births Total Number Percent Healthy Percent Died Gestation <=28 week s 0.88 22,808 0.44 0.38 28 weeks < Gestation <=32 weeks 1.44 37,358 0.76 0.07 32 weeks < Gestation <=36 weeks 7.05 183,478 0.92 0.02 36 weeks < Gestation 90.64 2,358,511 0.97 0.01 1999 Prematurity Percentage of Births Total Number Percent Healthy Percent Died Gestation <=28 week s 0.96 28,127 0.42 0.34 28 weeks < Gestation <=32 weeks 1.65 48,076 0.83 0.04 32 weeks < Gestation <=36 weeks 9.21 269,161 0.95 0.01 36 weeks < Gestation 88.18 2,575,586 0.98 0.00 Table 2-6. Results of the second model (1) (2) Healthy Died Coefficient Std Error Coefficient Std Error noCON89 0.4942 (0.1535) -0.0293 (0.0560) noCON99 -0.0857 (0.2148) 0.0430 (0.0259) Weight 0.5397 (0.0118) -0.2952 (0.0074) Age -0.0189 (0.0028) 0.0084 (0.0012) Year99 0.9354 (0.0593) -0.2786 (0.0177) Premature -7.6778 (0.2647) 2.9625 (0.0780) Premature x noCON89 1.9781 (0. 9056) -0.2087 (0.2041) Premature x noCON99 0.3842 (0 .6014) 0.2386 (0.0889) Constant 79.2237 (0.3660) 10.4124 (0.2379) Observations 5523105 5523105 Number of state 45 45 R-squared 0.06 0.05

PAGE 37

28 Table 2-7. Results of the third model (1) (2) Healthy Died Coefficient Std Error Coefficient Std Error noCON89 0.4761 (0.1507) -0.0185 (0.0535) noCON99 -0.1009 (0.1973) 0.0525 (0.0182) Weight 0.3196 (0.0075) -0.1576 (0.0043) Age -0.0095 (0.0024) 0.0022 (0.0010) Year99 0.9090 (0.0547) -0.2615 (0.0159) P28 -47.7169 (0.8620) 34.3691 (0.8272) P32 -17.3646 (0.5657) 4.0656 (0.1837) P36 -3.0095 (0.2340) 0.2526 (0.0480) P28 x noCON89 2.3086 (3.4791) -0.5584 (3.0722) P32 x noCON89 3.3508 (2.2495) -0.8877 (0.5830) P36 x noCON89 1.6457 (0.6683) -0.1384 (0.1504) P28 x noCON99 -3.1427 (2.3017) 3.2114 (0.8427) P32 x noCON99 1.1753 (0.6270) -0.8233 (0.3561) P36 x noCON99 0.7822 (0.4249) 0.0286 (0.0559) Constant 86.5317 (0.2382) 5.8538 (0.1431) Observations 5523105 5523105 Number of state 45 45 R-squared 0.11 0.15

PAGE 38

CHAPTER 3 HOSPITAL COMPETITION AND PRICES Originally mandated by the Federal gove rnment, Certificate of Need (CON) regulations were designed to control hos pital competition. With the goal of limiting redundancy and tempering a medical arms race, Certificate of Need regulations restricted hospital competition. The repeal of these laws creates new opportunities for the hospital markets; new entrants have a lowe r barrier to entry and existing players can expand services more easily. To explore the relationship between the re peal of CON, hospital competition and prices charged by hospitals, I employ two techniques. First, a hospital-specific Hirschman-Herfindahl Index (HHI) is constr ucted to capture each hospitals competitive environment. Second, I use quantile regressi on to estimate the effect of changes in competition on prices. The extreme skewness of price data could unduly influence the econometrics otherwise. While prior hospita l competition research focused on Medicare data, this paper uses the fu ll sample of obstetrics patient s in Pennsylvania between 1994 and 2004. These patients differ greatly in terms of both age and general health from Medicare patients providing a new perspect ive on the impact of hospital competition. Given the type of data, it is impossible to identify a causal relations hip between the repeal of CON and hospital competition. However, I find that while the repeal of CON is not correlated with changes in competition, changes in the competitive landscape have an impact on prices. This effect is larger for the lower quantiles, implying that competition matters more for the easier/cheaper medical care. 29

PAGE 39

30 History of Certif icate of Need Certificate of Need regulations affect th e number of hospitals in a community as well as the number and type of services that a hospital can provide. Certificate of Need regulations started in New York in 1965 with the intent of controlling health care costs. At the time, hospitals were paid on a cost-plus basis. That is, their costs were covered plus a given percentage as profit. Because of this regulatory struct ure, hospitals had no incentives to control costs since they were al l but guaranteed to cover them. There was also a fear that as insurance dulled the price sensitivity of consumers, hospitals in competitive markets would engage in a m edical arms race and supply a socially excessive amount of medical care. These regulations are designed to limit competition and in the process, control costs. The Certificate of Need process requires hos pitals to receive state approval of large capital projects, including either the expansion of existing faci lities or the introduction of new services. For example, the purpose of Missouris CON statute is cost containment through health cost management, assurance of community need and the prevention of unnecessary duplication of hea lth care services. CON is based on a goal of public accountability through public review of propos ed health care services, value promotion and negotiation among competing interests (Mei er, July 2001). States differ on the level of investment or changes in the level of serv ice (regardless of cha nges in capital) that requires prior approval. Often a decrease in the level of services as well as an increase must go through approval process. Federal regulation in 1972 required hospitals to obtain state approval of capital improvements in order to rece ive Medicare/Medicaid payment for health care that used these improvements. In 1974, federal regulati on required all states to implement a CON

PAGE 40

31 program by 1980. The Reagan revolution in the early 1980s removed the federal requirement for CON programs and, subseque ntly, some states eliminated their CON laws. However, most did not, and some e xpanded their programs. Currently, 15 states, including Pennsylvania, do not have a CON program. There is a perception within the health care industry that CON programs protect incumbents (i.e., Stiglers capture theory of regulation). The director of government relations for the Missouri State Medical Asso ciation said, The Certificate of Need program has outlived its usefulness. It doesn t do anything but stifle competition and innovation. Its extremely bureau cratic, and no one relishes ha ving to go through it. Its the people who have existing projects who want it to continue. It helps them keep competition out of their bac kyard (Meier, July 2001). Economic theory suggests that Certificate of Need regulation will lead to inefficient results in competitive markets. However, the hospital market is far from competitive. There is a mix of non-profit and profit hospitals in most geogr aphical markets. There are also areas with only one hospital. It is hypothesized that either prices/costs will be higher than they would be in the absence of pure comp etition or that quality will be lower. This can occur because incumbents are protect ed from competition because competitors (either new entrants or exis ting hospitals that wish to expand) must go through the Certificate of Need process. This proce ss is costly and, in many states, an existing hospital is allowed to testify against a potential entrant. As with any market barrier, it is reasonable to expect less strenuous comp etition and possibly less innovation. Indeed, previous studies have shown that prices/costs tend to be higher in states with CON requirements.

PAGE 41

32 Previous Literature The existing literature on Certificate of Need and hospital competition covers two distinct areas. The earliest research looked at the impact of both CON regulation and its subsequent repeal in some st ates, as well as general competition within the health care market, on costs. 1 In general, the early studies find that CON has no statistically significant effect on health care costs and most ag ree that if there were to be an effect, it would increase costs rather than lower co sts. CON proponents argued that these early studies were flawed becaus e of their limited time horiz on immediately following the introduction of these programs. It was argue d that the programs needed time to get up and running before they could be expected to control costs. More recent studies have l ooked at longer periods of tim e and evaluated the effects of repeal of the programs. 2 These studies suggest that, if anything, CON programs have tended to increase costs (Sloan, Morrisey and Valvona, 1988) and found that the repeal of CON had no effect on hospital costs per capita (Sherman, 1988). Conover and Sloan (1998) look at the effect of CON removal on state-level per-capita hospital spending among other things for the years 1976 to 1993. They find that CON laws had no effect on per-capita health expenditures, however CON did reduce spending on acute care by 5%. They find no effects on spending from the removal of CON laws. Antel, Ohsfeldt, and Becker (1995) use state-level average hospital costs to investigate the affect of stat e regulations over the years 1968 to 1990. Specifically, they look at rate-setting regulations and Certificate of Need regulations; as well as procedure ____________ 1 Sloan (1988) provides a detaile d review of the early literature. 2 A good survey of the more recent work is Morrisey (2000).

PAGE 42

33 controls like Peer Review Or ganizations and the interactions of these regulations. After controlling for state fixed effects, they find th at regulations in isol ation do not have an effect on costs. There are some effects from different interactions of regulations, but the magnitudes are small. Health care costs and prices charged increase with the duration of the CON program. Using Medicare price and cost data from 1977 to 1978, Noether (1988) found prices were higher relative to costs and theref ore profits were higher in states with a CON program. CON proponents also ar gued that hospitals were di fferent than other industries and therefore competition did not, and could not, work. However, the empirical evidence does not support this. A number of studies have investigated th e effects of hospital competition on prices and costs. The early research tended to us e a Hirschman-Herfindahl Index (HHI) as a measure of competition faced by a hospital. Because of data limitations, this index was usually calculated at the count y level as finer geographic in formation was unavailable. There are two drawbacks to this approach. On e, this assigns every hospital in a county the same measure of competition, ignoring the fact that there is frequently a dominant hospital and a couple of smaller hospitals. Two, it assumes that a patient would be willing to go to any hospital in a county or stated differently, that every hospital is on equal footing in competing for patients and that no patients cross county borders for medical care.

PAGE 43

34 Because of these strong a nd unrealistic assumptions, a hospital-specific HHI is used in this paper. 3 A hospital specific HHI is calculated using the patients zip code as the measure of the market. The market shares of the competing hospitals were calculated for each zip code. These market shares were th en used to calculate a HHI at the zip code level. A weighted sum of these HHIs was then calculated for each hospital where the weights are the hospitals share of patients coming from that zip code. A number of papers have used hospital-specific HHIs as a measure of hospital competition. Melnick, et al. (1992) looked at transaction prices that a large California preferred provider organization (PPO) negotiated with hospitals in 1987. Wh ile controlling for many factors, they found that the PPO paid mo re in the less competitive markets. This suggests that competition does work to lower prices in hospital markets. The authors compare the results when competition is measured with a hospital-specific HHI and a county-level HHI. They conclude that th e hospital-specific measure of competition performs better in explaini ng the price differences. Zwanziger and Melnick (1988) use hospital-le vel data from California for the years 1980 to 1985 to investigate the effects of hos pital competition as well as the introduction of Medicares Prospective Payment System (P PS) on hospital costs. They find that the introduction of the PPS caused hos pitals to significantly reduce their costs. They also find that hospitals in less competitive markets had higher costs prior to the introduction of the PPS, a result consistent with a medical arms race. After the PPS was introduced, hospital competition no longer had a significant effect on costs. ____________ 3 See Morrisey, Sloan and Valvona (1988) and Zwanziger, Melnick and Mann (1990) for reviews of defining hospital markets.

PAGE 44

35 Using hospital-level data for the United St ates, Bamezai, et al. (1999) look at the effects of managed care, as well as market structures, on hospital operating costs. The direct effect of hospital competition is not statistically significant. However, when interacted with measures of managed care penetration, it is highly significant. This implies that greater hospital competition is e ffective only in areas with high levels of managed care penetration. Kessler and McClellan (2000) look at th e effects of hospital competition on Medicare patients heart atta ck care. They use predicted patient flows to calculate hospital-specific HHIs. They find that e xpenditures were 8% hi gher in the hospitals facing the least competition relative to those hospitals facing the most competition. A paper close in spirit to this one is Zwanziger and Mooney (2005). They look at the effects of the deregulati on of hospital prices in New York. They investigate HMOhospital transaction prices and show that negotiated prices were lower in more competitive markets. This effect be comes larger after deregulation. Description of data This paper uses patient-level hospital discharge data on obstetric s patients from the state of Pennsylvania for the years 1994 to 2004. 4 The data includes information on every obstetrics patient discharged in Pe nnsylvania for those years, approximately 375,000 observations. The data contain detailed billing information for each patient as well as information on who paid for the medi cal care. While the actual amount paid is almost certainly less than the billed am ount, unfortunately that amount is unknown. ____________ 4 For financial reasons, the data is for the 1st quarter of each year.

PAGE 45

36 There is also detailed diagnosis and proce dure information as well as the patients zip code. While it would be preferable to know the actual price paid, we can be relatively certain that the price reported is a list price rather than a negotiated price that differs for each patient or payer. Figure 3-1 plots the average price of a caesarean section by payer type. We see that the there ha s been a large increas e in the price of a c-section, but there is not a large difference between payer types. We see the same results for premature births in Figure 3-2. Again, there has been a large increase in prices, but there is no difference between payer types. Because we expect that different organizations will pay different prices for a given procedure, the lack of differences suggests that the prices used in this paper are list prices. It is reasonabl e to believe that the list prices are related to actual prices paid. Because it is doubtful that anyone woul d pay more than the billed price, any results can be in terpreted as a lower bound. Traditionally, hospital competition is measur ed with a Hirschman-Herfindahl Index (HHI). This index measures competition as the sum of the squared market shares in a given geographic market. Greater levels of competition are represented by lower numbers. Figure 3-3 shows the HHI calculate d at the county-level for Pennsylvania in 1994. The lighter shaded areas have the grea test amounts of compe tition. The two areas with the most competition are Allegheny Count y in the west and Philadelphia County in the east. Figure 3-4 shows the comparable HHI for 2004. It is apparent from these maps that, by this measure, there has not been a large change in competition over the intermediate years. Some of the smaller c ounties have lost hospita ls and therefore the market has become more concentrated.

PAGE 46

37 These maps also show the num ber of hospitals in a given county for that year. A number of counties have had decreases in th e number of hospitals. Using the number of hospitals in a county as a measure of compe tition, it would appear that there have been changes that may not have been picked up by the HHI. As can be seen from Figure 3-5, most counties did not see a change in the number of hospitals. 5 44 counties, out of 67 total, did not see any change at all in the number of hospitals. Only two counties gained a hospital. We also observe that most countie s only have one or two hospitals. None of the counties with only one hospital in 1994 lost a hospital, and only one county with two hospitals lost one. The larger counties did lose hospitals, in some cases more than one. It is important to note, that not al l of the losses are necessarily du e to hospital closures. It is possible that some of these hospitals di scontinued their obstetrics service while continuing to provide other forms of care. Small hospitals (as measured by the number of patients) were more likely to close as compared to larger hospitals. Figure 3-6 is a scatter plot of the number of patients per hospital in 1994 and the number of patients per hospital in 2004. 6 The number of patients per hospital in 1994 for those hospi tals that closed between 1994 and 2004 is plotted along the x-axis. It is clear that the large hospitals have a tendency to get bigger. It is also clear that the majority of hospitals have less than 300 patients in a given year. As well, it appears that most hospitals did not change their patient counts very much. ____________ 5 The number of counties in a cell is represented by the size of the circle. All cells greater than 1 have the number of counties inside the circle. 6 Only one hospital had more than 1200 patients in one year, the number of patients did not vary greatly from 1994 to 2004.

PAGE 47

38 We also see from Figure 3-7, that the hosp itals that closed were those that were facing the higher levels of competition (as measured by the hospital-specific HHI). Again, the hospitals that closed between 1994 and 2004 are plotted along the x-axis. This figure also makes apparent that a number of hospitals faced dramatically less competition in 2004 as compared to 1994. This is almost cer tainly due to the hospitals that left the market. In order to rectify these discrepancies, a different approach is required. The traditional Hirschman-Herfindahl Index is calc ulated using county borders as the borders of the market. As mentioned, this approach makes several strong assumptions. The assumption that no patients cross county borders for care can be shown to be false. Table 3-1 shows that in any given year, approximate ly 25% of patients receive treatment in a county other than their own. Allegheny and Philadelphia counties are broken out of the total, as they are the largest counties in Pennsylvania. Allegheny County is similar to the state as a whole, while in Philadelphia County approximately 16% of patients are from another county. For ease of display, the hosp ital-specific HHIs are averag ed at the county level for the years 1994 and 2004 (Figures 3-8 and 3-9) These maps show greater amounts of competition as compared with the traditional HHI. As expected, competition is greatest in those areas with the larg est population. Figure 3-10 show s the MSAs as well as the location of the major cities in Pennsylvania. The level of hospital competition is seen to be greater on average in these areas. Somewhat surprisingly, there is very little change over the time period in the levels of hospital competition. Figure 3-6 plots the average of the hospital-specific HHI for

PAGE 48

39 both counties in and out of a MSA (metro a nd non-metro counties). It is difficult to detect any effect from the repeal of CON. However, there is a slight upward trend over time in both areas. This implies that hospital competition has actually decreased on average over this time period, which is the opposite of what we would expect from the removal of CON. Of course, it is possible that these aver ages obscure the individual hospitals changing competitive environment. Related to the slight increase in the hospital-specific HHI, we see in Table 32 that the actual number of hospitals in Pennsylvania peaked in 1996. The number of hospitals decreases dramatically over the following years. Again, Allegheny and Philade lphia counties are broken out. We see that the number of hospitals peaks in A llegheny County in 1996 and peaks in 1997 for Philadelphia County. Although it is impossible to say with certainty, the evidence above makes it is likely that the repeal of CON led some of the weaker hos pitals to discontinue their obstetrics service. Figure 3-11 shows the average price charged for metropolitan and non-metro counties. Hospitals in metro counties charge more relative to non-metro counties. The difference increases dramatically over the time period. This di fference is consistent with the Medical Arms Race hypothesis in that the areas facing the greate st competition also charge the most. It is also possible that more complex medical cases are admitted to hospitals in metro areas and thus lead to highe r charges. In this case, using price as a proxy for medical complexity, we would expe ct to see the greates t difference between metro and non-metro areas at the top of the di stribution and very little difference at the bottom of the distribution. Figures 3-8 and 3-9 plot the distribution of prices for metro counties and non-metro counties.

PAGE 49

40 We would also expect to see the average le ngth of stay to be longer in metro areas if this is true. Figure 3-14 shows the average length of stay for the two areas. We see that the average stay in non-metro counties ha s been relatively flat over this period, while the length of stay has increased in th e metro counties. Table 3-3 shows the 90th and 99th percentiles of the length of stay for both counties over the time period. We see that the 90th percentile in non-metro c ounties has stayed constant at three days; the comparable percentile for metro counties is four days. At the 99th percentile, the le ngth of stay is again greater in the metro counties as compar ed to the non-metro counties. It appears that some of the difference in the average price may be due to the sorting of complex medical cases. These figures also suggest that only i nvestigating the effects of competition on mean prices may be misleading. Because hospita ls that face the most competition tend to be located where the patients are, that is in metro areas, simply regressing prices on a Herfindahl index would lead to the inference that increased levels of competition increase prices. This paper circumvents this problem by including hospital fixed effects as well as quantile regression. The fixed effects control for unobserved heterogeneity among hospitals, while quantile regression allows me to investigate the effects of CON removal as well as the effects of changing compe tition on prices at different points on the distribution. Because of the concern that part of the di fference in prices in metro counties versus non-metro counties may be due to the complexity of the patient care, it is important to control for premature births and caesarean sections. Figure 3-15 shows the percent of births delivered by caesarean section over the time period. Th ere is not a large difference

PAGE 50

41 in caesarean rates between metro and non-metr o counties. Figure 3-16 shows the percent of births that were born prem aturely. At the beginning of the time period, metro counties had a higher incidence of prematurity but, th is difference has disappeared by the end of the time period. Managed care, i.e., Preferred Provider Organizations (PPOs) and Health Maintenance Organizations (HMOs) is believed to play an important part in controlling health care costs, it is important to contro l for this. Figure 3-17 shows the percent of patients whose primary payer was a PPO or HMO. 7 While managed care organizations have increased their penetration in both type s of counties, the increase has been more dramatic in metro counties. By 2004, more th an 60% of patients are in managed care in the metro counties compared with less than 40% in non-metro counties. It is also reasonable to belie ve that Medicaid patients will affect the price a hospital charges. Figure 3-18 shows the percent of mothers whose primary payer was Medicaid. 8 There is a small increase in the number of Medicaid patients in non-metro counties but the increase is much larger in the metro c ounties, however in all years, the share of patients on Medicaid is larger in the non-metro counties. The Repeal of Certificate of Need and Hospital Competition The first model estimated considers the relationship between the repeal of the Certificate of Need regulation and hospita l competition as measured by the hospitalspecific Hirschman-Herfindahl Index (scaled to range from 0 100). In order for the CON repeal to have an effect in a given area, there must be multiple hospitals competing ____________ 7 The data does not distinguish between the two types before 2000. 8 Some Medicaid patients are enrolled in a managed care plan.

PAGE 51

42 with each other. This is more likely to o ccur in the metro areas as there are a greater number of potential patients Although we do not have a truly exogenous source of variation, and therefore can not argue that this procedure will estimate a causal relation, we can use this idea to test the effect of the CON repeal on hospital competition. By comparing the change in the HHI in metro count ies after the repeal to any change in nonmetro counties, we can get an approximation of the effect of the repeal. The model estimated is, hhiit = 1 + 2 metro + years aftert + metro x years aftert + it (2-1) where hhiit is the hospital-specific HHI for hospital i in year t metro is an indicator equal to one if the hospital is in a metropolitan county, yeart is a vector of indicators for the years after repeal, and metro x yeart is a vector of indicators of metro-specific years after repeal. If there is a differe ntial effect between the two types of counties, then the coefficients on the metro-specific year indicator s will be statistically different from zero. We would expect to see that if anything, the repeal of C ON has increased competition in the metro areas and therefore the estimated coef ficients will be nega tive. Column 1 of Table 3-4 shows the estimated OLS coefficients for this model. The only coefficients statistically different from zero are the constant and the metro indicator. This implies that there was not an effect from the repeal on competition. Because it is possible that any effect would be greater in th e larger or smaller areas, thr ee quantile regressions were estimated. Columns 2 report the es timated coefficients for the 75th, 50th, and 25th quantiles. In no case are any coefficients st atistically significant, again except for the constant and metro indicator. While we can not say with certainty th at the repeal of CON had no effect on hospital competition as measured by the hospital-specific HHI, the fact

PAGE 52

43 that the repeal is not correlated with changes in competition is suggestive of the idea that there was not an effect. Hospital Competition and Prices The basic model estimated is chargesijt = 1 hhijt + 2 hmoijt + 3 medicaidijt + 4 csecijt + 5prematureijt + j + t + ijt (2-2) The dependent variable is the total charges for patient i that the hospital j reported to the state in year t Where hhi is the hospital-specific HHI, hmo is an indicator that the patients primary payer was a HMO, medicaid is an indicator th at the patient has Medicaid, csec is an indicator that a caesar ean section was performed, premature is an indicator of a premature birth, and metro is an indicator that the hospital is in a metro county, while j is a hospital fixed-effect and t is a year fixed-effect. The variable of interest is hhi. Beca use the Herfindahl index decreases as competition increases, if hospital competition reduces prices then the coefficient on hhi will be negative. As a baseline, the model was estimated using OLS. These results are reported in the first column of Table 3-5. The coefficient on hhi is statistically insignificant, and the magnitude is small, implying that competition does not affect hospital prices. As expected, both c-sections and premature bi rths cost more. Somewhat surprisingly, Medicaid patients are charged more as well. The second model estimated is chargesijt = 1 hhijt + 2 hmoijt + 3 medicaidijt + 4 csecijt + 5 prematureijt + 6 yeart + j + ijt (2-3) This model substitutes a linear time trend in stead of individual y ear fixed-effects. Again, this model was estimated using OLS. These results are reported in column 2 of

PAGE 53

44 Table 3-5. Again, we see that the coeffi cient on hhi is small and statistically insignificant, implying that competition does not have an affect on hospital prices. The other coefficients are no t markedly different. The third model estimated is chargesijt = 1 hhijt + 2 hmoijt + 3 medicaidijt + 4 csecijt + 5 prematureijt + 6 yeart + 7 metro x yeart + j + ijt (2-4) The third model adds a metro-county specifi c time trend. This controls for the different growth trajectory in metro counties s een in Figures 3-8 and 3-9 (distribution of charges). The results of this estimation are reported in column 3 of Table 3-5. The coefficient on hhi is again small and insignificant, while the other coefficients are qualitatively the same. As can be seen from Figures 3-8 and 39, the distribution of prices is heavily skewed by the top of the distribution. It is reasonable to believe that only a couple of hospitals in given area are responsible for the highest charges as these would be the most medically complex. Given this, it is not surprising to find that competition does not affect average hospital prices as the average in this case is dominated by the upper half of the distribution where there is little compe tition. For this reason, quantile regression is employed. It is reasonable to believe that competit ion would have the grea test effect at the bottom of the price distribution. These are the simplest medical cases It is likely that more complicated cases can potentially be ad mitted to only a subset of an areas hospitals, because not every hos pital would have the staff or facilities to manage a complex medical case. If this is the case, then we would expect to see competition have a

PAGE 54

45 greater effect on the lower qua ntiles. The first model is estimated at fi ve different quantiles: .90, .75, .50, .25 and .10 in order to investigate the hypothesis that hospital competition does not affect the upper part of the distribution, but ma y affect the lower part. Table 3-6 reports the results for the resp ective quantile regressions. All of the coefficients are statistically si gnificant and, the coefficients are increasing as the quantiles get smaller. This supports the hypothesis that competition has a greater effect for the least complicated medical cases. For exam ple, a one standard-deviation increase in competition would reduce the price by a pproximately $225 at the median price. 9 Relative to the average charge of $5950 in 1999, this implies a 3.8% change in price. If we compare the competition effect between the 75th and 25th quantiles, we see that competition matters much more for the lower quantile. A one standard deviation increase in competition would reduce the price by 1.7% at the 75th percentile; a similar change would reduce the price by 8.0% at the 25th percentile. 10 It does appear that hospital competition matters more for the lower quantiles. The results of the quantile regressions of the second model are reported in Table 37. As with the OLS version, the results are qua litatively similar to the first model. The coefficients on hhi are all sta tistically significant but not very large. The third model is reported in Table 3-8. Again, th e results are very similar to the other models. All three ____________ 9 The standard deviation of hhi is 15.6; this multiplied by the coefficient on hhi for the median regression (14.67) implies a price change of $229. 10 The calculation for the 75th percentile is: 8.14*15.6 = $127 as compar ed to a price of $7200 in 1999. This is a difference of 1.7%. A similar calculation using an average price of $3118 in 1999 yields a difference of 8.0% for the 25th percentile.

PAGE 55

46 models suggest that hospital competition has largest effect on the lower half of the price distribution. This estimated effect can be interpre ted as an upper bound. Given that the dependent variable in all three models is the billed price, not the actual price paid, it is likely that some patients paid less. It is difficult to believe that anyone would pay more than the billed price. If that is the case, th e effect of competition could be even larger. Conclusion This paper looks at the effect of the repeal of Pennsylvanias Certificate of Need regulations in 1996 on hospital competition and hospital prices. I us e a hospital-specific Hirschman-Herfindahl Index to measure hos pital competition. Although the CON repeal does not appear to affect hospital competiti on, I show that there were changes in the number of hospitals in Pennsylvania and th erefore the hospital-specific HHIs. The hospitals that were small and faced high levels of competition, as measured by the hospital-specific HHI, were more likely to discontinue their obstetrics service. The changes in the individual HHIs was th en used to determine the effects of competition on prices charged. A quantile re gression approach was used because the hospital prices are heavily skewed to the ri ght. Using this approach, increased hospital competition was found to reduce prices charged. Hospital competition is found to matter much more for the lower half of the price di stribution as compared to the upper half. A one standard deviation increase in competiti on, as measured by the hospital-specific HHI, would reduce prices by 1.7% at the 75th percentile. A comparable increase in competition would reduce prices by 8.0% at the 25th percentile.

PAGE 56

47 Table 3-1. Percentage of patients crossing county borders Year Total Allegheny Philadelphia 1994 0.259 0.233 0.169 1995 0.252 0.238 0.168 1996 0.259 0.253 0.171 1997 0.265 0.250 0.179 1998 0.257 0.248 0.172 1999 0.264 0.254 0.189 2000 0.261 0.244 0.182 2001 0.267 0.250 0.183 2002 0.262 0.267 0.166 2003 0.267 0.266 0.167 2004 0.272 0.270 0.160 Table 3-2. Number of hospitals in the state Year Metro Non-Metro Allegheny Philadelphia 1994 111 40 14 17 1995 114 39 13 20 1996 116 41 14 19 1997 110 40 11 20 1998 104 40 12 16 1999 103 39 13 16 2000 103 40 12 15 2001 101 39 11 15 2002 97 38 10 13 2003 97 36 11 11 2004 89 36 10 11 Table 3-3. Percentiles of lengt h of stay (by county type) 90th percentile 99th percentile year non-metro metro non-metro metro 1994 3 4 7 9 1995 3 3 6 9 1996 3 3 6 8 1997 3 4 6 8 1998 3 4 6 8 1999 3 4 5 9 2000 3 4 5 8 2001 3 4 6 8 2002 3 4 6 8 2003 3 4 6 9 2004 3 4 6 8

PAGE 57

48 Table 3-4. Effect of CON repeal on the HHI (standard errors in parentheses) (1) (2) (3) (4) OLS Quantile (.75) Quantile (.50) Quantile (.25) metro -19.576 -24.277 -19.759 -18.233 (1.484) (2.105) (2.291) (2.836) 1 year after x metro 0.102 7.585 1.641 -4.536 (2.977) (4.211) (4.574) (5.675) 2 years after x metro 0.892 4.430 1.116 0.014 (2.995) (4.241) (4.600) (5.714) 3 years after x metro -0.260 6.197 -0.721 -1.669 (3.018) (4.204) (4.636) (5.665) 4 years after x metro 1.424 9.044 -0.779 -3.525 (2.998) (4.236) (4.605) (5.708) 5 years after x metro 1.458 4.043 3.850 -0.193 (3.025) (4.219) (4.645) (5.685) 6 years after x metro 1.226 6.387 1.229 -1.429 (3.059) (4.287) (4.697) (5.776) 7 years after x metro -0.865 9.518 -2.305 -6.416 (3.106) (4.394) (4.766) (5.920) 8 years after x metro -0.166 8.817 -4.207 -0.640 (3.135) (4.434) (4.810) (5.975) 1 year after 0.518 -3.630 0.185 3.370 (2.552) (3.613) (3.914) (4.868) 2 years after 0.674 -0.910 1.997 0.401 (2.552) (3.613) (3.914) (4.868) 3 years after 1.420 -2.551 0.818 1.581 (2.577) (3.575) (3.950) (4.817) 4 years after -0.303 -4.551 0.912 4.078 (2.552) (3.613) (3.914) (4.868) 5 years after 0.616 -1.210 -1.329 2.131 (2.577) (3.575) (3.950) (4.817) 6 years after 1.805 -1.718 2.306 3.448 (2.602) (3.633) (3.988) (4.895) 7 years after 2.948 -4.928 3.714 8.554 (2.657) (3.759) (4.069) (5.065) 8 years after 2.418 -3.635 4.114 4.037 (2.657) (3.759) (4.069) (5.065) Constant 58.600 70.605 57.398 46.683 (1.276) (1.810) (1.969) (2.439) Observations 1573 1573 1573 1573

PAGE 58

49 Table 3-5. Regression results using OLS (with hospital fixed-effects) (1) (2) (3) hhi 37.53 34.91 31.48 (25.94) (23.57) (24.35) hmo 278.69 96.50 10.82 (156.08) (165.38) (170.00) csec 5096.58 5141.49 5133.17 (472.12) (479.65) (477.60) premature 3603.95 3597.61 3608.21 (578.50) (576.26) (577.70) medicaid 415.96 463.09 431.73 (105.66) (107.89) (103.62) year 560.73 153.54 (89.04) (37.42) metro_year 484.06 (113.80) Constant 1621.50 493.10 627.19 (1299.44) (1405.22) (1402.66) Observations 374770 374770 374770 Number of Hospitals 169 169 169 R-squared 0.05 0.05 0.05 F-Stat 29.11 45.01 39.55 Hospital Fixed Effects Yes Yes Yes Year Fixed Effects Yes No No Table 3-6. Quantile regression results for model 1 (with Hospital Fixed-Effects) (.90) (.75) (.50) (.25) (.10) hhi 7.80 8.14 14.67 16.03 12.61 (1.34) (0.77) (0 .59) (0.52) (0.55) hmo 205.40 167.42 177.15 162.60 125.47 (14.59) (8.19) (6.49) (6.06) (6.69) csec 5557.68 4129.68 3210.34 2663.22 2287.21 (14.81) (8.31) (6.50) (5.93) (6.42) premature 6779.11 2171.95 847.43 459.40 313.57 (23.73) (13.28) (10.38) (9.49) (10.28) medicaid 340.14 199.82 141.81 87.25 41.94 (14.67) (8.09) (6.25) (5.66) (6.08) Observations 374770 374770 374770 374770 374770 Number of Hospitals 169 169 169 169 169 Pseudo R-squared 0.46 0.43 0.38 0.34 0.30 Hospital Fixed Effects Yes Yes Yes Yes Yes Year Fixed Effects Yes Yes Yes Yes Yes

PAGE 59

50 Table 3-7. Quantile regression results fo r model 2 (with hospital fixed-effects) (.90) (.75) (.50) (.25) (.10) hhi 8.18 9.29 15.58 15.59 11.74 (1.72) (1.00) (0 .64) (0.50) (0.55) hmo 102.50 99.39 125.68 142.83 121.03 (15.35) (9.66) (6.96) (5.94) (6.62) csec 5760.95 4157.28 3270.89 2664.86 2285.00 (16.13) (10.04) (7.03) (5.85) (6.39) premature 6672.35 2185.47 883.98 462.61 315.51 (25.84) (16.04) (11.23) (9.37) (10.27) medicaid 388.66 255.43 172.34 90.08 44.33 (15.93) (9.76) (6.75) (5.58) (6.07) year 381.99 326.30 273.31 236.84 207.54 (2.69) (1.59) (1 .05) (0.87) (0.98) Observations 374770 374770 374770 374770 374770 Number of Hospitals 169 169 169 169 169 Pseudo R-squared 0.45 0.42 0.38 0.34 0.30 Hospital Fixed Effects Yes Yes Yes Yes Yes Year Fixed Effects No No No No No Table 3-8. Quantile regression results fo r model 3 (with hospital fixed-effects) (.90) (.75) (.50) (.25) (.10) hhi 3.71 5.76 11.64 12.39 9.86 (1.67) (0.93) (0 .61) (0.52) (0.53) hmo 57.70 51.18 87.84 109.49 102.16 (15.12) (9.07) (6.63) (6.13) (6.34) csec 5739.92 4157.34 3260.90 2662.78 2281.28 (15.75) (9.38) (6.68) (6.06) (6.15) premature 6633.41 2193.67 881.48 467.64 321.49 (25.24) (14.98) (10.68) (9.68) (9.88) medicaid 380.64 246.02 147.51 80.34 37.27 (15.54) (9.12) (6.42) (5.77) (5.83) year 198.59 149.53 114.76 89.56 77.34 (6.17) (3.49) (2 .36) (2.09) (2.15) metro x year 236.64 230.34 208.61 189.67 165.98 (6.70) (3.79) (2 .54) (2.26) (2.33) Observations 374770 374770 374770 374770 374770 Number of Hospitals 169 169 169 169 169 Pseudo R-squared 0.46 0.42 0.38 0.35 0.30 Hospital Fixed Effects Yes Yes Yes Yes Yes Year Fixed Effects No No No No No

PAGE 60

51 5000 10000 15000 20000Average Price of C-Section 1994 1996 1998 2000 2002 2004 Year Blue-Cross Commercial Medicaid Figure 3-1. Average price of a C-S ection (by Payer Type and Year) 5000 10000 15000 20000Average Price of a Premature Birth 1994 Figure 3-2. Average price of a premat ure birth (by payer type and year) 1996 Blue-Cross Commercial Medicaid 1998 2000 2002 2004 Year

PAGE 61

1 14 1 1 1 3 4 2 5 1 3 2 1 3 1 2 1 2 2 2 3 4 1 6 1 2 1 1 1 1 2 2 5 2 1 4 4 2 1 3 1 1 9 1 1 1 17 1 2 2 1 1 1 1 1 2 1 6 1 3 [0.00,0.20] (0.20,0.30] (0.30,0.40] (0.40,0.50] (0.50,1.00]HHI 52 Figure 3-3. County-level HHI for 1994 ( number of hospitals in county)

PAGE 62

53 1 10 1 1 1 2 3 2 4 1 2 2 1 5 1 2 1 2 2 2 2 3 1 4 1 2 1 1 1 1 2 3 2 2 1 3 4 1 1 3 1 1 8 1 1 1 11 1 1 2 1 1 1 1 1 2 1 3 1 3 [0.00,0.20] (0.20,0.30] (0.30,0.40] (0.40,0.50] (0.50,1.00]HHI Figure 3-4. County-level HHI for 2004 ( number of hospitals in county)

PAGE 63

54 303030303030303030303030303030303030303030303030303030303030 11111111111111111111113 33 2 23 33 0 2 4 6 8 10 12 14 16 18 20# of Hospitals per CountyI in 2004 0 2 4 6 8 10 12 14 16 18 20 # of Hospitals per County in 1994 Figure 3-5. Number of hospita ls per county (1994 vs. 2004) 0 100 200 300 400 500 600 700 800 900 1000 1100 1200# of patients in 2004 0 100 200 300 400 500 600 700 800 900 1000 1100 1200 # of patients in 1994 Existing HospitalsDrop-out Hospitals Figure 3-6. Number of patient s per hospital (1994 vs. 2004)

PAGE 64

55 Figure 3-7. Hospital-specific HHIs (1994 vs. 2004) 10 20 30 40 50 60 70 80 90 100HHI in 2004 10 20 30 40 50 60 70 80 HHI in 1994 Existing HospitalsDro p-out Hospitals 90 100

PAGE 65

[0.00,0.35] (0.35,0.45] (0.45,0.55] (0.55,0.65] (0.65,1.00]HHI 56 Figure 3-8. Hospital-specific HHI av eraged at the county-level for 1994

PAGE 66

[0.00,0.35] (0.35,0.45] (0.45,0.55] (0.55,0.65] (0.65,1.00]HHI 57 Figure 3-9. Hospital-specific HHI av eraged at the county-level for 2004

PAGE 67

58 Pittsburgh Reading Altoona Johnstown StateCollege Harrisburg Erie Scranton Lancaster Allentown Williamsport Sharon Philadelphia York Figure 3-10. Map of MSAs and major cities in Pennsylvania

PAGE 68

59 35 40 45 50 55 60 Average HHI 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Metro County Non-Metro County Figure 3-11. Average ho spital-specific HHI (by type of county) 4000 6000 8000 10000 12000 Average Total Charge 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Metro County Non-Metro County Figure 3-12. Average hospital ch arge (by type of county)

PAGE 69

60 2500 5000 7500 10000 12500 15000 17500 20000 22500 25000Percentiles of Total Charges 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year 90th 75th 50th 25th 10th Percentiles Figure 3-13. Distribution of charges for metro counties 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000Percentiles of Total Charges 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year 90th 75th 50th 25th 10th Percentiles Figure 3-14. Distribution of charges for non-metro counties

PAGE 70

61 2 2.2 2.4 2.6 2.8 Average Length of Stay 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Metro County Non-Metro County Figure 3-15. Averagelength of stay (by type of county) .2 .22 .24 .26 .28 Percent C-Section 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Metro County Non-Metro County Figure 3-16. Percentage of births delivered by caesarean section (by type of county)

PAGE 71

62 .05 .06 .07 .08 .09 Percent premature 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Metro County Non-Metro County Figure 3-17. Percentage of births deli vered prematurely (by type of county) 0 .2 .4 .6 .8 Percent hmo 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Metro County Non-Metro County Figure 3-18. Percentage of payers who are HMOs (by type of county)

PAGE 72

63 .25 .3 .35 .4 .45 Percent medicaid 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 200 4 year Metro County Non-Metro County Figure 3-19. Percentage of payers w ho are Medicaid (by type of county)

PAGE 73

CHAPTER 4 MALPRACTICE LAWSUITS AN D MEDICAL PROCEDURE USE Americas health care professionals should be focused on fighting illnesses, not on fighting lawsuits. Junk lawsuits change the way docs do their job. If youre worried about getting sued, youre going to do everything you can to make sure you dont get sued. Thats why doctors prac tice whats called defensive medicine. That means theyre writing prescriptions or ordering tests th at really arent necessary, just to reduce the pot ential of a future lawsuit. George W. Bush 1 Defensive medicine is a serious public policy concern and a potential contributor to the increasing cost of medical care in th e United States. Anecdotal evidence indicates that the problem may be growing in severi ty. However, the academic literatures understanding of defensive medicine remain s incomplete. The process of determining what constitutes defensive medicine has been confounded by data limitations. Current research is based on medical costs, malpract ice premiums, or legal reforms not actual medical decisions and outcomes. Thus far, the practice of defensive medicine has been inferred from proxies for the fear of malpractice lawsuits or more specifically the legal environment. This paper adds to the lite rature on defensive medicine by examining the effect of malpractice laws uits on specific physicians medical procedure use. I hypothesize that defensive medicine is due to bo th the fear of being sued as well as to physicians responses to actually being sued. Using a panel of obstetricians, a difference in difference approach circumvents the pr oblem of unobserved physician heterogeneity. ____________ 1 Speech on January 5, 2005. 64

PAGE 74

65 While the legal environment may be the prim ary impetus for defensive medicine, I argue a doctors response to litigation is in teresting in its own right. In this paper, I use hospital discharge da ta for maternity patients at all of the hospitals in Allegheny and Philadelp hia counties for the years 1994 2004 2 These data identify the physician by medical license num ber which was then used to identify the physician by name. The doctors name wa s then matched to medical malpractice lawsuits in Allegheny County for the year s 1995 2004 and Philadelphia County for the years 1980-2004 3 By following the individual doctors behavior over multiple years (both before and after a lawsuit for those su ed), I am able to identify the impact of malpractice lawsuits on a physicians choi ce of obstetrical procedures. Using a difference in difference approach to deal w ith the problem of doctor heterogeneity, the results show that after being sued, a doctor increases the use of caesarean sections (csection) by approximately 5.5% relative to doctors who were not sued in Allegheny County. The estimated effect is approximate ly 1.5% in Philadelphia County. It is possible that lawsuits may lead to an in crease in the number of labor inductions; however, these estimates lack statistical precision. Previous Literature Malpractice insurance premiu ms are not experience rate dcontrary to economic theory. That is, premiums are not based on past claims history. Rather, they are set at the community level (often at the state level) and adjusted for medical specialty and limits of ____________ 2 These Counties were chosen because their court systems identify lawsuits as medical malpractice cases. Due to the cost to obtain the data, the data is actually the 1st quarter of the years 1994 2004. 3 The earliest court records that ar e online for Allegheny County begin in 1995, while the earliest court records that are online for Philadelphia County begin in 1980.

PAGE 75

66 coverage (Danzon 2000) 4 One implication of this aspect of malpractice is that doctors are somewhat isolated from the costs of their own behavior. In fact it could be argued that the major costs, to a doctor, of a malp ractice suit are time and reputation, neither of which are insurable. The laws uit does not affect malpract ice insurance premiums, except in extreme cases. Indeed, the individual docto r has very little to no control over his own malpractice insurance rates. From a purely economic point of view, a doctor provides prudent medical care to minimize these time and reputation costs, not to control his insurance rates. This leads to the hypothesis that a doctor will seek to minimize his exposure to the court system. If a doctor perfectly forecas ts the true costs of being sued, in terms of time and reputation, then a lawsuit will have no effect on his procedure use. However, without perfect foresight, it is likely that at least some doctors will underestimate these costs. It is these doctors who I expect to change their behavior after being sued. A secondary hypothesis is that the doctors with the least exposure (either themselves or their colleagues) to the courts will change their behavior the most after being sued themselves. There has a long line of rese arch that discusses the issue of bounded rationality and decision making. It has been shown that people in general do not forecast risks, or deal with probabilities, very well. Tver sky and Kahneman (1974) document many biases in judgment that people make. They argue th at people tend to use he uristics, or rules of thumb, to make decisions under uncertainty. Th is article led to a long line of research that attempted to formalize economics models of decision making that take into account ____________ 4 Danzon (2000) presents a thorough review of medical malpractice issues.

PAGE 76

67 these biases. 5 Kahneman and Tversky (1973) show that people do not behave as a statistician would when presented with prob abilities. Instead they rely on a limited number of heuristics which sometimes yield r easonable judgments and sometimes lead to severe and systematic errors. In this sa me vein, Camerer and Lowenstein (2004) argue that Bayes Rule is unlikely to be used corre ctly because it has seve ral features that are cognitively unrealistic. Lowenstein and Mather (1990) attempt to assess how accurate risk perceptions are over time as the underlying source of different risks chan ge. They find that in many cases, the public perception of risks changes dramatica lly although the un derlying factor has changed very little. Lowenstein, O Donoghue and Rabin (2003) develop a model of what they term projection bias. They argue that people project their current preferences onto their future selves. Lowenstein (2005) discusses pr ojection bias in the realm of medical decision ma king. He argues that many medi cal decisions involve fear, pain and discomfort and therefore affect pe oples decision making. These decisions can be exacerbated by the fact that the doctor is does not adequately pe rceive the patients mental state. This research leads one to the conclusion that people do not behave in a utility-maximizing manner. It is therefore likely that physic ians do not base their medical decisions on accurate estimates of future litigation costs. Within the existing research on defensiv e medicine, Kessler and McClellan (1996) is one of the more prominent papers. Using data from all Medicare patients treated for serious heart disease in addition to information about change s in state malpractice laws (for example, caps on punitive damages or changes to the statute of limitations), they ____________ 5 See Kahneman (2003) for a review.

PAGE 77

68 show that states which enact malpractice reforms have lower health care spending without a change in mortality or medical co mplications. They conclude that this is evidence that defensive medicine exists. Ke ssler and McClellan (1997) use a survey from the American Medical Association to inve stigate how malpractic e reforms influence physician perceptions and change self-repor ted behavior. While the authors do not observed health outcomes or medical cost s (and so cannot precisely determine the existence of defensive medicine), they find the legal environment affects both the likelihood of being sued as well as physician behavior. One paper that deals specifically with defe nsive medicine in obstetrics is Dubay, et al. (1999) which uses birth certificate data for the years 1990-1992. The authors use the fact that medical malpractice insurance prem iums are not experience rated. Then based on the assumption that premiums are an accurate measure of the likelihood of malpractice lawsuits in a geographic area, the premiums are used as a proxy for the likelihood of a doctor being sued. With this, the authors are able to investigate the relationship between the legal environment (and, therefore, the fear of lawsuits) and the county-level rate of caesarean sections. They show that there is an increase in the number of c-sections due to malpractice fears without a concurrent increa se in the health of the babies. They argue that this is evidence of defensive medicine but the overall effect is small. A second obstetrics related paper is Duba y, et. al (2001) which again us es the same birth certificate data and malpractice insurance premiums to s how that a reasonable decrease in premiums would lead to an increase in prenatal care. While they argue that malpractice pressure reduces the supply of prenatal care, this is no t shown to affect the health of newborns.

PAGE 78

69 Baicker and Chandra (2004) look at the pot ential costs of malpractice on patient care. They find no evidence of a change in treatment patterns in response to increases in malpractice premiums. They also find no ev idence that malpractice costs affect the overall number of doctors. The influential Harvard Medical Practic e Study (Brennan, et al. 1991) reviewed 7,743 medical charts for evidence of neglig ence. They found 1,278 adverse events of which 306 were attributed to negligence. The authors found 47 malpractice claims from these cases but, only 8 of these claims were f ound to have evidence of malpractice. More importantly, 40 percent of cases without negl igence resulted in a payment. While the authors draw the conclusion that there are not enough malpractice la wsuits (i.e., only 15% of negligent doctors are sued), the ev idence also points to the randomness of malpractice lawsuits (i.e., only 17% of ma lpractice claims involve actual cases of negligence). Given this apparent lack of correlation between actual malpractice by the physician and the likelihood of a lawsuit, this implies that malpractic e suits can be treated as a random event. This paper uses hospital discharge data fo r all births in Alle gheny and Philadelphia Counties, Pennsylvania for the years 1994-2004. While earlier researchers have shown that doctors appear to respond to genera l malpractice pressure (i.e., the legal environment), none have been able to invest igate individual doctors responses to being sued for medical malpractice. With this dataset, I am able to isolate a doctors response to new litigation from the preex isting legal environment. A nave approach to the problem of docto rs responses to lawsuits would use a cross-section of doctors and their responses to malpractice cases. However, the

PAGE 79

70 probability of being sued is almost certainly correlated with the difficulty of the medical cases, and hence with the procedures used. While diagnosis information is available, it is difficult to summarize and therefore difficult to control for in a regression model. Instead, to circumvent this problem of unobserved heterogeneity, a difference in difference approach is used. That is, the effect of being sued is determined by each doctors change in behavior. If malpractice lawsuits are a random event, controlling for patient characteristics, then it is straight forward to determine the effect of lawsuits on physician behavior. The average number of c-sections or other poten tially defensive procedures performed can be compared before and after a doctor is sued. A ny changes in behavior can be attributed to the lawsuit. One potential problem with this approach is the large changes in procedure use over the period. Given these time trends, it is possible that the effect of lawsuits would be overstated. The difference in difference approach, by using the un-sued doctors as a control group, removes this i ssue by assuming that both groups of doctors have the same underlying time trend. Description of the Data The hospital data includes detailed information for a pproximately 100,000 mothers in Allegheny and Philadelphia Counties, Pennsylvania for the years 1994-2004. Each mother has up to three doctors (referring, attending, and operating physicians) listed on her record. The doctors are identified by their Pennsylvani a medical license number. While three doctors are possible, frequently one doctor is listed in multiple roles. While most records have both a referring and a ttending doctor listed, many do not have an

PAGE 80

71 operating doctor listed. 6 For the majority of the data, only one license number is listed for a given mother. This paper reports only th e attending physician re sults, as the results are qualitatively similar for the other docto rs. Each of these license numbers was matched to the doctors name and the date the license was first issued using the State of Pennsylvanias license verification website 7 The doctor names then were matched to court data on medical malp ractice cases using the respective countys Prothonotarys website 8 Annual summary statistics for doctors and patients are provided in Table 4-1 and Table 4-2. While the number of births declines in both counties over the data period, the number of doctors remains relatively c onstant in Allegheny and decreases in Philadelphia. The number of c-sections increases over the years in the sample, from an average of 19% to an average of 27% ( Figure 4-1). We also see a larg e increase in the number of labor inductions in Allegheny County and a slight rise in Philadelphia County ( Figure 42). These changes in procedure use are a f actor in the rising concern of defensive medicine. However, it is possible that this shift reflects cross-sec tional changes in the composition of mothers. Mothers could be older and exhibit different c-section preferences in later periods. In addition, the use of other obstetric procedures has changed during this sample time. ____________ 6 The only time there is an operating doctor listed is if there was a surgical procedure performed. 7 http://licensepa.state.pa.us/ 8 A Prothonotary is the chief legal clerk in a county in Pennsylvania. It is comparable to a clerk of the court in other states.

PAGE 81

72 The percentage of vaginal births after caesarean sections (VBAC) has decreased ( Figure 4-3). Given that a vaginal birth is higher risk after a previous c-section, it is reasonable to believe that some of this re duction may due to malpractice fears, i.e., defensive medicine. Grady (2004) presents anecdotal evidence that so me hospitals in the U.S. no longer allow VBACs to be performed. Likewise, there is a simultaneous decrease in the percentage of pregnancies that are prolonged i.e., those that have gestational periods longer than 42 weeks. As prolonged pregnancies increase the risk to the mother and the baby, it is not surprising that doctors have reduced their occurrence. There are no prolonged pregnancies after 2002 compared to approximately 4% of the births in 1994 ( Figure 4-4). Part of this decrease is almost certainly due to the increase in labor inductions during this period. Over the period of the data, the occurrence of neonatal distress decreases markedly ( Figure 4-5), while there is not a large ch ange in the incidence of neonatal deaths ( Figure 4-6). There is no significant change in the incidence of premature births in either county (Figure 4-7). Two possible explanations for the increase in caesarean sections are the increase in the number of breech births ( Figure 4-8) and the increase in mothers over the age of 35 (Figure 4-9), although these increases were larger in Allegheny County than in Philadelphia County. While there is some evidence for a relationship between older mothers and breech births, the possible effect is small (Rayl, Gibson, and Hickok 1996). It is also possible that mothers preferences are likely to have a la rge effect on caesarean sections. Anecdotally, older mo thers are thought to be more likely to request a c-section,

PAGE 82

73 however this is difficult to determine from the data as the positive association between age and c-sections my also be due to increased pregnancy risk. While the two counties look very similar over the time period, there is one very large difference. Both counties started the pe riod with approximately 30% of the mothers on Medicaid. This percentage stayed the same in Allegheny County while it doubled in Philadelphia County by the end of the period ( Figure 4-10). Physicians and Medical Malpractice Lawsuits While the total number of doctors is similar in each county in 1994, by the end of the period, the relative number in Philadelphia decreased ( Table 4-3). The number of lawsuits per doctor differs dramatically be tween the two counties. The two counties diverge in another way as well; the total number of licenses issued in Allegheny County is almost four times the number issued in Philadelphia County. The court data for Allegheny County begins in 1995. This unfortunately limits our sample, as this paper focuses on the effect of a physician being sued the first time. To capture this initial reaction, th e first medical malpractice laws uit must be identified for each doctor. Given the limitations of the c ourt data, it is impossible to accurately determine the total number of malpractice lawsuits e xperienced by a physician who received his license before 1995. In this pa per, young doctors refers to those licensed in 1994 9 to the present, while old doctors refers to those that received their license before 1994. In a later part of the paper, us ing the data from Phila delphia, I divide the old doctors into old and senior doctors. In this case, the old doctors are those who ____________ 9 1994 is used as the cutoff because I have 1st quarter data. A doctor who finishes school in the spring will not appear in the hospital data until 1995.

PAGE 83

74 received their license between 1980 and 1993, while the senior doctors are those who received their license before 1980. 10 Philadelphia County court data are ava ilable beginning in 1980. Although this provides much greater detail for a given doc tor, the hospital data begin in 1994. Now while we can not evaluate physicians response s to lawsuits over this whole panel, we can exploit the variation in a physicians laws uit history. For the Philadelphia data, the physicians are segmented into age and lawsuits categories, i.e., an old doctor is one who received his license between 1980 and 1993. The old doctors then are grouped according to their past malpractice lawsuit hi story, i.e., those who have been sued and those who never have experienced a lawsuit. Obstetricians/gynecologists are among the hi ghest risk of malpractice lawsuits of all medical specialties. However, even with this heightened probability of legal action, the probability of a lawsuit varies greatly within the same state. Comparing Figure 4-11 and Figure 4-12, substantial differences are clea r between Allegheny and Philadelphia Counties. Figure 4-11 shows the cumulative probability of a lawsuit for a given number of years of practice in Allegheny County. Af ter practicing for five years, approximately 15% of doctors have been sued at least once 11 In Philadelphia, the likelihood of a lawsuit is more than 20% for doctors with the same experience ( Figure 4-12). By the 10th ____________ 10 This division is based on the court data for Philadelphia County. 11 Because the majority of doctors received their license s at the end of the period (when it is impossible for a doctor to have practiced for more than 5 or 6 years, the cumulative distribution is calculated with just the doctors from the 1st half of the period.

PAGE 84

75 year of practice, only 20% of Allegheny doctors have been sued, while close 40% of the Philadelphia doctors have been sued at least once 12 Responses to Lawsuits It is reasonable to think that malpracti ce lawsuits are driven by doctor and patient characteristics. Known doctor characterist ics are limited to experience and procedure use. I assume that malpractice lawsuits are exogenous, as earlier research has shown that lawsuits, while not entirely random, are not based on the medical record in most cases (see Harvard study). Given this exogeneity of lawsuits, the difference in difference estimate is an unbiased estimate of the effect of being sued on doctors behavior. Obstetrical Procedures Of the obstetric procedures investigated as potentially defensive in nature, csections would seem to be the likeliest can didate. A c-section transforms a potentially litigious situation (a vaginal birth) into a controlled medical procedure. There is the added benefit that c-sections can be sc heduled in advance, which is probably psychologically reassuring to both the doctor as well as the mother. There is evidence that inducing labor on a mother who has previously had a csection is dangerous, because of this, it is im portant to control for previous c-sections. However, if a doctor is fearful of a malpracti ce lawsuit, then he is likely to perform a csection on a high-risk mother. The effect of malpractice lawsuits on inductions is an empirical question. While there is anecdotal evidence that fear of lawsuits has reduced ____________ 12 These results are likely understated as it is reasonable to believe that some doctors would have stopped practicing before they appeared in the hospital data.

PAGE 85

76 the use of VBACs, there are not enough women in the data that have had a previous csection to effectively estimate an eff ect of a lawsuit on physician behavior. To test the hypothesis that doctors respond to their first lawsuit, model 1 was estimated separately for each county. procedurei = 1 sued1i + 2 sued2i + 3 sued3i + 4 sued4i + 1 breechi + 2 previousi + 3 oldi + 4 medicaidi + i (3-1) where procedure is an indicator of either a caesarean section or an induced labor, sued1 sued4 are indicators of the numb er of lawsuits a patients doctor has experienced. 13 These indicators can change for each doctor from year to year depending on his lawsuit history. For example, a doctor wh o is never sued for malpractice will have all of the sued indicators equal to zero. A doctor who is sued for the first time in 1999 will have the sued1 equal one for the years af ter 1999. If he is sued a second time, the sued2 indicator will switch to a one, while the sued1 indicator will not change. Sued3 and sued4 are indicators of three and four lawsuits that are coded similarly. These indicators are difference in difference estimat es of the effect of a given lawsuit on a doctors behavior. The variables breech, prev ious, old, and Medicaid are indicators of a breech birth, a previous c-section, a mother over 35, and a mother on Medicaid. Because there are not any doctors that appear in both Philadelphia and Allegheny Counties, and the model includes doctor and year level fixed effects 14 estimating this with both ____________ 13 The sued indicators switch from 0 to 1 when a doctor gets sued the respective time. The indicators never switch off after being turned on. 14 The model was also estimated with hospital fixed effects. There is no qualitative difference in the estimates.

PAGE 86

77 counties jointly is not possible, unless the assumption is made that doctors in both counties react in the same manner to a lawsuit. In order for the parameter of interest ( 1) to be identified, the time trend must be the same for the untreated group (i .e., doctors who are not sued) 15 Because it appears that lawsuits are random, it is assumed that this condition is met. After controlling for doctor, year, and hospital effects, 1 is the estimate of the treatm ent effect of being sued. If doctors react to lawsuits by increasing the number of c-sections, then 1 should be positive. The same will be true for 2, 3 and 4 if the response is not negligible for additional lawsuits. Because the variation in lawsuits occurs at the doctor level, the standard errors were clustere d at the doctor level. The va riables of interest in these regressions are the sued variables. If the hypothesis that at doctor only responds to a first lawsuit is true, then the coefficients on sued2 sued4 should not be significantly different from zero. With a binary dependent variable, a logit or probit model might be assumed. In actuality, the combination of doctor level fixed effects as well as a doctor level treatment effect implies that the dependent variable is the average of the docto rs procedure use in a given year. The same model could have been estimated by collapsi ng the data to doctor and year means and then applying ordinary l east squares. I chose to estimate these models at the individual patient level to ensu re the most precise estimates possible. The estimates are not qualitatively different when averages are used. ____________ 15 Because of the strong time trend and the fact that there are not a large nu mber of lawsuits in a given year, this is difficult to show.

PAGE 87

78 Caesarean Sections in Philadelphia The results for all of the doctors in Philade lphia are reported in the first column of Table 4-4. We see that as expected, breech births and previous c-sections are highly correlated with a c-section, as is the indicator for older mothers. A breech birth increases the probability of a c-section by 52%, while a pr evious c-section increases the probability of a c-section by 44%. We also see that Me dicaid mothers are less likely to receive a csection, although the coefficient is not very la rge. This may be because they are less demanding of their doctors or the doctors are less respons ive to the demands of less wealthy patients. It is also possible that Medicaid reimbursement rates affect the doctors choice of a c-section. We also see that, as predicted, doctors do respond to a malpractice lawsuit. The estimated impact of a first lawsuit on physician behavior is an increase in the c-section rate of 1.4%. Additional lawsuits ha ve no effect. Given these results are for all of the Philadelphia doctors, young and old, it is possible that this underestimates the effect. Some of these doctors have been su ed before the hospital data begins and may have adjusted their behavior already. To address this issue, the model then wa s estimated with subsets of doctors. As mentioned before, the doctors were divided in to young, old, and senior categories in Philadelphia County. Thes e categories corres pond to the hospital and court data. Young doctors received their license after 1994 and hence, hospital and court data is available for their whole career. Old docto rs received their lic ense between 1980 and 1993; court data is available for their whole career but not the hos pital data. Lastly, senior doctors received their licenses before 1980 and thus have missing court information from the beginning of their career.

PAGE 88

79 The expectation is that laws uits will have the greate st effect on the young doctors with a possible effect on the old doctors who were not sued early in their career. The predicted effect of a lawsuit on the senior doctors is nil, it is assumed that they either have been sued previously or are experienced enough not to react in a significant manner. The results from these additional regressions are reported in columns 2 4 of Table 4-4. Somewhat surprisingly, there is no statis tically significant effect of a first lawsuit on young doctors while old doctors do respond to a la wsuit. As expected the senior doctors do not respond. Interestingly, when estimat ed separately, both the young and old doctors respond to a second lawsuit. The young doctors increase the rate of c-sections while the old doctors actually reduce their rate. It is possible that the old doctors change their patient mix in order to avoid high risk cases ; however it is not po ssible to test this hypothesis. It is also possible to segment the old a nd senior doctors into groups based on their earlier lawsuit history. Table 4-5 reports the results of regr essions with the old doctors segmented into groups based on whether they were sued between 1980 and 1993. Column 1 reports the results for old doctors who were not sued before 1993. The effect of a first lawsuit on these doctors is positive but not statistically significant. Again, we find the old doctors respond to the second laws uits by reducing the number of c-sections. They do not respond to additional lawsuits. By comparison, column 2 reports the results for the old doctors who were sued before 1993. It is these doctors who we do not expect to change their behavior in response to a lawsuit after 1994 and yet, we find that they increase their c-section rate by 3.5%. These doctors do not change their behavior in response to additional lawsuits. Columns 3 and 4 perform the same regressions for the

PAGE 89

80 senior doctors (i.e., those who received their licenses before 1980). As expected, in both cases, the senior doctors do not respond to a lawsuit. Caesarean Sections in Allegheny When the same model is estimated with the data from Allegheny County, we see that with all of the doctors, the coefficients on the control variables are similar. Although it does appear that doctors in Allegheny County have a greater propensity to perform a csection relative to Philadelphi a County, the coefficients on br eech and previ ous are larger in Allegheny County. That is, c-sections are mo re likely simply due to patient attributes. Interestingly, it appears that Medicaid mother s are less likely to receive a c-section in Allegheny versus Philadelphia County. The variab les of interest are the sued variables. With Allegheny County there is no effect of a lawsuit on behavior (column 1 of Table 46). 16 Again, these results are probably underestimated because some of these doctors have almost certainly been sued before. This model again is estimated with the doctors segmented into different groups. For comparability purposes, the same divi sions (young and old) are made with the Allegheny doctors even though the court data do not begin until 1995. Column 4 groups the old and senior doctors together since, they are indistinguishable in Allegheny County for all practical purposes. The results of thes e regressions are reporte d in columns 2 of Table 4-6. Young doctors respond in a dramatic fash ion to a first lawsuit. The estimated effect is a 5.6% increase and is highly statistically signi ficant. The young doctors do not ____________ 16 While it is possible that doctors are responding to a third lawsuit and no others, this is probably an artifact of the data.

PAGE 90

81 respond to a second lawsuit. The old doctors do not respond to a fi rst or second lawsuit but again, appear to res pond to third lawsuit. Labor Inductions in Philadelphia Next, I evaluate the use of labor inductions. As mentioned previously, the predicted effect of a lawsuit on a physicians behavior is uncertain. A doctor may want to reduce the number of vaginal births he perf orms and so will be less willing to induce labor. However, it is also possible that th e potential risk from a prolonged pregnancy will lead the doctor to induce labor inst ead of waiting for spontaneous labor. We see in the first column of Table 4-7 that doctors in Philadelphia do not appear to change the number of inductions performed in response to a lawsuit. As before, it is possible that this understates the effect becau se a number of these doctors have probably been sued before. When, the doctors are di vided into groups based on when their license was issued, it does not appear that any of the groups of doc tors respond to a lawsuit by changing their behavior. Columns 2, 3 and 4 of Table 4-7 show the results of these separate regressions. Segmenting doctors based on their lawsuit history shows differences among old doctors. Columns 1 and 3 of Table 4-8 report the results for doctors who were not sued before 1993. The senior doctors respond to a first lawsuit by reduc ing the number of inductions by almost 8%, while the old docto rs do not change their behavior. These results may be driven by the small sample size. Columns 2 and 4 of Table 4-8 report the results for the doctors who had previously been sued. As expected, these doctors do not change the rate of inductions. Again, it is interesting to note that the doctors who had previously been sued appear to be less likely to induce labor for breech births and

PAGE 91

82 previous c-sections. This may imply that th ey have already made adjustments to their practices. Labor Inductions in Allegheny It appears that docto rs in Allegheny County change th eir induction rate in response to malpractice lawsuits ( Table 4-9). A first lawsuit leads to a reduction of inductions by 1.5%, while additional lawsuits do not appe ar to matter. Columns 2, 3 and 4 of Table 4-9 report the results for the separa te age groups. While the coefficients of the first lawsuit are negative in all three cases, only the one for the old doctors is statistically significant. Again, additional lawsuits do not appear to matter. Conclusions While it is impossible to say that these results are proof of the existence of defensive medicine in obstetrics, evidence exis ts that doctors change their procedure use after being sued for malpractice. There is also evidence of differences in procedure use across geographical areas. It appears that the age of a doctor influences how he responds to a lawsuit. In Allegheny County, when all doctors are in cluded in the sample, there is no effect of lawsuits on procedure use. This lack of an effect is not surprising given that many of the doctors have been practici ng medicine since the 1980s. It is likely that many of them have been sued previously and have already changed their medical practices. However, when the sample is limited to young doctors (who we can be relatively certain have not been sued before entering the data), it is estimated that a doctor will increase the number of c-sections performed by approximately 5% in response to a lawsuits. This is a large change given that the average c-section ra te is 20% in 1994 and 28% in 2004. It is possible that the use of induc tions is affected by lawsuits but, these estimated effects

PAGE 92

83 suffer from a lack of precision. Doctor s in Philadelphia County, who have more Medicaid patients and much higher risk of litigation appear much less sensitive to lawsuits than doctors in Allegheny County. They also appear to be less likely to use csections in potentially risky situations. In order to draw conclusions about the pr esence of defensive medicine in response to malpractice lawsuits, measurable outcomes are needed. Two obvious ones are maternal and fetal deaths. Thankfully (from societys point of vi ew), these both occur infrequently; however the rar ity of these outcomes makes them less than ideal for my purposes. One thing is certain; physicians do appear to respond to malpractice lawsuits by changing their practice patterns.

PAGE 93

84 Table 4-1. Number of do ctors by type and county gg Philadelphia Pittsburg Year Total Drs Young Dr s Total Drs Young Drs 1994 197 1 194 2 1995 188 7 188 7 1996 188 24 190 11 1997 207 31 203 22 1998 236 40 198 28 1999 158 38 205 41 2000 180 56 192 50 2001 169 57 194 63 2002 155 74 190 62 2003 126 59 192 74 2004 130 70 186 71 Table 4-2. Number of patient s by doctor type and county Philadelphia Pittsburg Year Old Drs Young Drs Total Old Drs Young Drs Total 1994 5,689 18 5,707 4,761 241 5,002 1995 5,213 188 5,401 4,758 212 4,970 1996 4,383 511 4,894 4,228 437 4,665 1997 3,914 720 4,634 3,983 473 4,456 1998 3,586 957 4,543 3,817 470 4,287 1999 2,548 995 3,543 3,602 747 4,349 2000 3,194 1,471 4,665 3,417 1,037 4,454 2001 2,973 1,398 4,371 3,065 1,227 4,292 2002 2,363 1,940 4,303 2,986 1,154 4,140 2003 2,044 2,203 4,247 2,959 1,313 4,272 2004 1,871 2,608 4,479 2,827 1,485 4,312

PAGE 94

85 Table 4-3. Number of licenses issued and num ber of doctors sued by county and year the license was issued. Pittsburg Philadelphia Year # issued # Sued Percent S ued # issued # Sued Percent Sued 1980 63 29 0.460 20 10 0.500 1981 82 33 0.402 19 12 0.632 1982 52 31 0.596 22 17 0.773 1983 67 26 0.388 18 10 0.556 1984 89 37 0.416 9 2 0.222 1985 87 40 0.460 23 15 0.652 1986 89 50 0.562 15 4 0.267 1987 92 38 0.413 27 12 0.444 1988 64 21 0.328 18 7 0.389 1989 75 27 0.360 12 8 0.667 1990 86 27 0.314 18 11 0.611 1991 77 22 0.286 29 10 0.345 1992 87 31 0.356 33 18 0.545 OLD DOCTORS 1993 120 36 0.300 25 10 0.400 1994 78 18 0.231 29 11 0.379 1995 103 30 0.291 34 15 0.441 1996 81 18 0.222 29 12 0.414 1997 92 21 0.228 21 4 0.190 1998 73 13 0.178 23 7 0.304 1999 74 15 0.203 15 4 0.267 2000 60 3 0.050 18 2 0.111 2001 53 10 0.189 18 6 0.333 2002 41 2 0.049 9 2 0.222 YOUNG DOCTORS 2003 28 2 0.071 7 0 0.000 Total 1813 132 0.073 491 209 0.426

PAGE 95

86 Table 4-4. Philadelphia results with c-sections as the dependent variab le (standard errors in parentheses) C-section all doctors young doctors old doctors senior doctors sued1 0.0140 0.0130 0.0232 0.0080 (0.0070) (0.0163) (0.0103) (0.0142) sued2 -0.0072 0.0379 -0.0203 0.0066 (0.0078) (0.0177) (0.0109) (0.0174) sued3 0.0045 -0.0293 0.0110 0.0083 (0.0091) (0.0220) (0.0118) (0.0217) sued4 0.0047 0.0079 0.0068 -0.0216 (0.0099) (0.0288) (0.0121) (0.0238) medicaid -0.0253 -0.0200 -0.0298 -0.0292 (0.0039) (0.0075) (0.0057) (0.0080) old 0.0537 0.0579 0.0587 0.0411 (0.0048) (0.0098) (0.0069) (0.0097) previous 0.4380 0.4830 0.4252 0.4280 (0.0050) (0.0103) (0.0072) (0.0101) breech 0.5249 0.4810 0.5293 0.5612 (0.0064) (0.0128) (0.0090) (0.0134) Constant 0.1087 0.1405 0.0980 0.1227 (0.0059) (0.1261) (0.0076) (0.0090) Observations 50787 13009 24804 11926 # of Doctors 577 174 257 141 R-squared 0.23 0.23 0.24 0.24 F stat 840.1 209.3 419.6 202.1

PAGE 96

87 Table 4-5. Philadelphia segmented results with c-sections as the dependent variable (standard errors in parentheses) C-section old doctors not sued before 1994 old doctors sued before 1994 senior doctors not sued before 1994 senior doctors sued before 1994 sued1 0.0177 0.0358 -0.0064 0.0231 (0.0146) (0.0165) (0.0291) (0.0193) sued2 -0.0342 -0.0236 -0.4248 0.0085 (0.0174) (0.0152) (0.2030) (0.0191) sued3 0.0057 0.0095 0.0000 0.0069 (0.0177) (0.0164) 0.0000 (0.0227) sued4 -0.0017 0.0158 0.0000 -0.0043 (0.0194) (0.0163) 0.0000 (0.0252) medicaid -0.0313 -0.0255 -0.0185 -0.0389 (0.0070) (0.0100) (0.0120) (0.0109) old 0.0660 0.0491 0.0535 0.0305 (0.0093) (0.0103) (0.0142) (0.0134) previous 0.4079 0.4522 0.3954 0.4540 (0.0091) (0.0115) (0.0152) (0.0136) breech 0.5131 0.5536 0.5690 0.5567 (0.0115) (0.0144) (0.0205) (0.0178) Constant 0.0986 0.0997 0.1103 0.1329 (0.0099) (0.0120) (0.0138) (0.0120) Observations 15349 9455 5385 6541 # of doctors 178 79 61 80 R-squared 0.22 0.26 0.22 0.25 F stat 239.4 182 96.26 118.8

PAGE 97

88 Table 4-6. Allegheny results with c-sections as the dependent variable (standard errors in parentheses) C-section all doctors young doctors old doctors all old doctors sued1 0.0068 0.0568 -0.0034 0.0014 (0.0062) (0.0220) (0.0083) (0.0065) sued2 -0.0133 -0.0169 -0.0046 -0.0097 (0.0081) (0.0353) (0.0105) (0.0084) sued3 0.0322 N/A 0.0229 0.0332 (0.0116) N/A (0.0142) (0.0116) sued4 -0.0048 N/A -0.0150 -0.0040 (0.0214) N/A (0.0249) (0.0212) medicaid -0.0321 -0.0349 -0.0388 -0.0316 (0.0044) (0.0100) (0.0062) (0.0049) old 0.0317 0.0320 0.0340 0.0319 (0.0041) (0.0107) (0.0057) (0.0045) previous 0.4772 0.4708 0.4592 0.4778 (0.0048) (0.0128) (0.0066) (0.0052) breech 0.6380 0.5968 0.6558 0.6476 (0.0057) (0.0134) (0.0078) (0.0063) Constant 0.1115 0.0310 0.1157 0.1102 (0.0056) (0.0779) (0.0073) (0.0055) Observations 48696 8293 25241 40403 # of doctors 425 141 181 284 R-squared 0.33 0.3 0.33 0.33 F stat 1302 218.9 695.3 1112

PAGE 98

89 Table 4-7. Philadelphia results with inductions as the dependent variab le (standard errors in parentheses) Induce all doctors young doctors old doctors senior doctors sued1 0.0033 -0.0126 0.0122 -0.0300 (0.0063) (0.0149) (0.0090) (0.0124) sued2 0.0088 0.0376 0.0110 0.0113 (0.0069) (0.0163) (0.0095) (0.0153) sued3 0.0143 -0.0357 0.0344 -0.0050 (0.0081) (0.0201) (0.0102) (0.0190) sued4 0.0097 0.0299 0.0105 0.0036 (0.0088) (0.0264) (0.0105) (0.0208) medicaid -0.0049 -0.0001 -0.0078 -0.0085 (0.0034) (0.0068) (0.0050) (0.0070) old 0.0152 0.0134 0.0151 0.0196 (0.0043) (0.0090) (0.0060) (0.0085) previous -0.0538 -0.0713 -0.0484 -0.0485 (0.0045) (0.0094) (0.0062) (0.0089) breech -0.0444 -0.0533 -0.0447 -0.0294 (0.0057) (0.0117) (0.0078) (0.0118) Constant 0.1037 0.0806 0.1072 0.0870 (0.0052) (0.1154) (0.0066) (0.0079) Observations 50787 13009 24804 11926 # of Doctors 577 174 257 141 R-squared 0.01 0.01 0.01 0.01 F stat 21.5 7.575 12.69 5.037

PAGE 99

90 Table 4-8. Philadelphia segmented results with inductions as the dependent variable(standard errors in parentheses) Induce old doctors not sued before 1994 old doctors sued before 1994 senior doctors not sued before 1994 senior doctors sued before 1994 sued1 0.0110 0.0046 -0.0783 -0.0008 (0.0125) (0.0147) (0.0285) (0.0151) sued2 -0.0171 0.0203 -0.0902 0.0134 (0.0149) (0.0136) (0.1993) (0.0149) sued3 0.0471 0.0169 0.0000 -0.0036 (0.0152) (0.0146) 0.0000 (0.0177) sued4 0.0154 -0.0012 0.0000 -0.0090 (0.0166) (0.0145) 0.0000 (0.0197) medicaid -0.0128 0.0031 0.0051 -0.0157 (0.0060) (0.0089) (0.0118) (0.0085) old 0.0197 0.0105 0.0138 0.0244 (0.0080) (0.0092) (0.0139) (0.0104) previous -0.0388 -0.0630 -0.0447 -0.0514 (0.0078) (0.0102) (0.0149) (0.0107) breech -0.0363 -0.0584 -0.0140 -0.0397 (0.0099) (0.0129) (0.0201) (0.0139) Constant 0.1207 0.0928 0.1023 0.0735 (0.0085) (0.0107) (0.0135) (0.0094) Observations 15349 9455 5385 6541 # of doctors 178 79 61 80 R-squared 0.01 0.01 0.01 0.01 F stat 6.615 7.455 2.594 4.15

PAGE 100

91 Table 4-9. Allegheny results with inductions as the dependent variable (standard errors in parentheses) Induction all doctors young doctors old doctors all old doctors sued1 -0.0154 -0.0243 -0.0186 -0.0132 (0.0064) (0.0228) (0.0085) (0.0066) sued2 0.0096 0.0252 0.0081 0.0085 (0.0084) (0.0365) (0.0108) (0.0086) sued3 0.0058 0.0000 0.0014 0.0059 (0.0120) 0.0000 (0.0146) (0.0119) sued4 -0.0111 0.0000 -0.0075 -0.0118 (0.0220) 0.0000 (0.0258) (0.0218) medicaid -0.0199 -0.0331 -0.0155 -0.0165 (0.0045) (0.0103) (0.0064) (0.0050) old 0.0128 0.0173 0.0126 0.0120 (0.0042) (0.0111) (0.0058) (0.0046) previous -0.0733 -0.1094 -0.0696 -0.0667 (0.0049) (0.0132) (0.0069) (0.0053) breech -0.0639 -0.0638 -0.0622 -0.0644 (0.0059) (0.0139) (0.0081) (0.0065) Constant 0.1226 0.2561 0.1146 0.1223 (0.0058) (0.0805) (0.0075) (0.0057) Observations 48696 8293 25241 40403 # of doctors 425 141 181 284 R-squared 0.01 0.02 0.01 0.01 F stat 35.86 7.82 18.29 30.55

PAGE 101

92 .18 .2 .22 .24 .26 .28 Percent csec 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 200 4 year Pittsburg Philadelphia Figure 4-1. Percentage of births that are caesarean sections (by county) .1 .15 .2 Percent induce 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Pittsburg Philadelphia Figure 4-2. Percentage of births th at are labor inductions (by county)

PAGE 102

93 .2 .3 .4 .5 Percent vbac 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Pittsburg Philadelphia Figure 4-3. Percentage of birt hs that are vaginal births after c-sections (by county) 0 .01 .02 .03 .04 .05 Percent prolong 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Pittsburg Philadelphia Figure 4-4. Percentage of births that are prolonged pregnancies (by county)

PAGE 103

94 0 .05 .1 .15 Percent distress 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Pittsburg Philadelphia Figure 4-5. Percentage of births that experience fe tal distress (by county) .004 .006 .008 .01 .012 Percent death 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Pittsburg Philadelphia Figure 4-6. Percentage of births that lead to ne onatal death (by county)

PAGE 104

95 .06 .07 .08 .09 .1 .11 Percent premature 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Pittsburg Philadelphia Figure 4-7. Percentage of births that are premature (by county) .05 .06 .07 .08 .09 .1 Percent breech 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Pittsburg Philadelphia Figure 4-8. Percentage of births that are breech (by county)

PAGE 105

96 .1 .12 .14 .16 .18 .2 Percent old 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 200 4 year Pittsburg Philadelphia Figure 4-9. Percentage of births to mothers over 35 (by county) .2 .3 .4 .5 .6 Percent medicaid 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 year Pittsburg Philadelphia Figure 4-10. Percentage of births to medicaid mothers (by county)

PAGE 106

97 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 01234567891011 Year of 1st lawsuit Year License IssuedCum Prob license issued 1994 2004 license issued 1994 1999 Figure 4-11. Cumulative probability of a first lawsuit in allegheny county 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 01234567891011121314151617 Year of 1st lawsuit Year License IssuedCum Prob License Isssued 1980-2004 License Issued 1980-1990 License Issued 1991-2004 Figure 4-12. Cumulative probability of a first lawsuit in Philadelphia county

PAGE 107

CHAPTER 5 CONCLUSION My dissertation is investigates the some of the determinants of physician procedure use as well as the effects on prices and health outcomes. The use of data on obstetrics patients provides information on a popul ation that has not been studied in detail by health economists. I show that rem oving regulatory constr aints increases the probability of newborn baby being born healthy. I also show that increased levels of hospital competition lead to lower prices bei ng charged by hospitals. Lastly, I show that medical malpractice lawsuits induce physicia ns to increase their use of caesarean sections, a result consistent with th e defensive medicine hypothesis. 98

PAGE 108

LIST OF REFERENCES Almond, Douglas, Kenneth Y. Chay, and David S. Lee. (2002). Does Low Birth Weight Matter? Evidence from the U.S. Population of Twin Births. Center for Labor Economics Working Paper No. 53, University of California, Berkeley. Antel, John J., Robert L. Ohsfeldt, and Ed mund R. Becker. (1995). State Regulation and Hospital Costs. Review of Economics and Statistics Vol. 77, Issue 3, 416. Baicker, K., & Chandra, A. (2004). The Effect of Malpractice Liabil ity on the Delivery of Health Care. NBER Working Paper 10709. Brennan,R.A., Leape,L.L., Laird,N.M., Hebert,L., Localio,A.R., Lawthers,A G., & Newhouse,J P. et al. Incidence of Advers e Events and Negligence in Hospitalized Patients: Results of the Harv ard Medical Practice Study I. The New England Journal of Medicine (February 7, 1991) 370. Casey, Brian M., Donald D. McIntire, and Kenneth J. Leveno (2001). The Continuing Value of the Apgar Score for the Assessment of Newborn Infants. The New England Journal of Medicine (February 15,2001) 467. Camerer C. and G. Lowenstein (2004). Behavioral Economics: Past, Present, Future. Advances in Behavioral Economics Princeton, NJ: Princeton University Press, 2004. 2. Caudill, Steven B. and Jon M. Ford, David L. Kasserman. (1995). Certificate-of-Need Regulation and the Diffu sion of Innovations: A Random Coefficient Model , Journal of Applied Econometrics Vol. 10, 73. Conover, Christopher J. and Frank A. Sloa n. (1998). Does Removing Certificate-ofNeed Regulations Lead to a Surg e in Health Care Spending? Journal of Health Politics, Policy and Law Vol. 23 No. 3 (June 1998), 455. Danzon, P (2000). Liability for Medical Malpractice. Handbook of Health Economics Amsterdam: Elsevier Science North-Holland, 2000, 1339. Dubay, L., Kaestner, R., & Waidmann, T. (2001) Medical Malpracti ce Liability and its Effect on Prenatal Care Utilization and Infant Health. Journal of Health Economics, 20(4), 591. Dubay, L., Kaestner, R., & Waidmann, T. (1999). The Impact of Malpractice Fears on Caesarean Section Rates. Journal of Health Economics 18(4), 491. 99

PAGE 109

100 Eakin, Kelly B. (1991). Allocative Inefficiency in the Production of Hospital Services, Southern Economic Journal Vol. 58, 240. Gaynor, Martin and Deborah Haas-Eilson (1999). Change, Consolidation, and Competition in Health Care Markets, Journal of Economic Perspectives, Vol. 13 No. 1 (Winter 1999), 141. George, J. (2001). Study: City Malpractice Juries Generous. Philadelphia Business Journal October 26, 2001. Grady, D. (2004). Trying to Avoid 2nd Caesarean, Many Find the Choice Isnt Theirs. New York Times November 29, 2004. Ho, Vivian and Barton H. Hamilton. (2000). Hospital Mergers and Acquisitions: Does Market Consolidation Harm Patients? Journal of Health Economics 19: 767. Kahneman D. and Tversky A. (1973). On The Psychology of Prediction. Psychological Review 80, 273. Kahneman, D. (2003). Maps of Bounded Rationality: Psychology for Behavioral Economics. American Economic Review (December 2003), 1449. Kaplan, Mark E (1991). Comment: An Economic Analysis of Floridas Hospital Certificate Of Need Program A nd Recommendations For Change. Florida State University Law Review Vol. 19 (Fall 1991), 475 Kessler, Daniel and Mark McCl ellan. (1996). Do doctors practice defensive medicine? The Quarterly Journal of Economics 111(2), 353. Kessler, Daniel and Mark McCl ellan. (1997). The Effects of Malpractice Pressure and Liability Reforms on Physicians' Perceptions of Medical Care. Law and Contemporary Problems 60(1), 81. Kessler, Daniel and Mark B. McClellan. (2000). Is Hospital Competition Socially Wasteful? Quarterly Journal of Economics Volume 115, 2, 577 Lowenstein, G. (2005). Projection Bias in Medical Decision Maki ng. Medical Decision Making (Jan-Feb 2005), 96. Lowenstein, G. and J. Mather (1990). D ynamic Processes in Risk Perception. Journal of Risk and Uncertainty 3:155. Lowenstein, G., T. ODonoghue, and M. Rabi n (2003). Projection Bias in Predicting Future Utility. Quarterly Journal of Economics Volume 118, 4, 1209. McGinley, Patrick John (1995). Comment: Beyond Health Care Reform: Reconsidering Certificate Of Need Laws In a "Managed Competition" System. Florida State University Law Review Vol. 23 (Summer 1995), 141.

PAGE 110

101 McIntire, Donald D., Steven L. Bloom, Br ian M. Casey, and Kenneth J. Leveno (1999). Birth Weight in Relation to Morbidit y and Mortality among Newborn Infants. The New England Journal of Medicine (April 22, 1999) 1234. Meier, Conrad F. (2001). Missouri De bates Future of Certificate-of-Need. Health Care News July 2001. Melnick, G. A., J. Zwanziger, A. Bamezai, and R. Pattison. (1992). "The Effects of Market Structure and Bargaining Position on Hospital Prices." Journal of Health Economics 11 (3): 217. Moulton, Brent R. (1990). An Illustration of a Pitfall in Esti mating the Effects of Aggregate Variables on Micro Units. The Review of Economics and Statistics 72 (2): 334 (May). Morrisey, Michael A. (2000). S tate Health Care Reform: Protecting the Provider, in Feldman, Roger D., ed. American Health Care: Government, Market Processes, and the Public Interest 229. Oakland, CA: The Independent Institute. Noether, Monica. (1988). Competition among Hospitals. Journal of Health Economics 7(3): 259 (September). Rayl, J., Gibson, J., Hickok, D. (1996). A P opulation-Based Case-Control Study of Risk Factor for Breech Presentation. American Journal of Obstetrics and Gynecology, 174(1), 28. Robinson, J.L., D.B. Nash, E. Moxey and J. OConnor. (2001). ISTAHC Annu. Meet. 2001; 17: abstract no. 206 Ryan, Barbara A. (2000). Hospital Regul ation and Antitrust Paradoxical Policies, in Feldman, Roger D., ed. American Health Care: Government, Market Processes, and the Public Interest 171. Oakland, CA: The Independent Institute. Sherman, Daniel. (1988). The Effect of State Certificate-of-Need Laws on Hospital Costs: An Economic Policy Analysis. Washington: Federal Trade Commission Staff Report (January). Shortell, S.M. and E.F. Hughes (1988). T he Effects of Regulation, Competition, and Ownership on Mortality Rates among Hospital Inpatients. The New England Journal of Medicine (April 28,1988) 1100 Simpson, James B. (1985). State Certifi cate-of-Need Programs: The Current Status, American Journal of Public Health Vol. 75, No. 10, 1225.

PAGE 111

102 Sloan, Frank A. (1988). Containing Hea lth Expenditures: Lessons Learned from Certificate-of-Need programs, in Frank A. Sloan, James F. Blumstein, and James M. Perrin, eds. Cost, Quality and Access in Health Care: New Rules for Health Planning in a Competitive Environment 44. San Francisco: Jossey-Bass Publishers. Sloan, Frank A., Michael A. Morrisey, and Joseph Valvona. (1988). Effects of the Medicare Prospective Payment System on Hospital Cost Containment: An Early Appraisal. Milbank Quarterly 66(2): 191. Tancredi, Laurence R. (1988). Defining, M easuring, and Evaluating Quality of Care, in Frank A. Sloan, James F. Blumstein, and James M. Perrin, eds. Cost, Quality and Access in Health Care: New Rules for Health Planning in a Competitive Environment 91. San Francisco: Jossey-Bass Publishers. Tversky, A. and D. Kahneman (1974). Judgmen t Under Uncertainty: Heuristics and Biases. Science (September 1974) 1124. U.S. Dept. of Health and Human Services, National Center for Health Statistics. NATALITY DETAIL FILE, 1998: [UNITED STATES] [Computer file]. Hyattsville, MD: U.S. Dept. of Health and Human Services, National Center for Health Statistics [producer], 1999. Ann Arbor, MI: Inter-university Consortium for Political and Social Rese arch [distributor], 2002. Vaughan-Sarrazin, Mary S., Edward L. Hannan, Carol J. Gormley and Gary E. Rosenthal (2002). Mortality in Medicare Beneficiaries following Coronary Artery Bypass Graft Surgery in States With and Wit hout Certificate of Need Regulation, Journal of the American Medical Association, Vol. 288 No. 15 (October 16, 2002) 1859 1866. Wysocki, Bernard (2002). Care, Costs a nd Competition Medical, Business Interests Spar Over State Hospital-Growth Rules. The Wall Street Journal May 7, 2002 p. A4.

PAGE 112

BIOGRAPHICAL SKETCH Before beginning the Ph.D. program at the University of Florida, Scott Hankins received a B.A. and an M.A. in economics fr om the University of South Florida. He became interested in health economics through his mother, a labor and delivery nurse. He has accepted a visiting position with the Martin School of Public Policy at the University of Kentucky. 103


Permanent Link: http://ufdc.ufl.edu/UFE0015120/00001

Material Information

Title: Three Essays on Health Economics
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0015120:00001

Permanent Link: http://ufdc.ufl.edu/UFE0015120/00001

Material Information

Title: Three Essays on Health Economics
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0015120:00001


This item has the following downloads:


Full Text












THREE ESSAYS ON HEALTH ECONOMICS


By

SCOTT HANKINS













A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


2006

































Copyright 2006

by

Scott Hankins
















TABLE OF CONTENTS



L IS T O F T A B L E S ............................................ ....................................... ................ .. v

L IST O F F IG U R E S .... ...... ................................................ .. .. ..... .............. vii

ABSTRACT ........ .............. ............. ...... ...................... ix

CHAPTER

1 INTRODUCTION ......... ........... .. ................ ...................

2 CERTIFICATE OF NEED REGULATIONS AND HEALTH OUTCOMES ............5

In tro d u ctio n ............... ........... .......... ......... .................. ................ .. 5
Previous Literature............................. .................. ...... .. ............. ....
Certificate of N eed and Costs......... ............................................................. ... 8
Certificate of Need and Quality............................. ........................9
Quality of Neonatal Health ....... ................. ................. 11
D description of D ata............ ................................................................ ........ .. ... 13
M odel Specification ........ ....................................................................... ... ........ 13
Results .............. ........ .................................16
C on clu sion .................................................................................................... 2 3

3 HOSPITAL COMPETITION AND PRICES........................... ....................29

H history of Certificate of N eed ............................................................................. 30
P previous L iterature........ ............................................. ............ ..32
Description of data............................ ........................... ......... ..............35
The Repeal of Certificate of Need and Hospital Competition...................... ..41
H hospital Com petition and Prices ........................................ .......................... 43
C conclusion ........ ... ... .. ..... .......... .............. ............ 46

4 MALPRACTICE LAWSUITS AND MEDICAL PROCEDURE USE...................64

P rev iou s L iteratu re ......................................................................................... .. 6 5
Description of the Data ........................ ................ ........ ................. 70
Physicians and Medical Malpractice Lawsuits........... ..... .................73
R responses to Law suits ........................... .. .......... ...... ....... .....75
O bstetrical Procedures ......... ................. .................................... ............... 75









Caesarean Sections in Philadelphia ........................................ ........................ 78
Caesarean Sections in A llegheny........................................... .......... ............... 80
L abor Inductions in Philadelphia........................................... .......... ............... 81
Labor Inductions in Allegheny ............................................................................82
C o n c lu sio n s........................................................................................................... 8 2

5 C O N C L U SIO N ........ ............................................................................. 98

L IST O F R E F E R E N C E S ........................................................................... ............... 99

B IO G R A PH IC A L SK ETCH .................................................................. ...............103
















LIST OF TABLES


Table page

2-1 States without Certificate of Need (CON) .................................... ............... 25

2-2 A pgar scores ............................................................... ... .... ......... 25

2-3 Summary statistics (standard deviations in parentheses)......................................26

2-4 R results of the first m odel .............................. ......... .............................. ..............26

2-5 H health outcome e if prem ature............................................................ ...............27

2-6 R results of the second m odel.............................. ......... ............... .. ............. 27

2-7 Results of the third m odel ............................................. ...............................28

3-1 Percentage of patients crossing county borders ...................................47

3-2 Num ber of hospitals in the state ..................... ....... ..................... ............. 47

3-3 Percentiles of length of stay (by county type)............... ....................................47

3-4 Effect of CON repeal on the HHI (standard errors in parentheses) .......................48

3-5 Regression results using OLS (with hospital fixed-effects).............................. 49

3-6 Quantile regression results for model 1 (with Hospital Fixed-Effects)....................49

3-7 Quantile regression results for model 2 (with hospital fixed-effects)...................50

3-8 Quantile regression results for model 3 (with hospital fixed-effects)...................50

4-1 Number of doctors by type and county ........................................ ............... 84

4-2 Number of patients by doctor type and county ......................................................84

4-3 Number of licenses issued and number of doctors sued by county and year the
license w as issued ................................................ ................ 85

4-4 Philadelphia results with c-sections as the dependent variable (standard errors in
parentheses) ............................................................... ... .... ........ 86









4-5 Philadelphia segmented results with c-sections as the dependent variable
(standard errors in parentheses)...................... ..... ............................. 87

4-6 Allegheny results with c-sections as the dependent variable (standard errors in
parentheses) ............ ........ ........... ............. ............ .......... 88

4-7 Philadelphia results with inductions as the dependent variable (standard errors in
parentheses) ............. .. ................................................89

4-8 Philadelphia segmented results with inductions as the dependent
variable(standard errors in parentheses)........................ ........ ... ............. 90

4-9 Allegheny results with inductions as the dependent variable (standard errors in
p aren th eses) ...................... .. .. ......... .. .. ......... ................................... 9 1
















LIST OF FIGURES


Figure page

3-1 Average price of a C-Section (by Payer Type and Year)............... ...................51

3-2 Average price of a premature birth (by payer type and year) ................................51

3-3 County-level HHI for 1994 (number of hospitals in county)............................. 52

3-4 County-level HHI for 2004 (number of hospitals in county)............................. 53

3-5 Number of hospitals per county (1994 vs. 2004).................. ................................54

3-6 Number of patients per hospital (1994 vs. 2004).............. .................... 54

3-7 Hospital-specific HHIs (1994 vs. 2004)...................................... ....................... 55

3-8 Hospital-specific HHI averaged at the county-level for 1994..............................56

3-9 Hospital-specific HHI averaged at the county-level for 2004...............................57

3-10 Map of MSAs and major cities in Pennsylvania .............. ....................................58

3-11 Average hospital-specific HHI (by type of county) ...........................................59

3-12 Average hospital charge (by type of county) ................................. ............... 59

3-13 Distribution of charges for metro counties.................................... ............... 60

3-14 Distribution of charges for non-metro counties ....................................... .......... 60

3-15 Averagelength of stay (by type of county) ............... ...... .. ............... ......... 61

3-16 Percentage of births delivered by caesarean section (by type of county) ................61

3-17 Percentage of births delivered prematurely (by type of county)............................62

3-18 Percentage of payers who are HMOs (by type of county) ................. ................62

3-19 Percentage of payers who are Medicaid (by type of county)..............................63

4-1 Percentage of births that are caesarean sections (by county)...............................92









4-2 Percentage of births that are labor inductions (by county)................... .................92

4-3 Percentage of births that are vaginal births after c-sections (by county) ................93

4-4 Percentage of births that are prolonged pregnancies (by county) ..........................93

4-5 Percentage of births that experience fetal distress (by county).............................94

4-6 Percentage of births that lead to neonatal death (by county) .............. ...............94

4-7 Percentage of births that are premature (by county) ...........................................95

4-8 Percentage of births that are breech (by county).............. ...................95

4-9 Percentage of births to mothers over 35 (by county) ................. ....... ............96

4-10 Percentage of births to medicaid mothers (by county) .......................... ..........96

4-11 Cumulative probability of a first lawsuit in allegheny county .............. ...............97

4-12 Cumulative probability of a first lawsuit in Philadelphia county.............................97















Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

THREE ESSAYS ON HEALTH ECONOMICS

By

Scott Hankins

August 2006

Chair: David Figlio
Major Department: Economics

There are approximately 4 million babies born every year in the United States. My

research investigates two distinct issues that may affect both health outcomes of

newborns and mothers as well as the costs of their care. The first chapter introduces the

issues of hospital competition. In the second chapter, I find that increased levels of

hospital competition lead to better outcomes for some premature babies. The third

chapter investigates the relationship between competition and prices. I find that

increased competition leads to lower prices and this effect is greater for the lower cost,

i.e., least complicated patients. The fourth chapter investigates physician decision

making in response to medical malpractice lawsuits. I show that doctors respond to

lawsuits by increasing the number of caesarean sections performed, a result consistent

with defensive medicine.














CHAPTER 1
INTRODUCTION

With approximately 4 million babies born every year in the United States,

Caesarean sections, and other obstetrics procedures, are some of the most common

medical procedures performed. At the same time, obstetrics patients are rarely the focus

of health economics research. Little is known about the effects of competition on

procedure use or health outcomes in this patient population. Health care costs are a

growing concern in America. Much academic research has investigated the causes of

growing medical costs and, more recently, the benefits and determinants of procedure

use. Frequently, health economists rely on Medicare data as it is readily available for all

states. By construction, Medicare patients are older and this data ignores a large section

of hospital admissions. My research uses data on obstetrics admissions, both mothers

and babies. The vast majority of these patients are healthy and young which leads to a

considerably different population to study. With this data, I study the effects of hospital

regulation on health outcomes and prices. I also use this data to investigate the effects of

malpractice lawsuits on physician procedure use.

Researchers have documented vast differences in procedure use among different

geographic areas that lead to higher health care costs without corresponding health

improvements. It is unclear whether the tendency to use Medicare data skews these

conclusions. It is reasonable to assume that the underlying distribution of health

outcomes differs between Medicare Patients and new mothers and babies. The cause of

procedure use variation is still an open question. While it is possible that some of these









differences are due to patient and physician preferences, the wide range of legal and

competitive environments may be responsible. Obstetrics data sheds new light on these

issues.

In my first chapter, I use the repeal of Certificate of Need (CON) regulations as a

proxy for hospital competition and study the effects of this competition on newborn

health outcomes. These regulations started in New York in 1965 with the intent of

controlling health care costs. At the time, hospitals were paid on a cost-plus basis. That

is, their costs were covered plus a given percentage as profit. Because of this regulatory

structure, hospitals had no incentives to control costs since they were all but guaranteed

of covering them. There was also a fear that as insurance dulled the price sensitivity of

consumers, hospitals in competitive markets would engage in a "medical arms race" and

supply a socially excessive amount of medical care.

Federal regulation in 1972 required hospitals to obtain state approval of capital

improvements in order to receive Medicare/Medicaid payment for health care that used

these improvements. In 1974, federal regulation required all states to implement a CON

program by 1980. In the early 1980s the federal requirement for CON programs was

repealed. In the following year, 14 states eliminated their CON laws. This framework

allows me to study the relationship between changes in hospital competition and health

outcomes.

I find that most babies benefit from being born in a state without Certificate of

Need. It appears that the repeal of these laws does not have an immediate effect on

newborn health, but outcomes shift over a period of time. In the short run, very









premature babies may be harmed by the repeal. However, it is possible that this effect is

due to small sample issues.

The second chapter uses detailed hospital discharge data from Pennsylvania for

all new mothers and babies in the years surrounding Pennsylvania's repeal of Certificate

of Need regulations. I examine hospital competition as the regulatory regime changed

and determine the type of hospital most likely to be affected by the repeal.

To measure hospital competition, I calculate a hospital-specific Hirschman-

Herfindahl Index (HHI). Instead of using county borders as the definition of the

hospital's market, the hospital-specific HHI uses a patient's zip code as the geographic

market and constructs a weighted average of these smaller geographic areas. I argue that

the hospital-specific HHI is superior to the standard HHI as, it measures each hospital's

competitive environment at a much finer geographic level. I find that the repeal of CON

did not have a measurable effect on hospital competition although, it appears that small

hospitals in more competitive areas are more likely to close. I use the changing levels of

hospital competition over these years to investigate the effects of competition on the

prices charged by hospitals. I also find that increased levels of competition lead to lower

prices.

The third chapter uses the same Pennsylvania discharge data to explore the effect

of medical malpractice lawsuits on physician behavior. While there are a number of

papers that investigate the issue of litigation induced medical procedures or "defensive

medicine", this chapter directly measures how doctors respond when they are sued for

malpractice. This is distinct from previous research which was not able to link an









individual doctor's lawsuit experience with their procedure use and therefore looked at

the how changes in the legal environment might be related to changing procedure use.

Using the data from Philadelphia and Allegheny Counties, I show that doctors

increase the number of caesarean sections they perform after being sued for malpractice.

This is consistent with the hypothesis of defensive medicine. I also find that doctors in

Allegheny County respond to a lawsuit in a more dramatic fashion relative to doctors in

Philadelphia County. It is possible that the larger number of lawsuits relative to

Allegheny County causes doctors in Philadelphia to be hyper-sensitive to the threat of

lawsuits given.














CHAPTER 2
CERTIFICATE OF NEED REGULATIONS AND HEALTH OUTCOMES

Introduction

Certificate of need regulations affect the number of hospitals in a community as

well as the number and type of services that a hospital can provide. Initially designed to

control costs, these regulations may limit competition and potentially impact the quality

of healthcare. While there has been much research on the impact of Certificate of Need

laws (CON) on health care costs, few have looked at the effect of these laws on health

outcomes. In this paper, I document a negative correlation between Certificate of Need

laws and health outcomes of newborn babies. Specifically, I use data from the U.S.

Department of Health and Human Services Linked Birth/Infant Death File (which

includes information from almost every birth certificate matched to the corresponding

death certificate), and a difference-in-difference approach to show that babies in states

with CON laws are less likely to be healthy (as measured by the 5-minute Apgar score).

By comparing health outcomes in states that repealed their laws at different times, I also

show that the positive effects, i.e., greater probabilities of a healthy birth, of removing

CON restrictions is due to those states which repealed their laws at least 10 years earlier.

In addition, these positive effects, of increased hospital competition, are greater for most

premature births.

Certificate-of-Need regulations started in New York in 1965 with the intent of

controlling health care costs. At the time, hospitals were paid on a cost-plus basis. That

is, their costs were covered plus a given percentage as profit. Because of this regulatory









structure, hospitals had no incentives to control costs since they were all but guaranteed

of covering them. There was also a fear that as insurance dulled the price sensitivity of

consumers, hospitals in competitive markets would engage in a "medical arms race" and

supply a socially excessive amount of medical care.

Certificate of need requires state approval of a hospital's large capital projects,

including either the expansion of existing facilities or the introduction of new services.

For example, the purpose of Missouri's CON statute "is cost containment through health

cost management, assurance of community need and the prevention of unnecessary

duplication of health care services. CON is based on a goal of public accountability

through public review of proposed health care services, value promotion and negotiation

among competing interests" (Meier, July 2001). States differ on the level of investment

or changes in the level of service (regardless of changes in capital) that requires prior

approval. Often a decrease in the level of services as well as an increase must go through

approval process.

Federal regulation in 1972 required hospitals to obtain state approval of capital

improvements in order to receive Medicare/Medicaid payment for health care that used

these improvements. In 1974, federal regulation required all states to implement a CON

program by 1980. The "Reagan revolution" in the early 1980s removed the federal

requirement for CON programs and, subsequently, some states eliminated their CON

laws. However, most did not, and some expanded their programs. Currently, 14 states

do not have a CON program (Table 2-1).

There is a perception within the health care industry that CON programs protect

incumbents (i.e., Stigler's capture theory of regulation). The director of government









relations for the Missouri State Medical Association said, "The Certificate of Need

program has outlived its usefulness. It doesn't do anything but stifle competition and

innovation. It's extremely bureaucratic, and no one relishes having to go through it. It's

the people who have existing projects who want it to continue. It helps them keep

competition out of their backyard" (Meier, July 2001).

Economic theory suggests that certificate of need regulation will lead to inefficient

results in competitive markets. However, the hospital market is far from competitive.

There is a mix of non-profit and profit hospitals in most geographical markets. There are

also areas with only one hospital. While there has been research on competition and

medical care, there is not a generally accepted theory of how competition affects

prices/costs and quality. As will be discussed, others have looked at the effects of

competition and prices/costs. It is hypothesized that either prices/costs will be higher

than they would be in the absence of pure competition or that quality will be lower. This

can occur because incumbents are protected from competition because competitors

(either new entrants or existing hospitals that wish to expand) must go through the

certificate of need process. This process is costly and, in many states, an existing hospital

is allowed to testify against a potential entrant. As with any market barrier, it is

reasonable to expect less strenuous competition and possibly less innovation. If these

barriers limit competition along both the price as well as the quality dimension, then it is

reasonable to expect sub-optimal health outcomes in the states with CON requirements.

Previous studies have shown that prices/costs tend to be higher in states with CON

requirements; this paper shows that there is also a decrease in health outcomes in those

same states.









Previous Literature

The existing literature on certificate of need and hospital competition covers two

distinct areas. The earliest research looked at the impact of both CON regulation and its

subsequent repeal in some states, as well as general competition within the health care

market, on costs. More recently, researchers have begun to investigate the effect of

competition, and specifically CON programs, on the quality of health care.

Certificate of Need and Costs

Sloan (1988) provides a detailed review of the early literature. In general, the early

studies find that CON has no statistically significant effect on health care costs and most

agree that if there were to be an effect, it would increase costs rather than lower costs.

CON proponents argued that these early studies were flawed because of their limited time

horizon immediately following the introduction of these programs. It was argued that the

programs needed time to "get up and running" before they could be expected to control

costs.

More recent studies have looked at longer periods of time and evaluated the effects

of repeal of the programs (a good survey is Morrisey, 2000). These studies suggest that,

if anything, CON programs have tended to increase costs (Sloan, Morrisey and Valvona,

1988) and found that the repeal of CON had no effect on hospital costs per capital

(Sherman, 1988). Conover and Sloan (1998) found that "mature" CON programs result

in a slight reduction in bed supply but higher costs and higher profits.

Antel, Ohsfeldt, and Becker (1995) analyzed hospital costs allowing for interaction

between regulation programs other than CON. Using state data on hospitals costs per

day, per admission, and per capital, they found that CON had no statistically significant

effect in any of their empirical specifications. Health care costs and prices charged both









increase with the duration of the CON program. Using Medicare price and cost data from

1977-78, Noether (1988) found prices were higher relative to costs and therefore profits

were higher in states with a CON program. CON proponents also argued that hospitals

were different than other industries and therefore competition did not, and could not,

work. However, the empirical evidence does not support this. Melnick, et al. (1992)

looked at transaction prices that a large California preferred provider organization (PPO)

negotiated with hospitals in 1987. While controlling for many factors, they found that the

PPO paid more in the less competitive markets. This suggests that competition does

work to lower prices in hospital markets.

Certificate of Need and Quality

While much research has been done on the determinants of health care quality (see

Tancredi, 1988 for a review), very little work has been devoted to examining the effects

of health care regulation or competition on quality. In one of the few papers, Ho and

Hamilton (2000) looked at the effects of hospital consolidation on health care quality.

Analyzing California hospital care before and after mergers and acquisitions between

1992 and 1995, they looked at several proxies for quality of care. They find 90-day

readmission rates for heart attack patients and discharges within 48 hours for normal

newborn babies increased in some cases. While mortality and readmission rates are

reasonable, though imperfect, metrics for health care quality, it is unclear that an increase

in "early discharges" should be considered as such. The authors equate early discharge

with cost cutting measures of hospitals, but the connection with quality is not discussed.

Furthermore, they find no measurable effect on inpatient mortality for heart attack and

stroke patients.









A recent paper by Kessler and McClellan (2000) looks at hospital competition and

Medicare beneficiaries' heart attack care from 1985 to 1994. They find the welfare

effects, i.e., expenditures and treatment and patient health outcomes, of competition to be

ambiguous in the 1980s. However, in the 1990s, "competition unambiguously improves

social welfare".

In addition to these papers on competition and the quality of healthcare, two papers

have focused on the effect of CON regulations on Coronary Artery Bypass Graft (CABG)

quality. Robinson, et al. (2001) looked at outcomes in Pennsylvania for the years 1994 -

1999. The elimination of the state's CON program is found to increase the number of

open-heart surgery programs by 25% without a significant increase in the number of

surgeries performed. With this limited sample, quality (as measured by the mortality

rate) was not impacted by this reallocation of volume. Vaughan-Sarrazin, et al. (2002)

consider a larger dataset of Medicare beneficiaries who underwent CABG surgery

between 1994 and 1999. The authors find higher mortality rates for CABG patients in

states without CON programs (5.1% compared to 4.4%). There is one potential problem

with this paper, it compares cross-sectional differences across states. The authors cite

several papers that point to a negative connection between hospital volume and mortality

for CABG surgery. While they do not state such a hypothesis, it would appear that there

is a type of "learning by doing" in open-heart surgery. This is a reasonable hypothesis as

open-heart surgery is a complicated and lengthy process in which the skills of the surgeon

and the other medical professionals could have a great impact on patient outcomes.

While the previous literature looked at both CON regulation and hospital

competition as determinants of health care costs, the question how CON regulations









affect health care quality has only begun to be addressed. This paper will attempt to

evaluate the issue of certificate of need regulation, hospital competition and the quality of

care. Specifically, the effect of CON regulation on the quality of neonatal care will be

investigated.

Quality of Neonatal Health

Birth certificates in the United States record many different kinds of information

about the parents as well as the baby. Information about both parent's socioeconomic

background, as well as factors affecting the mother and baby's health are collected. One

variable recorded is the "5-minute Apgar score" for the newborn baby. I use the Apgar

score as a proxy for the quality of neonatal health.

The Apgar score is a subjective measure of the infant's condition based of heart

rate, respiratory effort, muscle tone, reflex irritability, and color. Each of these factors is

given a score of 0, 1, or 2; the sum of these 5 values is the Apgar score, which ranges

from 0 to 10. A score of 10 is optimal, and a very low score raises a flag about the

subsequent health and the survival of the infant. The Apgar score was designed to be a

useful measure of the need for resuscitation and a predictor of the infant's chances of

surviving the first year of life. A recent paper by Casey, et al. (2001) shows that for

premature infants (26 to 36 weeks of gestation), the neonatal mortality rate was 315 per

1000 births for an infant with an Apgar score of 0-3, as compared with 5 per 1000 births

for an infant with an Apgar score of 7-10. Similar results are demonstrated for full term

babies.

Despite the emphasis placed on low birth weight and poor health outcomes, in the

popular press, there is some evidence that low birth weight is not itself the sole predictor

of infant mortality. One paper concludes that the threshold weight below which mortality









is significantly greater is the 3rd percentile (McIntire, et al. (1999)). A recent working

paper by Almond, et al. (2002) uses twins to compare the correlation between birth

weight and various health outcomes. The authors find that the heavier twin is no more

likely to survive past the first year than the lighter twin. They also find that the Apgar

score is more highly associated with infant mortality but is not correlated with birth

weight. This is because a low Apgar score may be caused by a birth trauma not related to

prematurity. For these reasons, low birth weight is not used as a health outcome.

After consultation with a labor and delivery nurse, I chose Apgar scores of 8 and

above to be healthy.1 The reasoning for this is as follows, while the Apgar score is a

subjective measure based on the health care professional's opinion, there appears to be

some agreement on what score denotes a healthy baby. For example, one nurse may give

a baby an Apgar score of 9 while another may give a score of 8, the score is different but

both professionals would agree that the baby was healthy. It is highly unlikely that one

professional would score a baby a 6 and another score the same child as an 8. In

addition, while the percentage of Apgar scores either 9 or 10 have changed from 1983 to

1999, the percentage of Apgar scores greater than or equal to 8 has stayed relatively

stable.2 While the break between score of 8 and 9 appears to be the "natural" break (see

Table 2-2), I have chosen to be conservative and defined the break point for health to be

between 7 and 8.




1 As a robustness check, the same models were estimated with a cutoff of 7 or 9. The conclusions do not
change with the different cutoff points.
2 One potential reason for the shift from Apgar scores of 10 to Apgar scores of 9 is the increasing use of
pain medication over this time period, specifically epidurals. These medications tend to "dull" the baby's
responses and therefore affect the Apgar score.









Description of Data

The data set and the description of the variables are from the National Center for

Health Statistics Matched Birth/Death Certificates file for 1983 and 1999. The beginning

of the sample was chosen to be 1983 because in years preceding that time the Matched

Birth/Death was incomplete for many of the states for earlier years. The data for 1983

are based on the total number of births in 46 states. In 1983, California, Delaware,

Oklahoma and Texas did not report Apgar scores on their birth certificate therefore these

four states were excluded from the sample. Unfortunately, California and Texas are two

of the 15 states without a CON program. New Mexico was also dropped from the sample

because it did not record gestational age in 1983. Louisiana was excluded from the

analysis because it was the only state to not implement a CON regulation while it was

required by the federal government3. Arizona, California, Delaware, and Georgia as well

as the District of Columbia only reported a 50 percent random sample of their births.

The data set is limited to those births for which all variables of interest, described

in the model specification section, are available. In order to control for differences in

prenatal care in other countries, an observation was also dropped if the mother was a

resident of another country4. We are left with 2,602,155 observations in 1983 and

2,920,950 observations in 1999, for a total of 5,523,105 observations.

Model Specification

The basic model estimated is,

outcome = Xp + noCON89a1 + noCON99a2 + year99ac3 + uj +



3 Louisiana actually started its CON regulations in 1991, after 11 states had repealed their CON laws.

4 Most of the discarded observations are lacking an Apgar score or gestational age.









where outcome is an indicator of either a healthy birth or a neonatal death. For the

models that estimate the probability of a healthy birth, the dependent variable, is a

dummy variable that takes the value of one if a baby has a 5-minute Apgar score greater

than or equal to 8 depending on the model. Because the 5-minute Apgar score is a

measure of health at the 5-minute mark, it is an imperfect measure of hospital quality at

best. It is likely that many quality related problems would occur after the 5-minute mark

and would not be picked up in this data. Nonetheless, it is reasonable to believe that some

quality related problems would occur before the Apgar score. Because the vast majority

of problems will not be captured in the Apgar score, the estimates can be considered a

lower bound. The fact that we find any results with the healthy dependent variable,

suggests that these laws have an effect.

Because the Apgar score may not pick up all health problems, the same model is

estimated with the dependent variable an indicator of neonatal death. That is, the variable

is equal to one if the baby died in its first year of life and is zero otherwise. If CON laws

have an effect, we would expect the results to have opposite signs because of the nature

of the indicators. For example, if CON laws are beneficial then the estimated coefficients

on the noCON indicators will be negative if the dependent variable is healthy (indicating

that dropping a state's CON program reduces the probability of a healthy baby).

Similarly, the estimated coefficients will be positive for the noCON indicators if the

dependent variable is death (indicating that dropping a state's CON program increases the

probability of a baby dying in its first year).

The variables of interest are the noCON variables; noCON89 is a dummy variable

that takes the value of one if the state has dropped its Certificate of Need program by









1989 and noCON99 is an indicator that is equal to one if the state dropped its CON

program between 1990 and 1999. The noCON variables are the difference in differences

estimates of the effect of dropping a state's certificate of need program either in the

1980s or the 1990s relative to the states that still maintain these regulations. That is,

noCON89 is the difference between 1983 and 1999 in the difference between states with

a CON program and those that dropped their CON program in the 1980s (a similar

interpretation holds for noCON99). As previously mentioned, of the 45 states in the

sample and Washington D.C., 14 had dropped their CON programs by 1999 (11 in the

1980s and 3 in the 1990s).

The obvious question is why did only some of the states repeal their regulations in

the 1980s. And then why did the second wave of repeals occur in the 1990s. On the

surface, endogeneity appears to be a problem. However, it is likely to be less of a

concern than is initially apparent.

Omitted variables are the most common form of endogeneity; to the extent to

which the underlying factors do not change much over time, the omitted variables

problem is dealt with using state level fixed effects. If there are time-varying omitted

variables, we may have a reverse causality problem. That is to say, the dependent

variable influences the variable of interest. In this case, health outcomes, or underlying

health trends, would have to affect the repeal of a state's certificate of need regulations.

This is highly unlikely, in order for this to be true one would have to argue that

politicians or bureaucrats observe a downward trend in health outcomes, relative to other

states, and conclude that the way to fix this problem is through the repeal of the state's

CON regulation. One could argue that obvious solution from a planner's perspective









would be to argue for more regulation in this case, rather than less. If this is the case,

then the potential bias in this case works against finding any effect from the repeal of

CON.

The mother and child characteristics contained in X are: the mother's age at the

time of birth and the baby's birth weight rounded to the nearest 100 grams5. (See Table

2-3). Year99 is an indicator that is equal to one if the year is 1999; while, uj represents

state fixed effects and F is a random error.

All models are estimated as a linear probability model with state fixed effects to

control for unobserved state specific variation. As the certificate of need laws vary at the

state level, all of the standard errors are corrected for clustering of the errors at the state

level as described in Moulton (1990). State fixed effects and state level laws imply that

we are estimating probabilities at the state level, i.e., the average probability of a healthy

baby in a state, conditional on observed characteristics. Because the vast majority of

births in the United States are healthy ones, we are trying to explain a rare event, that of

an unhealthy birth or a neonatal death. The rarity of this event, combined with the

limited information recorded on the birth certificate, makes one suspect that a goodness

of fit measure such as R2 is going to be poor. Indeed, this is what is found for all of these

models.

Results

As mentioned in the previous section, the dependent variable (Healthy) is an

indicator equal to one if the Apgar score is greater than or equal to 8, and zero otherwise.



5 While birth certificates record other useful information, i.e mother's education, number of prenatal visits,
etc. not all states report these variables. The model was limited to mother's age and birth weight to
maximize the number of states included in the data. Gestational age is used in later models.









The coefficients all have the expected sign (see Table 2-4, column 1). The variable

year99 is positive and significant, which suggests that the likelihood of a healthy baby

increased from 1983 to 1999, given the advances in medical technology (such as fetal

monitoring), this intuitively makes sense. The coefficient on birth weight is positive and

statistically significant. Babies with higher birth weight are more likely to have an Apgar

score of 8 or higher, an increase of one standard deviation (600 grams or 21 ounces) leads

to an increase in the probability of a healthy birth by 4 %.

The variables of interest in this paper are the noCON indicators. This variable takes

the value of one if a state does not have a Certificate of Need program and zero

otherwise. The noCON89 coefficient is both statistically and economically significant.

A baby born in a state that removed its CON program in the 1980s is .59 % more likely to

be healthy than a baby born in a state with a CON program. While this may appear

economically insignificant, the estimated increase in the probability of a healthy birth

only increased 1.2% from 1983 to 1999. As the vast majority of babies are born healthy,

it is impossible for the effect of CON repeal to lead to a large change in the likelihood of

a healthy birth. The noCON99 coefficient is negative but not statistically significant.

This implies that the states which dropped their restrictions on hospital competition in the

1990s did not see an increase in the likelihood of a healthy birth. This result is not

surprising, since the entire 2nd wave of CON removals occurred in the later part of the

1990s.6 The differential impact between the early and later groups implies a time lag

between the removal of competition barriers and an increase in hospital competition or

the increased competition's effect on health outcomes. Given the time required to


6 The three states that dropped their CON regulations in the 1990s did it in either 1995 or 1996









finance and construct major capital improvements, a delayed effect on health outcomes is

not surprising. The lack of an effect implies that it takes time for there to be an increase

in hospital competition and for this increase to have an effect on health outcomes. Given

the length of time it takes to build new a hospital, a lag in the effect of removing

constraints on competition is to be expected.

Because the Apgar score is an imperfect measure of a newborn baby's health, the

same basic model was estimated with an indicator of neonatal death as the dependent

variable. Again, the coefficients have the expected signs. An increase in the baby's

weight of 600 grams (one standard deviation) leads to a 2% reduction in the probability

of death. While the coefficient on mother's age is positive, indicating increasing age

increases the chance of death, it is difficult to argue that the coefficient is economically

significant. The coefficient on the year99 indicator is negative, which again makes sense

given the improvements in medical technology. The coefficients on both of the noCON

variables are statistically and economically insignificant. This is not completely

unexpected since very few babies are so sick that they are in danger of dying. However,

it raises the question of the effects of CON regulation on high-risk, i.e., premature, births.

Because premature births are more likely to have health problems and are more likely to

die, there is the possibility of more aggressive medical interventions and hence, more

quality related issues.

The Interactions of Certificate of Need regulation and Premature Births:

Because CON programs may affect neonatal health by limiting Neonatal Intensive Care

Units (NICUs), the same model is estimated with the inclusion of different indicators of

premature birth interacted with the certificate of need indicators as before. These high-









risk births are more likely than a normal birth to require the services of a NICU. While

not all premature births need a NICU, the presence of a NICU indicates that the hospital

staff is more prepared to handle these high-risk births.

The average length of gestation is 39 weeks, while the definition of full term

pregnancy is 40 weeks. Anything less than 37 weeks is considered premature and

increases the risk of medical complications for the infant and the mother. Premature

births are relatively rare in the data (approximately 10% are born at less than 36 weeks in

1983 and 12% in 1999). Of these premature births, the majority of premature births are

born between 36 and 32 weeks (variable P36), this represents 7% of the total number of

births in 1983 and 9% in 1999. As can be seen in Table 2-5, all types of premature births

have increased relative to 1983.

Babies born premature are less likely to be healthy and more likely to die.

Although these babies are less healthy, their prospects have improved over the time

period. For those births with a gestational age between 32 and 36 weeks, the probability

of being healthy has increased from 92% to 95%, while the mortality has decreased from

2% to 1%.

To test the effect of CON regulations on premature births, an indicator of

prematurity is created that equals one if the baby had a gestational age of 36 weeks or

less. The first of these models estimated is'

outcome = Xp + noCON89a1 + noCON99a2 +
PREPf + PRE x noCON89P32 + PRE x noCON99P3 + year99a3 + uj + 6


The secondary interactions are not reported but are available upon request.









where PRE is an indicator of premature birth (i.e., PRE equals 1 if gestation is less

than or equal to 36 weeks). The coefficient on the interacted term can be interpreted as a

"difference in difference in differences" estimate. For example, the interaction of PRE

with one of the noCON indicators gives us the "difference" between these high-risk

births and other births in the "difference" between states with a certificate of need

program and those without a program. Given that noCON is a difference in difference

estimate, the coefficient on the interaction term is the difference in difference in

differences estimate.

First, the model was estimated with health as the dependent variable. As expected,

being born prematurely significantly reduces the probability of being healthy (-7.7%).

The coefficient on the interaction of the noCON indicators with the premature indicator is

positive for both the states that dropped their CON in the 1980s and in the 1990s. Again,

we see that only the interaction with noCON89 is statistically significant. The estimated

impact on premature babies is 1.98% (Column 1, Table 2-6) in those states that repealed

their CON programs in the 1980s. These positive coefficients indicate that removing

barriers to competition results in a higher likelihood of a healthy birth. As before, the

higher probability of a healthy birth in the states that removed their CON restrictions in

the 1980s versus those that removed the restrictions in the 1990s is reasonable given the

time needed for increased competition to have an effect on health care. Again, this may

appear to be a small effect. However, when compared to the average change in the

probability of being healthy over this time period, the effect of CON repeal leads to a

large increase.









Next, the same model was estimated with death as the dependent variable. As

would be expected, being born prematurely increases the chance of death (2.96%). The

interactions of the premature indicator and the noCON indicators are not as easy to

interpret as with the health regression. Premature babies born in states that removed their

CON restrictions in the 1990s are slightly more likely (.23%) to die relative to those

states that still maintain CON restrictions. There is no statistically significant effect on

those babies born in the states that removed their CON restrictions in the 1980s. This

seems to imply that there are short run costs to the removal of CON, although this effect

is small from an economic significance point of view.

Because it is possible that the effect of certificate of need regulation varies with the

gestational age of the baby, indicators of prematurity by specific weeks are created. The

variables P28, P32, P36 are indicators of premature births (i.e., P28 indicates gestation

less than or equal to 28 weeks while P32 indicates gestation greater than 28 and less than

or equal to 32 weeks).

These indicators of premature birth are interacted with the noCON indicators and

the following model estimated is8:

outcome = Xp + noCON89a1 + noCON99a2 + year99a3
+ P28p, + P32P2 + P36P3
+ P28 x noCON89, + P32 x noCON8982 + P36 x noCON89 3
+ P28 x noCON99y) + P32 x noCON99y2 + P36 x noCON99y3

+ u +


8 As before, the secondary interactions are not reported.









Not surprisingly, the probability of a healthy birth is monotonically increasing in

the prematurity indicators and the probability of death is monotonically decreasing (Table

2-7). The extremely premature births (P28) are much less likely to be born healthy (47%)

and more likely to die (34%) than a full term birth. The births with a gestational age

between 28 and 32 weeks (P32) have a better chance of a healthy birth (17%) and are less

likely to die (4%) than the extremely premature births. The babies born between 32 and

36 weeks are only slightly more likely to be unhealthy (3%) or to die (.25%) than a full

term birth.

For all but the oldest premature births, the removal of CON restrictions does not

appear to have an effect. For the P36 births, the interactions of the noCON indicators are

positive for the health outcomes regardless of when the CON restrictions were removed.

As before, the effect of removing the CON restrictions is larger for the states that

repealed their laws in the 1980s versus those that repealed in the 1990s. These results are

also economically significant given the reduced probability of being healthy if bom

prematurely (Table 2-5).

Again, we see an effect in the short run when the dependent variable is death. The

interaction of noCON99 and P28 is positive (3.07%); indicating the repeal of Certificate

of Need in the 1990s has led to an increase in mortality for the extremely premature

births. We also see a decrease in mortality for the babies born between 28 and 32 weeks.

The interaction of noCON99 and p32 is -.82%. It appears that these babies benefit from

the repeal. There is no effect from the repeal of CON in the 1980s.

It is possible that these contradictory results are due to the small number of very

premature babies, especially in those states that dropped their laws in the 1990s, in this









data set. However there is a possible economic explanation for these results. The

certificate of need process was not established simply to hinder competition but also to

maintain acceptable levels of quality in the hospital industry. One argument in favor of

the certificate of need process is to maintain a critical volume at area hospitals, especially

in services like NICUs. It is conceivable that, in the short-term, the removal of CON

leads to worse outcomes for very high-risk infants by expanding the number of places

where a premature baby can be born within a geographical area. Newly opened hospital

facilities may not have high enough volumes of these special needs of births to allow the

medical staff to maintain their skills of a more established hospital.

A baby born with less than 28 weeks of gestation is extremely small and is likely to

have health problems. A baby of this size is difficult to intubate or administer an I.V. to

when such steps are necessary and problems in these processes could impact the 5-minute

Apgar score. These are also tasks for which there may be "learning by doing" present. If

experience leads to greater medical proficiency, learning by doing could result in better

health outcomes in those hospitals that perform a large number of these procedures. This

hypothesis would lead us to expect that very high-risk infants would be harmed by the

repeal of CON regulations, as seen in these empirical findings. In the long-run, the repeal

of CON regulations allows competitor hospitals to establish NICUs and to achieve high

enough volume levels to improve outcomes. It is important to remember that the vast

majority of births benefit immediately from the repeal of CON regulations and given time

extremely premature births benefit as well.

Conclusion

Restricting competition within the hospital market appears to negatively affect the

quality of care as measured by health outcomes. Complementing the earlier findings that









certificate of need regulation does not control health care costs, this paper shows that

CON regulations are associated with lower Apgar scores and a slightly higher incidence

of neonatal deaths. However, a small subsection of high-risk births may benefit from a

CON program. More research to determine the cause of this is necessary.

As this data does not include information on costs, it is impossible to say whether

or not the certificate of need process is socially efficient. Yet, given the negative impact

of these laws on the overwhelming majority of births, Certificate of Need regulations

may not be the optimal social policy.










Table 2-1. States without Certificate of Need (CON)
State Year Dropped
Arizona 1985
California* 1987
Colorado 1987
Idaho 1983
Kansas 1985
Minnesota 1984
New Mexico 1983
North Dakota 1995
Ohio 1995
Pennsylvania 1996
South Dakota 1988
Texas* 1985
Utah 1984
Wyoming 1985
* does not record Apgar scores

Table 2-2. Apgar scores
1983 1999
Percent Total Percent Total Percent Total Percent Total
Apgar of Births Number Dead Died of Births Number Dead Died
0 0.06 1,547 0.61 950 0.07 2,164 0.55 1180
1 0.21 5,543 0.77 4284 0.18 5,263 0.82 4303
2 0.14 3,629 0.55 1989 0.08 2,415 0.56 1356
3 0.14 3,752 0.36 1333 0.09 2,607 0.32 826
4 0.20 5,285 0.23 1225 0.14 4,079 0.18 739
5 0.38 9,889 0.15 1468 0.25 7,357 0.12 872
6 0.79 20,680 0.08 1754 0.59 17,184 0.08 1354
7 1.99 51,743 0.04 1821 1.51 43,961 0.04 1596
8 8.72 226,889 0.01 2618 7.06 206,303 0.01 2140
9 62.62 1,629,369 0.00 6699 82.05 2,396,692 0.00 5492
10 24.74 643,829 0.00 2102 7.97 232,925 0.00 439
Total 2,602,155 1.01 26,243 2,920,950 0.69 20,297










Table 2-3. Summary statistics (standard deviations in parentheses)
Variable 1983 1999 Total
Mom's Age 25.554 27.209 26.429
(5.307) (6.156) (5.831)
Birth Weight (100s grams) 33.514 33.144 33.319
(5.944) (6.152) (6.058)
Gestation (weeks) 39.405 38.772 39.070
(2.825) (2.620) (2.737)
Healthy 0.961 0.971 0.966
(0.194) (0.168) (0.181)
Died 0.010 0.007 0.008
(0.100) (0.083) (0.091)
(P28) Gestation <= 28 weeks 0.009 0.010 0.009
(0.093) (0.098) (0.096)
(P32) 28 weeks < Gestation <=32 weeks 0.014 0.016 0.015
(0.119) (0.127) (0.123)
(P36) 32 weeks < Gestation <=36 weeks 0.071 0.092 0.082
(0.256) (0.289) (0.274)


Table 2-4. Results of the first model


noCON89
noCON99
Weight
Age
Year99
Constant
Observations
Number of state
R-squared


(1)
Healthy
Coefficient Std Error
0.5808 (0.2040)
-0.0807 (0.2612)
0.6696 (0.0132)
-0.0226 (0.0028)
1.2182 (0.0690)
74.2231 (0.4208)
5523105
45
0.05


(2)
Died
Coefficient Std Error
-0.0197 (0.0613)
0.0795 (0.0324)
-0.3350 (0.0076)
0.0091 (0.0012)
-0.4593 (0.0246)
12.0051 (0.2452)
5523105
45
0.05










Table 2-5. Health outcome if premature


Prematurity
Gestation <=28 weeks
28 weeks < Gestation <=32 weeks
32 weeks < Gestation <=36 weeks
36 weeks < Gestation


Prematurity
Gestation <=28 weeks
28 weeks < Gestation <=32 weeks
32 weeks < Gestation <=36 weeks
36 weeks < Gestation


Table 2-6. Results of the second model


noCON89
noCON99
Weight
Age
Year99
Premature
Premature x noCON89
Premature x noCON99
Constant
Observations
Number of state
R-squared


(1)
Healthy
Coefficient Std Error
0.4942 (0.1535)
-0.0857 (0.2148)
0.5397 (0.0118)
-0.0189 (0.0028)
0.9354 (0.0593)
-7.6778 (0.2647)
1.9781 (0.9056)
0.3842 (0.6014)
79.2237 (0.3660)
5523105
45
0.06


(2)
Died
Coefficient Std Error
-0.0293 (0.0560)
0.0430 (0.0259)
-0.2952 (0.0074)
0.0084 (0.0012)
-0.2786 (0.0177)
2.9625 (0.0780)
-0.2087 (0.2041)
0.2386 (0.0889)
10.4124 (0.2379)
5523105
45
0.05


Percentage
of Births
0.88
1.44
7.05
90.64


Percentage
of Births
0.96
1.65
9.21
88.18


1983
Total
Number
22,808
37,358
183,478
2,358,511

1999
Total
Number
28,127
48,076
269,161
2,575,586


Percent
Healthy
0.44
0.76
0.92
0.97


Percent
Healthy
0.42
0.83
0.95
0.98


Percent
Died
0.38
0.07
0.02
0.01


Percent
Died
0.34
0.04
0.01
0.00










Table 2-7. Results of the third model


noCON89
noCON99
Weight
Age
Year99
P28
P32
P36
P28 x noCON89
P32 x noCON89
P36 x noCON89
P28 x noCON99
P32 x noCON99
P36 x noCON99
Constant
Observations
Number of state
R-squared


(1)
Healthy
Coefficient Std Error
0.4761 (0.1507)
-0.1009 (0.1973)
0.3196 (0.0075)
-0.0095 (0.0024)
0.9090 (0.0547)
-47.7169 (0.8620)
-17.3646 (0.5657)
-3.0095 (0.2340)
2.3086 (3.4791)
3.3508 (2.2495)
1.6457 (0.6683)
-3.1427 (2.3017)
1.1753 (0.6270)
0.7822 (0.4249)
86.5317 (0.2382)
5523105 5523105
45 45
0.11 0.15


(2)
Died
Coefficient
-0.0185
0.0525
-0.1576
0.0022
-0.2615
34.3691
4.0656
0.2526
-0.5584
-0.8877
-0.1384
3.2114
-0.8233
0.0286
5.8538


3td Error
(0.0535)
(0.0182)
(0.0043)
(0.0010)
(0.0159)
(0.8272)
(0.1837)
(0.0480)
(3.0722)
(0.5830)
(0.1504)
(0.8427)
(0.3561)
(0.0559)
(0.1431)














CHAPTER 3
HOSPITAL COMPETITION AND PRICES

Originally mandated by the Federal government, Certificate of Need (CON)

regulations were designed to control hospital competition. With the goal of limiting

redundancy and tempering a "medical arms race," Certificate of Need regulations

restricted hospital competition. The repeal of these laws creates new opportunities for the

hospital markets; new entrants have a lower barrier to entry and existing players can

expand services more easily.

To explore the relationship between the repeal of CON, hospital competition and

prices charged by hospitals, I employ two techniques. First, a hospital-specific

Hirschman-Herfindahl Index (HHI) is constructed to capture each hospital's competitive

environment. Second, I use quantile regression to estimate the effect of changes in

competition on prices. The extreme skewness of price data could unduly influence the

econometrics otherwise. While prior hospital competition research focused on Medicare

data, this paper uses the full sample of obstetrics patients in Pennsylvania between 1994

and 2004. These patients differ greatly in terms of both age and general health from

Medicare patients providing a new perspective on the impact of hospital competition.

Given the type of data, it is impossible to identify a causal relationship between the repeal

of CON and hospital competition. However, I find that while the repeal of CON is not

correlated with changes in competition, changes in the competitive landscape have an

impact on prices. This effect is larger for the lower quantiles, implying that competition

matters more for the easier/cheaper medical care.









History of Certificate of Need

Certificate of Need regulations affect the number of hospitals in a community as

well as the number and type of services that a hospital can provide. Certificate of Need

regulations started in New York in 1965 with the intent of controlling health care costs.

At the time, hospitals were paid on a cost-plus basis. That is, their costs were covered

plus a given percentage as profit. Because of this regulatory structure, hospitals had no

incentives to control costs since they were all but guaranteed to cover them. There was

also a fear that as insurance dulled the price sensitivity of consumers, hospitals in

competitive markets would engage in a "medical arms race" and supply a socially

excessive amount of medical care. These regulations are designed to limit competition

and in the process, control costs.

The Certificate of Need process requires hospitals to receive state approval of large

capital projects, including either the expansion of existing facilities or the introduction of

new services. For example, the purpose of Missouri's CON statute "is cost containment

through health cost management, assurance of community need and the prevention of

unnecessary duplication of health care services. CON is based on a goal of public

accountability through public review of proposed health care services, value promotion

and negotiation among competing interests" (Meier, July 2001). States differ on the level

of investment or changes in the level of service (regardless of changes in capital) that

requires prior approval. Often a decrease in the level of services as well as an increase

must go through approval process.

Federal regulation in 1972 required hospitals to obtain state approval of capital

improvements in order to receive Medicare/Medicaid payment for health care that used

these improvements. In 1974, federal regulation required all states to implement a CON









program by 1980. The "Reagan revolution" in the early 1980s removed the federal

requirement for CON programs and, subsequently, some states eliminated their CON

laws. However, most did not, and some expanded their programs. Currently, 15 states,

including Pennsylvania, do not have a CON program.

There is a perception within the health care industry that CON programs protect

incumbents (i.e., Stigler's capture theory of regulation). The director of government

relations for the Missouri State Medical Association said, "The Certificate of Need

program has outlived its usefulness. It doesn't do anything but stifle competition and

innovation. It's extremely bureaucratic, and no one relishes having to go through it. It's

the people who have existing projects who want it to continue. It helps them keep

competition out of their backyard" (Meier, July 2001).

Economic theory suggests that Certificate of Need regulation will lead to inefficient

results in competitive markets. However, the hospital market is far from competitive.

There is a mix of non-profit and profit hospitals in most geographical markets. There are

also areas with only one hospital. It is hypothesized that either prices/costs will be higher

than they would be in the absence of pure competition or that quality will be lower. This

can occur because incumbents are protected from competition because competitors

(either new entrants or existing hospitals that wish to expand) must go through the

Certificate of Need process. This process is costly and, in many states, an existing

hospital is allowed to testify against a potential entrant. As with any market barrier, it is

reasonable to expect less strenuous competition and possibly less innovation. Indeed,

previous studies have shown that prices/costs tend to be higher in states with CON

requirements.









Previous Literature

The existing literature on Certificate of Need and hospital competition covers two

distinct areas. The earliest research looked at the impact of both CON regulation and its

subsequent repeal in some states, as well as general competition within the health care

market, on costs.1 In general, the early studies find that CON has no statistically

significant effect on health care costs and most agree that if there were to be an effect, it

would increase costs rather than lower costs. CON proponents argued that these early

studies were flawed because of their limited time horizon immediately following the

introduction of these programs. It was argued that the programs needed time to "get up

and running" before they could be expected to control costs.

More recent studies have looked at longer periods of time and evaluated the effects

of repeal of the programs.2 These studies suggest that, if anything, CON programs have

tended to increase costs (Sloan, Morrisey and Valvona, 1988) and found that the repeal of

CON had no effect on hospital costs per capital (Sherman, 1988). Conover and Sloan

(1998) look at the effect of CON removal on state-level per-capita hospital spending

among other things for the years 1976 to 1993. They find that CON laws had no effect

on per-capita health expenditures, however CON did reduce spending on acute care by

5%. They find no effects on spending from the removal of CON laws.

Antel, Ohsfeldt, and Becker (1995) use state-level average hospital costs to

investigate the affect of state regulations over the years 1968 to 1990. Specifically, they

look at rate-setting regulations and Certificate of Need regulations; as well as procedure



1 Sloan (1988) provides a detailed review of the early literature.
2 A good survey of the more recent work is Morrisey (2000).









controls like Peer Review Organizations and the interactions of these regulations. After

controlling for state fixed effects, they find that regulations in isolation do not have an

effect on costs. There are some effects from different interactions of regulations, but the

magnitudes are small.

Health care costs and prices charged increase with the duration of the CON

program. Using Medicare price and cost data from 1977 to 1978, Noether (1988) found

prices were higher relative to costs and therefore profits were higher in states with a CON

program. CON proponents also argued that hospitals were different than other industries

and therefore competition did not, and could not, work. However, the empirical evidence

does not support this.

A number of studies have investigated the effects of hospital competition on prices

and costs. The early research tended to use a Hirschman-Herfindahl Index (HHI) as a

measure of competition faced by a hospital. Because of data limitations, this index was

usually calculated at the county level as finer geographic information was unavailable.

There are two drawbacks to this approach. One, this assigns every hospital in a county

the same measure of competition, ignoring the fact that there is frequently a dominant

hospital and a couple of smaller hospitals. Two, it assumes that a patient would be

willing to go to any hospital in a county or stated differently, that every hospital is on

equal footing in competing for patients and that no patients cross county borders for

medical care.









Because of these strong and unrealistic assumptions, a hospital-specific HHI is

used in this paper.3 A hospital specific HHI is calculated using the patient's zip code as

the measure of the market. The market shares of the competing hospitals were calculated

for each zip code. These market shares were then used to calculate a HHI at the zip code

level. A weighted sum of these HHIs was then calculated for each hospital where the

weights are the hospital's share of patients coming from that zip code. A number of

papers have used hospital-specific HHIs as a measure of hospital competition.

Melnick, et al. (1992) looked at transaction prices that a large California preferred

provider organization (PPO) negotiated with hospitals in 1987. While controlling for

many factors, they found that the PPO paid more in the less competitive markets. This

suggests that competition does work to lower prices in hospital markets. The authors

compare the results when competition is measured with a hospital-specific HHI and a

county-level HHI. They conclude that the hospital-specific measure of competition

performs better in explaining the price differences.

Zwanziger and Melnick (1988) use hospital-level data from California for the years

1980 to 1985 to investigate the effects of hospital competition as well as the introduction

of Medicare's Prospective Payment System (PPS) on hospital costs. They find that the

introduction of the PPS caused hospitals to significantly reduce their costs. They also

find that hospitals in less competitive markets had higher costs prior to the introduction of

the PPS, a result consistent with a medical arms race. After the PPS was introduced,

hospital competition no longer had a significant effect on costs.



3 See Morrisey, Sloan and Valvona (1988) and Zwanziger, Melnick and Mann (1990) for reviews of
defining hospital markets.









Using hospital-level data for the United States, Bamezai, et al. (1999) look at the

effects of managed care, as well as market structures, on hospital operating costs. The

direct effect of hospital competition is not statistically significant. However, when

interacted with measures of managed care penetration, it is highly significant. This

implies that "greater hospital competition is effective only in areas with high levels of

managed care penetration".

Kessler and McClellan (2000) look at the effects of hospital competition on

Medicare patients' heart attack care. They use predicted patient flows to calculate

hospital-specific HHIs. They find that expenditures were 8% higher in the hospitals

facing the least competition relative to those hospitals facing the most competition.

A paper close in spirit to this one is Zwanziger and Mooney (2005). They look at

the effects of the deregulation of hospital prices in New York. They investigate HMO-

hospital transaction prices and show that negotiated prices were lower in more

competitive markets. This effect becomes larger after deregulation.

Description of data

This paper uses patient-level hospital discharge data on obstetrics patients from the

state of Pennsylvania for the years 1994 to 2004.4 The data includes information on

every obstetrics patient discharged in Pennsylvania for those years, approximately

375,000 observations. The data contain detailed billing information for each patient as

well as information on who paid for the medical care. While the actual amount paid is

almost certainly less than the billed amount, unfortunately that amount is unknown.


4 For financial reasons, the data is for the 1st quarter of each year.









There is also detailed diagnosis and procedure information as well as the patient's zip

code.

While it would be preferable to know the actual price paid, we can be relatively

certain that the price reported is a list price rather than a negotiated price that differs for

each patient or payer. Figure 3-1 plots the average price of a caesarean section by payer

type. We see that the there has been a large increase in the price of a c-section, but there

is not a large difference between payer types. We see the same results for premature

births in Figure 3-2. Again, there has been a large increase in prices, but there is no

difference between payer types. Because we expect that different organizations will pay

different prices for a given procedure, the lack of differences suggests that the prices used

in this paper are list prices. It is reasonable to believe that the list prices are related to

actual prices paid. Because it is doubtful that anyone would pay more than the billed

price, any results can be interpreted as a lower bound.

Traditionally, hospital competition is measured with a Hirschman-Herfindahl Index

(HHI). This index measures competition as the sum of the squared market shares in a

given geographic market. Greater levels of competition are represented by lower

numbers. Figure 3-3 shows the HHI calculated at the county-level for Pennsylvania in

1994. The lighter shaded areas have the greatest amounts of competition. The two areas

with the most competition are Allegheny County in the west and Philadelphia County in

the east. Figure 3-4 shows the comparable HHI for 2004. It is apparent from these maps

that, by this measure, there has not been a large change in competition over the

intermediate years. Some of the smaller counties have lost hospitals and therefore the

market has become more concentrated.









These maps also show the number of hospitals in a given county for that year. A

number of counties have had decreases in the number of hospitals. Using the number of

hospitals in a county as a measure of competition, it would appear that there have been

changes that may not have been picked up by the HHI. As can be seen from Figure 3-5,

most counties did not see a change in the number of hospitals.5 44 counties, out of 67

total, did not see any change at all in the number of hospitals. Only two counties gained a

hospital. We also observe that most counties only have one or two hospitals. None of

the counties with only one hospital in 1994 lost a hospital, and only one county with two

hospitals lost one. The larger counties did lose hospitals, in some cases more than one. It

is important to note, that not all of the losses are necessarily due to hospital closures. It is

possible that some of these hospitals discontinued their obstetrics service while

continuing to provide other forms of care.

Small hospitals (as measured by the number of patients) were more likely to close

as compared to larger hospitals. Figure 3-6 is a scatter plot of the number of patients per

hospital in 1994 and the number of patients per hospital in 2004. 6 The number of

patients per hospital in 1994 for those hospitals that closed between 1994 and 2004 is

plotted along the x-axis. It is clear that the large hospitals have a tendency to get bigger.

It is also clear that the majority of hospitals have less than 300 patients in a given year.

As well, it appears that most hospitals did not change their patient counts very much.





5 The number of counties in a cell is represented by the size of the circle. All cells greater than 1 have the
number of counties inside the circle.
6 Only one hospital had more than 1200 patients in one year, the number of patients did not vary greatly
from 1994 to 2004.









We also see from Figure 3-7, that the hospitals that closed were those that were

facing the higher levels of competition (as measured by the hospital-specific HHI).

Again, the hospitals that closed between 1994 and 2004 are plotted along the x-axis. This

figure also makes apparent that a number of hospitals faced dramatically less competition

in 2004 as compared to 1994. This is almost certainly due to the hospitals that left the

market.

In order to rectify these discrepancies, a different approach is required. The

traditional Hirschman-Herfindahl Index is calculated using county borders as the borders

of the market. As mentioned, this approach makes several strong assumptions. The

assumption that no patients cross county borders for care can be shown to be false. Table

3-1 shows that in any given year, approximately 25% of patients receive treatment in a

county other than their own. Allegheny and Philadelphia counties are broken out of the

total, as they are the largest counties in Pennsylvania. Allegheny County is similar to the

state as a whole, while in Philadelphia County approximately 16% of patients are from

another county.

For ease of display, the hospital-specific HHIs are averaged at the county level for

the years 1994 and 2004 (Figures 3-8 and 3-9). These maps show greater amounts of

competition as compared with the traditional HHI. As expected, competition is greatest

in those areas with the largest population. Figure 3-10 shows the MSAs as well as the

location of the major cities in Pennsylvania. The level of hospital competition is seen to

be greater on average in these areas.

Somewhat surprisingly, there is very little change over the time period in the levels

of hospital competition. Figure 3-6 plots the average of the hospital-specific HHI for









both counties in and out of a MSA (metro and non-metro counties). It is difficult to

detect any effect from the repeal of CON. However, there is a slight upward trend over

time in both areas. This implies that hospital competition has actually decreased on

average over this time period, which is the opposite of what we would expect from the

removal of CON. Of course, it is possible that these averages obscure the individual

hospital's changing competitive environment. Related to the slight increase in the

hospital-specific HHI, we see in Table 3-2 that the actual number of hospitals in

Pennsylvania peaked in 1996. The number of hospitals decreases dramatically over the

following years. Again, Allegheny and Philadelphia counties are broken out. We see

that the number of hospitals peaks in Allegheny County in 1996 and peaks in 1997 for

Philadelphia County. Although it is impossible to say with certainty, the evidence above

makes it is likely that the repeal of CON led some of the weaker hospitals to discontinue

their obstetrics service.

Figure 3-11 shows the average price charged for metropolitan and non-metro

counties. Hospitals in metro counties charge more relative to non-metro counties. The

difference increases dramatically over the time period. This difference is consistent with

the Medical Arms Race hypothesis in that the areas facing the greatest competition also

charge the most. It is also possible that more complex medical cases are admitted to

hospitals in metro areas and thus lead to higher charges. In this case, using price as a

proxy for medical complexity, we would expect to see the greatest difference between

metro and non-metro areas at the top of the distribution and very little difference at the

bottom of the distribution. Figures 3-8 and 3-9 plot the distribution of prices for metro

counties and non-metro counties.









We would also expect to see the average length of stay to be longer in metro areas

if this is true. Figure 3-14 shows the average length of stay for the two areas. We see

that the average stay in non-metro counties has been relatively flat over this period, while

the length of stay has increased in the metro counties. Table 3-3 shows the 90th and 99th

percentiles of the length of stay for both counties over the time period. We see that the

90th percentile in non-metro counties has stayed constant at three days; the comparable

percentile for metro counties is four days. At the 99th percentile, the length of stay is

again greater in the metro counties as compared to the non-metro counties. It appears

that some of the difference in the average price may be due to the sorting of complex

medical cases.

These figures also suggest that only investigating the effects of competition on

mean prices may be misleading. Because hospitals that face the most competition tend to

be located where the patients are, that is in metro areas, simply regressing prices on a

Herfindahl index would lead to the inference that increased levels of competition increase

prices. This paper circumvents this problem by including hospital fixed effects as well as

quantile regression. The fixed effects control for unobserved heterogeneity among

hospitals, while quantile regression allows me to investigate the effects of CON removal

as well as the effects of changing competition on prices at different points on the

distribution.

Because of the concern that part of the difference in prices in metro counties versus

non-metro counties may be due to the complexity of the patient care, it is important to

control for premature births and caesarean sections. Figure 3-15 shows the percent of

births delivered by caesarean section over the time period. There is not a large difference









in caesarean rates between metro and non-metro counties. Figure 3-16 shows the percent

of births that were born prematurely. At the beginning of the time period, metro counties

had a higher incidence of prematurity but, this difference has disappeared by the end of

the time period.

Managed care, i.e., Preferred Provider Organizations (PPOs) and Health

Maintenance Organizations (HMOs) is believed to play an important part in controlling

health care costs, it is important to control for this. Figure 3-17 shows the percent of

patients whose primary payer was a PPO or HMO.7 While managed care organizations

have increased their penetration in both types of counties, the increase has been more

dramatic in metro counties. By 2004, more than 60% of patients are in managed care in

the metro counties compared with less than 40% in non-metro counties.

It is also reasonable to believe that Medicaid patients will affect the price a hospital

charges. Figure 3-18 shows the percent of mothers whose primary payer was Medicaid.8

There is a small increase in the number of Medicaid patients in non-metro counties but

the increase is much larger in the metro counties, however in all years, the share of

patients on Medicaid is larger in the non-metro counties.

The Repeal of Certificate of Need and Hospital Competition

The first model estimated considers the relationship between the repeal of the

Certificate of Need regulation and hospital competition as measured by the hospital-

specific Hirschman-Herfindahl Index (scaled to range from 0 100). In order for the

CON repeal to have an effect in a given area, there must be multiple hospitals competing



SThe data does not distinguish between the two types before 2000.

8 Some Medicaid patients are enrolled in a managed care plan.









with each other. This is more likely to occur in the metro areas as there are a greater

number of potential patients. Although we do not have a truly exogenous source of

variation, and therefore can not argue that this procedure will estimate a causal relation,

we can use this idea to test the effect of the CON repeal on hospital competition. By

comparing the change in the HHI in metro counties after the repeal to any change in non-

metro counties, we can get an approximation of the effect of the repeal. The model

estimated is,

hhiit = al + a2 metro + y years after + 6 metro x years after + Eit (2-1)

where hhiit is the hospital-specific HHI for hospital i in year t, metro is an indicator equal

to one if the hospital is in a metropolitan county, year is a vector of indicators for the

years after repeal, and metro x year is a vector of indicators of metro-specific years after

repeal. If there is a differential effect between the two types of counties, then the

coefficients on the metro-specific year indicators will be statistically different from zero.

We would expect to see that if anything, the repeal of CON has increased competition in

the metro areas and therefore the estimated coefficients will be negative. Column 1 of

Table 3-4 shows the estimated OLS coefficients for this model. The only coefficients

statistically different from zero are the constant and the metro indicator. This implies that

there was not an effect from the repeal on competition. Because it is possible that any

effect would be greater in the larger or smaller areas, three quantile regressions were

estimated. Columns 2-4 report the estimated coefficients for the 75th, 50th, and 25th

quantiles. In no case are any coefficients statistically significant, again except for the

constant and metro indicator. While we can not say with certainty that the repeal of CON

had no effect on hospital competition as measured by the hospital-specific HHI, the fact









that the repeal is not correlated with changes in competition is suggestive of the idea that

there was not an effect.

Hospital Competition and Prices

The basic model estimated is

chargesijt = 1 hhijt + 32 hmoijt + 13 medicaidijt + 04 csecijt + 35prematureijt
(2-2)
+ Yj+ 6+ Eijt

The dependent variable is the total charges for patient i that the hospital reported

to the state in year t. Where hhi is the hospital-specific HHI, hmo is an indicator that the

patients primary payer was a HMO, medicaid is an indicator that the patient has

Medicaid, csec is an indicator that a caesarean section was performed, premature is an

indicator of a premature birth, and metro is an indicator that the hospital is in a metro

county, while yj is a hospital fixed-effect and 6t is a year fixed-effect.

The variable of interest is hhi. Because the Herfindahl index decreases as

competition increases, if hospital competition reduces prices then the coefficient on hhi

will be negative. As a baseline, the model was estimated using OLS. These results are

reported in the first column of Table 3-5. The coefficient on hhi is statistically

insignificant, and the magnitude is small, implying that competition does not affect

hospital prices. As expected, both c-sections and premature births cost more. Somewhat

surprisingly, Medicaid patients are charged more as well.

The second model estimated is

chargesijt = 1 hhijt + 32 hmoijt + 13 medicaidijt + 34 csecijt + 35 prematureijt +
(2-3)
06 year + yj + Eijt

This model substitutes a linear time trend instead of individual year fixed-effects.

Again, this model was estimated using OLS. These results are reported in column 2 of









Table 3-5. Again, we see that the coefficient on hhi is small and statistically

insignificant, implying that competition does not have an affect on hospital prices. The

other coefficients are not markedly different.

The third model estimated is

chargesijt = 1 hhijt + 32 hmoijt + 13 medicaidijt + 34 csecijt + 35 prematureijt +
(2-4)
06 year + 37 metro x year + yj + ijt

The third model adds a metro-county specific time trend. This controls for the

different growth trajectory in metro counties seen in Figures 3-8 and 3-9 (distribution of

charges). The results of this estimation are reported in column 3 of Table 3-5. The

coefficient on hhi is again small and insignificant, while the other coefficients are

qualitatively the same.

As can be seen from Figures 3-8 and 3-9, the distribution of prices is heavily

skewed by the top of the distribution. It is reasonable to believe that only a couple of

hospitals in given area are responsible for the highest charges as these would be the most

medically complex. Given this, it is not surprising to find that competition does not

affect average hospital prices as the average in this case is dominated by the upper half of

the distribution where there is little competition. For this reason, quantile regression is

employed.

It is reasonable to believe that competition would have the greatest effect at the

bottom of the price distribution. These are the "simplest" medical cases. It is likely that

more complicated cases can potentially be admitted to only a subset of an area's

hospitals, because not every hospital would have the staff or facilities to manage a

complex medical case. If this is the case, then we would expect to see competition have a









greater effect on the lower quantiles. The first model is estimated at five different

quantiles: .90, .75, .50, .25 and .10 in order to investigate the hypothesis that hospital

competition does not affect the upper part of the distribution, but may affect the lower

part.

Table 3-6 reports the results for the respective quantile regressions. All of the

coefficients are statistically significant and, the coefficients are increasing as the quantiles

get smaller. This supports the hypothesis that competition has a greater effect for the

least complicated medical cases. For example, a one standard-deviation increase in

competition would reduce the price by approximately $225 at the median price.9

Relative to the average charge of $5950 in 1999, this implies a 3.8% change in price. If

we compare the competition effect between the 75th and 25th quantiles, we see that

competition matters much more for the lower quantile. A one standard deviation increase

in competition would reduce the price by 1.7% at the 75th percentile; a similar change

would reduce the price by 8.0% at the 25th percentile. 10 It does appear that hospital

competition matters more for the lower quantiles.

The results of the quantile regressions of the second model are reported in Table 3-

7. As with the OLS version, the results are qualitatively similar to the first model. The

coefficients on hhi are all statistically significant but not very large. The third model is

reported in Table 3-8. Again, the results are very similar to the other models. All three




9 The standard deviation of hhi is 15.6; this multiplied by the coefficient on hhi for the median regression
(14.67) implies a price change of $229.

10 The calculation for the 75th percentile is: 8.14*15.6 = $127 as compared to a price of $7200 in 1999. This
is a difference of 1.7%. A similar calculation using an average price of $3118 in 1999 yields a difference
of 8.0% for the 25th percentile.









models suggest that hospital competition has largest effect on the lower half of the price

distribution.

This estimated effect can be interpreted as an upper bound. Given that the

dependent variable in all three models is the billed price, not the actual price paid, it is

likely that some patients paid less. It is difficult to believe that anyone would pay more

than the billed price. If that is the case, the effect of competition could be even larger.

Conclusion

This paper looks at the effect of the repeal of Pennsylvania's Certificate of Need

regulations in 1996 on hospital competition and hospital prices. I use a hospital-specific

Hirschman-Herfindahl Index to measure hospital competition. Although the CON repeal

does not appear to affect hospital competition, I show that there were changes in the

number of hospitals in Pennsylvania and therefore the hospital-specific HHIs. The

hospitals that were small and faced high levels of competition, as measured by the

hospital-specific HHI, were more likely to discontinue their obstetrics service.

The changes in the individual HHIs was then used to determine the effects of

competition on prices charged. A quantile regression approach was used because the

hospital prices are heavily skewed to the right. Using this approach, increased hospital

competition was found to reduce prices charged. Hospital competition is found to matter

much more for the lower half of the price distribution as compared to the upper half. A

one standard deviation increase in competition, as measured by the hospital-specific HHI,

would reduce prices by 1.7% at the 75th percentile. A comparable increase in

competition would reduce prices by 8.0% at the 25th percentile.












Table 3-1. Percentage of patients crossing county borders


Total
0.259
0.252
0.259
0.265
0.257
0.264
0.261
0.267
0.262
0.267
0.272


Number of
Metro
111
114
116
110
104
103
103
101
97
97
89


Allegheny
0.233
0.238
0.253
0.250
0.248
0.254
0.244
0.250
0.267
0.266
0.270


Philadelphia
0.169
0.168
0.171
0.179
0.172
0.189
0.182
0.183
0.166
0.167
0.160


hospitals in the state
Non-Metro Allegheny
40 14
39 13
41 14
40 11
40 12
39 13
40 12
39 11
38 10
36 11
36 10


Table 3-3. Percentiles of length of stay


90th percentile


year
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004


non-metro
3
3
3
3
3


metro
4
3
3
4
4


(by county type)


99th percentile
non-metro metro
7 9
6 9
6 8
6 8
6 8


Year
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004


Table 3-2.
Year
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004


Philadelphia
17
20
19
20
16
16
15
15
13
11
11











Table 3-4. Effect of CON repeal on the HHI (standard errors in parentheses)


metro

1 year after x metro

2 years after x metro

3 years after x metro

4 years after x metro

5 years after x metro

6 years after x metro

7 years after x metro

8 years after x metro

1 year after

2 years after

3 years after

4 years after

5 years after

6 years after

7 years after

8 years after

Constant

Observations


(1)
OLS
-19.576
(1.484)
0.102
(2.977)
0.892
(2.995)
-0.260
(3.018)
1.424
(2.998)
1.458
(3.025)
1.226
(3.059)
-0.865
(3.106)
-0.166
(3.135)
0.518
(2.552)
0.674
(2.552)
1.420
(2.577)
-0.303
(2.552)
0.616
(2.577)
1.805
(2.602)
2.948
(2.657)
2.418
(2.657)
58.600
(1.276)
1573


(2)
Quantile (.75)
-24.277
(2.105)
7.585
(4.211)
4.430
(4.241)
6.197
(4.204)
9.044
(4.236)
4.043
(4.219)
6.387
(4.287)
9.518
(4.394)
8.817
(4.434)
-3.630
(3.613)
-0.910
(3.613)
-2.551
(3.575)
-4.551
(3.613)
-1.210
(3.575)
-1.718
(3.633)
-4.928
(3.759)
-3.635
(3.759)
70.605
(1.810)
1573


(3)
Quantile (.50)
-19.759
(2.291)
1.641
(4.574)
1.116
(4.600)
-0.721
(4.636)
-0.779
(4.605)
3.850
(4.645)
1.229
(4.697)
-2.305
(4.766)
-4.207
(4.810)
0.185
(3.914)
1.997
(3.914)
0.818
(3.950)
0.912
(3.914)
-1.329
(3.950)
2.306
(3.988)
3.714
(4.069)
4.114
(4.069)
57.398
(1.969)
1573


(4)
Quantile (.25)
-18.233
(2.836)
-4.536
(5.675)
0.014
(5.714)
-1.669
(5.665)
-3.525
(5.708)
-0.193
(5.685)
-1.429
(5.776)
-6.416
(5.920)
-0.640
(5.975)
3.370
(4.868)
0.401
(4.868)
1.581
(4.817)
4.078
(4.868)
2.131
(4.817)
3.448
(4.895)
8.554
(5.065)
4.037
(5.065)
46.683
(2.439)
1573











Table 3-5. Regression results using OLS (with hospital fixed-effects)


hhi

hmo

csec

premature

medicaid


year

metro_year

Constant

Observations
Number of Hospitals
R-squared
F-Stat
Hospital Fixed Effects
Year Fixed Effects


Table 3-6. Quantile regression results
(.90)
hhi 7.80
(1.34)
hmo 205.40
(14.59)
csec 5557.68 4
(14.81)
premature 6779.11 2
(23.73)
medicaid 340.14
(14.67)
Observations 374770 37
Number of Hospitals 169
Pseudo R-squared 0.46
Hospital Fixed Effects Yes
Year Fixed Effects Yes


for model 1 (with Hospital Fixed-Effects)


(.75)
8.14
(0.77)
167.42
(8.19)
129.68
(8.31)
171.95
(13.28)
199.82
(8.09)
4770
169
0.43
Yes
Yes


(.50)
14.67
(0.59)
177.15
(6.49)
3210.34
(6.50)
847.43
(10.38)
141.81
(6.25)
374770
169
0.38
Yes
Yes


(.25)
16.03
(0.52)
162.60
(6.06)
2663.22
(5.93)
459.40
(9.49)
87.25
(5.66)
374770
169
0.34
Yes
Yes


(.10)
12.61
(0.55)
125.47
(6.69)
2287.21
(6.42)
313.57
(10.28)
41.94
(6.08)
374770
169
0.30
Yes
Yes


(1)
37.53
(25.94)
278.69
(156.08)
5096.58
(472.12)
3603.95
(578.50)
415.96
(105.66)


1621.50
(1299.44)
374770
169
0.05
29.11
Yes
Yes


(2)
34.91
(23.57)
96.50
(165.38)
5141.49
(479.65)
3597.61
(576.26)
463.09
(107.89)
560.73
(89.04)


493.10
(1405.22)
374770
169
0.05
45.01
Yes
No


(3)
31.48
(24.35)
10.82
(170.00)
5133.17
(477.60)
3608.21
(577.70)
431.73
(103.62)
153.54
(37.42)
484.06
(113.80)
627.19
(1402.66)
374770
169
0.05
39.55
Yes
No











Table 3-7. Quantile regression results for model 2 (with hospital fixed-effects)


hhi

hmo

csec

premature

medicaid

year

Observations
Number of Hospitals
Pseudo R-squared
Hospital Fixed Effects
Year Fixed Effects


Table 3-8. Quantile regression results for model
(.90) (.75)
hhi 3.71 5.76
(1.67) (0.93)
hmo 57.70 51.18
(15.12) (9.07)
csec 5739.92 4157.34
(15.75) (9.38)
premature 6633.41 2193.67
(25.24) (14.98)
medicaid 380.64 246.02
(15.54) (9.12)
year 198.59 149.53
(6.17) (3.49)
metro x year 236.64 230.34
(6.70) (3.79)
Observations 374770 374770
Number of Hospitals 169 169
Pseudo R-squared 0.46 0.42
Hospital Fixed Effects Yes Yes
Year Fixed Effects No No


3 (with hospital fixed-effects)
(.50) (.25) (.10)
11.64 12.39 9.86
(0.61) (0.52) (0.53)
87.84 109.49 102.16
(6.63) (6.13) (6.34)
3260.90 2662.78 2281.28
(6.68) (6.06) (6.15)
881.48 467.64 321.49
(10.68) (9.68) (9.88)
147.51 80.34 37.27
(6.42) (5.77) (5.83)
114.76 89.56 77.34
(2.36) (2.09) (2.15)
208.61 189.67 165.98
(2.54) (2.26) (2.33)
374770 374770 374770
169 169 169
0.38 0.35 0.30
Yes Yes Yes
No No No


(.90)
8.18
(1.72)
102.50
(15.35)
5760.95
(16.13)
6672.35
(25.84)
388.66
(15.93)
381.99
(2.69)
374770
169
0.45
Yes
No


(.75)
9.29
(1.00)
99.39
(9.66)
4157.28
(10.04)
2185.47
(16.04)
255.43
(9.76)
326.30
(1.59)
374770
169
0.42
Yes
No


(.50)
15.58
(0.64)
125.68
(6.96)
3270.89
(7.03)
883.98
(11.23)
172.34
(6.75)
273.31
(1.05)
374770
169
0.38
Yes
No


(.25)
15.59
(0.50)
142.83
(5.94)
2664.86
(5.85)
462.61
(9.37)
90.08
(5.58)
236.84
(0.87)
374770
169
0.34
Yes
No


(.10)
11.74
(0.55)
121.03
(6.62)
2285.00
(6.39)
315.51
(10.27)
44.33
(6.07)
207.54
(0.98)
374770
169
0.30
Yes
No

































1996 1998 2000 2002
Year


- Blue-Cross


-- Commercial Medicaid


Figure 3-1. Average price of a C-Section (by Payer Type and Year)



20000


m
a
| 15000
E

0


10000





5000


1994


1996 1998 2000 2002
Year

SBlue-Cross -- Commercial Medicaid


Figure 3-2. Average price of a premature birth (by payer type and year)


20000-



o
0


1 -
c 15000
6

a,


i 10000-


5000-
1994


2004


2004















HHI
o [0.00,0.20]
o (0.20,0.30]
* (0.30,0.40]
* (0.40,0.50]
M (0.50,1.00]


Figure 3-3. County-level HHI for 1994 (number of hospitals in county)















HHI
o [0.00,0.20]
o (0.20,0.30]
M (0.30,0.40]
M (0.40,0.50]
M (0.50,1.00]


Figure 3-4. County-level HHI for 2004 (number of hospitals in county)



































I I I I I I I I
0 2 4 6 8 10 12 14
# of Hospitals per County in 1994
Figure 3-5. Number of hospitals per county (1994 vs. 2004)


1200-
1100-
1000-
900-
800-
700-
600-
500-
400-
300-
200-
100-
0-


0 100 200 300 400 500 600 700 800
# of patients in 1994


1 18 2
16 18 20


900 1000 1100 1200


Existing Hospitals A Drop-out Hospitals

Figure 3-6. Number of patients per hospital (1994 vs. 2004)


0










-o o
-' 0 o



,o o
a0
0
,0' 0
0 0' 0


0 0


S A. .-i'
.7 e
.- A
.'




dKAl L -I A A AA A
9 *
h Pe aa












"""

**


*.*.*^**. *
o *o







A A A^A AinAAAA A &A A A A


10 20 30 40 50
HHI in 1994


60 70 80 90 100


Existing Hospitals A Drop-out Hospitals

Figure 3-7. Hospital-specific HHIs (1994 vs. 2004)











HHI
= [0.00,0.35]
= (0.35,0.45]
- (0.45,0.55]
m (0.55,0.65]
m (0.65,1.00]


Figure 3-8. Hospital-specific HHI averaged at the county-level for 1994











HHI
o [0.00,0.35]
r (0.35,0.45]
- (0.45,0.55]
S(0.55,0.65]
m (0.65,1.00]


Figure 3-9. Hospital-specific HHI averaged at the county-level for 2004
































Figure 3-10. Map of MSAs and major cities in Pennsylvania



















Io
(2)
0)

I&


I I I I I I I I I I I
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
year

Metro County -- Non-Metro County

Figure 3-11. Average hospital-specific HHI (by type of county)


o
0
0-


O
CO
co
0



o
CD 0



oo
> 0D


0
0
O<


1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
year

Metro County Non-Metro County

Figure 3-12. Average hospital charge (by type of county)


....................... .................


~=L------"/













25000-

22500-

20000-

17500-
0

0
S15000-
4--
o 12500-

S10000-

S7500-







9000-







cn 8000-

.l 7000-
2500-

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
year
Figure 3-13. Distribution of charges for metro counties

10000

9000-

8000

c 7000-

0 6000-

5000-

4000



2000-

1000-
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
year
Figure 3-14. Distribution of charges for non-metro counties


Percentiles
- 90th
75th
-- 50th
A 25th
--- 10th




















Percentiles
- 90th
75th
-- 50th
-- 25th
x 10th












CO
(N




Co


4-
0
0c

a) (Nl
.j
)C
0)


<(N


1 I I I I I I I
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
year

-- Metro County Non-Metro County

Figure 3-15. Averagelength of stay (by type of county)


tu-
CO










C -
V5


I I I I I I I I I I I
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
year

SMetro County Non-Metro County


Figure 3-16. Percentage of births delivered by caesarean section (by type of county)




















C,
3
CO

1-
E
0.0
ao
c
(D
V
(D
0-


L19
1994


1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
year

Metro County -- Non-Metro County


Figure 3-17. Percentage of births delivered prematurely (by type of county)




co







0
E
I
C, r
V-/^ -* ^ ~


1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
year

Metro County Non-Metro County

Figure 3-18. Percentage of payers who are HMOs (by type of county)





















o
O, n
E co











I I I I I I I I I
1994 1995 1996 1997 1998 1999 2000 2001 2002
year

Metro County Non-Metro County

Figure 3-19. Percentage of payers who are Medicaid (by type of county)














CHAPTER 4
MALPRACTICE LAWSUITS AND MEDICAL PROCEDURE USE

"America's health care professionals should be focused on fighting illnesses, not on
fighting lawsuits. Junk lawsuits change the way docs do their job. ... If you're
worried about getting sued, you're going to do everything you can to make sure
you don't get sued. That's why doctors practice what's called defensive medicine.
That means they're writing prescriptions or ordering tests that really aren't
necessary, just to reduce the potential of a future lawsuit."

-George W. Bush1

"Defensive medicine" is a serious public policy concern and a potential contributor

to the increasing cost of medical care in the United States. Anecdotal evidence indicates

that the problem may be growing in severity. However, the academic literature's

understanding of defensive medicine remains incomplete. The process of determining

what constitutes defensive medicine has been confounded by data limitations. Current

research is based on medical costs, malpractice premiums, or legal reforms not actual

medical decisions and outcomes. Thus far, the practice of defensive medicine has been

inferred from proxies for the fear of malpractice lawsuits or more specifically the legal

environment. This paper adds to the literature on defensive medicine by examining the

effect of malpractice lawsuits on specific physician's medical procedure use. I

hypothesize that defensive medicine is due to both the fear of being sued as well as to

physician's responses to actually being sued. Using a panel of obstetricians, a difference

in difference approach circumvents the problem of unobserved physician heterogeneity.




1 Speech on January 5, 2005.









While the legal environment may be the primary impetus for defensive medicine, I argue

a doctor's response to litigation is interesting in its own right.

In this paper, I use hospital discharge data for maternity patients at all of the

hospitals in Allegheny and Philadelphia counties for the years 1994 20042. These data

identify the physician by medical license number which was then used to identify the

physician by name. The doctor's name was then matched to medical malpractice

lawsuits in Allegheny County for the years 1995 2004 and Philadelphia County for the

years 1980-20043. By following the individual doctors' behavior over multiple years

(both before and after a lawsuit for those sued), I am able to identify the impact of

malpractice lawsuits on a physician's choice of obstetrical procedures. Using a

difference in difference approach to deal with the problem of doctor heterogeneity, the

results show that after being sued, a doctor increases the use of caesarean sections (c-

section) by approximately 5.5% relative to doctors who were not sued in Allegheny

County. The estimated effect is approximately 1.5% in Philadelphia County. It is

possible that lawsuits may lead to an increase in the number of labor inductions;

however, these estimates lack statistical precision.

Previous Literature

Malpractice insurance premiums are not experience rated- contrary to economic

theory. That is, premiums are not based on past claims history. Rather, they are set at the

community level (often at the state level) and adjusted for medical specialty and limits of



2 These Counties were chosen because their court systems identify lawsuits as medical malpractice cases.
Due to the cost to obtain the data, the data is actually the 1st quarter of the years 1994 2004.

3 The earliest court records that are online for Allegheny County begin in 1995, while the earliest court
records that are online for Philadelphia County begin in 1980.









coverage (Danzon 2000)4. One implication of this aspect of malpractice is that doctors

are somewhat isolated from the costs of their own behavior. In fact, it could be argued

that the major costs, to a doctor, of a malpractice suit are time and reputation, neither of

which are insurable. The lawsuit does not affect malpractice insurance premiums, except

in extreme cases. Indeed, the individual doctor has very little to no control over his own

malpractice insurance rates. From a purely economic point of view, a doctor provides

prudent medical care to minimize these time and reputation costs, not to control his

insurance rates.

This leads to the hypothesis that a doctor will seek to minimize his exposure to

the court system. If a doctor perfectly forecasts the true costs of being sued, in terms of

time and reputation, then a lawsuit will have no effect on his procedure use. However,

without perfect foresight, it is likely that at least some doctors will underestimate these

costs. It is these doctors who I expect to change their behavior after being sued. A

secondary hypothesis is that the doctors with the least exposure (either themselves or

their colleagues) to the courts will change their behavior the most after being sued

themselves.

There has a long line of research that discusses the issue of "bounded rationality"

and decision making. It has been shown that people in general do not forecast risks, or

deal with probabilities, very well. Tversky and Kahneman (1974) document many biases

in judgment that people make. They argue that people tend to use heuristics, or rules of

thumb, to make decisions under uncertainty. This article led to a long line of research

that attempted to formalize economics models of decision making that take into account


4 Danzon (2000) presents a thorough review of medical malpractice issues.









these biases. Kahneman and Tversky (1973) show that people do not behave as a

statistician would when presented with probabilities. Instead "they rely on a limited

number of heuristics which sometimes yield reasonable judgments and sometimes lead to

severe and systematic errors." In this same vein, Camerer and Lowenstein (2004) argue

that Bayes' Rule is unlikely to be used correctly because "it has several features that are

cognitively unrealistic".

Lowenstein and Mather (1990) attempt to assess how accurate risk perceptions

are over time as the underlying source of different risks change. They find that in many

cases, the public perception of risks changes dramatically although the underlying factor

has changed very little. Lowenstein, O'Donoghue and Rabin (2003) develop a model of

what they term "projection bias". They argue that people project their current

preferences onto their future selves. Lowenstein (2005) discusses projection bias in the

realm of medical decision making. He argues that many medical decisions involve fear,

pain and discomfort and therefore affect people's decision making. These decisions can

be exacerbated by the fact that the doctor is does not adequately perceive the patient's

mental state. This research leads one to the conclusion that people do not behave in a

utility-maximizing manner. It is therefore likely that physicians do not base their medical

decisions on accurate estimates of future litigation costs.

Within the existing research on defensive medicine, Kessler and McClellan (1996)

is one of the more prominent papers. Using data from all Medicare patients treated for

serious heart disease in addition to information about changes in state malpractice laws

(for example, caps on punitive damages or changes to the statute of limitations), they


SSee Kahneman (2003) for a review.









show that states which enact malpractice reforms have lower health care spending

without a change in mortality or medical complications. They conclude that this is

evidence that defensive medicine exists. Kessler and McClellan (1997) use a survey from

the American Medical Association to investigate how malpractice reforms influence

physician perceptions and change self-reported behavior. While the authors do not

observed health outcomes or medical costs (and so cannot precisely determine the

existence of defensive medicine), they find the legal environment affects both the

likelihood of being sued as well as physician behavior.

One paper that deals specifically with defensive medicine in obstetrics is Dubay, et

al. (1999) which uses birth certificate data for the years 1990-1992. The authors use the

fact that medical malpractice insurance premiums are not experience rated. Then based

on the assumption that premiums are an accurate measure of the likelihood of malpractice

lawsuits in a geographic area, the premiums are used as a proxy for the likelihood of a

doctor being sued. With this, the authors are able to investigate the relationship between

the legal environment (and, therefore, the fear of lawsuits) and the county-level rate of

caesarean sections. They show that there is an increase in the number of c-sections due

to malpractice fears without a concurrent increase in the health of the babies. They argue

that this is evidence of defensive medicine, but the overall effect is small. A second

obstetrics related paper is Dubay, et. al (2001) which again uses the same birth certificate

data and malpractice insurance premiums to show that a reasonable decrease in premiums

would lead to an increase in prenatal care. While they argue that malpractice pressure

reduces the supply of prenatal care, this is not shown to affect the health of newborns.









Baicker and Chandra (2004) look at the potential costs of malpractice on patient

care. They find no evidence of a change in treatment patterns in response to increases in

malpractice premiums. They also find no evidence that malpractice costs affect the

overall number of doctors.

The influential Harvard Medical Practice Study (Brennan, et al. 1991) reviewed

7,743 medical charts for evidence of negligence. They found 1,278 adverse events of

which 306 were attributed to negligence. The authors found 47 malpractice claims from

these cases but, only 8 of these claims were found to have evidence of malpractice. More

importantly, 40 percent of cases without negligence resulted in a payment. While the

authors draw the conclusion that there are not enough malpractice lawsuits (i.e., only

15% of negligent doctors are sued), the evidence also points to the randomness of

malpractice lawsuits (i.e., only 17% of malpractice claims involve actual cases of

negligence). Given this apparent lack of correlation between actual malpractice by the

physician and the likelihood of a lawsuit, this implies that malpractice suits can be treated

as a random event.

This paper uses hospital discharge data for all births in Allegheny and Philadelphia

Counties, Pennsylvania for the years 1994-2004. While earlier researchers have shown

that doctors appear to respond to general malpractice pressure (i.e., the legal

environment), none have been able to investigate individual doctor's responses to being

sued for medical malpractice. With this dataset, I am able to isolate a doctor's response

to new litigation from the preexisting legal environment.

A naive approach to the problem of doctors' responses to lawsuits would use a

cross-section of doctors and their responses to malpractice cases. However, the









probability of being sued is almost certainly correlated with the difficulty of the medical

cases, and hence with the procedures used. While diagnosis information is available, it is

difficult to summarize and therefore difficult to control for in a regression model.

Instead, to circumvent this problem of unobserved heterogeneity, a difference in

difference approach is used. That is, the effect of being sued is determined by each

doctor's change in behavior.

If malpractice lawsuits are a random event, controlling for patient characteristics,

then it is straight forward to determine the effect of lawsuits on physician behavior. The

average number of c-sections or other potentially defensive procedures performed can be

compared before and after a doctor is sued. Any changes in behavior can be attributed to

the lawsuit. One potential problem with this approach is the large changes in procedure

use over the period. Given these time trends, it is possible that the effect of lawsuits

would be overstated. The difference in difference approach, by using the "un-sued"

doctors as a control group, removes this issue by assuming that both groups of doctors

have the same underlying time trend.

Description of the Data

The hospital data includes detailed information for approximately 100,000 mothers

in Allegheny and Philadelphia Counties, Pennsylvania for the years 1994-2004. Each

mother has up to three doctors (referring, attending, and operating physicians) listed on

her record. The doctors are identified by their Pennsylvania medical license number.

While three doctors are possible, frequently one doctor is listed in multiple roles. While

most records have both a referring and attending doctor listed, many do not have an









operating doctor listed.6 For the majority of the data, only one license number is listed

for a given mother. This paper reports only the attending physician results, as the results

are qualitatively similar for the other doctors. Each of these license numbers was

matched to the doctor's name and the date the license was first issued using the State of

Pennsylvania's license verification website7. The doctor names then were matched to

court data on medical malpractice cases using the respective county's Prothonotary's

website8.

Annual summary statistics for doctors and patients are provided in Table 4-1 and

Table 4-2. While the number of births declines in both counties over the data period, the

number of doctors remains relatively constant in Allegheny and decreases in

Philadelphia.

The number of c-sections increases over the years in the sample, from an average

of 19% to an average of 27% (Figure 4-1). We also see a large increase in the number of

labor inductions in Allegheny County and a slight rise in Philadelphia County (Figure 4-

2). These changes in procedure use are a factor in the rising concern of defensive

medicine. However, it is possible that this shift reflects cross-sectional changes in the

composition of mothers. Mothers could be older and exhibit different c-section

preferences in later periods. In addition, the use of other obstetric procedures has

changed during this sample time.




6 The only time there is an operating doctor listed is if there was a surgical procedure performed.

Shttp://licensepa.state.pa.us/

8 A Prothonotary is the chief legal clerk in a county in Pennsylvania. It is comparable to a clerk of the court
in other states.









The percentage of vaginal births after caesarean sections (VBAC) has decreased

(Figure 4-3). Given that a vaginal birth is higher risk after a previous c-section, it is

reasonable to believe that some of this reduction may due to malpractice fears, i.e.,

defensive medicine. Grady (2004) presents anecdotal evidence that some hospitals in the

U.S. no longer allow VBACs to be performed. Likewise, there is a simultaneous

decrease in the percentage of pregnancies that are "prolonged" i.e., those that have

gestational periods longer than 42 weeks. As prolonged pregnancies increase the risk to

the mother and the baby, it is not surprising that doctors have reduced their occurrence.

There are no prolonged pregnancies after 2002 compared to approximately 4% of the

births in 1994 (Figure 4-4). Part of this decrease is almost certainly due to the increase in

labor inductions during this period.

Over the period of the data, the occurrence of neonatal distress decreases

markedly (Figure 4-5), while there is not a large change in the incidence of neonatal

deaths (Figure 4-6). There is no significant change in the incidence of premature births in

either county (Figure 4-7).

Two possible explanations for the increase in caesarean sections are the increase in

the number of breech births (Figure 4-8) and the increase in mothers over the age of 35

(Figure 4-9), although these increases were larger in Allegheny County than in

Philadelphia County. While there is some evidence for a relationship between older

mothers and breech births, the possible effect is small (Rayl, Gibson, and Hickok 1996).

It is also possible that mother's preferences are likely to have a large effect on caesarean

sections. Anecdotally, older mothers are thought to be more likely to request a c-section,









however this is difficult to determine from the data as the positive association between

age and c-sections my also be due to increased pregnancy risk.

While the two counties look very similar over the time period, there is one very

large difference. Both counties started the period with approximately 30% of the mothers

on Medicaid. This percentage stayed the same in Allegheny County while it doubled in

Philadelphia County by the end of the period (Figure 4-10).

Physicians and Medical Malpractice Lawsuits

While the total number of doctors is similar in each county in 1994, by the end of

the period, the relative number in Philadelphia decreased (Table 4-3). The number of

lawsuits per doctor differs dramatically between the two counties. The two counties

diverge in another way as well; the total number of licenses issued in Allegheny County

is almost four times the number issued in Philadelphia County.

The court data for Allegheny County begins in 1995. This unfortunately limits our

sample, as this paper focuses on the effect of a physician being sued the first time. To

capture this initial reaction, the first medical malpractice lawsuit must be identified for

each doctor. Given the limitations of the court data, it is impossible to accurately

determine the total number of malpractice lawsuits experienced by a physician who

received his license before 1995. In this paper, "young doctors" refers to those licensed

in 19949 to the present, while "old doctors" refers to those that received their license

before 1994. In a later part of the paper, using the data from Philadelphia, I divide the

"old doctors" into old and senior doctors. In this case, the old doctors are those who




9 1994 is used as the cutoff because I have 1st quarter data. A doctor who finishes school in the spring will
not appear in the hospital data until 1995.









received their license between 1980 and 1993, while the senior doctors are those who

received their license before 1980.10

Philadelphia County court data are available beginning in 1980. Although this

provides much greater detail for a given doctor, the hospital data begin in 1994. Now

while we can not evaluate physician's responses to lawsuits over this whole panel, we

can exploit the variation in a physician's lawsuit history. For the Philadelphia data, the

physicians are segmented into age and lawsuits categories, i.e., an "old doctor" is one

who received his license between 1980 and 1993. The old doctors then are grouped

according to their past malpractice lawsuit history, i.e., those who have been sued and

those who never have experienced a lawsuit.

Obstetricians/gynecologists are among the highest risk of malpractice lawsuits of

all medical specialties. However, even with this heightened probability of legal action,

the probability of a lawsuit varies greatly within the same state. Comparing Figure 4-11

and Figure 4-12, substantial differences are clear between Allegheny and Philadelphia

Counties. Figure 4-11 shows the cumulative probability of a lawsuit for a given number

of years of practice in Allegheny County. After practicing for five years, approximately

15% of doctors have been sued at least once". In Philadelphia, the likelihood of a

lawsuit is more than 20% for doctors with the same experience (Figure 4-12). By the 10th








10 This division is based on the court data for Philadelphia County.

11 Because the majority of doctors received their licenses at the end of the period (when it is impossible for
a doctor to have practiced for more than 5 or 6 years, the cumulative distribution is calculated with just the
doctors from the 1st half of the period.









year of practice, only 20% of Allegheny doctors have been sued, while close 40% of the

Philadelphia doctors have been sued at least once12

Responses to Lawsuits

It is reasonable to think that malpractice lawsuits are driven by doctor and patient

characteristics. Known doctor characteristics are limited to experience and procedure

use. I assume that malpractice lawsuits are exogenous, as earlier research has shown that

lawsuits, while not entirely random, are not based on the medical record in most cases

(see Harvard study). Given this exogeneity of lawsuits, the difference in difference

estimate is an unbiased estimate of the effect of being sued on doctors' behavior.

Obstetrical Procedures

Of the obstetric procedures investigated as potentially defensive in nature, c-

sections would seem to be the likeliest candidate. A c-section transforms a potentially

litigious situation (a vaginal birth) into a controlled medical procedure. There is the

added benefit that c-sections can be scheduled in advance, which is probably

psychologically reassuring to both the doctor as well as the mother.

There is evidence that inducing labor on a mother who has previously had a c-

section is dangerous, because of this, it is important to control for previous c-sections.

However, if a doctor is fearful of a malpractice lawsuit, then he is likely to perform a c-

section on a high-risk mother. The effect of malpractice lawsuits on inductions is an

empirical question. While there is anecdotal evidence that fear of lawsuits has reduced






12 These results are likely understated as it is reasonable to believe that some doctors would have stopped
practicing before they appeared in the hospital data.









the use of VBACs, there are not enough women in the data that have had a previous c-

section to effectively estimate an effect of a lawsuit on physician behavior.

To test the hypothesis that doctors respond to their first lawsuit, model 1 was

estimated separately for each county.

procedures = a( suedli + a2 sued2i + a3 sued3i + (04 sued4i +
(3-1)
1i breech + 32 previous + 13 oldi + 34 medicaid + Ei

where procedure is an indicator of either a caesarean section or an induced labor,

suedl sued4 are indicators of the number of lawsuits a patient's doctor has

experienced.13 These indicators can change for each doctor from year to year depending

on his lawsuit history. For example, a doctor who is never sued for malpractice will have

all of the sued indicators equal to zero. A doctor who is sued for the first time in 1999

will have the suedl equal one for the years after 1999. If he is sued a second time, the

sued2 indicator will switch to a one, while the suedl indicator will not change. Sued3

and sued4 are indicators of three and four lawsuits that are coded similarly. These

indicators are difference in difference estimates of the effect of a given lawsuit on a

doctor's behavior. The variables breech, previous, old, and Medicaid are indicators of a

breech birth, a previous c-section, a mother over 35, and a mother on Medicaid. Because

there are not any doctors that appear in both Philadelphia and Allegheny Counties, and

the model includes doctor and year level fixed effects14, estimating this with both






13 The sued indicators switch from 0 to 1 when a doctor gets sued the respective time. The indicators never
switch off after being turned on.
14 The model was also estimated with hospital fixed effects. There is no qualitative difference in the
estimates.









counties jointly is not possible, unless the assumption is made that doctors in both

counties react in the same manner to a lawsuit.

In order for the parameter of interest (al) to be identified, the time trend must be the

same for the untreated group (i.e., doctors who are not sued)15. Because it appears that

lawsuits are random, it is assumed that this condition is met. After controlling for doctor,

year, and hospital effects, aL is the estimate of the treatment effect of being sued. If

doctors react to lawsuits by increasing the number of c-sections, then aL should be

positive. The same will be true for a2, a3 and a4 if the response is not negligible for

additional lawsuits. Because the variation in lawsuits occurs at the doctor level, the

standard errors were clustered at the doctor level. The variables of interest in these

regressions are the sued variables. If the hypothesis that at doctor only responds to a first

lawsuit is true, then the coefficients on sued2 sued4 should not be significantly different

from zero.

With a binary dependent variable, a logit or probit model might be assumed. In

actuality, the combination of doctor level fixed effects as well as a doctor level treatment

effect implies that the dependent variable is the average of the doctor's procedure use in a

given year. The same model could have been estimated by collapsing the data to doctor

and year means and then applying ordinary least squares. I chose to estimate these

models at the individual patient level to ensure the most precise estimates possible. The

estimates are not qualitatively different when averages are used.






15 Because of the strong time trend and the fact that there are not a large number of lawsuits in a given year,
this is difficult to show.









Caesarean Sections in Philadelphia

The results for all of the doctors in Philadelphia are reported in the first column of

Table 4-4. We see that as expected, breech births and previous c-sections are highly

correlated with a c-section, as is the indicator for older mothers. A breech birth increases

the probability of a c-section by 52%, while a previous c-section increases the probability

of a c-section by 44%. We also see that Medicaid mothers are less likely to receive a c-

section, although the coefficient is not very large. This may be because they are less

demanding of their doctors or the doctors are less responsive to the demands of less

wealthy patients. It is also possible that Medicaid reimbursement rates affect the doctor's

choice of a c-section. We also see that, as predicted, doctors do respond to a malpractice

lawsuit. The estimated impact of a first lawsuit on physician behavior is an increase in the

c-section rate of 1.4%. Additional lawsuits have no effect. Given these results are for all

of the Philadelphia doctors, young and old, it is possible that this underestimates the

effect. Some of these doctors have been sued before the hospital data begins and may

have adjusted their behavior already.

To address this issue, the model then was estimated with subsets of doctors. As

mentioned before, the doctors were divided into "young", "old", and "senior" categories

in Philadelphia County. These categories correspond to the hospital and court data.

Young doctors received their license after 1994 and hence, hospital and court data is

available for their whole career. Old doctors received their license between 1980 and

1993; court data is available for their whole career but not the hospital data. Lastly,

senior doctors received their licenses before 1980 and thus have missing court

information from the beginning of their career.









The expectation is that lawsuits will have the greatest effect on the young doctors

with a possible effect on the old doctors who were not sued early in their career. The

predicted effect of a lawsuit on the senior doctors is nil, it is assumed that they either

have been sued previously or are experienced enough not to react in a significant manner.

The results from these additional regressions are reported in columns 2 4 of Table

4-4. Somewhat surprisingly, there is no statistically significant effect of a first lawsuit on

young doctors while old doctors do respond to a lawsuit. As expected the senior doctors

do not respond. Interestingly, when estimated separately, both the young and old doctors

respond to a second lawsuit. The young doctors increase the rate of c-sections while the

old doctors actually reduce their rate. It is possible that the old doctors change their

patient mix in order to avoid high risk cases; however it is not possible to test this

hypothesis.

It is also possible to segment the old and senior doctors into groups based on their

earlier lawsuit history. Table 4-5 reports the results of regressions with the old doctors

segmented into groups based on whether they were sued between 1980 and 1993.

Column 1 reports the results for old doctors who were not sued before 1993. The effect

of a first lawsuit on these doctors is positive but not statistically significant. Again, we

find the old doctors respond to the second lawsuits by reducing the number of c-sections.

They do not respond to additional lawsuits. By comparison, column 2 reports the results

for the old doctors who were sued before 1993. It is these doctors who we do not expect

to change their behavior in response to a lawsuit after 1994 and yet, we find that they

increase their c-section rate by 3.5%. These doctors do not change their behavior in

response to additional lawsuits. Columns 3 and 4 perform the same regressions for the









senior doctors (i.e., those who received their licenses before 1980). As expected, in both

cases, the senior doctors do not respond to a lawsuit.

Caesarean Sections in Allegheny

When the same model is estimated with the data from Allegheny County, we see

that with all of the doctors, the coefficients on the control variables are similar. Although

it does appear that doctors in Allegheny County have a greater propensity to perform a c-

section relative to Philadelphia County, the coefficients on breech and previous are larger

in Allegheny County. That is, c-sections are more likely simply due to patient attributes.

Interestingly, it appears that Medicaid mothers are less likely to receive a c-section in

Allegheny versus Philadelphia County. The variables of interest are the "sued" variables.

With Allegheny County there is no effect of a lawsuit on behavior (column 1 of Table 4-

6). 16 Again, these results are probably underestimated because some of these doctors

have almost certainly been sued before.

This model again is estimated with the doctors segmented into different groups.

For comparability purposes, the same divisions (young and old) are made with the

Allegheny doctors even though the court data do not begin until 1995. Column 4 groups

the old and senior doctors together since, they are indistinguishable in Allegheny County

for all practical purposes. The results of these regressions are reported in columns 2-4 of

Table 4-6. Young doctors respond in a dramatic fashion to a first lawsuit. The estimated

effect is a 5.6% increase and is highly statistically significant. The young doctors do not






16 While it is possible that doctors are responding to a third lawsuit and no others, this is probably an
artifact of the data.









respond to a second lawsuit. The old doctors do not respond to a first or second lawsuit

but again, appear to respond to third lawsuit.

Labor Inductions in Philadelphia

Next, I evaluate the use of labor inductions. As mentioned previously, the

predicted effect of a lawsuit on a physician's behavior is uncertain. A doctor may want

to reduce the number of vaginal births he performs and so will be less willing to induce

labor. However, it is also possible that the potential risk from a prolonged pregnancy will

lead the doctor to induce labor instead of waiting for spontaneous labor.

We see in the first column of Table 4-7 that doctors in Philadelphia do not appear

to change the number of inductions performed in response to a lawsuit. As before, it is

possible that this understates the effect because a number of these doctors have probably

been sued before. When, the doctors are divided into groups based on when their license

was issued, it does not appear that any of the groups of doctors respond to a lawsuit by

changing their behavior. Columns 2, 3 and 4 of Table 4-7 show the results of these

separate regressions.

Segmenting doctors based on their lawsuit history shows differences among old

doctors. Columns 1 and 3 of Table 4-8 report the results for doctors who were not sued

before 1993. The senior doctors respond to a first lawsuit by reducing the number of

inductions by almost 8%, while the old doctors do not change their behavior. These

results may be driven by the small sample size. Columns 2 and 4 of Table 4-8 report the

results for the doctors who had previously been sued. As expected, these doctors do not

change the rate of inductions. Again, it is interesting to note that the doctors who had

previously been sued appear to be less likely to induce labor for breech births and









previous c-sections. This may imply that they have already made adjustments to their

practices.

Labor Inductions in Allegheny

It appears that doctors in Allegheny County change their induction rate in response

to malpractice lawsuits (Table 4-9). A first lawsuit leads to a reduction of inductions by

1.5%, while additional lawsuits do not appear to matter. Columns 2, 3 and 4 of Table 4-9

report the results for the separate age groups. While the coefficients of the first lawsuit

are negative in all three cases, only the one for the old doctors is statistically significant.

Again, additional lawsuits do not appear to matter.

Conclusions

While it is impossible to say that these results are proof of the existence of

defensive medicine in obstetrics, evidence exists that doctors change their procedure use

after being sued for malpractice. There is also evidence of differences in procedure use

across geographical areas. It appears that the "age" of a doctor influences how he

responds to a lawsuit.

In Allegheny County, when all doctors are included in the sample, there is no effect

of lawsuits on procedure use. This lack of an effect is not surprising given that many of

the doctors have been practicing medicine since the 1980s. It is likely that many of them

have been sued previously and have already changed their medical practices. However,

when the sample is limited to young doctors (who we can be relatively certain have not

been sued before entering the data), it is estimated that a doctor will increase the number

of c-sections performed by approximately 5% in response to a lawsuits. This is a large

change given that the average c-section rate is 20% in 1994 and 28% in 2004. It is

possible that the use of inductions is affected by lawsuits but, these estimated effects









suffer from a lack of precision. Doctors in Philadelphia County, who have more

Medicaid patients and much higher risk of litigation appear much less sensitive to

lawsuits than doctors in Allegheny County. They also appear to be less likely to use c-

sections in potentially risky situations.

In order to draw conclusions about the presence of defensive medicine in response

to malpractice lawsuits, measurable outcomes are needed. Two obvious ones are

maternal and fetal deaths. Thankfully (from society's point of view), these both occur

infrequently; however the rarity of these outcomes makes them less than ideal for my

purposes. One thing is certain; physicians do appear to respond to malpractice lawsuits

by changing their practice patterns.










Table 4-1. Number of doctors by type and county
gg Philadelphia Pittsburg
Year Total Drs Young Drs Total Drs Young Drs


1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004


197
188
188
207
236
158
180
169
155
126
130


1
7
24
31
40
38
56
57
74
59
70


Table 4-2. Number


Year
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004


Old Drs
5,689
5,213
4,383
3,914
3,586
2,548
3,194
2,973
2,363
2,044
1,871


of patients by
Philadelphia
Young Drs
18
188
511
720
957
995
1,471
1,398
1,940
2,203
2,608


doctor type and county
Pittsburg
Total Old Drs Young Drs
5,707 4,761 241
5,401 4,758 212
4,894 4,228 437
4,634 3,983 473
4,543 3,817 470
3,543 3,602 747
4,665 3,417 1,037
4,371 3,065 1,227
4,303 2,986 1,154
4,247 2,959 1,313
4,479 2,827 1,485


Total
5,002
4,970
4,665
4,456
4,287
4,349
4,454
4,292
4,140
4,272
4,312











Table 4-3. Number of licenses issued and number of doctors
license was issued.


sued by county and year the


Year
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
Total


# issued
63
82
52
67
89
87
89
92
64
75
86
77
87
120
78
103
81
92
73
74
60
53
41
28
1813


Pittsburg
# Sued
29
33
31
26
37
40
50
38
21
27
27
22
31
36
18
30
18
21
13
15
3
10
2
2
132


Percent Sued
0.460
0.402
0.596
0.388
0.416
0.460
0.562
0.413
0.328
0.360
0.314
0.286
0.356
0.300
0.231
0.291
0.222
0.228
0.178
0.203
0.050
0.189
0.049
0.071
0.073


# issued
20
19
22
18
9
23
15
27
18
12
18
29
33
25
29
34
29
21
23
15
18
18
9
7
491


Philadelphia
# Sued PE
10
12
17
10
2
15
4
12
7
8
11
10
18
10
11
15
12
4
7
4
2
6
2
0
209


percent Sued
0.500
0.632
0.773
0.556
0.222
0.652
0.267
0.444
0.389
0.667
0.611
0.345
0.545
0.400
0.379
0.441
0.414
0.190
0.304
0.267
0.111
0.333
0.222
0.000
0.426










Table 4-4. Philadelphia results with c-sections as the dependent variable (standard errors


in parentheses)
C-section all doctors


sued

sued2

sued3

sued4

medicaid

old

previous

breech

Constant

Observations
# of Doctors
R-squared
F stat


0.0140
(0.0070)
-0.0072
(0.0078)
0.0045
(0.0091)
0.0047
(0.0099)
-0.0253
(0.0039)
0.0537
(0.0048)
0.4380
(0.0050)
0.5249
(0.0064)
0.1087
(0.0059)
50787
577
0.23
840.1


young doctors


0.0130
(0.0163)
0.0379
(0.0177)
-0.0293
(0.0220)
0.0079
(0.0288)
-0.0200
(0.0075)
0.0579
(0.0098)
0.4830
(0.0103)
0.4810
(0.0128)
0.1405
(0.1261)
13009
174
0.23
209.3


old doctors
0.0232
(0.0103)
-0.0203
(0.0109)
0.0110
(0.0118)
0.0068
(0.0121)
-0.0298
(0.0057)
0.0587
(0.0069)
0.4252
(0.0072)
0.5293
(0.0090)
0.0980
(0.0076)
24804
257
0.24
419.6


senior doctors
0.0080
(0.0142)
0.0066
(0.0174)
0.0083
(0.0217)
-0.0216
(0.0238)
-0.0292
(0.0080)
0.0411
(0.0097)
0.4280
(0.0101)
0.5612
(0.0134)
0.1227
(0.0090)
11926
141
0.24
202.1










Table 4-5. Philadelphia segmented results with c-sections as the dependent variable
(standard errors in parentheses)
old doctors old doctors senior doctors senior doctors
C-section not sued sued not sued sued
before 1994 before 1994 before 1994 before 1994
sued 0.0177 0.0358 -0.0064 0.0231
(0.0146) (0.0165) (0.0291) (0.0193)
sued2 -0.0342 -0.0236 -0.4248 0.0085
(0.0174) (0.0152) (0.2030) (0.0191)
sued3 0.0057 0.0095 0.0000 0.0069
(0.0177) (0.0164) 0.0000 (0.0227)
sued4 -0.0017 0.0158 0.0000 -0.0043
(0.0194) (0.0163) 0.0000 (0.0252)
medicaid -0.0313 -0.0255 -0.0185 -0.0389
(0.0070) (0.0100) (0.0120) (0.0109)
old 0.0660 0.0491 0.0535 0.0305
(0.0093) (0.0103) (0.0142) (0.0134)
previous 0.4079 0.4522 0.3954 0.4540
(0.0091) (0.0115) (0.0152) (0.0136)
breech 0.5131 0.5536 0.5690 0.5567
(0.0115) (0.0144) (0.0205) (0.0178)
Constant 0.0986 0.0997 0.1103 0.1329
(0.0099) (0.0120) (0.0138) (0.0120)
Observations 15349 9455 5385 6541
# of doctors 178 79 61 80
R-squared 0.22 0.26 0.22 0.25
Fstat 239.4 182 96.26 118.8










Table 4-6. Allegheny results with c-sections as the dependent variable (standard errors in


parentheses)
C-section all doctors


sued


sued2

sued3

sued4

medicaid

old

previous

breech

Constant

Observations
# of doctors
R-squared
F stat


0.0068
(0.0062)
-0.0133
(0.0081)
0.0322
(0.0116)
-0.0048
(0.0214)
-0.0321
(0.0044)
0.0317
(0.0041)
0.4772
(0.0048)
0.6380
(0.0057)
0.1115
(0.0056)
48696
425
0.33
1302


young
doctors
0.0568
(0.0220)
-0.0169
(0.0353)
N/A
N/A
N/A
N/A
-0.0349
(0.0100)
0.0320
(0.0107)
0.4708
(0.0128)
0.5968
(0.0134)
0.0310
(0.0779)
8293
141
0.3
218.9


old doctors


all old
doctors
0.0014
(0.0065)
-0.0097
(0.0084)
0.0332
(0.0116)
-0.0040
(0.0212)
-0.0316
(0.0049)
0.0319
(0.0045)
0.4778
(0.0052)
0.6476
(0.0063)
0.1102
(0.0055)
40403
284
0.33
1112


-0.0034
(0.0083)
-0.0046
(0.0105)
0.0229
(0.0142)
-0.0150
(0.0249)
-0.0388
(0.0062)
0.0340
(0.0057)
0.4592
(0.0066)
0.6558
(0.0078)
0.1157
(0.0073)
25241
181
0.33
695.3










Table 4-7. Philadelphia results with inductions as the dependent variable (standard errors
in parentheses)
senior
Induce all doctors young doctors old doctors o
doctors
sued 0.0033 -0.0126 0.0122 -0.0300


sued2

sued3

sued4

medicaid

old

previous

breech

Constant

Observations
# of Doctors
R-squared
F stat


(0.0063)
0.0088
(0.0069)
0.0143
(0.0081)
0.0097
(0.0088)
-0.0049
(0.0034)
0.0152
(0.0043)
-0.0538
(0.0045)
-0.0444
(0.0057)
0.1037
(0.0052)
50787
577
0.01
21.5


(0.0149)
0.0376
(0.0163)
-0.0357
(0.0201)
0.0299
(0.0264)
-0.0001
(0.0068)
0.0134
(0.0090)
-0.0713
(0.0094)
-0.0533
(0.0117)
0.0806
(0.1154)
13009
174
0.01
7.575


(0.0090)
0.0110
(0.0095)
0.0344
(0.0102)
0.0105
(0.0105)
-0.0078
(0.0050)
0.0151
(0.0060)
-0.0484
(0.0062)
-0.0447
(0.0078)
0.1072
(0.0066)
24804
257
0.01
12.69


(0.0124)
0.0113
(0.0153)
-0.0050
(0.0190)
0.0036
(0.0208)
-0.0085
(0.0070)
0.0196
(0.0085)
-0.0485
(0.0089)
-0.0294
(0.0118)
0.0870
(0.0079)
11926
141
0.01
5.037










Table 4-8. Philadelphia segmented results with inductions as the dependent
variable(standard errors in parentheses)


Induce


sued

sued2

sued3

sued4

medicaid

old

previous

breech

Constant

Observations
# of doctors
R-squared
F stat


old doctors
not sued
before 1994
0.0110
(0.0125)
-0.0171
(0.0149)
0.0471
(0.0152)
0.0154
(0.0166)
-0.0128
(0.0060)
0.0197
(0.0080)
-0.0388
(0.0078)
-0.0363
(0.0099)
0.1207
(0.0085)
15349
178
0.01
6.615


old doctors
sued
before 1994
0.0046
(0.0147)
0.0203
(0.0136)
0.0169
(0.0146)
-0.0012
(0.0145)
0.0031
(0.0089)
0.0105
(0.0092)
-0.0630
(0.0102)
-0.0584
(0.0129)
0.0928
(0.0107)
9455
79
0.01
7.455


senior doctors
not sued
before 1994
-0.0783
(0.0285)
-0.0902
(0.1993)
0.0000
0.0000
0.0000
0.0000
0.0051
(0.0118)
0.0138
(0.0139)
-0.0447
(0.0149)
-0.0140
(0.0201)
0.1023
(0.0135)
5385
61
0.01
2.594


senior doctors
sued
before 1994
-0.0008
(0.0151)
0.0134
(0.0149)
-0.0036
(0.0177)
-0.0090
(0.0197)
-0.0157
(0.0085)
0.0244
(0.0104)
-0.0514
(0.0107)
-0.0397
(0.0139)
0.0735
(0.0094)
6541
80
0.01
4.15










Table 4-9. Allegheny results with inductions as the dependent variable (standard errors in


parentheses)
Induction all doctors


sued


sued2

sued3

sued4

medicaid

old

previous

breech

Constant

Observations
# of doctors
R-squared
F stat


-0.0154
(0.0064)
0.0096
(0.0084)
0.0058
(0.0120)
-0.0111
(0.0220)
-0.0199
(0.0045)
0.0128
(0.0042)
-0.0733
(0.0049)
-0.0639
(0.0059)
0.1226
(0.0058)
48696
425
0.01
35.86


young
doctors
-0.0243
(0.0228)
0.0252
(0.0365)
0.0000
0.0000
0.0000
0.0000
-0.0331
(0.0103)
0.0173
(0.0111)
-0.1094
(0.0132)
-0.0638
(0.0139)
0.2561
(0.0805)
8293
141
0.02
7.82


old doctors


all old
doctors
-0.0132
(0.0066)
0.0085
(0.0086)
0.0059
(0.0119)
-0.0118
(0.0218)
-0.0165
(0.0050)
0.0120
(0.0046)
-0.0667
(0.0053)
-0.0644
(0.0065)
0.1223
(0.0057)
40403
284
0.01
30.55


-0.0186
(0.0085)
0.0081
(0.0108)
0.0014
(0.0146)
-0.0075
(0.0258)
-0.0155
(0.0064)
0.0126
(0.0058)
-0.0696
(0.0069)
-0.0622
(0.0081)
0.1146
(0.0075)
25241
181
0.01
18.29