The data on new physician practice location choice come from the Center for Health Workforce Studies of the State University of New York at Albany, which has conducted the Annual Survey of Residents Completing Training each May and June for New York since 1998. Physicians are surveyed once and there is no subsequent follow-up. We obtained data on graduating residents from 1998 to 2003 for New York State. The dataset provides information on demographics, educational debt, education, residency training, specialty, practice setting and location (zip code), future job plans, and other variables on a new cohort of graduating residents each year of the survey (Center for Health Workforce Studies 2004a
;). Because we are interested in location choices for physicians beginning their professional career in patient care we did not include observations for graduating residents pursuing fellowship training or non–patient-care-oriented positions, which represented 55.1 percent of the total sample of 17,890 physicians. Physicians who at the time of the survey had not yet accepted a job, failed to report the practice location, or entered an invalid zip code or misspelled city names also were excluded from the sample (approximately 27 percent of the 9,854 entering patient care fields). After these exclusions the remaining sample consisted of 7,212 physicians.
Resident survey data are supplemented with malpractice premium data from the Medical Liability Monitor (MLM). Since 1991, the MLM has conducted an annual national rate survey of physician malpractice insurance premiums of major professional liability insurance companies. The MLM data report premiums for three specialties: internal medicine, general surgery, and obstetrics–gynecology. Annual premium information is presented as the average charged by each company for an entire state, regions within a state, counties, or metropolitan areas, though in most cases (34 states in 2003) the premium information is available only as a statewide average. Using the most geographically appropriate level of detail available in the MLM survey we calculate MSA and non-MSA average premiums for the three specialties contained in the data. For cases in which an area has multiple insurance companies we calculate the simple mean of the premiums. This is a limitation because a more appropriate measure would be the weighted average of premiums based on enrollment, but information on enrollment is unavailable.
displays population weighted averages of malpractice insurance premiums over time. Premiums grew quite slowly between 1998 and 2000, but they increased rapidly after 2000, with OB/GYN premiums increasing roughly 50 percent by 2003 and surgeon and internal medicine premiums increasing over 70 percent by 2003. In 2003 OB/GYN premiums were still nearly 1.5 times greater than premiums faced by surgeons and nearly five times greater than premiums faced by internal medicine.
Weighted Mean of Annual Malpractice Premiums by Specialty and Year
In order to most appropriately match physicians to the malpractice insurance premiums they would likely face in different locations, we limit the sample of physicians to three groups: OB/GYNs, surgeons, and physicians who are plausibly likely to face premiums best approximated by those for internal medicine. We term this last group “primary care physicians,” which includes general internal medicine physicians, general pediatricians, and family physicians (Cooper 1994
). Of the full 7,212 sample of physicians, 7 percent of the sample members are OB/GYNs, 10 percent are surgeons, and 35 percent are PCPs. Thus, our final sample consists of 3,758 physicians.
presents information on the sample by year and specialty. Note that for all three specialty groups in New York the absolute number of graduating residents entering patient care is falling over time, despite the fact that the overall survey response rate stayed roughly constant over the period. Because it is possible that there was a trend toward seeking greater subspecialty training or otherwise postponing entry into patient care fields, we examined the response to the question “What do you expect to be doing after completion of your current training program?” (presented in ). A downward trend in the fraction of graduates entering patient care is evident, but it is not large enough to explain the decrease in the number of graduating residents going into patient care.3
While the overall response rate to the survey is a respectable 67 percent, we are nonetheless concerned about potential biases that might result from the somewhat low response rate. However, we have no reason to suspect that lack of response to the survey would be correlated with location choice; hence, we believe the extent of any selection bias to be minimal.
Summary of Survey Response Rate and Sample Sizes and Patient Care Percentage by Year and Specialty
Information on the enactment of state medical liability law limiting noneconomic damage awards in malpractice cases was also acquired for our analysis (Hellinger and Encinosa 2003
; Encinosa and Hellinger 2005
;). By 2003, 25 states enacted legislation capping the awards on noneconomic damages. The amount of the cap for noneconomic damages varies across these states, ranging from US$250,000 to US$750,000 (Studdert, Yang, and Mello 2004
). Because our empirical model contains location fixed effects, only the five states that changed their laws regarding damage award limits during the period of our analysis (1998–2003) provide identifying information in our model. The five states that enacted laws were Mississippi and Nevada in 2002, and Florida, Ohio, and Texas in 2003. Research by Kilgore, Morrisey, and Nelson (2006)
has demonstrated that malpractice insurance premiums are responsive to the enactment of damage caps. By including both premium information and damage award caps in our model we are in effect hypothesizing that new physicians might view the enactment of a damage cap as a signal that the state represents a more “friendly” medico-legal environment to physicians (or vice versa) above and beyond the effect of the current level of malpractice insurance premiums.
To account for federal policies designed to encourage physicians to practice in areas with perceived shortages, which are federally designated HPSAs, we use the coding available in the area resource file (ARF).4
Scholarship and loan repayment programs through the NHSC, rural Health Clinic reimbursements, and Medicare incentive payments for physicians are available for primary care (including OB/GYN) physicians willing to locate in designated HPSAs.
A number of additional local market characteristics were obtained from the ARF. These variables include the ratio of physicians to population, hospitals per capita, hospital beds per capita, total medical residents per capita, total births per capita, per capita income, area population, and the unemployment rate. In order to control for the possibility that salaries might compensate for high malpractice insurance premiums, we obtained information on average physician hourly wages at the MSA level from the 1998–2003 Occupational Employment Statistics Survey conducted by the Bureau of Labor Statistics. The physicians' hourly wage in 1999–2003 was calculated for eight specialties, including internists, generalists, obstetricians and gynecologists, and surgeons.5
Because the dataset does not include hourly wages for non-MSA areas, we used the average physician hourly wage by state as a proxy for the hourly wage in the non-MSA regions of the state.
Several interaction effects are included in the model. To determine whether physicians with relatively high educational debt are more likely to practice in an HPSA than in a non-HPSA, we include an interaction term between our HPSA measure and educational debt. To examine the effect of the ethnic composition of the practice area, an effect documented in Polsky et al. (2002)
, interaction terms were created between variables for the physicians' own race and the corresponding proportion of the population who were white, black, Asian, or Hispanic.