Salary Equity among Male and Female Internists in Pennsylvania

In the past generation, women have made up an increasing percentage of the U.S. physician workforce. Yet women have lagged behind men in measures of career success. As of 19941995, women accounted for 41% of medical students, 33% of residents, and 25% of full-time medical school faculty (1). Among faculty, full professorships were held by 10% of women compared with 31% of men (1). This gender gap has been explained by the later entry into medicine by large numbers of women, fewer working hours or lower productivity among women, or gender bias; various studies have reached different conclusions (2-4). An important marker of gender equity at work is salary. Studies conducted in the 1970s through the early 1990s, when the number of women entering the physician workforce rapidly increased (5), indicated that men earned more than women after adjustment for specialty, number of hours worked, and practice setting, among other factors (6-12). However, studies published in the mid-1990s reported conflicting results. Baker (13) found no overall earnings differences after adjusting for practice factors between young male and female physicians who responded to a national survey. These results have been interpreted as indicating that the gender gap in earnings for physicians has been eliminated (14). However, Kaplan and colleagues (15) found substantial earnings differences by gender in a national survey of academic pediatricians. We sought to reexamine physicians' salaries by gender almost a decade after the aforementioned surveys were conducted among internists working in nonacademic and academic settings. We report the results of a 1998 survey sent to Board-certified female internists and an equal number of male internists practicing in Pennsylvania. Methods Participants The study was approved by the institutional review board of the University of Pittsburgh. Our overall strategy was to maximize the internal validity of gender-specific comparisons while maintaining reasonable generalizability. We therefore surveyed Board-certified internists from a single state and from a defined age cohort: those currently practicing in Pennsylvania who had graduated from medical school 10 to 30 years ago. We specifically chose physicians at least 10 years past graduation to account for different lengths of residency and to ensure a more stable earning pattern among respondents. We obtained a mailing list of 4901 Board-certified physicians from the American Board of Medical Specialists. This list is updated quarterly by using direct mailings to physicians to ascertain new address information. The inclusiveness of the list was validated by comparing it with the American Medical Association Master File Report of the number of Board-certified physicians in Pennsylvania who were within the same age range (16). The number of internists estimated from the American Medical Association Master File was within 10% of the number in the mailing list that we used. When only physicians who graduated from medical school between 1967 and 1987 were included, 3721 internists made up the sampling frame. From this sampling frame, we sought to survey relatively equal numbers of men and women. We therefore selected all 628 female internists and an equal number of randomly selected male internists (constituting about a 20% sampling). Because the mailing list did not disclose gender, female and male internists were selected on the basis of first names. For the 102 internists on the list whose first names could be either male or female and the 80 for whom only initials or unfamiliar first names were available, we selected a 20% random sample. Female or male gender was then determined with certainty on the basis of self-designation of gender in the survey. Data Collection After an extensive process of survey construction and pilot testing, the survey instrument was sent to physicians. The survey, which took 15 to 20 minutes to complete, asked questions about gender; age; marital status; number and ages of children; training; Board certification; leave from work; practice type and size; partnership; and faculty appointment, type, and level of promotion. In addition, physicians were asked about the number of hours worked per week (including work-related activities done at home) and how those hours were divided among patient care, research, patient-related paperwork, administration, and teaching. Salary information was obtained from the answer to the question, What is your yearly take-home salary from all professional sources? Questionnaires and a stamped return envelope were mailed to all 1256 internists surveyed. For physicians with incomplete mailing addresses or those whose surveys were returned as undeliverable by the postmaster, we sought updated mailing and telephone information from the American Medical Association master list on the Internet (www.ama-assn.org). Second mailings were sent to all nonrespondents, followed by telephone calls made by the authors or third mailings if telephone numbers were not available. Three hundred seventeen internists were excluded because they had an incomplete mailing address and had no telephone information (n =65) or were no longer working in Pennsylvania or were retired (n =252). Of the 939 eligible respondents, 595 (63.4%) completed the survey. Response rates for men and women were 62.4% and 65.7%, respectively. An additional 17 physicians were excluded because they had graduated from medical school less than 10 years ago, were currently in a fellowship program, reported working less than 20 hours per week, or reported earning less than $10 per hour. These exclusions paralleled those made by Baker (13). We eliminated these physicians because their earning potential may have been affected by their status as trainees or as part-time workers. Excluding such physicians had a very limited impact on our estimates of effect. The salary question was the one that most often remained unanswered; 90 (15.6%) respondents did not answer it, whereas 43 others had missing data on all other questions combined. Internists who did not answer the salary question did not statistically significantly differ from those who answered the question with respect to age, years in practice, hours worked per week, and specialty, but they were more likely to be in nonacademic practices and to be men. We present the data set for the 232 men and 213 women for whom information on all variables was complete. Descriptive characteristics for this subset were very similar to those for the group as a whole. Statistical Analysis Descriptive characteristics of respondents are given as the median (25th percentile, 75th percentile) or frequency. Continuous measures were compared by using the MannWhitney U test, and frequencies were compared by using the chi-square test. Hourly earnings for each respondent were calculated by dividing the yearly take-home salary by the number of hours worked. Because the distribution of hourly earnings was skewed, a log transformation was used to normalize the data and analyses were performed on the transformed data. For presentation, the geometric means of the hourly wages and corresponding male-to-female ratios are shown. The geometric mean, like the median, reflects the typical salary, wage, or income more accurately than does the simple arithmetic average when, as in our study, a few physicians report unusually large values (17). This calculation was done by back-transforming the averages of the transformed salaries and then calculating the male-to-female ratios. In this way, the ratios reflect the degree to which the actual wages, rather than the log of wages, differed between men and women. Adjusted values for the log of hourly earnings were calculated by using analysis of covariance. Overall, the log of hourly earnings was adjusted for training (fellowship, Board certification), practice description (specialty, academic affiliation, years in practice, partnership), and personal data (marital status, children, time off work or employed part-time). Specialties were divided into high-earning (cardiology, hematology/oncology, gastroenterology, pulmonology), low-earning (allergy and immunology, endocrinology, geriatrics, infectious disease, nephrology, rheumatology), and general internal medicine. Mean hourly earnings for the high-earning specialties ($56.9 per hour), low-earning specialties ($43.9 per hour), and general internal medicine ($48.8 per hour) were clustered and distinct among categories. A single question asked about taking time off from work or working part-time. We separately analyzed women who took time off or worked part-time for more than 12 months and for 12 months or fewer; because the mean hourly earnings of these women were similar ($46.55 and $43.52, respectively), this distinction was not made in the final analysis. Separate analyses of covariance were used to obtain the adjusted log of wages within strata defined by each covariate separately. The adjustments included all of the variables mentioned above, except for the covariate creating the strata. Confidence intervals around the male-to-female ratios were computed within each strata. Of note, this technique often produced wide confidence intervals because some strata included small samples. For these analyses, we used the GLM procedure in SAS software (SAS Institute, Inc., Cary, North Carolina); this procedure generates adjusted estimates, ratios, and confidence intervals and tests for differences. Results The internists surveyed ranged in age from 33 to 63 years; 85% were in their forties or fifties. On average, they graduated from medical school 19 years previously and had been in practice for 13 years. Fifteen percent were in solo practice; the remainder were in group practice. Group practice arrangements were described as specialty by 46.0% of internists, mixed specialty by 19.6%, academic by 27.1%, hospital staff by 11.4%, and health maintenance o