Gender Differences in the Annual Income of Psychiatrists

We used survey data to conduct a cross-sectional retrospective analysis of white psychiatrists' annual incomes. Because of low numbers of black respondents to the surveys, we were unable to examine the influence of race on psychiatrists' incomes. Between 1992 and 2001 the American Medical Association (AMA) conducted regular telephone surveys of physicians that collected a broad variety of individual physician-level data, including weeks and hours of practice, number of completed patient visits, provider characteristics, practice characteristics, and physician incomes ( 14, 15, 16, 17, 18, 19, 20, 21 ). The survey was designed to provide representative information on the population of all actively practicing, nonfederal physicians who spend the greatest proportion of their time in patient-care activities. Weights for each respondent were calculated to correct for potential bias created by unit nonresponse and survey eligibility and to ensure that physician responders reflected the national distribution of physicians ( 21 ). This study was approved by Dartmouth Medical School's Committee for the Protection of Human Subjects.

Each year, the telephone-administered survey was conducted with a random sample of physicians from the AMA Masterfile who are eligible for the survey. The following physicians were excluded: doctors of osteopathy, foreign medical graduates with temporary licensure, inactive physicians, physicians sampled during the past five years, physicians on the "do not contact" list, physicians not practicing in the United States, and physicians without a license. In addition, after initial screening, federally employed physicians and physicians who spent less than 20 hours each week in patient-care activities were excluded.

The following field procedures were developed to minimize nonresponse bias: two weeks before data collection advance letters were sent describing the process and the survey, many specialty organizations provided endorsement letters, and summaries of the types of practice expenses to be reviewed were provided in advance of the survey. In addition, a minimum of four callbacks to respondents were made before abandoning interview efforts, letters encouraging participation were sent to physicians who initially refused participation, and attempts to secure participation were made by select interviewers ( 21 ).

Survey weights were derived by first dividing the AMA Physician Masterfile population and survey respondents into 200 cells defined by specialty, years since the respondent received an M.D. degree, AMA membership status, and board certification status. Unit response rates were constructed as the ratio of the number of physicians in the population to the number of respondents in each cell. Second, an eligibility correction was employed, because only nonfederal physicians involved in direct patient care, excluding residents, were eligible. The eligibility correction divided the subset of the population for which eligibility was known into 40 cells—according to years in practice, AMA membership status, gender, and board certification—and calculated the proportion of physicians in each cell who were eligible. This defined the eligibility weight. The overall weight applied for a given respondent was the product of the unit-response weight and the eligibility weight ( 21 ).

Although the survey had been conducted for much longer, this analysis was limited to data collected between 1992 and 2001 for two reasons. First, during the study period the survey methods allowed us to categorize physicians into well-defined specialty groups in a way that allowed for the disaggregation of responses from psychiatrists and other medical specialists. Second, these were the most recent data available for analysis, and therefore, they were likely to be the most relevant to the currently practicing physician workforce.

To ensure that all the physicians analyzed were comparable, that variables critical to the analysis were available for each respondent, and that extreme outliers did not drive results, a sequential process of elimination of survey respondents was used. First, only self-identified white physicians who were identified as practicing psychiatry in an office-based practice were included in the study, thereby eliminating a handful of black physicians and a minority of physicians who worked primarily doing research, as medical educators, as administrators, or in hospital settings. In addition, only respondents who provided information on key variables were included, and extreme outliers in annual patient visits and net incomes were excluded. This process left 740 white male and 236 white female psychiatrists available for analysis. By using survey weights, these respondents represented 709 white male and 232 white female psychiatrists.

From the AMA data set, three types of independent variables that were likely to influence the dependent variable—net annual income—were extracted.

Physician workload as measured by annual hours worked. It has been demonstrated that hours worked is an important variable in analysis of physician incomes ( 5, 6, 7, 8, 22 ). Indeed, in the study sample, there was a modest linear relationship between inflation-adjusted annual physician incomes and annual hours worked (r=.42, p<.001).

Provider characteristics. When making gender comparisons of physician incomes, age has commonly been used as an adjustment factor ( 5, 6, 7, 8 ). Over the working lifetime, incomes demonstrate an "inverted U" pattern ( 23 ) that typically peaks near age 55 for primary care physicians ( 24, 25 ), or after 20 to 25 years of practicing medicine. To dispel a concern that gender might influence the age at which a physician entered medical school and therefore bias results, the number of years that respondents had been practicing medicine were incorporated into the analysis instead of physician age. In the study sample the number of years practicing medicine was highly correlated with age (r=.88, p<.001). In addition, because practice arrangements, such as having an ownership interest in the practice, has been associated with differences in annual income among physicians ( 9 ), employment status—that is, whether the physician was an employee, as opposed to a full or partial owner of the practice—was included in the analysis. Finally, because board certification has been associated with higher physician incomes ( 11 ), board certification status was also included as an independent variable in the analysis.

Practice characteristics. Physicians who live in different U.S. census regions have been shown to have modestly different annual incomes ( 14, 15, 16, 17, 18, 19, 20, 21 ); therefore, information on the U.S. census region in which the practice was located was collected. In addition, because one study showed that physicians who live in poorly populated settings have lower incomes ( 26 ) and another study showed that these physicians have higher incomes ( 27 ), this analysis categorized responding physicians' county codes into three categories of metropolitan settings (less than 50,000, between 50,000 and 500,000, or greater than 500,000). Finally, because treating persons who are medically indigent and those with relatively poor insurance has been hypothesized to decrease physician incomes ( 28 ), variables likely to reflect those factors were incorporated into the analysis: whether the practice provides Medicare services and the reported proportion of patients in the practice receiving Medicaid.

We used the consumer price index ( 29 ) to adjust reported net annual income to constant 2004 dollars (inflation-adjusted annual incomes). For instance, to inflate income reported for 1995 to 2004 dollars, we multiplied the reported income in 1995 by the consumer price index in 2004 (188.9) and then divided that figure by the consumer price index in 1995 (152.4). The reported number of weeks worked in the past year was multiplied by the total number of hours worked in the past week and the total number of visits seen in the past week to calculate the annual number of hours worked and the annual number of visits, respectively. Because of the inverted-U relationship between number of years practicing medicine and annual incomes, dummy variables were constructed that reflected the categorization of years practicing medicine into five-year increments, from 0 to five years practicing to 40 or more years practicing. Although these dummy variables were used in the regression analysis, they were aggregated into ten-year increments through 30 or more years practicing for the purposes of demographic comparisons.

To determine the association between gender and physician income, after adjustment for practice and provider characteristics, we used a linear regression model that simultaneously entered the independent variables detailed above and used the annual income adjusted by the consumer price index as the dependent variable. For each independent variable we used logistic regression models to calculate dollar-denominated regression coefficients and 95% confidence intervals. For dummy variables the coefficient represents the change in annual income associated with the presence of the dummy variable, compared with a referent group or state. For continuous variables the coefficient represents the change in annual income associated with the slope of that variable: therefore, for annual hours worked the coefficient represents the incremental annual income for another hour worked in the year; for proportion of Medicaid patients, the coefficient represents the incremental annual income associated with increasing the percentage of the practice population that is on Medicaid by 1%. In addition, we calculated standardized partial regression coefficients to help readers evaluate the relative importance of each of the variables in accounting for variance in income. For all analyses we used SPSS (version 11.5.1, Base and Advanced Models add-on module using the general linear models procedure) and the survey weights described above.