I am disappointed that the BMJ chose to publish a article that reported that black male physicians had lower incomes than their white male counterparts (1) because the methods that the authors used are severely flawed.
The authors used two datasets to arrive at their conclusions. The first, ‘The US Census American Community Survey’ (ACS), did not include information on physician specialty or physician practice characteristics, both of which are critical to the analysis of physician incomes. (2-5) Worse, that dataset truncates annual earned income at $200,000, causing the authors to impute earnings for 47% of the white male physicians they analyzed. I find it hard to imagine that BMJ would publish any other study in which the primary outcome was estimated for one-half of the study population. Any findings derived from this dataset should be disregarded.
The second dataset was derived from Community Tracking Study physician surveys for 2000-01 and 2004-05 and the 2008 Health Tracking physician survey. These data are better because they include physician specialty and physician and practice characteristics. However, in the 2008 dataset, income data are available only in $50,000 wide categories and have an upper value of "> $300,000" (the category for 19.4% of 2008 respondents). (6) The authors again impute values to generate their primary outcome variable – bizarrely applying the median level of what the income category in 2004-05 would have been to estimate a value of the categories in 2008. The authors subject these multiply imputed…and adjusted….and normalized estimates of key variables generated from wide income and work hour categories to multiple regression; they then provide absurdly specific adjusted annual income estimates.
The values that the authors report in the abstract are from their analysis of the ACS and demonstrate the largest race- and sex-based gaps of any of the 9 analyses they completed. These are the values that the media will report and the ones that the public will fixate upon. The authors claim that ‘patterns [shown in the abstract] were unaffected by adjustment’. But the degree of differences certainly were. The last table of the online-only available appendix shows results from the Community Tracking Studies that adjust for specialty and other variables known to predict physician incomes: the income gap between black and white male physicians drops from the $65,000 reported in the abstract (a 26% discount from median white male incomes) to $21,000 (a 9.5% discount). Further, the difference between the upper bound of the black male physician income estimate and the lower bound of the white male physician income estimate decreases from $43,000 (reported in the abstract at p<0.001) to a barely statistically significant $1,100. Pretty big differences. Pretty important to note. And it is possible that analysis of data that were not repeatedly massaged would have found no statistical differences whatsoever.
Particularly at a time where there is a high level of racial tension in the US – perhaps most evident in the Black Lives Matter movement and in the nomination of Donald Trump – editors and researchers have a responsibility to be exceedingly careful when conducting research that might inflame those tensions. I appreciate that publishers want to publish articles that sell issues and generate media interest. But it was irresponsible of the editors of BMJ to publish such sensationalistic and poor quality work.
.
Facebook
Twitter
Instagram