Porter (1) screened three leading medical journals, British Medical Journal (BMJ), the Lancet and the New England Journal of Medicine (NEJM), for articles having misused correlation and regression analysis in articles published during 1997. Thirteen articles from BMJ, 5 from the Lancet and 6 from NEJM were reviewed.
Fifteen errors were identified, 8 of which considered to be important because of their frequent occurrence or potentially major consequences. These were included: failure to define clearly the relevant sample number (BMJ=3), the display of potentially misleading scatterplots (BMJ=2, Lancet=4, NEJM=5), attachment of unwarranted importance to significance levels (BMJ=2, Lancet=2, NEJM=2), the omission of confidence intervals for correlation coefficients and regression lines (BMJ=6, Lancet=4, NEJM=5), the appearance of apparent heteroscedasticity (which violates an underlying assumption) with no comment in the text (BMJ=2, Lancet=1, NEJM=1).
The author concludes that editors and referees should demand a more disciplined, standardized and structured approach to the use and presentation of correlation and linear bivariate regression statistics.
1. Porter AM. Misuse of correlation and regression in three medical journals. J R Soc Med 1999;92:123-128.