Misuse of correlation and regression in three medical journals

Errors relating to the use of the correlation coefficient and bivariate linear regression are often to be found in medical publications. This paper reports a literature search to define the problems. All the papers and letters published in the British Medical Journal, The Lancet and the New England Journal of Medicine during 1997 were screened for examples. Fifteen categories of errors were identified of which eight were important or common. These included: failure to define clearly the relevant sample number; the display of potentially misleading scatterplots; attachment of unwarranted importance to significance levels; and the omission of confidence intervals for correlation coefficients and around regression lines.