Selecting the correct variables is crucial in epidemiological analyses for avoiding biased parameter estimates. Old methods like stepwise regression analysis is becoming increasingly criticized, and several new methods have been developed, like shrinkage and penalized regression. Walter and Tiemeier (1) reviewed 300 articles published in American Journal of Epidemiology, Epidemiology, European Journal of Epidemiology and the International Journal of Epidemiology during 2008 in order to assess how often different variable selection techniques were applied in contemporary epidemiological analyses.

In 83 (28%) articles, the authors selected covariates for multivariable models based on prior knowledge. Stepwise selection procedures, extensively criticized in modern literature, were used in 59 (20%) articles. Not a single article presented use of shrinkage methods such as LASSO, and as many as 105 (35%) publications did not describe the method, which the authors see as an indication of low quality of the information presented in the methods sections.

The authors conclude that variable selection methods which have been formally criticized as flawed still prevail in the scientific literature.


  1. Walter S, Tiemeier H. Variable selection: current practice in epidemiological studies. Eur J Epidemiol  2009;24:733–736.