Prediction by multiple regression how many variables to enter?

Multiple regression techniques have been used in a number of outcome prediction problems in psychiatric research with results that are encouraging, but far from satisfactory in terms of cross-validation. The authors draw attention to the increased risk of Type 1 error that accompanies entry of a large number of variables into a multiple regression equation. An approach is proposed that consists of a step-wise sliding scale of F-values to enter variables into the equation according to the assumed number of predictor dimensions, the number of variables that have already been entered and the degree of risk that the experimenter is willing to assume that a variable being entered is not a true predictor.