METHODS FOR SEARCHING FOR THE MOST INFORMATIVE SET OF ATTRIBUTES IN REGRESSION-ANALYSIS

The currently popular statement of the problem of searching for an inf ormative set of attributes is discussed in relation to the paper by A. N. Kolmogorov ''Concerning the Applicability of Statistically Determi ned Prediction Formulas'' (1933), which shows that the exhaustive sear ch for predictor sets leads to an inadequate ''inflation'' of the corr elation coefficient between the response and the ''best'' linear predi ctor function. The errors into which some present-day authors fall wer e analyzed by A.N. Kolmogorov more than 60 years ago.