Premium
THE SELECTION OF VARIABLES IN MULTIPLE REGRESSION ANALYSIS
Author(s) -
HALINSKI RONALD S.,
FELDT LEONARD S.
Publication year - 1970
Publication title -
journal of educational measurement
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.917
H-Index - 47
eISSN - 1745-3984
pISSN - 0022-0655
DOI - 10.1111/j.1745-3984.1970.tb00709.x
Subject(s) - statistics , selection (genetic algorithm) , multivariate statistics , mathematics , set (abstract data type) , regression analysis , population , sample (material) , variables , stepwise regression , sampling (signal processing) , data set , regression , sample size determination , econometrics , computer science , artificial intelligence , demography , chemistry , filter (signal processing) , chromatography , sociology , computer vision , programming language
4 different procedures are commonly employed with sample data to reduce a set of predictor variables. In the present study these procedures were repeatedly applied to computer‐simulated samples to provide comparative data pertaining to two questions: (a) Which procedure can be expected to produce an equation that yields the most accurate predictions for the population? (b) Which procedure is most likely to identify the optimal set of independent variables? The samples were drawn from 12, mathematically defined, multivariate normal populations. Each population consisted of 1 criterion and 10 predictor variables. Five or fewer independent variables constituted the optimal set in each case. With respect to both questions small differences among the procedures were observed. However, the forward selection and stepwise procedures consistently produced more favorable results than the 2 backward elimination procedures. The question of the number of sampling units to use is discussed.