Comparing Regression Lines From Independent （从独立比较回归直线）.pdf

Comparing Regression Lines From Independent Samples The analysis discussed in this document is appropriate when one wishes to determine whether the linear relationship between one continuously distributed criterion variable and one or more continuously distributed predictor variables differs across levels of a categorical variable (and vice versa). For example, school psychologists often are interested in whether the predictive validity of a test varies across different groups of children. Poteat, Wuensch, and Gregg (1988) investigated the relationship between IQ scores (WISC-R full scale, the predictor variable) and grades in school (the criterion variable) in independent samples of black and white students who had been referred for special education evaluation. Within each group (black students and white students) a linear model for predicting grades from IQ was developed. These two models were then compared with respect to slopes, intercepts, and scatter about the regression line. Such an analysis, when done by a school psychologist, is commonly referred to as a Potthoff (1966) analysis. Poteat et al. found no significant differences between the two groups they compared, and argued that the predictive validity of the WISC-R does not differ much between white and black students in the referred population from which the samples were drawn. Weaver and Wuensch (2013) have provided SAS and SPSS code for conducting a Potthoff analysis (see “Testing the difference between two independent regression coefficients”). Watkins and Hetrick (1999) have provided a Macintosh app to do the same. In the simplest case, a Potthoff analysis is essentially a multiple regression analysis of the following form: Y = a + b C + b G + b CG, where Y is the criterion variable, C is the continuously 1 2 3 distributed predictor variable, G is the dichotomou