The Use of the Johnson-Neyman Confidence Bands and Multiple Regression Models to Investigate Interaction Effects: Important Tools for Educational Researchers and Program Evaluators

Abstract

When investigating the impact of predictor variables on an outcome variable or measuring the effectiveness of an educational program, educational researchers and program evaluators cannot ignore the possible influences of interaction effects. The purpose of this paper is to present a procedure that educational researchers can follow in order to increase their understanding of the nature of  the interaction effect between a dichotomous treatment variable and a continuous independent variable. This technique involves the use of three separate analytical techniques implemented in three steps. First, the interaction effect is statistically tested using a multiple regression model. Second, the interaction effect is plotted, and if the interaction effect is disordinal, the intersection point of the regression lines is calculated. Third, the Johnson-Neyman confidence limits are calculated. A list of the computer commands that can be used in conjunction with the SPSS/PC+ Statistics and the SPSS® for Windows computer software to calculate the Johnson-Neyman confidence limits is provided. In addition, this three-step analytical procedure is applied to a set of efficacy data that was collected in a study of the FOCUS instructional model in order to illustrate how it can be used by researchers and program evaluators.