Significance testing for round-robin data

Abstract

To understand two-person social processes, psychologists have analyzed round-robin research data with a Social Relations Model (Kenny & La Voie, 1984). Although conclusions about social processes have been reached with tests of the statistical significance of the parameters of this model, the validity of the significance tests has never been established. This dissertation reports a series of computer simulations on round-robin significance tests. Six different methods of significance testing were assessed on eight different configurations of population effects, two population distributions, and two round-robin group sizes. The significance testing methods were compared in terms of empirical Type I error rates and power levels. A test based on work by Bond and Lashley (In press) was the best all-purpose inferential method. For significance tests of one Social Relations parameter (the dyadic covariance), a second method performed acceptably. For studies of multiple round-robins, a widely-used t-test was also acceptable.