REGRESSION AND ANOVA MODELS FOR RELIABILITY
Reliability can be assessed through either a regression or an analysis of variance approach. We first consider a general result from the regression model. In bivariate regression one interesting but perhaps little known point is that two different regressions with the same number of observations and exactly the same fitting errors do not necessarily have the same correlation coefficient. Consider the two lines shown in Figure 1. Even if the fitting errors and consequently their sum of squared fitting errors (SSE) are the same in both cases, line (b) will yield a higher correlation coefficient (reliability) than will line (a) because the total sum of squares (SSTO) will be larger for line (b). With R- being defined as 1 - SSE/SSTO, the steeper line will be given the larger correlation.