**ERIC Number:**ED409325

**Record Type:**Non-Journal

**Publication Date:**1997-Mar

**Pages:**38

**Abstractor:**N/A

**Reference Count:**N/A

**ISBN:**N/A

**ISSN:**N/A

The Error of Accuracy for Two Regression Techniques: Does Psychometric Parallelism Matter?

Chang, Te-Sheng; Brookshire, William

The question of least-squares weights versus equal weights has been a subject of great interest to researchers for over 60 years. Several researchers have compared the efficiency of equal weights and that of least-squares weights under different conditions. Recently, S. V. Paunonen and R. C. Gardner stressed that the necessary and sufficient condition for equal-weights aggregation is that the predictors satisfy the requirements of psychometric parallelism. In this study, the effect of psychometric parallelism on the error of accuracy for equal weights and least-squares weights was investigated with the combination of different numbers of predictors, sample sizes, and intercorrelations. The findings indicate that equal weights always perform more precisely than least-squares weights as long as the following situations are satisfied: (1) the number of predictors is small; (2) the ratio of observation to predictor is small, less than or equal to 10; and (3) the magnitude of the mean intercorrelation is high, at least 0.6. Least-squares weights may perform more accurately than equal weights in the opposite situations of a large number of predictors, a high ratio of observation to predictor, and low intercorrelations. Nevertheless, the combination of a large number of predictors, large sample sizes, and a low mean of intercorrelation does not guarantee that least-squares weights are more accurate than equal weights. Equal weights are still more accurate than least-squares weights for the sample with a relatively high level of psychometric parallelism. (Contains 16 tables and 34 references.) (Author/SLD)

**Publication Type:**Reports - Evaluative; Speeches/Meeting Papers

**Education Level:**N/A

**Audience:**N/A

**Language:**English

**Sponsor:**N/A

**Authoring Institution:**N/A