Enhancing predictive power by unamalgamating multi-item scales.

Area of Science:

• Social Sciences
• Psychometrics
• Statistical Modeling

Background:

• Small correlation coefficients in social sciences often lead to issues in theory testing and prediction.
• Traditional formulas like Spearman's attenuation formula and Cronbach's alpha are commonly used to assess scale reliability and predictive power.

Purpose of the Study:

• To investigate the underlying causes of small correlation coefficients in social science research.
• To examine how interitem correlations affect a scale's predictive power under different treatment conditions (scale components vs. separate items).

Main Methods:

• Re-evaluation of traditional psychometric equations, including Spearman's (1904) attenuation formula, Cronbach's (1951) alpha, and Guilford and Fruchter's (1973) equation.
• Analysis of the differential impact of interitem correlations on predictive power.

Main Results:

• Interitem correlations can decrease predictive power when items are treated as multi-item scale components.
• Conversely, high interitem correlations can increase predictive power when items are treated as separate predictors.
• Findings challenge conventional practices regarding the interpretation of interitem correlations.

Conclusions:

• The way items are treated (as components or separate) significantly alters the effect of interitem correlations on predictive power.
• Understanding these nuances is crucial for accurate theory testing and prediction in the social sciences.
• The study's implications are broad, affecting scale construction and data interpretation.