Optimal methods for disattenuating correlation coefficients under realistic measurement conditions with single-form, self-report instruments

Correlational analyses have long been important tools for elucidating relationships between individual difference variables in educational and psychological research. Such research often involves correlating scores from instruments designed to measure such individual differences. Charles Spearman’s disattenuation formula is commonly used to estimate the true score correlation between measures by correcting correlations for the attenuating effects of measurement error. Spearman’s original intent was to include correlation and reliability terms within the formula that reflected administration of pairs of parallel forms for the two instruments across multiple administration occasions. However, the common lack of parallel forms for psychological instruments results in researchers employing single-occasion study designs, potentially introducing inaccuracies due to failure to account for the attenuating effects of transient measurement error and correlations between such errors within occasions.

The purpose of the present simulation-based study was to evaluate the accuracy and precision of a set of disattenuation methods designed to allow for accurate estimation of disattenuated correlations using data from administration of a single form of X and of Y on two administration occasions. Fourteen disattenuation methods, including variations on the Spearman formula, as well as methods based on multivariate generalizability theory and confirmatory factor analysis (CFA), were employed. Overall, two-occasion CES Spearman variants and two-occasion CFA models that accounted for all major sources of measurement error and their intercorrelations yielded more accurate estimates than did methods incapable of appropriately accounting for such error, including single-occasion and test-retest Spearman variants.