Do the Redundant and Locally Dependent Items of the LS/CMI Contribute in Any Meaningful Way to Its Reliability and Its Potential to Predict Criminal Recidivism?

Abstract

This article studies the effects of local dependence within the items of the first section of the LS/CMI on its reliability. Analysis were done to identify the dependent items namely through their correlations before and after Rasch modeling. Seven items were thus discarded, deemed dependent and redundant, and Cronbach's alpha was calculated with all 43 items and then with the 36 items deemed independent. Test information and predictive validity were also compared. Removing the seven redundant items did not seem to have major effects on the reliability of the LS/CMI or the psychometric information it provided, and no tangible effects were observed on its predictive validity. The reliability of an instrument should be assessed with items that contribute each in its own way. However, it is hazardous to report the reliability of an instrument known to be multidimensional with means meant to be used with unidimensional instruments.