Evaluating and Comparing Profiles of Burglars Developed Using Three Statistical Classification Techniques: Latent Class Analysis, Cluster Analysis, and Multidimensional Scaling

• In Event: Innovations in Quantitative Modeling

Abstract

We all rely on classifications to make sense of the heterogeneity in the concepts and individuals that we study. In criminology, there are a variety of statistical classification methods available, with the most prominent being Latent Class Analysis (LCA), Cluster Analysis (CA), and Multidimensional Scaling (MDS). These techniques all aim to uncover homogeneous groups within a dataset, and have each been highly useful for classification of offenders, offending behaviors, and more. However, the math, underlying assumptions, and interpretation of the three analyses are considerably different. Therefore, the current study aims to conduct a comparative analysis of LCA, CA, and MDS on a single sample of offenders to determine similarities and differences in the resultant typologies, and the strengths, weaknesses, and overall accuracy of the three classification methods.Through this study, we demonstrate several benefits and drawbacks of each classification method, how to select the technique that is best fit for available data, and how the choice of analytical technique can have considerable impact on the results of our research.

Authors

• Melanie Valentin Rosa, University of South Florida

• Bryanna Hahn Fox, University of South Florida