Premium
Improvement of Thematic Classification in Offender Profiling: Classifying Serbian Homicides Using Multiple Correspondence, Cluster, and Discriminant Function Analyses
Author(s) -
Goodwill Alasdair M.,
Allen Jared C.,
Kolarevic Dag
Publication year - 2014
Publication title -
journal of investigative psychology and offender profiling
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.479
H-Index - 22
eISSN - 1544-4767
pISSN - 1544-4759
DOI - 10.1002/jip.1416
Subject(s) - offender profiling , discriminant function analysis , centroid , linear discriminant analysis , profiling (computer programming) , thematic map , multiple correspondence analysis , typology , psychology , crime scene , cluster (spacecraft) , correspondence analysis , thematic analysis , artificial intelligence , computer science , pattern recognition (psychology) , machine learning , criminology , geography , sociology , cartography , visualization , qualitative research , social science , archaeology , programming language , operating system
This paper investigates thematic classification of homicides for the purpose of behavioural investigative analysis (e.g. offender profiling). Previous research has predominantly used smallest space analysis (SSA) to conceptualise and classify offences into thematic groups based on crime scene behaviour data. This paper introduces a combined approach utilising multiple correspondence analysis (MCA), cluster analysis (CA), and discriminant function analysis (DFA) to define and differentiate crime scenes into expressive or instrumental and impersonal or personal crimes. MCA is used to derive the latent structural dimensions in the crime data and produce quantitative scores for each offence along these dimensions. Two‐step CA was then utilised to classify offences. Offence dimensional scores were then used to predict cluster membership under DFA, producing cluster centroids corresponding to MCA dimensions. Centroids were plotted on the MCA correspondence map to simultaneously conceptualise crime classification and the latent structure of the Serbian crime data. Classification of offences based on MCA dimensional scores were 91.5% accurate. This MCA–CA–DFA approach may reduce some of the more subjective aspects of SSA methodology used in classification, whilst producing a product more amenable to objective and cumulative review. Implications for offender profiling research utilising SSA and this approach are discussed. Copyright © 2014 John Wiley & Sons, Ltd.