Using Chernoff face plots for burglary suspect prioritisation

Most good quantitative analysis begins with data exploration. However, when there are many variables how can one summarise and compare cases in terms of all values? The first two Principal Components are often plotted which can, for example, reveal clusters. However, implementation and interpretation of such a method requires a moderate level of statistical skill and there is no guarantee that meaningful components will emerge. This paper presents another graphical method, namely Chernoff Face Plots, in the context of matching suspects to crimes. Sizes of facial features are mapped to values of variables. The underlying reasoning is that humans have a highly developed ability to recognise and distinguish faces. Similar cases produce similar faces whereas outliers ‘stand out from the crowd’. In this study a face is produced for 43 suspects based on crime characteristics. Faces for 3 new crimes are drawn and an attempt made to identify correct suspects.