The use of jackknifing for the evaluation of geographic profiling reliability

The use of geographic profiling (GP), based on “Rossmo's formula”, a technique derived from criminology, has been previously proven to be effective in assessing the origin of invading species. The application on Caulerpa taxifolia showed the most probable center of spread of the invasion. This article discusses a method of assessing the degree of robustness of the results obtained with Rossmo's method.To provide an evaluation of the reliability of geographic profiling results we used the jackknife technique, randomly eliminating part of the data set for a given number of replicates (500) in order to analyze the obtained result for each replicate. In GP the results are a series of images with geoprofiling prioritization, each produced with one of the replicates. These images can be summarized in three different ways: (1) OR, depicting all the high probability pixels from the series of replicates; (2) AND, depicting only those high probability pixels present in every replicate; and (3) MEAN, depicting the mean color value for each pixel calculated from all the replicates. We show that jackknifing can be a useful method to increase robustness of GP analysis in criminology, epidemiology and biological invasions. Summarizing jackknifing results with the OR logical operator yields the highest sensitivity and worst specificity, while the use of the AND operator increases specificity but reduces sensitivity. Using the mean of the pixel values maintains the visualization of the areas of highest priority (specificity), while also showing the surrounding area with varying colors, analogous to confidence limits.