Semester

Summer

Date of Graduation

2020

Document Type

Dissertation

Degree Type

PhD

College

Eberly College of Arts and Sciences

Department

Forensic and Investigative Science

Committee Chair

Jacqueline Speir

Committee Member

Casey Jelsema

Committee Member

Tatiana Trejos

Committee Member

Glen Jackson

Abstract

Footwear evidence holds tremendous forensic value, owing to its ability to formulate linkages between victims, suspects and scenes. Naturally, the strength of these linkages is a function of the perceived clarity, quality and rarity of class, subclass and randomly acquired characteristics (RACs), which are the fundamental outsole features used to formulate source associations. In order to reach a conclusion when performing a footwear comparison, forensic examiners must assign value to the observed similarities and differences that exist between questioned crime scene and test impressions. Embedded within this process is an evaluation of the random association between unrelated shoes as a function of both class and acquired wear characteristics. To date, weight of evidence within this space has been largely informed by the training and subjective casework experience accumulated by an examiner over the life of his or her career. In pursuit of supporting the foundational validity of this comparison process, this research sought to quantify the chance association of RACs on unrelated shoes and the spatial distribution of these features on outsoles, with the long-term goal of aiding weight of evidence assessments in forensic footwear examinations.

Using a large-scale database of 1,300 unrelated outsoles, the position and shape of 72,306 RACs was investigated. Features with consistent position and shape-classification were pairwise compared and sorted using a numerical estimate of similarity. Based on this assessment, more than 91,000 of the most quantitatively similar features were visually evaluated in order to model the relationship between numerical similarity and visual indistinguishability. Using this model, more than 1 million additional feature comparisons were evaluated in order to predict the potential for visual confusion. Subsequently, empirical and modeled probabilities of indistinguishability were combined with the chance for positional overlap to yield location- and shape-specific estimates of chance association. The results indicated that RACs exhibit high discriminating potential, with median chance associations ranging from 1 in 541,276 to 1 in 18,031,824, depending upon shape. However, additional inspection revealed that chance association was not constant across an outsole. Given this secondary observation, the spatial distribution of RACs on outsoles was further investigated. In order to conduct this analysis, a set of over 1.7 million null and 1.9 million alternative contact-modified synthetic distributions were simulated for comparison against the collected empirical data. Results indicated that Poisson null distributions (both synthetic and modeled) well-describe the frequency of RACs across approximately 64% of an outsole. Moreover, the regions not well represented by a random distribution were highly localized to three general areas (ball of the toe, arch, and edge of the heel). Based upon this observation, it was purported that an important theoretical or practical factor was additionally required to improve prediction in these locations.

Therefore, spatial regression modeling was utilized in order to assess the impact of spatial effects on RAC distributions. Under optimal conditions, 87% of location-specific RAC counts were well predicted using contact area and incorporating neighboring cells’ data for contact and accidentals (a 67% performance increase over non-spatial predictions). Based upon a visual inspection of the remaining 13% of cells with persisting residual correlation, it was hypothesized that wear (the intersection of contact and use) may further improve model predictions and a proof of concept study was conducted to evaluate this theory. After incorporation of contact-localized wear as a predictor in the spatial models, nearly 96% of the outsole was well described. Considered collectively, the results from this work indicate that RACs are sufficiently rare, owing to variability in position, shape, and geometry, to differentiate shoes, as evidenced by the low probabilities of stochastic chance association between unrelated features. Furthermore, the majority of feature frequencies across the outsole can be adequately described by tread contact alone, irrespective of position. However, positional considerations for evidentiary value must be incorporated for features occurring in three specific areas including the ball of the toe, the arch, and the edge of the heel. Ultimately, the results from this study provide fundamental knowledge about the practical and theoretical/statistical factors that underpin the spatial distribution and subsequent weight of evidence of RACs for footwear evidence.

Embargo Reason

Publication Pending

Available for download on Wednesday, July 21, 2021

Share

COinS