Sex estimation from handprints in a Croatian population sample: developing a tool for sex identification in criminal investigations

Methods: A cross-sectional study was conducted on 100 adult volunteers from a Croatian population (50 males and 50 females) aged between 20 and 45 years. Using a fingerprint ink, we collected handprints of both hands on a paper sheet. We scanned handprints and took 13 measurements. Bilateral asymmetry and sexual dimorphism of the measurements was analyzed and sex estimation models were developed using linear discriminant analysis.


Introduction
Handprints are a conclusive type of evidence often available at the scenes of various crime types. If they are recovered from the scene and used in forensic examinations, they can help establish the identity of a person included in the crime. Unfortunately, the identification is often not possible as it is required that both the recovered (questioned) prints and those collected from perpetrators/suspect (known prints) are available for comparison [1-Cite as: Kolić A, Jerković I, Anđelinović Š. Sex estimation from handprints in a Croatian population sample: developing a tool for sex identification in criminal investigations. ST-OPEN. 2020; 1: e2020.1919.34. 3]. Although in such instance, handprints cannot directly reveal the perpetrator's identity, their dimensions can still provide valuable clues on the biological features of an individual like sex and stature and narrow down the list of potential suspects [1,2,[4][5][6].
In forensic sciences, studies that employ the statistical models to classify sex using body dimensions have been extensively conducted on skeletal material or body parts [7][8][9][10][11][12].
The primary aim of those studies has been to aid the identification process of the human remains in crimes, disasters, or wars by estimating the sex of individuals. On the other hand, a similar approach is more rarely applied in forensic science to reveal the sex of the donor of prints found at the crime scenes. Until now, studies that consider handprint measurements have been conducted only in Western Australian [6] and French [13] population samples but showed great potential for application in forensic investigations. Both studies revealed that dimensions of handprints demonstrate differences between males and females and thus could be used to develop sex classification models. The standards they developed could estimate sex from handprints correctly in more than 90% of cases when multiple variables are used, while accuracy of 88%-90% could be achieved even with a single variable [6,13]. However, like most of the anthropometric methods, their major limitation is sensitivity to the interpopulation differences in body size, robusticity, and sexual dimorphism [14,15]. Therefore, they must be developed and validated on each population separately, especially considering the level of scientific rigor inherent in modern forensic science.
Except for the sex, previous studies showed that handprints could also be used to estimate the height of a person, thus providing even more information about the print donor [2,4,5,16]. Nonetheless, as many anthropological studies demonstrated that the error of the height estimates is smaller when males and females are considered separately [17,18], height estimation equations for handprints are also developed to be sex-specific [2,5,16].
For this reason, the initial identification of sex is an inevitable prerequisite for their application.
Since sex estimation standards for handprints were not developed for the Croatian population, our study aimed to test if handprint measurements show sexual dimorphism in a Croatian population sample and, if so, to establish population-specific sex classification standards that could be employed in forensic investigations to examine handprints found at crime scenes.

Participants
The cross-sectional study was conducted at the University Department of Forensic Sciences

st-open.unist.hr
Using a brief questionnaire, we collected basic information on participants' sex, age, and handedness as well as information on orthopedic or dermatological health issues used as exclusion criteria.

Handprint analysis
Using fingerprint ink Dacty ink® (BVDA International, Amsterdam, The Netherlands), we collected paper handprints of both hands. Impressions were scanned with a CANON C3320i at 600dpi. All images were imported into and measured in Adobe Photoshop (version CC 2019, Adobe Systems, San Jose, CA, USA). The images were calibrated, measured using a ruler tool, and recorded using the measurement log. All measurements were taken by the first author (AK).
For each impression, we performed following 13 measurements (Figure 1

Statistical analysis
Bilateral asymmetry was tested using a paired sample t-test. For male and female print measurements, we calculated descriptive statistics and analyzed sexual dimorphism by independent samples t-test.
Statistical models for sex classification were developed using linear discriminant analysis with an equal prior probability for male and female measurements. Cut off values for univariate discriminant functions were calculated as a mean of male and female measurements. Multivariate discriminant functions were obtained using the stepwise procedure.
For those functions, the sex estimation equations were derived using unstandardized function coefficients, and cut off values were calculated as a mean of functions' centroids.
Those equations were provided as a linear combination of the variables in the following form [19,20]: where x represents variables used in the study, β weighted coefficients, and c a constant.
The sex can be estimated by comparing the F score with the provided cut off value. If the score is greater than the cut off, it is estimated that print originates from a male, while if it is smaller, it is estimated that print originates from a female individual [19,20].
Sex estimation accuracy was examined using a leave-one-out cross-validation (LOOCV) algorithm that provides realistic classification results. In LOOCV, a classification model is tested on each specimen using the functions calculated from all remaining cases except that one [21]. The sex estimation accuracy was given separately for males, females, and for overall results as a proportion of correctly classified and a total number of individuals.
Statistical analyses were performed using the IBM SPSS (version 22, SPSS Inc., Chicago, IL, USA) with a statistical significance set at P≤0.05.

Results
All the participants met inclusion criteria since they did not report orthopedic or dermatological health issues. The median age of male subjects was 27.5 (range 20-45), while for the females, it was 25 (range 20-39). The sample comprised 92 right-handed and seven left-handed individuals, and one person was ambidextrous.
From 13 measurements, 12 measurements were larger on the right prints, while four of them showed statistically significant bilateral differences (P<0.05, Table 1). For this reason, we analyzed sexual dimorphism for the left and the right prints separately.    Cut off values for univariate discriminant functions and accuracy rates are provided in Table 4 and Table 5. If the measurement is greater than the cut off value, the print belongs to a male individual, otherwise it belongs to a female. For both hands, the highest sex estimation accuracy was achieved for handprint breadth (92%), palm length (87% and st-open.unist.hr 7 88%), and handprint length (86%). Accuracy for complete finger lengths ranged from 78% to 81%, while for distal and middle finger length, it was between 75% and 80%.

Discussion
This study showed that selected handprint measurements exhibited statistically significant sexual dimorphism in a Croatian population sample and provided statistical models that could estimate sex from handprints in Croatian population. This paper presents the first study of this type conducted in a Croatian population, and it is one of the two studies published on samples from European populations [13].
Right hand measurements were significantly greater than left in four variables as has been reported earlier with handprint measurements [4,5], but also in the other types of anthropometric studies [22]. This type of asymmetry is often attributed to the use of the dominant hand [9], which conforms the fact that most of our participants were right-handed. However, it is a complex issue since previous studies showed that dimensions of the right side could be larger not only in right-handed individuals but also in left-handed and bimanual ones [9,22]. So, to avoid additional sources of error, we examined prints from each side separately.
The handprint variables showed a degree of sexual dimorphism similar to the previous study by Ishak et al. [6]. In both that and our study, the measurements with the highest degree of sexual dimorphism were handprint breadth, handprint length, and palm length, whereas sexual dimorphism of complete finger lengths was less pronounced. The dimorphism of distal and middle finger lengths that were not previously used to estimate sex was also statistically significant but less pronounced in comparison to the other variables.
The studies that used handprints to estimate height showed that those variables were less correlated to the stature in comparison to the handprint length [2,16]. So, they are probably more prone to intra-sex variations, and less reflect body size differences between males and females.
The overall accuracy level (75%-93%) was also in line with the previous studies [6,13], where it was 89%-91% [6] and 92% [13] when the same statistical procedures were applied. In the present study, the univariate discriminant function of the handprint breadth achieved the highest accuracy of 92% with no sexing bias, which is also a variable that performed best in the previous study. However, in that study, only handprint length and breadth achieved accuracy greater than 80% with sexing bias smaller than 5%, so researchers did not include the remaining variables for sex estimation [6]. On the contrary, in our research, five variables for the left hand and four variables for the right hand reached accuracy level equal or greater to 80% with a sexing bias less than 5%, which is why they could also be applicable for sex estimation. In practice, it means that sex can be reliably classified even if the partial handprint is available, e. g., if an interdigital area of the palm is available (handprint breadth), or if finger lengths can be measured.
Using a stepwise analysis, we developed one multivariate discriminant function with two variables that could estimate sex accurately for 93% of prints, which was also in accor-dance with previous studies [6,13]. However, due to the high sexing bias (10%), we suggest estimating sex using the univariate function of the handprint breadth that had the highest accuracy and smallest sexing bias. When results of the present research are compared to recent research conducted on the direct hand measurements, it is evident that they follow a similar pattern of sexual dimorphism and show a similar relative contribution of variables in sex estimation models [6,10]. Specifically, the most important variables were hand breadth, palm length, and hand length, which is probably one of the reasons why some studies use only those variables [12].
As a study limitation, we should consider that the sex estimation standards were developed on the limited sample size, which could be attributed to the type of material used in the study. Specifically, handprints are sensitive biometric data that can be misused, which is probably why many refuse to participate in the research. Due to the convenience sampling strategy, we also could not claim that sample uniformly included people from all regions, even though we collected most of the samples at the second largest university in Croatia that is attended by the students from all over the country. For this reason, the sample should be furtherly extended to target different regions, or the method should be validated on an independent sample of known regional structure. It is also important to stress that, like in previous research of this type [4,6], handprints were taken in controlled conditions that are not always found in real-life crime scenes. For example, a hand can be in a different position when leaving a handprint; it can be arched, bent, loosened, etc. [1], and the amount of hand pressure could also be different [6]. Additional factors that could impact shape or dimensions could be hand movement to the latent print, as well as visibility of the latent print and print developing method. So, the results of the study should be implemented with caution.
In the present study, we developed statistical models that could be used to classify sex from complete or partial handprints. However, as crime scene can often contain even smaller segments of the prints left and recovered in different conditions, the present research could not cover all the possibilities. Therefore, we plan to extend our research to the others part of the hand, e. g., fingerprints [23] and other isolated palmar regions [24,25], and to test methods in real-life situations considering the factors that could impact the features of the print.
Provenance: Submitted. Based on the master's thesis by Andrea Kolić "Estimation of stature and sex from handprint dimensions in Croatian population", the first author on this article, at the University of Split Department for Forensic Sciences.
Peer review: Externally peer reviewed.