Estimation of sex in forensic examinations using logistic regression and likelihood ratios
2020
Verma, Rajesh | Dewett, Kewal Krishan | Rani, Deepika | Kumar, Ajay | Sharma, Vishal | Shrestha, Rijen | Kanchan, Tanuj
Estimation of sex is an essential parameter which helps in establishing the biological profile of the deceased in forensic and archaeological examinations. There are two types of methods useful in the estimation of sex –morphological and morphometric. While morphological analysis is subjective and involves visual inspection of the dimorphic traits, morphometric analysis involves statistical analysis and comparison of the measurements to develop a probabilistic estimation of sex. Most studies estimate sex using Discriminant Function Analysis (DFA) that has many restrictive conditions, however, the logistic regression does not have these restrictions and is suitable for a two class problem like estimation of sex.In the present investigation, an attempt has been made to estimate sex by developing logistic regression equations using a cross-sectional sample of 344 young adults (172 females and 172 males). Eight anthropometric measurements were taken on the participants with standard procedures. Hand breadth shows the highest sexual dimorphism and is the best classifier of sex, followed by foot length and hand length. A combined regression model using these three variables has also been proposed that can correctly classify sex with an accuracy of 90.1 % and 91.3 % for males and females, respectively. The likelihood ratio is calculated from the normal probability density curve constructed from the mean and standard deviation for various predictors obtained for the population. It has been concluded that the likelihood ratios corroborate the classification well and the likelihood ratio with its threshold at '1′ can be used as a classifier when there are only two classes. The performance of the likelihood ratio classifier in the present study has been found to be equal to the logistic regression classifier.
显示更多 [+] 显示较少 [-]