Stature estimation in forensic examinations using regression analysis: A likelihood ratio perspective

Abstract For the identification of decomposed dead bodies and skeletal remains, estimation of stature is of utmost importance. In estimating stature from the percutaneous bone lengths or skeletal remains, regression analysis has been the method of choice. Technically, as the regression line is just the best fit to the data, the estimate of stature differs from the actual stature, and this difference is reported in terms of the root mean square error (RMSE). The estimate using regression analysis is probabilistic in nature and there is a confidence interval about the prediction. This interval is mostly not reported in various studies, neither are the probable errors in estimation. In the likelihood ratio approach, absolute “matching” or “identification” is not possible and therefore, only the quantitative strength of evidence (strength of association in the present context) is given. The sample for the present study comprised of 344 young adults (172 females and 172 males) residing in the Shimla city of Himachal Pradesh State of north India. The data have been generated from anthropometric measurements of seven percutaneous bone lengths besides stature. The present study reports the regression equations for stature estimation as traditionally calculated, but additionally, it reveals the determination of the confidence intervals of the regression line as well as the confidence intervals of the prediction of stature. The study also shows that for the likelihood ratio estimation, the regression analysis needs to be flipped, i.e. instead of using regression of stature versus bone length; the regression of bone length versus stature should be used. Practical application of finding the likelihood ratio of a recovered bone to be associated with two missing persons of known stature has also been discussed in the current study. This approach is further extended to a general scenario, where the likelihood of a person of known stature is evaluated with reference to any other person chosen at random from the relevant population. The likelihood ratios arrived at, have been found to be generally low, meaning that the strength of association of stature and percutaneous bone length is not very high, and combining likelihood ratios from different methods of personal identification may be needed.