Just PMI Estimation Research_Special Release_013

Just Science

Dr. LaMotte and Dr. Wells discuss their NIJ funded research on PMI and how it can help crime scene investigators. This is their abstract submitted for the R&D Symposium: "To our knowledge an estimate of time since death is almost never accompanied by the kind of mathematically explicit probability statement that is the standard in most scientific disciplines. This has been a problem both for death investigation casework (and court testimony) and for research, because scientists have not known how to design decomposition experiments to provide adequate statistical power for postmortem interval (PMI) estimation. We have been developing methods for calculating statistical confidence limits about a PMI estimate based on either continuous quantitative or categorical data. The examples we present are from forensic entomology, but the approach is suitable for any postmortem variable. To do this we extended and adapted the time-tested statistical method of inverse prediction (IP, also called calibration) to the PMI estimation setting. Methods to produce valid p-values for this process are known for single, quantitative y and x that follow a linear regression relation and with y having constant variance. Some exist for multivariate y, but only for settings where y has constant variance. Many measurements used for PMI estimation do not fit these criteria. The current project builds on earlier work in which we developed IP methods for non-constant variance of a single, quantitative y (e.g. estimating carrion maggot age using a single size measurement, Wells and LaMotte 1995), and in which we developed the first ever method for IP based on categorical data (e.g. estimating PMI based on carrion insect succession, LaMotte and Wells 2000). One possible barrier to the adoption of these new inverse prediction methods by researchers and death investigators has been that they are not implemented in statistical software packages. In this presentation we will show how IP using categorical data can be done by simply reading a table. Concerning quantitative data we will show how inverse prediction of PMI can be performed using statistical analysis software already widely available for general linear mixed models, where the statistical theory and methodology are well-established. We will show how flexible models using polynomial splines can be fit for both the means and variance-covariance matrices, and how to use dummy variables over a grid of values of x to get the p-values required for confidence sets automatically. Attendees familiar with mixed models and their applications will be able to implement these methods in standard statistical packages. Statistical Methods for Combining Multivariate and Categorical Data in Postmortem Interval Estimation 2013‐DN‐BXK042 Lynn R. LaMotte,1 and Jeffrey D. Wells2 1Biostatistics Program, LSU School of Public Health To learn more visit www.ForensicCOE.org