Reliability Assessment of Current Methods in Bloodstain Pattern Analysis
Terry Laber, Paul Kish, Michael Taylor, Glynn Owens, Nikola Osborne, James Curran
ABSTRACT
To date there have been relatively few error rate or validation studies in BPA and none has investigated the role that contextual information might have on analysts’ conclusions. This study was designed to produce the first baseline measure of reliability for the major BPA method of pattern recognition. The approach used was designed to help define the upper limit of pattern classification reliability by focusing attention on method reliability rather than analyst competency.
A panel of experienced bloodstain pattern analysts examined over 730 patterns in two phases of the study, one focussing on three rigid non-absorbent surfaces (painted wood, wallpaper and chipboard) representing commonly encountered crime scene surfaces and the other on three fabric surfaces (cotton sweatpants, polyester trousers and demin jeans) representing clothing. Six different pattern types, blunt force impact spatter, firearms (back and forward) spatter, cast-off, satellite stains from a drip pattern, transfer and expirated, were used over the two studies. The extent of available pattern, the nature of the substrate and the type of contextual information (positive, negative and neutral bias) were varied in a balanced experiment designed to determine the effect of these variables on pattern classification accuracy. As a small adjunct to the main focus on pattern recognition, a set of superimposed bloodstains prepared on non-absorbent rigid surfaces was also included for sequence of events determinations.
Where a bloodstain pattern classification was made, either by choosing a single pattern or by nominating more than one, 13.1% of these classifications did not include the correct pattern type for the rigid surfaces and 23.4% for fabric surfaces. These can be considered the first approximations of overall error rates for the pattern classification method. Some patterns were more reliably classified than others. In particular the error rate was 4% for expirated patterns on rigid surfaces and 8% for impact patterns on fabric surfaces. The highest rates of misclassification were 59% for satellite stains from a drip pattern on fabric surfaces and 19% for impact patterns on rigid surfaces. Generally speaking where the pattern was more difficult to recognise (e.g. less pattern available or a patterned substrate), analysts became more conservative in their judgment, choosing the inconclusive option.
Study results showed that where a scenario was offered that deliberately pointed analysts towards the correct classification, the proportion of misclassifications that resulted was significantly lower (8% rigid surfaces, 14% fabric surfaces) than that observed for patterns with neutral scenarios (11% rigid surfaces, 26% fabric surfaces). This is an example of the well- known phenomenon of confirmation bias. Where a scenario was offered that deliberately pointed analysts towards an incorrect classification, the proportion of misclassifications that resulted was significantly higher (20% rigid surfaces, 30% fabric surfaces) than that observed for patterns with neutral scenarios (11% rigid surfaces, 26% fabric surfaces).
The supplementary study on superimposed patterns showed that, for the current sequencing methods, the chances of incorrectly concluding the order of deposition in a spatter/transfer pattern combination is approximately 12% where spatter stains are deposited on top of transfer stains and 17% for the reverse sequence.
The implications for practitioners and agencies involved in BPA are discussed.