Table 1.
The four continuous models tested in this study and their modeling assumptions.
Table 2.
Summary of the LRs and p-values for the true contributors to the samples in the testing set.
Table 3.
Standards for verbal expression of likelihood ratio (Association of Forensic Science Providers, 2009).
Fig 1.
Variation of the LR within and between models.
The verbal expression corresponding to the LRs from five runs for the true contributors to all the samples in the testing set is shown. For each set of samples (i.e. 1-person, 2-person, and 3-person samples), samples are numbered starting from 0 in increasing order of the total template mass. The samples that resulted in inter-model variation are as follows: sample 9 in one-person profiles; samples 1, 3, 12 (Contributor 2), 3, 6, 7, 26 and 27 (Contributor 1) in two-person profiles; samples 1, 16, 25 (Contributor 3), 11, 12, 15, 20, 23 (Contributor 2), 2, 3, 5, 8, 9, 16, 18, 20, 22, 23, 24, 25, 26, 29 and 30 (Contributor 1) in three-person profiles.
Table 4.
Intra-model variation in the LR.
Fig 2.
LR verbal expression levels and the LRs and p-values from the four models for the true contributors in a 1:4:4, 0.28ng 3-person mixture.
Model D resulted in higher LRs for Contributor 1 (starting template mass: 0.03ng) than the other models—it resulted in an LR < 1 and showed ‘Weak’ support for the prosecution’s hypothesis. Models A, B and C resulted in LRs < 1 for Contributor 1. The p-values for Contributor 1 from all versions ranged between 10−1 and 10−4. Contributors 2 and 3 both had ‘Extremely strong’ verbal interpretations from all models and their p-values had an upper bound of 10−9.
Table 5.
Inter-model variation in the LR.
Fig 3.
LR verbal expression levels and the LRs and p-values from the four models for the true contributor in a 0.03ng 1-person sample.
The contributor had a LR < 1 with models A, B and C which consider forward stutter and had an LR > 1 under model D, which does not incorporate forward stutter. This occurred due to 100% forward stutter ratio at one locus, which had a low probability.
Fig 4.
EPG of locus D16S539 in the 0.03ng 1-person sample with the LRs shown in Fig 3.
Allele 11 belongs to the genotype of the contributor and has a height of 6 RFU. Allele 12 (in the forward stutter position) also has a height of 6 RFU.
Fig 5.
LR verbal expression levels and the LRs and p-values from the four models for the true contributors in a 1:19, 1ng 2-person sample.
Contributor 1 had an LR < 1 under model D but had an LR > 1 under other three models because inclusion of forward stutter gave a better explanation for the heights of the peaks at reverse and forward stutter position at two loci, since reverse stutter alone was not sufficient.
Fig 6.
LR verbal expression levels and the LRs and p-values from the four models for the true contributor in a 0.03ng 1-person sample.
Models A and B (normal noise distribution assumption) had LRs lower than 10−7 while Model C (lognormal noise distribution assumption) suggested ‘Strong’, ‘Moderately strong’ and ‘Moderate’ interpretations. LRs for Model D, fell both above and below 1 (log10(LR)s ranging from -2.81 to 2.18).
Fig 7.
LR verbal expression levels and the LRs and p-values from the four models for the true contributors in a 1:1:1, 0.047ng 3-person sample.
Models C and D (constant mixture ratio assumption) had LRs < 1. Model B (varying mixture ratio assumption) suggested ‘Weak’ interpretation. In Model A, the LRs fell both above and below 1 (log10(LR)s ranging from -0.03 to 0.11).