1. The evidence — synthetic mixed STR profile
Ten loci. Peak heights in RFU. Grey struck-through peaks have been called as stutter artefacts. Blue = an allele the suspect carries. Orange = the planted stutter trap. Dashed red = a suspect allele with no peak (possible drop-out).
12 1020 reported allele
11 102 called as stutter
18 330 suspect allele present
11 no peak suspect allele absent
2. Per-locus arithmetic
Minimum contributors = ceiling(number of called alleles / 2). Hp = the suspect plus one unknown person contributed. Hd = two unknown, unrelated people contributed.
4. Combined likelihood ratio
How it works — plain words
- STR mixture
- DNA from more than one person in one sample. The machine reports, at each locus, which allele sizes it saw and how tall each peak was (RFU). Taller peak = more DNA of that allele.
- Stutter
- A copying slip during PCR makes a small extra peak one repeat unit below a real allele. It is an artefact of the chemistry, not a person. Here any peak shorter than 15% of the peak one repeat above it is called stutter and set aside. The danger: a stutter peak can sit exactly where a suspect's allele would be, and then it looks like proof.
- Drop-out
- When very little DNA is present, a real allele can fail to show up at all. The drop-out probability (D, default 0.10) is how often we allow that to happen. If D = 0, a missing suspect allele is a flat exclusion; the larger D is, the more forgiving the model becomes.
- Likelihood ratio (LR)
- LR = (probability of seeing this profile if the suspect contributed) divided by (probability of seeing it if an unknown unrelated person contributed instead). LR = 50 means the evidence is 50 times more probable under the prosecution proposition than under the defence one. LR below 1 means the evidence favours the defence.
- Product rule
- The per-locus LRs are multiplied together, which assumes the loci are independent of one another.
- Prosecutor's fallacy
- Turning "the evidence is 1,000 times more probable if he contributed" into "there is a 1 in 1,000 chance he is innocent". Those are different questions. The LR is about the evidence; guilt depends on everything else known in the case as well.
- Verbal scale used here
- LR < 1 supports the defence · 1–10 limited · 10–100 moderate · 100–1,000 strong · above 1,000 very strong support.
The exact model this page computes