Meeting 21 January 2014

### Issues discussed:

- Puzzle about whether UAF in CGLM model makes the weight of evidence (WoE) more or less incriminating;
- Are the difference with and w/o UAF always so negligible?
- MLE is problematic, do a Bayesian approach. We have estimates of some parameters from experimental data (see Puch_solis; Buckleton for example) which could be used as informative priors. Use uniform prior on $\phi$ and integrate out?
- Extend to IBD and subpopulation: the challenge is the interaction between the cross-marker latent variables and the maximisation. Can G&M sec 3.4.2 ideas be used here?
- Extend to allow for paternity testing or other specific identity hypotheses.
- Likelihood always increases with increase in number of unknown contributors. (Nested model and using MLE) How to set a bound. Idea (
*Peter*) try get a**bound for infinite number of contributors.** - Multivariate UAF model.
- Considering the LR as random and considering LR disribution issues (as in Corradi, Ricciardi)
- In past we discussed issues of dependent markers e.g. Y haplotype markers

### In progress/done

**P**Extend DNAmixtureUAF to treat more than two individuals $I >2$.*Done but there are bugs***J**Example 5 Puch-Solis with $\theta=0.02$ so $M=(1-\theta)/\theta=49$;*Done***J**Check large M we converge to vanilla results.*Results look weird - check!*

### To Do List:

**P & J**Recheck Beta-Binomial algebra.**J**Check Hugin net for two actor examples.**J**Try UAF for two traces. Find suitable example with 2 actors at least one in common in both traces. Check program gives correct result when actors in two traces are different (independence case so likelihoods multiply).**P**Try CLM model with MLE, to try to distinguish whether reduction in sensitivity to UAF is due to model or method of inference