Dropout issues and differences in results of Euroforgen Exercise, using LRmix vs LRmixStudio

[Carlos Baeza Richer]

Hello,

I am Carlos Baeza Richer from the School of Medicine in the Complutense University, Madrid

I wanted to ask you some questions about the LRmixStudio and some differences I have found in the performance and the results between Lrmix and LRmix Studio. Especially regarding the dropout results and estimations. I have to give a lecture, the 28th of this month, to members of the law enforcement and I want to be sure I got right.

The first thing is that in some cases the programme gives a range of dropout that goes from 0.00 to another value, in cases where you need a value over 0 to explain the data under any of the two hypothesis. I suppose it is a problem of precision, but what value should I chose when that happens? Especially in the cases when the lower end of the range is the one most conservative. For instances this occurs in the Case 1 you gave us in the course.  It also happens in the first exercise of Euroforgen, where before we obtained a range of 0,01 to 0,17 now it is 0,00 to 0,17. I know it is not a big difference here but what is recommended to do here? I imagine, chose the smallest value 0,01?

The other issue is the option of fixing the dropout of a component and varying another. Here I have found some results quite different (especially in the dropout range) that can even change the conclusions of the Euroforgen collaborative exercise 2. I made 3 tests:

a) All components vary dropout (the conditions of the original exercise)

b) Suspect fixed 0,1 dropout and vary the rest. (I made this test because some people wrongly leaves this value (0,1) since it is the default value)

c) Suspect fixed 0 dropout and vary the rest.(This is a sensible thing to do, since all the alleles of the suspect are in the evidence)  

Results:

Range droput (in bold the value I would chose): a) 0,19 to 0,45   LR: 219511,597

                                                                    b) 0,29 to 0,78   LR: 34632,9965

                                                                    c) 0,45 to  0,86. LR: 13146, 3955. This last result is quite shoking to me, because at first sight i would not expect so                                                                         much dropout in the victim.   

The LR results may be not very different but the performance test makes me doubt a lot. I made 1000 iterations and I show the maximum value (log10)           

a) -0,479

b) 1,458

c) 2,049 vs 4,119 the actual value.

Giving the results of the performance test especially in case C, what can we make of this? Wouldn't it change the conclusion vs a?. What parameters would you recommend to chose for the dropout estimation test?

Thank you a lot in advance and sorry for the long and dense email...

Kind regards,

Carlos