Netherlands Forensic Institute; VU University Amsterdam.
Forensic Sci Int Genet. 2022 Jan;56:102592. doi: 10.1016/j.fsigen.2021.102592. Epub 2021 Sep 10.
When we compare a DNA mixture profile to a single person of interest, there are often just two competing explanations considered, and the comparison of how likely these are to lead to the observed mixture is summarized by a likelihood ratio. However, in more complex cases this does not suffice, e.g., when there are multiple persons of interest. One can then compute several likelihood ratios, corresponding to several pairs of hypotheses, and subsequently decide which one(s) to report. This may lead to the computation of a rather large number of such likelihood ratios. In this article we advocate a systematic approach that starts by describing all relevant hypotheses. For each hypothesis, we then compute its likelihood (i.e., the probability to see the genetic data if the hypothesis is true). Based on the likelihoods of all considered hypotheses, one can then make a summary of the findings to report. This may be on the level of the considered hypotheses and/or with likelihood ratios per person of interest. We illustrate with several examples how this approach assists interpretation. The likelihoods summarize how the trace can help to distinguish between the considered hypotheses, in the sense that they transform the prior odds on them into posterior odds, without having to assign prior probabilities on the hypotheses for the calculation of the likelihoods themselves. On the other hand likelihood ratios (LR's) for individual PoI's cannot be obtained without these priors. In many cases these LR's will be quite insensitive to the choice of prior probabilities but in other cases they will be; we give examples of both.We argue that the table of likelihoods of the considered hypotheses is a more natural analog of the LR provided in the simple case with one PoI and two considered hypotheses, compared to the computation of a LR per PoI. We end with a discussion of the choice of prior probabilities, of the existing recommendations for this situation, and on reporting.
当我们将 DNA 混合物图谱与单一的感兴趣个体进行比较时，通常只考虑两种相互竞争的解释，并且通过似然比来总结这些解释导致观察到的混合物的可能性。然而，在更复杂的情况下，这并不足够，例如，当存在多个感兴趣的个体时。然后，人们可以计算几个似然比，对应于几个假设对，然后决定报告哪一个或哪几个。这可能导致计算相当大量的似然比。在本文中，我们提倡一种系统的方法，首先描述所有相关的假设。对于每个假设，我们计算其似然率（即，如果假设为真，观察到遗传数据的概率）。然后，基于所有考虑的假设的似然率，人们可以对要报告的结果进行总结。这可以是在考虑的假设层面上进行，也可以是针对每个感兴趣的个体的似然比进行。我们通过几个例子来说明这种方法如何辅助解释。似然率总结了痕迹如何帮助区分所考虑的假设，因为它们将先验概率转换为后验概率，而无需为似然率的计算分配假设的先验概率。另一方面，如果不使用这些先验概率，则无法获得针对单个感兴趣个体的似然比 (LR)。在许多情况下，这些 LR 对先验概率的选择不太敏感，但在其他情况下则不然；我们给出了这两种情况的例子。我们认为，与为每个 PoI 计算 LR 相比，考虑的假设的似然率表是更自然的模拟，而简单情况下只有一个 PoI 和两个考虑的假设。最后，我们讨论了先验概率的选择、这种情况下现有的建议以及报告问题。
