What statistical methods are appropriate to comparing e.g. VOT across different groups?

Question

As part of forensic linguistics practice, I must assess the relative age of multiple individuals based only voice. One crucial variable is VOT, which is known to decrease with age.

So, let's say hypothetically that I classify the individuals into two groups: one with a longer VOT than the other. However, I also want to do some statistical analysis to make the nuance of my results clearer by the jury.

What statistical techniques are optimal to these ends, in the case of (non)-normally distrubted data - how will normality be characterized? ANOVA, T, Kruskal-Wallis, etc.

Answer 1

I don't think you've given enough detail about what you want to do with 'some statistical analysis', but I'll take a stab. I am not a stats expert, so these rough ideas will need to be checked by someone who is; I present them for you to know what to ask about.

If you want to be able to say 'it's 85% likely that this person is from the longer VOT group', that sounds like a mixture model approach (there are many mixture model packages in your stats environment of choice). But you need enough data to figure out what the groups are, and you're collapsing what you described as a continuous variable into a 2-categorical variable.

Maybe a better approach would be a mixed-effects linear model (using e.g., LME4). With VOT as your output variable, this approach would allow you to consider the relative contributions of various factors: speaker_age, but also gender, language background/dialect, control factors like vowel/phonetic context/word_frequency, etc., and (crucially) estimate the interaction of these features. Using the resulting model, you could make predictions and construct confidence intervals for those predictions.