Comparing \(f_3\) versus \(f_4\)
Suppose you want to test for evidence of admixture in some population of interest assuming some demographic model (we’re talking general case here, think of any model you’d like). In our exercises earlier, we’ve seen that both \(f_3\) and \(f_4\) can be used for this purpose—in fact, they were both originally developed as admixture tests. What’s more, they’re both used in the literature for for this purpose in practical research.
Imagine you need to convince a reviewer that one or the other statistic is better for your application, or you need to pick an appropriate statistic to test for evidence of admixture a study you’re currently working on and want to justify your decision using simulations based on the specifics of your assumed model (after all, all of our inferences are based on some assumptions).
Build a simulation pipeline using the slendr modeling building blocks and the simulation techniques you’ve learned in this workshop and find answers to questions such as:
Under which conditions would the \(f_3\) or \(f_4\) be more applicable as an admixture test? Is one of them more suitable for some particular kind of model or some specific model parameters than the other? Proportion of admixture? Time of admixture? Effective population size?
Is one perhaps more noisy than the other? Is one more sensitive to detect evidence of admixture in some situation (but perhaps not in another situation)?
My suggestion is to start with a baseline model which is as simple as possible and generalize it into a function with parameters whose effect you’d like to investigate—perhaps one of the general four-population demographic models from our \(f\)-statistic explorations would make a good starting point.