I am trying to test for the effects of environmental variables (i.e. precip, mean temp, elevation) on genetic differentiation of a resident songbird, but I am unsure about the GESTE output. When running all factors individually, the model involving the constant always has the highest probability (i.e. constant = 0.502, factor = 0.492). When I carry out different scenarios with a subset of factors, the post probs range between 0.1 and 0.4 for the different models. Is there a cutoff where the model with a certain post prob or higher is significant? Any help on how to interpret GESTE results would be great! Thanks!
I am sorry for my answer here (Do not know if is really an answer!). I like to compare models using a one taking out method. Instead to use the posteriors I like to use de predictive probabilities - probabilities of a new observation -. Take the predictive distribution of the models and take one observation out of the sample. Use the remaining n-1 to adjust the built the predictive probability. Built the credible interval (or if need to be classification in groups classify it). Take each time one sample observation and repeat this to all sample points. Consider than the frequency of times you do correctly the prediction. The model with higher frequency is the better obviously. Recall that the predictive probability is the marginal f(x)=integral{f(x|teta)p(teta)}, f is the sample probability and p is the prior.
As Mr. PEREIRA, I don't if is an answer, but if the question is to know the confidence level to validate a posteriori, it depends on the nature of the random process associated with the variable that we want to correct. In the case of a Poisson process and in the context of engineering, if the level of correlation return of feedback on which to base correction (for the marginal integral presented by Mr. Pereira) is 75% or Moreover, we believe that this is "interested."
These website I believe will be helpful to you:
http://heidi.chnebu.ch/doku.php?id=geste_bayescan
http://www-leca.ujf-grenoble.fr/moyens-techniques/logiciels/article/population-genomics-software?lang=fr
This method is discussed in:
Foll, M., and O.E. Gaggiotti, 2006. Identifying the environmental factors that determine the genetic structure of Populations. Genetics 174 : 875-891.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Probability
- Probability Theory