|
Hello, This is more a technical question. I was looking for ways to sample from a T-Standard distribution with given mean and variance. Most programs offer only to modify the Degrees of Freedom, but not the mean nor the variance. My intuition, is that since the T-Distribution is a generalization of the Normal distribution, shifting it to the mean, and then scaling it times the Cholesky decomposition of the variance should be enough. Regards |
|
I seem to recall from Bayesian Data Analysis, that Gelman stated you can generate samples for the t-distribution by using a hierarchical model where you sample the variance from an inverse-chi-squared distribution: x_i ~ N(mu, V_i) V_i ~ Invchi^2(nu,sigma) You can sample from the inverse-chi-squared distribution by sampling from the chi-squared distribution X ~ chi2(nu) and transforming the results, V = frac{nu sigma}{X} . This is the only web resource I found find on sampling from the chi^2 distribution http://stattrek.com/probability-distributions/chi-square.aspx |
