Bayesian Inference for Logit-Model using Informative and Non-informative Priors

Abstract

In the field of econometrics analysis of binary data is widely done. When the sample is small Bayesian approach provides more appropriate results on classical approach (MLE). In Bayesian approach the results can be improved by using different Priors. It is known that the binary data can be modeled by using Logistic, Probit or Tobit Links. In our study, we use Logistic Link. When modeling binary data, the shape of the distribution of Regression coefficients is no more normal. To find the coefficient of Skewness of Regression coefficients we use different Priors. It is observed that Haldane Prior provides better results than Jeffreys Prior, while Informative Prior performs better than the other Non-informative Priors for the data set under consideration.