The use of Bayesian analysis is on the rise and is widespread across different disciplines including health, medicine, economics and other social sciences. The main difference of Bayesian analysis over the classical, frequentist method, is that a Bayesian assumes model parameters to be random and therefore having distributions while the frequentist method assumes model parameters to be unknown but fixed. Under Bayesian, one relies on a prior knowledge for the unknown and uses evidence from observed data (likelihood model) to get a posterior distribution of unknown parameter. The main advantages of the Bayesian approach is the ability to include prior information in the analysis. It also allows researchers to better handle repeated, missing, unbalanced and multivariate data.
Stata 14 offers 12 built-in likelihood models for different outcomes (continuous, binary, ordinal and count), the capability to write own likelihood models and the ability to use the 22 built-in priors and to take advantage of postestimation features. A new command, bayesmh, allows users to fit models using two different algorithms (Metropolis-Hastings algorithm, Gibbs algorithm or a combination of both) for univariate, multivariate and multiple-equations, both linear and nonlinear.
Upgrade your Stata license to learn more about Bayesian analysis with the all-new 261 page Stata Bayesian Analysis Reference Manual that is part of Stata 14’s extensive documentation.