problem with Bayes factors

[Ramona Paetzold]

Why do I get a Bayes factor that is a decimal point only (no values)? For example, in a comparison of 2 models, one value was 1.037 (base model), but the full model had a value that was just a decimal point. This is a simple Bayesian regression using the auto data provided by Stata. Below are the commands that I used.

sysuse auto
reg price mpg weight headroom
bayes, hpd : regress price mpg weight headroom
est sto full
bayes, hpd : regress price mpg weight
est sto base
bayesstats ic full base, bayesfactor

[Ben A. Dwamena]

See the methods and formula entry for bayesstats ic to understand use and interpretation of bayes factors. Bayes factors are used to compare two or more models applied to the same data using one as the baseline which latter will be assigned null bf value. The BF of two models is just the ratio of their marginal likelihoods calculated using the same dataset. BFs compute relative probabilities of how well each model fits the data compared with the base model. By default the first-listed model( in your case, full) is used as the baseline. if you had instead bayesstats ic base full , bayesfactor, the baseline will be base and will have a null(.) BF. You may also specify the baseline model using the basemodel() option. Hope this helps

[Jan Sadowski]

I don't have yet Stata 15. There is a non-standard likelihood function that is useful for survival analysis, an alternative to Cox model. I would like to code it in Stata. I wonder if I could next use Bayesian estimation in Stata for that likelihood model? Is it programmable in Stata 15?