Hello all!
It's been at least six years since I've been on here, but have been trying to learn a new series of techniques through Stata's Bayes package.
First, this seems like an enormous achievement to have such a sophisticated and flexible set of commands available in Stata. Has there been broader adoption of the Bayes package yet? Are there any strong reasons why, especially for likelihood modeling that is more predictive in nature and less causal, one would not use these routines?
Second, and more specifically, I have a setting where I am modeling events. There is some measurement error in whether an event is observed, but I impute zeros for any time the event is not observed for the group, under the assumption the measurement error can be treated like a random variable through a random effect + Bayes Poisson panel data model. My main question may come across overly general, so please point me to any instructional references: what's the best way to interpret all the results and diagnostics relating to: (a) coefficients, (b) model fit, (c) predictive power, and (d) other model parameters? My understanding is that (a) is similar to a Poisson or other non-linear model in the traditional frequentist sense, and I have gone through the manual covering (b) through (d) too, but finding it hard to get comprehensive enough answers that lead to mastery from the manual alone; perhaps there are lecture notes that link code/model output with explanation somewhere?
Many thanks again for your time and the hard work making the community thrive,
It's been at least six years since I've been on here, but have been trying to learn a new series of techniques through Stata's Bayes package.
First, this seems like an enormous achievement to have such a sophisticated and flexible set of commands available in Stata. Has there been broader adoption of the Bayes package yet? Are there any strong reasons why, especially for likelihood modeling that is more predictive in nature and less causal, one would not use these routines?
Second, and more specifically, I have a setting where I am modeling events. There is some measurement error in whether an event is observed, but I impute zeros for any time the event is not observed for the group, under the assumption the measurement error can be treated like a random variable through a random effect + Bayes Poisson panel data model. My main question may come across overly general, so please point me to any instructional references: what's the best way to interpret all the results and diagnostics relating to: (a) coefficients, (b) model fit, (c) predictive power, and (d) other model parameters? My understanding is that (a) is similar to a Poisson or other non-linear model in the traditional frequentist sense, and I have gone through the manual covering (b) through (d) too, but finding it hard to get comprehensive enough answers that lead to mastery from the manual alone; perhaps there are lecture notes that link code/model output with explanation somewhere?
Many thanks again for your time and the hard work making the community thrive,
Comment