It’s definitely #3. The posterior predictive has no ability to do “new inferences”.
The only thing it can do is to sample new values that follow a standard closed form distribution given the parameter values in the trace.
What do you mean by:
It’s definitely #3. The posterior predictive has no ability to do “new inferences”.
The only thing it can do is to sample new values that follow a standard closed form distribution given the parameter values in the trace.
What do you mean by: