When using the Bayesian method with multiple semiologies, we can do the following:
treat each dataset separately. This is what is done in #222 . Concretely, using the SS dataset, it combines the semiologies (e.g. inverse variance) then using posterior-TS it combines the semiologies (using inverse variance), then takes the mean of the two estimates.
Instead, integrate the SS and posterior-TS estimates per semiology first, then combine semiologies using inverse variances.
Could argue for both: the second method gives a single estimate for each semiology, then combines. The first method queries each database separately for all the semiologies, then combines the results at the end.
When using the Bayesian method with multiple semiologies, we can do the following:
treat each dataset separately. This is what is done in #222 . Concretely, using the SS dataset, it combines the semiologies (e.g. inverse variance) then using posterior-TS it combines the semiologies (using inverse variance), then takes the mean of the two estimates.
Instead, integrate the SS and posterior-TS estimates per semiology first, then combine semiologies using inverse variances.
Could argue for both: the second method gives a single estimate for each semiology, then combines. The first method queries each database separately for all the semiologies, then combines the results at the end.
Which is best?