Weighting observations and arviz.loo

[realkrantz]

Hi, @avehtari. Thanks for your explanation. My question is for those cases that the weights are part of the generative story.

For these cases, that the use of weights in the model block is reasonable, I would appreciate a rationale for:

@avehtari

you don’t weight win_chance_logit[n] .

Because, like @Calen_Pennington, I am not certain which approach is more appropriate here:

@realkrantz

I want to do predictions via generated quantities block. Should the weights be applied again in the generated quantities block in addition to the likelihood in the model block?

   model{
        ...
    	target +=  weights[i] * binomial_lpmf(...);
        ...
        }

    generated quantities{
        ...
        y_pred[i] = weights[i] * inv_logit(...);
        ...
        }

or not?

    model{
        ...
    	target +=  weights[i] * binomial_lpmf(...);
        ...
        }

    generated quantities{
        ...
        y_pred[i] = inv_logit(...);
        ...
        }

Thanks in advance for any help.