v0.2.3 Results of sample_posterior_predictive with kind='mean'

[johnmdusel]

Greetings,

I am looking for help understanding the output of sample_posterior_predictive with the option kind='mean'. My goal is to access the hit rates of the Bernoulli distributions that describe the model's responses. I ran this command:

idata_postpred = model.sample_posterior_predictive(
    idata=idata,  # mcmc chains from previous run of model.sample()
    data=df_pc,  # empirical p(c) per subject per condition
    inplace=False,
    kind="mean"
)

and got an InferenceData instance with a field "rt,response_mean".

In my case idata_postpred.posterior["rt,response_mean"].values are between -0.639 and 1.393. Can someone help me understand what these values are? I'd expect them to lie between -1 and +1 if they were mean responses. On the other hand, I'd expect them to be positive if they were mean response times.

I would prefer to directly access the hit rates that were found from MCMC sampling, rather than computing hit rates from samples obtained with kind="pps".

Respectfully,
John Dusel