Bias with trapezoidal AUC

[Leon-Alph]

I use an integration in my model (AUC is done with a trapezoidal method), but it seems to bias the likelihood estimator. I saw in a previous post that integration may interfere the sampling process.

How could I solve it? I tried to use ODE… but with some issues (see my previous today post :-)).

[Bob Carpenter]

The problem is when you get discontinuities. What you're
doing looks OK---it might not be very accurate at those
integrals or their derivatives.

I'm not sure what you mean by "bias the likelihood estimator".
Are you using optimization (for maximum likelihood estimation)
or sampling (for full Bayes)?

The bigger issue is that if you write code into the model
that affects the log density, then it'll have an effect on the
fit. If this is an after-the-fact calculation, then you can put
it in generated quantities.

On a more positive note, we'll be releasing our own integrator
soon.

- Bob

> On Apr 12, 2017, at 12:12 PM, Leon-Alph <leon.cor...@gmail.com> wrote:
>
> I use an integration in my model (AUC is done with a trapezoidal method), but it seems to bias the likelihood estimator. I saw in a previous post that integration may interfere the sampling process.
>
> How could I solve it? I tried to use ODE… but with some issues (see my previous today post :-)).
>
>
>

> --
> You received this message because you are subscribed to the Google Groups "Stan users mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-users+...@googlegroups.com.
> To post to this group, send email to stan-...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.
> <TKTD_AUC.stan>

[Virgile]

Thank you Bob,
Actually I realized that nothing was wrong with JAGS and the AUC. It was just on the last part of the code.
But as the AUC could be improved, I'm eager to have the stan integrator !
- Virgile