Censoring

[WallSt]

Is there any way to work with censored data (e.g. right censored data) where the right limit changes for each data sample?  The example in the manual assumes a single fixed right limit.  

I also looked at the Stan implementation of the "mice" example.  This example in openbugs/winbugs has multiple right limits in the data but the stan example looks like it has assumed all right limits are set at time=40, so I think there is a mistake in the stan version of the mice example.

I know you can use the direct implementation using the lp__ <- lp__ + ....  methodology but this mean calculating the log probability of the distribution which can get tricky.

[Jiqiang Guo]

Thanks for pointing out the problem in our mice example.  We should fix it soon. 

As we cannot declare a vector parameter with different bounds, one way is that you have to declare them individually. 

For the censored data, the typical way is not to model the censored data using the cdf, which is illustrated in the kidney example.  I am not sure we have weibull_cdf now, but working on lp__ Is not that tricky.  In the kidney example, the censoring points are different. 

Jiqiang

[Jiqiang Guo]

I meant to say to model the censored data using the cdf. 

[Bob Carpenter]

We're releasing Stan 1.0.1 early next week
that will:

-- fix sign error in denominator for truncation

-- declare the cdfs (including Weibull) so they're
accessible from Stan

Until then, censored models won't work, nor will
truncations or attempts to call _p (which we're
going to rename to _cdf for clarity).

- Bob

[Jiqiang Guo]

The mice example in Stan is corrected. 

Jiqiang

On Sat, Sep 8, 2012 at 9:42 AM, WallSt <how...@zail.org> wrote:

[Daniel Lee]

The corrected mice model will be available in v1.0.1.