Message-ID: <508539CA.6050906@mayo.edu>
Date: 2012-10-22T12:19:22Z
From: Terry Therneau
Subject: Expected number of events, Andersen-Gill model fit via coxph in package survival
In-Reply-To: <CAKO_XCMSAVFFXqjkjEVzU-RgL4QakuCEOvhMHhm==crcBD=2Yg@mail.gmail.com>
Therneau doesn't know the answer either.
The predictions are positively correlated since they all depend on the same beta-hat from
the original model. The same will be true for any model: logistic, poisson, linear, ...
Terry T
On 10/20/2012 06:06 PM, Omar De la Cruz C. wrote:
> I have a follow-up question (for either Dr. Therneau, or anybody who
> might know).
>
> sum(zz) (see below) estimates the number of events for the cohort.
> Now, how can I compute a confidence interval for sum(zz)? Or a
> standard error?
>
> My obvious choice, square root of the sum of the squares of the
> standard errors for zz provided by predict.coxph, turns out to be too
> small.
>
> Thank you for any suggestions.
>
> Omar.
>