All,
I'm analyzing some youth residential treatment data. So I know the admit and discharge dates and some covariates. Let's say I set up the model as a discrete time survival analysis. The results give me an intercept coefficient, coefficients for the time blocks (assume one month blocks) and coefficients for the covariates. Standard output, nothing special. But this seems to be where standard textbooks on survival end.
But now I have data for new kids and I want to predict expected time in treatment. So I pour the data into the regression equation. In normal distribution regression, the computed value is the desired number. With survival, not so, I think. The metric is log odds. The questions.
1. I think the expected time calculation is iterative in that, given the sum of the intercept and covariate coefficients-covariate value products, i must sum, in sequence, the time block coefficents until the cumulative sum is greater than 0. The expected duration is number of time blocks summed. Am I thinking correctly?
2. Must every cumulative sum be greater than 0?
3. I would like to put a 1 standard error confidence interval around each estimated duration. How do I do that computation?
The same data could be analyzed using a Cox regression model and since time is treated as continuous the estimated durations might have smaller confidence intervals. But, is such an estimated duration computation even possible using a Cox model (or other continuous time model)?
I know that survival analysis is not much discussed on this list but I'm hoping that there are some knowledgeable folks out there.
Gene Maguin
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
