Re: st: stcox and weighting

[Steve Samuels <sjsamuels@gmail.com>]

A Google search of "risk set sampling" will turn up many references. You don't
cite sources, as requested in the FAQ, but I imagine that some of yours are at
http://gking.harvard.edu/category/research-interests/methods/rare-events.

Stata has a command -sttocc- to turn -stset- data into case-control data.
Commands -stcox- and -cloglog- (for grouped or discrete data) do hazard models
and take sampling weights. As the weights for risk set sampling will vary over
time, one Stata approach can be found at:
http://www.stata.com/statalist/archive/2012-02/msg00888.html, part of a thread
started at: http://www.stata.com/statalist/archive/2012-02/msg00858.html.

Steve

On Sep 7, 2012, at 7:48 AM, Robert Davidson wrote:

Dear Statlist,

I am modeling a rare event in which the outcome variable is a 1 less
than 1% of the time.  Several papers document the significant downward
bias on coefficients when modeling these rare events and suggest
sampling a proportion of the population of 0s and then either applying
prior correction or sampling weights.  I have done this with a simple
logit model and the results have changed predictably.  However, my
data is more appropriately modeled with a hazard model (1s are deemed
failures).  Is there a way to apply sampling weights in a hazard model
or does this not make sense given the mechanics of the model?  Is
there another way to properly deal with rare events modeling in the
context of a hazard model?

Thank you
Rob
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/