st: re: logistic modeling with longitudinal data (NLSY)

[Matthew Aronson <maronson@rams.colostate.edu>]

In the context of my problem about an analysis of high school
completion as an outcome, and with a study sample where followup time
varies across respondents,

Brendan Halpin said:
<My first reaction to your question is that you have event history data,
<so you should use hazard rate models, but then I see you want to compare
<logistic regression with a hazard-rate model.

<My second reaction is to guess that your logistic regression will
<perform less well. But you should try it anyway, with a crude "period of
<observation" variable. The more you think about how to handle the years
<of observation, the more obvious a hazard-rate model seems.


Yes, this is in line with my thinking. The real crux for me is how to
most appropriately define "observation time" in the context of a
logistic model for whether respondents were ever detected to have
graduated high school. I'm wondering whether it's better to use: a)
"total observation time, regardless of whether R ever completed high
school," or b) "total observation time up to the point at which either
R completed high school or at which no further observations occurred
for the R"
*
*   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/