[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Modelling the passage of time: OLS vs survival analysis

From   n j cox <>
Subject   Re: st: Modelling the passage of time: OLS vs survival analysis
Date   Fri, 05 Oct 2007 11:36:17 +0100

In this situation, my guess is that you don't absolutely need
the -st- commands, but they could be tried out. I am fuzzy
about how bad it is to have some zeros.

I can't see that OLS -- or rather anything based on
an assumption of normal distributions -- is at all
attractive. Manifestly your response is constrained
to be zero or positive and right up against the limit of 0.

Alternatively, you could try fitting some non-normal
distribution directly, with and without your covariates.

Although some might be queasy about it, given
the nature of your data, -poisson-
might be a reasonable thing to try. More generally,
consider various kinds of -glm- or related commands.

Absent any censoring, what is best depends on what
kind of variability you have.


Christer Thrane

My dependent variable, gathered from a survey, is the variable "time
(measured in weeks) from from one started thinking about "X" until one
reached a decision about X." Also, everybody in the data reached this
decision. The data structure looks like this, and time ranges between 0 and
52 weeks (mean = 5.2 weeks):

id       gender        age        time     ...
1        0             30         7        ...
2        0             45         5        ...
3        1             36         0        ...
4        1             27         44       ...

and so on.

Q1: Is it theoretically meaningfull to think of this variable as suiteble
for survival analysis (SA), or is this "a job" for OLS regression

Q2: Provided that SA is meaningful, how should I set up the data in ths

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index