It is defenitely wrong to use OLS for the time till treatement. This just fits a linear line, and you will predict negative times till treatement for suitably large or small levels of the explanatory variables. In other words you would predict for some persons that they receive treatment in the ER before they arrived in the ER, this clearly does not make sense.
In survival analysis you model how explanatory variables influence the probability of receiving a treatment at each point in time (more precisely: not the probability but the hazard). The probability of receiving treatement is likely to change over time. With survival analysis you can make assumptions about the functional form of this effect of time on the probability of receiving treatement (-streg-). One such assumption is called lognormal, and is equivalent to an OLS regression of the log of time till treatement on the explanatory variables if you do not have any censored observations. However, this is just one possible assumption out of many, and the choise of assumption should be guided by theory, by testing assumptions, and by looking at various types of residuals. The choice of assumption should not be guided by whether or not it can be estimated by OLS (not with the current computing power and software like STATA that takes care of all the hard part, except thinking).
Maarten
- ----- Original Message -----
From: "Zhu, Carolyn" <[email protected]>
Date: Tuesday, October 5, 2004 1:54 pm
Subject: st: survival analysis vs ols
> We are trying to estimate the effect of ER room conditions on the
> time it takes for patients to receive treatment A. All patients
> received treatment A some time their ER admission so we don't have
> any right censoring problem. I was just told that in this case
> survival analysis and OLS give the same results. Is that correct?
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
