Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Determining Best Predictors


From   "Ángel Rodríguez Laso" <angelrlaso@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Determining Best Predictors
Date   Fri, 26 Dec 2008 13:45:43 +0100

Unless you have a theoretical basis to check directly if there are
turning points in your data (for example, "a determined injury is
produced by a deterioration in a physiological function which becomes
apparent at age whatever"), I would start by checking if age and
experience are associated with the risk of having an injury in a
linear way. After that, you can check if age or experience become
important at a determined point, as you propose. This would be
checking for non-linearity in your model. It can be achieved through
different methods that are well explained in Gould's "Linear splines
and piecewise linear functions"

http://www.stata.com/products/stb/journals/stb15.html

and in any textbook on logistic regression (see for example
"assumptions" in Logistic regression:

http://faculty.chass.ncsu.edu/garson/PA765/logistic.htm

Answering to the first part of your email, a more difficult thing
would be to model the interaction between age and experience. It can
be done by multiplying both quantitative variables to create an
interaction term that is then introduced in the model. The coefficient
of this term would be more difficult to interpret. It would be easier
if one of the two variables (for example experience) could be
dichotomised after checking at what point it becomes important. Then
you could use an interaction term:

experience (dichotomous)*age(quantitative)

and see if it is significant in your model.

HTH

Angel Rodriguez-Laso


2008/12/25  <jasonm@ucla.edu>:
> Sorry, this is a vague question. I'm wondering if anyone has any thoughts on
> this or can point me in the direction of some material which discusses my
> question.
> I first just want to look at a couple of situations and then go from there.
> I want to see who has a higher risk for injury:
> 1) someone who is older and just beginning a job vs. someone is younger and
> just beginning a job
> 2) someone who is younger and has lots of experience vs. someone who is
> older and has lots of experience
> To start off, I just want to use two continuous variables (age and
> experience) and see how they operate in these two situations for predicting
> injury risk. For 1) I also want to see when age starts to become important
> (if it actually ever is) and for 2) I alsoe want to see when experience
> starts to become important (if it actually ever is).
>
> *
> *   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index