Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: st: xtreg - continuous or discrete time

 From Maarten Buis <[email protected]> To [email protected] Subject Re: st: xtreg - continuous or discrete time Date Wed, 17 Aug 2011 09:19:13 +0200

```On Wed, Aug 17, 2011 at 2:49 AM, Ricardo Ovaldia wrote:
> I have a longitudinal data on children measured at
> ages 5, 10, 15 and 20. They were all measured
> within two weeks of their birthday.
> When using -xtreg-, I get very different results
> depending of whether I use time as a continuous or
> categorical variable.
<snip>
> What I do know is which is the most appropriate
> parametrization of time. Or how to decide.

The way to decide is to consider what it would mean when you enter
time/age as a continuous variable. In that case you are stating that
going from 5 to 10 years will have the same effect as going from 10 to
15 years, which will have the same effect as going from 15 to 20
years. Admittedly, these steps are the same in the sense that they are
all 5 years apart, but if we think of a child growing up, than we
would expect these steps to have wildly different implications for
many aspects that we could be studying. Growing up is just not a
linear process.

I don't know what you are studying, and even if I did I am probably
not an expert on that issue, so I cannot give you concrete advise on
what is right. However, I can make your live even harder by giving you
a third option: You can enter the main effect of time/age as a
discrete variable and enter the interactions as continuous variables:
growing up may be wildly non-linear, but differences between groups
may change (approximately) linear over time. Technically, you could
also enter the main effect of time as linear and the interactions as
discrete, but in my experience the interaction terms tend to be rather
sensitive to misspecification of the main effect. So I use as a rule
of thumb that the main effect can be more flexible (e.g. categorical,
or spline, or polynomial) than the interaction effect but not the
other way around. But that is just my rule of thumb.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------
*
*   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/
```