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

# Re: st: state trends

 From Maarten buis To statalist@hsphsun2.harvard.edu Subject Re: st: state trends Date Tue, 19 Oct 2010 08:23:39 +0100 (BST)

```--- On Mon, 18/10/10, sara borelli wrote:
> I am running a model with state and year fixed effects and
> I need to add state specific time trends. I created a dummy
> for each state and interacted it with the variable year,
> where year=1990,1991,1992,1993...
> The trend variables created are:  trendstate1
> trendstate2 ....trendstate51
>
> I also re-scaled the year variable so that
> yearnew =1 if year==1990; =2 if year==1991; =3 if
> year==1992  and so on. The trend variables in this case are:
> trendnewstate1...trendnewstate51
>
> I then regressed:
> xi: y  x  i.year i.state trendstate1-trendstate51
> and
> xi: y  x  i.yearnew  i.state trendnewstate1-trendnewstate51
>
> the coefficient of my X variable of interest is the same in
> both regressions but the coefficient on the state fixed
> effects are very different. It must be due to the rescaling
> of the year variable, but I am not sure why.

In the top regression you get for each state the deviation
from the reference state in the expected y if x = 0 in the year
0 (i.e. 2010 years ago). Which is probably a "not-so-mild"
extrapolation. In the bottom regression you get the same, but
now for 1989.

> I then tried to add a quadratic time trend. That is I
> interacted each state dummy with year2=year*year.
> Then I run
> xi: y  x  i.year i.state trendstate1-trendstate51
>
>    trendsquarestate1-trendsquarestate51
> but stata altogether drops the state fixed effects
>
> However, if I create   yearnew2=yearnew*yearnew and
> quadratic trends based on it and run
> xi: y  x  i.year   i.state
> trendnewstate1-trendnewstate51   trendnewsquarestate1-trendnewsquarestate51
>
> then stata does not drop the state fixed effects and
> estimates differ I am not sure what is going on or what
> I am doing wrong

The devil is in the detail, it probably depends on exactly
what variables are in the list trendstate1-trendstate51.

In simular situations I usually work these things out for
one state, and than start creating interactions. Actually,
trouble, and than go back and work it out step by step.

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/
```