Re: st: Confusion about collinearity

[Richard Williams <richardwilliams.ndu@gmail.com>]

See comments below.

At 12:40 AM 12/2/2013, Yarbrough, Kevin T CADET MIL USA USMA wrote:
> I'm having a problem with collinearity in my 
> difference-in-differences model. I'm using Stata 12 for Windows.
>
>
>
> I'm using pooled cross section from a 5% sample of ACS data from 
> 2011 to 2011 across all 50 states and Washington D.C. I'm attempting 
> to analyze a policy implementation with the 
> difference-in-difference(DID) identification method. 25 states have 
> implemented this policy in different years. I'm trying to measure 
> the effect on income. From my knowledge, a correct use of DID should 
> result in a model: y=B0+B1(Treatment_group)+B2(post-policy year)+ 
> B3(interaction term of treatment_group and post-policy year).
>
>
>
> I began by creating a dummy variable for each year that equals 1 if 
> a state had the program in that year. For example:
>
> gen eitc00=1 if year==2000 &  (statefip==08 |statefip==11 | 
> statefip==17 |statefip==19 | statefip==20 | statefip==23 | 
> statefip==24 |statefip==25 |statefip==27 |statefip==34 |statefip==36 
> |statefip==41 | statefip==44 |statefip==50 |statefip==55)
>
>
>
> I followed that up with this code to replace the missing values:
>
> replace eitc00=0 if year==2000 & eitc00==.

If I am reading this code right, then eitc00 still = missing for all 
years other than 2000. That means the regression will only include 
data from 2000. Is this really what you want? How about this instead?

replace eitc00 = 0 if eitc00 == .

> I then created a dummy variable for years:
>
> gen yr02=year==2002

In 2000, yr02 = 0.

> And the interaction term between the two:
>
> gen eitc00yr02=eitc00*yr02

In 2000, the interaction term = 0. In all other years it is missing 
since eitc00 is missing for all years besides 2000.

> My regression is
>
> Reg lnincwage eitc00 yr02 eitc00yr02

In 2000, yr02 and eitc00yr02 both equal 0. All other years get 
dropped because eitc00 is missing for all other years.

> Which results in the interaction term being omitted because of 
> collinearity. I cannot figure out why.

I'm surprised only the interaction gets dropped.

I may be making a mistake somewhere. But have you run descriptive 
statistics on your created vars?

As a sidelight I would use factor variable notation for my 
interactions and dummy variables.

-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW:    http://www.nd.edu/~rwilliam

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/