Difference-in-Difference using SPSS

[3J LEMA]

I wonder if we can perform Difference-in-Difference analysis in SPSS.  Appreciate your comments.

Thank you. 

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[Andy W]

Like many analyses, it is just setting up a regression equation in a
particular way.

Y = B0 + B1*(Treated Dummy) + B2*(Post Dummy) + B3*(Treated*Post Dummy) +
error

And B3 is the DiD effect.



-----
Andy W
[hidden email]
http://andrewpwheeler.wordpress.com/
--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Bruce Weaver]

There is a user-written Stata command called -diff- that performs
difference-in-difference analysis.  But as this short set of slides shows,
it gives results that are equivalent to the t-test for the Time x Treated
product term in a regression model.  

https://www.princeton.edu/~otorres/DID101.pdf

If you are presenting to an audience that expects the results presented as a
Difference-in-Differences analysis, you could just tailor your results
accordingly.  (And show any skeptics those slides by Oscar Torres-Reyna.)  
 


3J LEMA wrote
> I wonder if we can perform *Difference-in-Difference* analysis in SPSS.
> Appreciate your comments.
>
> Thank you.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to

> LISTSERV@.UGA

>  (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD





-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Art Kendall]

It would be possible to give a more focused response if you describe the
situation.   There are many ways to deal with the difference in differences.

If you are more familiar with an ANOVA terminology Difference-in-Difference
is often another way to say "interaction".

How many Independent variables do you have? What are their levels of
measurement?  How many repeats of measures (DVs)? What levels of measurement
are plausible for your DV's Are there variables to be controlled for
(Covariates)?  etc.

Are you more familiar with ANOVA or regression terminology?



-----
Art Kendall
Social Research Consultants
--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Nils Braakmann-2]

Depends on how sophisticated your DiD is supposed to be. The simple 2*2 design is basically just a regression with a treatment group effect, a post effect and their interaction. Designs with more units and time periods but where everyone who is treated is treated at the same time are straightforward extensions of that principle, i.e., more group fixed effects and/or more time effects and more interactions. If you get into differential treatment timing, it becomes more complicated. This (http://goodman-bacon.com/pdfs/ddtiming.pdf) paper by Andrew Goodman-Bacon (forthcoming in the Journal of Econometrics) is a good description of the problem. There are some new specialised estimators that deal with DiD designs in that context, e.g., https://www.aeaweb.org/articles?id=10.1257/aer.20181169, https://www.sciencedirect.com/science/article/pii/S030440762030378X or https://www.sciencedirect.com/science/article/pii/S0304407620303948. Most of these have code in R, some in Stata, probably none off the shelf in SPSS.

On Tue, Apr 27, 2021 at 9:10 AM 3J LEMA <[hidden email]> wrote:

I wonder if we can perform Difference-in-Difference analysis in SPSS.  Appreciate your comments.

Thank you. 

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

--
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD