diff in diff, xtreg, treatment variable, database construction

[Chiara Tasselli]

Good morning,
My question is very simple, but I have a strong doubt which I would like to clarify.
I am using a panel database of only two periods to set up a diff in diff ( 2 groups in two periods), to simply let's considered that regression:

xtreg DependVariable TREAT##FEMALE

The interaction with females is in because I wanna study that particular subgroup.
My question is: what is the correct form of constructing the variable "treat" that I use to see the treatment effect:

1. it must indicate the treated individuals (so for those that are treated assume value 1 in both periods and the difference is given from time comparison of in xtreg)
2. It must indicate the treatment (so assume value zero for all in T0 and value 1 for treated only in T1)

Many thanks in advance for your time and for replying at this very "beginners" question, but but a deep doubt has arisen in me.
Wishing you all a great weekend

[Jared Greathouse]

Please show your example data using dataex

[Carlo Lazzaro]

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

[Chiara Tasselli]

Carlo Many thanks for the useful advice: the interaction in that way is my common way to work whit simple reg.
My issue is to deal with the command xtreg, does that strategy hold? I have only two periods so the variable which corresponds to the treatment period is exactly the same as the one which indicates the Time in my xtset. it's here that I get stuck and I started on thinking in my assumption N° 2 as above (that in my case already corresponds to the interaction term)...

[Carlo Lazzaro]

Chiara:
you should have two categorical predictors for DID:
group (0=not treated; 1=treated);
time (before something=0; after something=1).

I do not see timevar in your code, that seems aimed at investigating whether gender mediates treatment effect: this is perfectly legal, but, in my opinion, is somethng different from a DID.

[Mead Over]

It seems to me that we need to understand why you are using -xtreg- instead of -regress- It would help to see an extract of your data as Jared Greathouse suggests and to know how your data is -xtset-.

Do your fixed effects include both a panel variable (like e.g. the cluster number) as well as a time variable or only a time variable. If you have a time variable with only two values, before and after, but you have no panel variable, then you don't need to use a fixed-effects (or random-effects) model. The analysis could simply be:

reg depvar  i.TIME##i.TREAT if female==0 
reg depvar  i.TIME##i.TREAT if female==1

Then you compare the treatment effect (the estimated interaction coefficient) across genders.

You will get similar results with a single combined model:

reg depvar  i.TIME##i.TREAT#i.female

But if you have a panel variable for clusters or households or countries etc, then you can -xtset panelvar- followed by using -xtreg- like this:

xtset panelvar 
xtreg depvar  i.TIME##i.TREAT if female==0 
xtreg depvar  i.TIME##i.TREAT if female==1

[Chiara Tasselli]

Many thanks to both for the useful advice, yes I think that I need to go more in deep the xtset comprehension and management, as a start I'll follow for sure your guides!
Once again thank you a lot and wishing you a pleasant week ahead.