What type of regression should I run

[Michael Fulton]

Hello,

I have panel data which has the following variables:

• Pidm - unique id for each person
• Gift - gift amount to non-profit
• Gift_Date - the date a gift was made
• Gift_Count - the total amount of gifts the individual has made at the current gift date (changes for every repeated Pidm row, when they give another gift, Gift_Count goes up by one)
• Extreme_Exclusion - an indicator that is 1 if they don't want to be solicited and 0 if they do not have an extreme exclusion
• Extreme_Exclusion_Date - the date they acquired this extreme exclusion

I want to measure the average change in gift amount and gift count after individuals acquire an Extreme_Exclusion (value becomes 1).

Is this just a standard xtreg that I should run? Or marginal effects? Or something else...

I assume I will only include those who have an extreme exclusion in my dataset, since I don't want to compare the difference in giving between those with extreme exclusions and those without.

Thank you

[Clyde Schechter]

I assume I will only include those who have an extreme exclusion in my dataset, since I don't want to compare the difference in giving between those with extreme exclusions and those without.

Actually, I wouldn't do it that way. If you do that, your extreme exclusion variable is confounded with any time trend that may exist in the amount of giving. And I would think that it is highly likely that gift giving shows time trends. So I would include people who never get an extreme exclusion also and do a (generalized) difference-in-differences analysis.

Here's a link to a slide set that describes the general approach.

[Michael Fulton]

Hi Clyde Schechter,

Thank you for your recommendation. I am scanning your post for a link but do not see one? Is it perhaps attached? Thanks.

[Clyde Schechter]

Hmm! I don't know why that link isn't there. I'm sure I pasted it in before hitting Post Reply. Anyway, let's try it again:

https://www.ipr.northwestern.edu/wor.../Day%204.2.pdf

[Michael Fulton]

Hi Clyde Schechter ,

I read the journal thoroughly and just have a couple questions. The equation for the DID contains a variable which is called Post*t which is set to 1 if the observation is from the post treatment period in either group.

My question is what if the treatment is applied to each individual in a 150,000 person dataset at different times? For the untreated group do I just mark each of their observations as 1 for post treatment?

I can give you a much more detailed description of the panel dataset if you would like.

Thank you

[Clyde Schechter]

For the untreated group do I just mark each of their observations as 1 for post treatment?

No. In generalized DID, for the untreated group post*t = 0 in every observation. In fact you should think of this post*t variable as meaning "1 if this entity is receiving treatment now; 0 otherwise."