Difference in differences and Propensity score matching for different event dates

[Luc Abbey]

Dear all,

I am new to Stata and now experiencing some hard time dealing with the following problem:
I want to test the causality between firms' mergers & acquisitions (dummy variable M&A) and firms' innovation propensity (variable Innov). Basically I have a panel data which ranges from 1999 to 2007 and covers 200 firms. All observations are annual.

For me the best identification strategy is to implement both propensity score matching and Difference in differences approach.

My treatment group is composed of firms that experiences M&A, my control group is composed of those that didn't experience any MA.
My time indicator is before and after M&A . I want to compare the innovation scores 2 years before and after M&A.


However, the challenges are as follows:

1. In my treatment group (firms that experiences M&A) , the event dates are different for each firm.
2. I want to match treatment and control groups on several characteristics such as ROA, leverage, size. The idea is that in my Diff in diff analysis, I don't want to compare apples with oranges.

Hence I want to know how I can arrange my dataset, what dummy variables I should create and How can I combine propensity score matching and Difference in differences in Stata. I know there exists -diff- command, however I don't know how to apply it to the analysis involving different event dates.

After browsing several posts on diff in diff on the forum, I still cannot find a comprehensive and applicable answer. Therefore I decided to create a new post here.



Thanks in advance.


Best regards,

Luc

[Luc Abbey]

Hi all,

I just added an example file here. The data is structured in this way at the moment.

In the example, firm 1 is considered in the control group while firm 2 is in the treatment group since it experienced a MA event.

My dependant variable is Innovation score (Innov) and I want to match treatment with control based on, say, leverage and ROA during the same period of time.

Then I need to run a Diff in diff base upon matched pairs.

Can someone tell me how to proceed precisely??? Thank you!!

Best,

[Marcos Almeida]

Welcome to the Stata Forum / Statalist.

You didn't get a helpful reply so far. Among the reasons: your question may be quite overarching. Indeed, you

want to know how I can arrange my dataset, what dummy variables I should create and How can I combine propensity score matching and Difference in differences in Stata. I know there exists -diff- command, however I don't know how to apply it to the analysis involving different event dates.

Also, the data set was shared as a snapshot in #2. Please take a look at the FAQ, particularly on how to share data which is "appplicable" for Stata users. There you may find a specific warning on not to use snapshots, and the reason is given.

That said, if you haven't yet, maybe you wish to take a look at this text. It seems to lack the approach of different dates, but some command was given for estimations within 2 periods.

To end, since I'm not well acquainted with this method, sorry, but I cannot provide a "comprehensive and applicable answer" for the whole scenario, daring as it may, even for those with much expertise, I suspect.

[Luc Abbey]

Hi Marcos,

Thanks for you reply. Here attached a csv file which contains my example file.

I read the slides that you mentioned. The biggest difference between my case and hers is that I have event 'shocks' over different years for different firms. Hence before conducting a diff in diff, I cannot predefine the dummy variable before/after for my control group.

For instance, for firm A, the shock happened in 2005, I want to find a control firm which didn't experience any shock and match them upon several criteria . However, I have no clue which firm is the best candidate and how I can generate the before/post dummy variable for the control group for my diff in diff analysis.

Next, for firm B, the shock happened in 2007, and the same situation etc etc.

I hope it is a bit clearer as question.

Thanks in advance!

[Luc Abbey]

Hi all,

Apparently no one can tell me at least how I can run the diff in diff with moving 'shock' date for each firm???

Thank you

[Ashish Tyagi]

Luc,

Moving shock date is not a problem since the dummy on treated can adapt to it. The problem results from longitudinal data.

The independent and identically distributed (i.i.d.) sampling assumption ensures that the outcome and treatment status of each individual are unrelated to the outcome and treatment status of all the other individuals in the population. Correlated data arising from hierarchical or longitudinal study designs do not meet this assumption.

- p. 44, second para under 'Caveats and Assumption Section", Stata Manual v13

Here is one post which addresses longitudinal aspect of matching in a similar problem & might be helpful.

The solution might be to use matching over some average values of covariates using Coarsened Exact Matching (or other estimator), then take the weights and use in panel diff-in-diff. Please see this post with similar topic under study. I hope it is helpful.

[CARA LIN]

Hi Macros,

It is very important for me to see your reply. But this link is invalid, showing "DNS_PROBE_FINISHED_NXDOMAIN".

Welcome to the Stata Forum / Statalist.

You didn't get a helpful reply so far. Among the reasons: your question may be quite overarching. Indeed, you

Also, the data set was shared as a snapshot in #2. Please take a look at the FAQ, particularly on how to share data which is "appplicable" for Stata users. There you may find a specific warning on not to use snapshots, and the reason is given.

That said, if you haven't yet, maybe you wish to take a look at this text. It seems to lack the approach of different dates, but some command was given for estimations within 2 periods.

To end, since I'm not well acquainted with this method, sorry, but I cannot provide a "comprehensive and applicable answer" for the whole scenario, daring as it may, even for those with much expertise, I suspect.

Can you update the link or share the file? Thank you for your kindness.

Best regrads,
Cara

[Marcos Almeida]

Hello Cara, welcome to the Stata Forum/Statalist.

The link I shared in #3 is not active anymore. But you may wish to take a look at this text.

Hopefully that helps.

[CARA LIN]

Hi Marcos,

The link I shared in #3 is not active anymore. But you may wish to take a look at this text.

Hopefully that helps.

It really help me understand DID-M.I have learned a lot from it, thank you for sharing.

Best regards,
Cara

[CARA LIN]

Hi there,

Here is a document detailing the principles and applications of the propensity score matching and difference-in-differences method. Please take a look at https://edge.edx.org/assets/coursewa...ifferences.pdf . And I got the data from here https://stanislaomaldonado.org/impact-evaluation-old/ . Hopefully that helps.

[Mjnh Pham]

Hi there,

Here is a document detailing the principles and applications of the propensity score matching and difference-in-differences method. Please take a look at https://edge.edx.org/assets/coursewa...ifferences.pdf . And I got the data from here https://stanislaomaldonado.org/impact-evaluation-old/ . Hopefully that helps.

Hi CARA,

Can you share the data for me? The link no longer works.

Thanks so much

[Paul Lavery]

Hi Luc

Albeit the post was a couple of years ago, I just wanted to ask whether or not you found a solution to your issue?

I have a similar issue with a similar dataset and have so far been unsuccessful in finding a solution. Forum post raised at: https://www.statalist.org/forums/for...nt-event-dates