How to use Propensity Score Matching on the treatment group and control group with different variables?

[Rogier Joris]

I was wondering if someone could help me with the following:

For my thesis I study the effects of secondary buyouts on operational performance (Revenue growth, EBIT growth, etc.) in target companies compared to a control group.
To study these effects I want to use propensity score matching to find the best comparable companies.
For the treatment group (secondary buyouts) I have values for variables around the deal date (event window -1,3) and converted these variables to year variables (i.e. I converted the variable for Revenue in year 3 after the deal date to Revenue in 2019). However these matches vary between companies, so, Company A has Revenue in year 3 after the deal date that matches with Revenue in 2019, but Company B has Revenue in year 3 after the deal date that matches with Revenue in 2017 (see screenshots: REV3B = Revenue in year 3 after Buyout & REV19 = Revenue in year 2019).

But for the control group, I have these values only in years (i.e. Revenue 2019).
Therefore I need to create a stata code that will use propensity score matching individually for each company in the treatment group on all companies in the control group as the event window differs between all companies.

Does anyone know a solution for this problem?
I would really appreciate it if someone could help me out.

Best,
Roger

[David Radwin]

See this post for suggestions: https://www.statalist.org/forums/for...19#post1500819.

[Rogier Joris]

Thanks for your quick response!
However I face similar issues as these posts: https://www.statalist.org/forums/for...nt-event-dates & https://www.statalist.org/forums/for...nt-event-dates

[David Radwin]

In general, there are a few major approaches for dealing with missing data:

1. listwise deletion (also known as complete case-analysis and case deletion) and sometimes pairwise deletion
2. imputation (many types)
3. modelling (such as the expectation-maximization algorithm with maximum likelihood methods), arguably including weighting

There is a large and detailed literature describing and evaluating these approaches. Stata can handle most if not all of them.

Matching offers another option: matching on missing values, which is described in the citation linked above.

Beyond that, there is no "solution" to your problem or the other posters' problems that I know of.