Closest distance with panel regression

[Simon Kweq]

Dear all,

i want to conduct the following regression:

I have a dataset of bank branches handing out credit sums over the years from 2010 to 2022 (on a monthly base). I want to conduct a panel regression if Worldbank projects closeby have an effect on the disbursed sums over the years.

I assigned the closest bank branch to each World Bank project to conduct a difference in difference approach. However, some bank branches are the closest to two or more World bankprojects. Hence, they are duplicates in my dataset.

e.g.

Worldbank project x ..... closest bank branch y ... year 2013 .... disbursed sum of bank branch y
Worldbank project z...... closest bank branch y.... year 2013..... disbursed sum of bank branch y

Therefore, i receive error messages of repeated time values for the branch. Also, the disbursed sum of bank branch y is of course two times the same in the year 2013 and cannot be divided (since it is the effect of the World Bank project on this sum, which i want to evaluate).

Any suggestions on how i could solve that without dropping one of the observations?

thank you very much,
Simon

[George Ford]

I would think if you're looking at projects generally, then the dummy variable for closest applies to a bank and not a project. Consequently, there should be only one observation per bank, and it is closest to a project or it is not, no matter how many projects it is closest to.

If you think the number matters, then the treatment becomes a count of how many projects it's closest to.