SDID with frequency weights

[Wahid Abdallah]

Hello everyone,
This is my first post, so please forgive my mistakes if it is outside the forum norms and suggest how I could improve.
I am analyzing whether properties owned by Institutional Investors (IIs) (e.g., venture capitalists, REIT etc.) charge a higher rent than the other owners (e.g., individuals, property management companies etc.). My dataset contains the year a rental property purchased by IIs and the average rents in those properties for each year. I am using the purchase year of a IIs as a treatment. I have a panel data of 15 years. I want to estimate the treatment effect using sdid (Clarke at al. 2023). Here are my questions:

1. The properties, both in the treatment and donor groups, are of various sizes (in terms of number of units in a property). Some properties have only 2-3 units, whereas others have 60+ units. Since my dependent variable is average rent of a unit in a property, I believe I need to adjust ATT based on the number of units in a property. I was just wondering how I can do that using SDiD. In a typical regression, I can use units as frequency weights, but SDiD does not seem to have that option. Or am I missing something? In case it does not, any suggestion how I can estimate the ATT weighted by number of units?
2. Any suggestions on the number of reps? The SDiD helpfile says it needs to be way more than 50, but how many? Any rule of thumb?

[Jared Greathouse]

SDID isn't normal regression. The idea is that the unit and time weights will make your trends parallel in the pre period. So long as you can do that given your data, then you don't need to do this.


I'd need to read the papers, but usually 1 to 2000 doesn't hurt, I'm unaware of a precise metric to know how to choose

[Wahid Abdallah]

Thanks so much for your quick response, Jared Greathouse.
I understand that for a single structure/building, this works. My concern is when I am aggregating the ATTs based on total number of post-treatment periods (equation 7, page 10, Clarke et al. 2023). Don't I need to adjust it based on number of units in a building? My dependent variable is rent per unit in a building. Hence, a simple aggregation would attach equal weight to a 1-unit building and a 100-unit building, no?
Would highly appreciate your expert comments.