I am studying a natural experiment that resulted in some cities randomly receiving different numbers of competitors in a certain sector. (Stores applied for licenses in a government-run lottery.) In particular, I'm interested in whether firms that, randomly, ended up as monopolies in their city received lower customer satisfaction scores than firms that ended up having competitors.

I understand that I need to control for the number of incumbents in a city, the number of applicants in a city to the lottery, and the number of winners the government would select for the broader region the city is in.

The equation I want to estimate is then: $satisfaction = \beta_0+\beta_1monopoly+\beta_2incumbents+\beta_3applicants+\beta_4winnersinregion+\epsilon$.

However, my dataset is very small, as I only observe data on 28 firms. Thus, my estimate of $\beta_1$ lacks precision and is statistically insignificant.

I am wondering if it is okay to combine my 'applicants' and 'winners in region' control variables into one control variable so as to increase the precision of my estimate. In particular, I would calculate the expected number of competitors in a city based on how many applicants the city had and how many licences were awarded in the city's region and estimate the following equation:

$satisfaction = \beta_0+\beta_1monopoly+\beta_2incumbents+\beta_3expectednumberofcompetitors+\epsilon$

Does this approach still adequately control for the potentially confounding variables 'applicants' and 'winners in region'?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.