Binary Vs Continuous Treatment Variable in Diff-in-Diff

[Nur Jahan]

Hello everyone,

I would to study the impact of a policy in a country, the data I have allow me to do so using difference-in-differences. To calculate the exposure to the policy at the regional level, I wonder if I should use a continuous treatment variable that captures the intensity of exposure to the policy or a binary one that only captures if a region is exposed or not?

My worry is that the continuous variable has little variation and is very close to zero.

What would you advise me to do then?

Many thanks

[George Ford]

you can use both, a dummy and the intensity.

[Nur Jahan]

and I compare the results of both afterwards?

[George Ford]

run with the dummy only and then both at the same time.

[Nur Jahan]

Thank you for this suggestion, I did both at the same time and the result of the interaction term of both had different signs. The dummy interaction is positive while the continuous one is negative in the same regression. What does this imply?

[George Ford]

is the model
y treat c.treat#c.intensity X

[Nur Jahan]

Thank you for this suggestion. I did the regression as you suggested y treat c.treat#c.intensity X.
When I do not add year fixed effects my interaction term is significant but as soon as I add year fixed effects, the results are no longer significant.