State Panel data - generating a year treatment variable with a month component?

[marilyn ibarra-caton]

I have annual state panel data. Each state implemented a state reform at different times (year). I have information on the month and year of implementation. I want to generate a treatment dummy variable for the timing (year) of the reform implementation. I can generate a dummy variable equal to 1 or 0 by the year. However, in some cases like when the reform was implemented in November 2014, would the dummy receive a 1 in 2014 or a 0? Is there a way to create a dummy variable that would allow me to somehow capture the month in generating the dummy treatment variable? Note I don't have monthly state data, it is a panel of state annual data so I need to keep the year dimension in the data set.

Thanks in advance!
Marilyn

[Clyde Schechter]

To the extent that I understand your question, the answer is no, you cannot do that. You will have to make a decision about what to do for each state in the year of reform implementation. That decision is not a statistical matter--it is a substantive matter and it depends on what your research goals are.

For example, if you are trying to estimate the effect of the reform implementation on some outcome, and you plan to use this indicator to represent the post-reform condition in a difference-in-differences analysis, and if you expect that the impact of the reform on the outcome would be delayed in any case, you would probably set the indicator to 0 for the year in which the reform was introduced. That's because you wouldn't expect the outcome to change in that year anyway.

If you think that the impact of reform on the outcome is likely to be more or less immediate, then you might do something like set the indicator for the reform year to 1 if the state introduced the reform in the first half of the year, but 0 if it was introduced only in the second year.

It also depends on how the outcome is measured in time. If the outcome is always measured at the end of the year, then it might make sense to set the indicator to 1 for the reform year regardless of which month the reform began if the effect of the reform on the outcome is more or less intermediate.

Another possibility, if the impact of the reform on the outcome is felt gradually, is not to use a simple pre-post indicator, but rather to work with the number of months since enactment of the reform or something like that.

You have to think through all these possibilities and make a decision. You must also be prepared to accept the fact that with coarse-grained data you sometimes have to make decisions that compromise the realism of your analysis.

[marilyn ibarra-caton]

Thanks Clyde, I was afraid that was going to be the response Thanks for confirming what I had done and for the many suggestions to go about this (subjectively).

Best,
Marilyn