Hi all,
I have got a question concerning a regression discontinuity design with time series data.
Among others, I am intending to assess the impact of the Mumbai metro opening (coded as a dummy variable) on the probability of exceeding a national air pollution threshold (another dummy). I have assembled a database with hourly data on air pollutant concentrations and weather variables (temperature, wind, humidity and cloud coverage) for a 2-year window around the opening date of the metro i.e. a 4-year period with roughly 34,000 observations.
The residuals of the regression of the dummy that indicates whether the threshold is exceeded (standard) on the current and 1-hour lagged quartics in weather covariates, a third-order polynomial time trend interacted with the opening dummy, and some other time controls (in addition standard error are clustered at a 5-week level), seem to have a logistic distribution.
I checked on many forums and article but it appears that running a logistic regression with time series data is not so common. I already tried with the following code, however I have no clue if this is a good estimation strategy.
Do you have any suggestions on how to deal with time series data, a discontinuity and a logistic approach?
Thanks a lot for your advice!
William
I have got a question concerning a regression discontinuity design with time series data.
Among others, I am intending to assess the impact of the Mumbai metro opening (coded as a dummy variable) on the probability of exceeding a national air pollution threshold (another dummy). I have assembled a database with hourly data on air pollutant concentrations and weather variables (temperature, wind, humidity and cloud coverage) for a 2-year window around the opening date of the metro i.e. a 4-year period with roughly 34,000 observations.
The residuals of the regression of the dummy that indicates whether the threshold is exceeded (standard) on the current and 1-hour lagged quartics in weather covariates, a third-order polynomial time trend interacted with the opening dummy, and some other time controls (in addition standard error are clustered at a 5-week level), seem to have a logistic distribution.
I checked on many forums and article but it appears that running a logistic regression with time series data is not so common. I already tried with the following code, however I have no clue if this is a good estimation strategy.
Code:
xi: logit standard i.opening*time_trend weather_covariates time_controls, vce(cluster week5)
Thanks a lot for your advice!
William
