I am trying to analyse causal inference associated with an intervenion using either Difference-in-Differences or Interrupted Time Series Analysis. I have a discrete time series consisting of data covering a four year period, which could either be aggregated by month [allowing for 24 observations in both the pre- and post-intervention periods] or quarterly [8 observations in each period]. Ideally, I assume that aggregation by month would be preferable....however there is a distinct possibility that there will be a large number of zero value observations, which I assume would make the use of regression techniques more complex.

Were aggregation to be undertaken by quarter there would likely be less zeros, but much potentially insufficient observations for ITS [as the preferred method]. Does anyone know of reliable models for use in regards to data with either a small number of observations or data with large numbers of zero values?

Any help hugely appreciated.....

  • $\begingroup$ Just to help yourself here: as you have "so little data" it would be relevant to do a small simulation study to see what can be estimated. If the simulation study, where we know the true ATE, is unable to pick up the correct answer then expecting to achieve this using real data is unreasonable. $\endgroup$
    – usεr11852
    Dec 10, 2021 at 14:48


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy