1
$\begingroup$

I am looking at the fantastic R package CausalImpact and had a couple questions hopefully someone can help with.

  1. What should be done when there are 0 values in a series (Y or X) due to impossible situations - for example if the series is retail sales and the stores are all closed on Sunday? Should it be coded as zero in the data or omitted?

  2. Is the x in the data only to be control series (e.g. sales for retail control markets) or can additional covariates be used - e.g. other promos, demographic profiles of the markets or other events in the pre-period of either X or y?

$\endgroup$
1
$\begingroup$
  1. A principled solution would be to model your outcome variable as a mixture distribution where one component is zero. CausalImpact doesn't currently implement such mixtures. I would set the values in question to NA.

  2. Covariates in X are time series that are predictive of the outcome time series y, and whose relationship with y is stable and not influenced by the intervention. Typical examples are sales in unaffected markets, stock market indices, or the GBP.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.