Question over CausalImpact's capabilities

[Ryan Beckham]

Does this function allow the same type of regression that a Synthetic control does?

I am not finding enough examples to see if I can use multiple test markets to use against the predicted market.

For example,  I have sales in Austin of three widgets.  I just introduced a fourth. 

I sell widgets in 10 other cities and have demographic data on them.  I want to use that demographic data and the sales from the other 10 cities to see if the how many of the first three widgets I would have sold had I not introduced the fourth.  Can someone help me with this?

[Mark Edmondson]

Dear Ryan, using multiple regions as control is pretty much the example used in the original paper, section 5. It certainly reads like it covers what you are asking.
Yours sincerely,
Mark

[Kay Brodersen]

Dear Ryan,

CausalImpact adopts a different approach than the one underlying classical synthetic control estimators. In these classical estimators, a vector of static observables is used to find the weights that form the synthetic control (e.g., demographic data). In CausalImpact, by contrast, we use the full response time-series in the pre-period to do the matching. This has several advantages. For details, see the third paragraph on p. 5 of the paper.

In your example, you'd use sales data from the 10 non-treated cities to predict the sales you would have expected to see in Austin if you hadn't introduced a fourth widget. The demographic data from the cities are not used here - the best predictors are chosen purely based on how predictive each city has been in the past.

Best wishes,

Kay

--
You received this message because you are subscribed to the Google Groups "CausalImpact" group.
To unsubscribe from this group and stop receiving emails from it, send an email to causalimpact...@googlegroups.com.
To post to this group, send email to causal...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/causalimpact/04afbaaa-9a6b-4f45-bfbc-0ce432992118%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.