Assumptions about the data distribution of the original data

[Claire Austin]

Are there any assumptions about the data distribution of the underlying original data that need to be satisfied when using the causalimpact package?

[Kay Brodersen]

You can find a summary of the assumptions in Figure 2 on page 11 of the accompanying manuscript. In essence, the model assumes that the response variable has normally distributed observation noise given its state and the values of the covariates. Given a decently sized set of predictor variables, this doesn't pose much of a restriction in practice.

On 25 February 2015 at 21:51, Claire Austin <claire...@gmail.com> wrote:

Are there any assumptions about the data distribution of the underlying original data that need to be satisfied when using the causalimpact package?

--
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/a3115387-ea9a-46fb-a4fc-2887823e49a6%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.