Feature@confidence interval walkthrough

facebookincubator / GeoLift

GeoLift is an end-to-end geo-experimental methodology based on Synthetic Control Methods used to measure the true incremental effect (Lift) of ad campaign.

https://facebookincubator.github.io/GeoLift/

MIT License

175 stars 54 forks source link

Feature@confidence interval walkthrough #167

Closed raphaeltamaki closed 1 year ago

raphaeltamaki commented 1 year ago

Add document explaining the two methods used by GeoLift to estimate confidence intervals This commit should just add 3 documents

vignettes/Confidence_Interval_Explanation.Rmd file containing the code that creates the output
website/documnets/.../Confidence_Interval_Explanation.md that is the immediate output of knitting the previous file, with the exception that (1) I use a relative path to grab an image, since using the command compiled by the knitting to generate an image didn't work, and (2) I corrected the output of the Knit for the numbered bullet points to work correctly in markdown

Below you can see how a screenshot of how the equations are going to be shown in the website Screenshot 2023-09-11 at 10 55 44

gufengzhou commented 1 year ago

Everything looks great, just probably a small typo on page 3 that says "credible interval". We're not touching any Bayesian frameworks here, right? So we should be consistently talking about confidence interval.

Regarding the conformal type, please ensure that there's no specific reason why the creators used different types for avg and daily. If no, then it's a good idea to unify.

raphaeltamaki commented 1 year ago

Everything looks great, just probably a small typo on page 3 that says "credible interval". We're not touching any Bayesian frameworks here, right? So we should be consistently talking about confidence interval.

Regarding the conformal type, please ensure that there's no specific reason why the creators used different types for avg and daily. If no, then it's a good idea to unify.

Agree, but I think this should ideally be in a separate PR