Hacker News Reader

Thanks!

I haven't read the paper yet, but noticed it uses a regression discontinuity design without showing raw data (i.e. a scatterplot).

There have been some cases where such discontinuity analyses were extremely misleading [1], so some caution is advised.

[1] https://statmodeling.stat.columbia.edu/2021/03/11/regression...

View on HN

Thanks for the article! I love methods :-)

Awesome!

Maybe you'll like this better method then: Bayesian counterfactual time-series analysis [1]. In essence, it compares a "neutral" time series and another one affected by a natural experiment and produces bayesian point-estimates for the causal impact.

Super powerful, there's a great R package, and really easy to use.

[1] https://research.google/pubs/pub41854/

Thanks! I'll check it out :-)

Reply on news.ycombinator.com