Regression Discontinuity Design (RDD) | Causal Inference in Data Science Part 3

This video is the third part of our mini course on application of Causal Inference in data science. We talked about the concept and implementation of Regression Discontinuity Design (RDD) via studying an example from DoorDash.
The refunding example in the video is inspired by a talk given by the Head of Analytics at DoorDash, Jessica Lachs (tinyurl.com/doordash-experimentation).
🔗 Slides by Yuan: rdd-demo.netlify.app/
🔗 Code by Yuan: tinyurl.com/rdd-notebook
🔗 Regression and Matching • Regression and Matchin...
🔗 Difference-in-differences and Synthetic Control • Difference-in-differen...
📃 Yuan's blog post on causal inference www.yuan-meng.com/posts/causa...
📚 References recommended by Yuan:
Causal Inference for The Brave and True (Chapter 16): matheusfacure.github.io/pytho...
The Effect (Chapter 20): theeffectbook.net/ch-Regressi...
🟢Get all my free data science interview resources
www.emmading.com/resources
🟡 Product Case Interview Cheatsheet www.emmading.com/product-case...
🟠 Statistics Interview Cheatsheet www.emmading.com/statistics-i...
🟣 Behavioral Interview Cheatsheet www.emmading.com/behavioral-i...
🔵 Data Science Resume Checklist www.emmading.com/data-science...
✅ We work with Experienced Data Scientists to help them land their next dream jobs. Apply now: www.emmading.com/coaching
// Comment
Got any questions? Something to add?
Write a comment below to chat.
// Let's connect on LinkedIn:
/ emmading001
====================
Contents of this video:
====================
00:00 Introduction
00:20 Natural Experiments
00:54 Regression Discontinuity Design
10:03 Regression Discontinuity Implementation
16:33 Topic of Next Video

[@xinyuan6649]

"the hungry customer is an angry customer" 😂

[@xinyaohui1919]

Thanks Emma and Yuan! very useful videos!

[@Petite_xiaoxiao]

Thanks Emma, all your videos have been very helpful. It would be great that if there's sequential label for the casual inference series then we can know which one to watch first and then next. Thanks!

[@emma_ding]

Great suggestion Qianyu!

[@GEMINIYYY]

Thanks for the tutorial and example. One question related to the regression result. Why is the estimated beta2 in 'regress All' closer to the true value(10) than the one in 'regress near'.(Intuitively, the latter one should provide better result. However, in this demo, even 95% C.I. of estimated beta2 in 'regress near' missed the true value)

[@richardcyc]

Thanks for sharing! The explanation is pretty clear. I have a question about bandwidth: how do we decide the bandwidth? Is there a formula or a method to decide how big the bandwidth is suitable? Thanks!

[@staiwow]

Very helpful! But I'm curious does the cutoff (in this video's example, cutoff being 30 min of order late) has to be a natural cutoff? If not, how could we decide the cutoff point? Thanks!

[@jennywu705]

Can we interpret the beta_3 to be how the causal impact to LTV from refund changes as the delay time increases?

[@guimaraesalysson]

I dont undersand beta 3. Why use (min_late - cutoff) * was_refund ?

[@brotherbig4651]

Nice video. However, if you invite someone with an Economic Phd background to talk about causal inference, the quality of this talk will be higher. Applied micro economists s are more obsessed with causal inference than researchers in any other fields. It seems your presenter was not focusing on causal inference study during his PhD period.

[@dataseance4041]

if you can provide some suggestions for improvement, that would be helpful. otherwise, per the causal markov assumption, i think the explanation is independent of education given the understanding.