Difference in Difference : Data Science Concepts

Video quality	The size	Download

Information Difference in Difference : Data Science Concepts

Title	:	Difference in Difference : Data Science Concepts
Lasting	:	6.32
Date of publication	:
Views	:	5 rb

If you are comparing test minus control dufferences for two treatments, I assume that is also a difference in difference analysis How do you stat test that? 4 sources of variation and use a pq/n formula if you are testing a percentage difference?
Comment from : Joel Rubinson

it seems so logical It's not the same level as your other videos
Comment from : E W

Parallel assumption seems like a very strong assumption
Comment from : C

Yes!! I think you should keep going on these econometric/ behavioural economics/ videos I think they are super useful
Comment from : CUENTAME LA PELICULA

Ok I am going to need to watch the video again when rested
Comment from : Paul Seidel

I actually got a similar question in a BI interview, it would have been helpful to have this video 2 weeks ago
Comment from : hpp496videos

Hi Rithvik, can you also explain about calibration error or expected calibration error?
Comment from : Surya Ashish

What if the price drop in B caused a drop in sales for A? Before Drop: A100 B500 & After Drop: A50 B1000 What is the assumption of B?
Comment from : James Lucas

Hey, can you recommend a method to test the statistical significance of the difference using this method ?
Comment from : shravone

Thank you so much Ritvikmath I learned this method at uni but i really struggled with it Can you show us how to run it on Python or R ? It will be great to also cover RDD, Instrumental variables, and RCT ? They all are impact evaluation methods Thank you brother
Comment from : Yass Chaachaa

It might be good to also pair diff and diff with regression discontinuity and synthetic control Get a whole econometrics series going
Comment from : Resilient Fighter

Good job ritvik, like it better than econometrics academy Snd Ben lambert videos brbrCurious question, since u love teaching stats so much, have you considered a PhD?
Comment from : Resilient Fighter

To explain parallel trends in a more generalized framework, PTA basically says that the non-treated units are good counterfactuals for intervention units In the real world, this rarely ever happens, and we must use synthetic controls, propensity score matching, weighting, or other more advanced methods to make PTA plausible
Comment from : Jared Greathouse

Cool hat!! Catch' em all
Comment from : Why Not?

So is this method only useful when growth is linear (additive) and not multiplicative? What I mean is that, for instance, if the total market size of pizzas had doubled and both pizzerias had remained at a constant market share, we would expect A to go from 100 to 200 and B from 500 to 1000 Since this growth is multiplicative, the trends are not parrallel although the growth rates in are the same Is there a workaround, like using a log scale ?
Comment from : Arnaud Ray