Rajdeep Chaudhuri

@jmwooldridge @UptonOrwell So the statement that no correlation implies no causation holds only if the treatment is randomly assigned?

@UptonOrwell In the example I gave, there is no co-movement even if we observed the entire population. What we know is that E(Y|W) = E(Y) for the observed Y and W — so no relationship — even though the average dose-response function, E[Y(w)], could depend strongly on w.

In this example I was using an informal notion of "correlation." For a treatment W, write the dose-response function as Y(w) = g(w) + U, where g(w) can be any function and U has zero mean (for simplicity doesn't vary with the treatment level w). We observe Y = Y(W) = g(W) + U.

@causalinf But for a 3*3 scenario, there are 4 independent basis vectors for the treatment indicator, and their linear combinations as well, these combinations could have the "bad" comparisons. How do we extend this framework to the continuous case, if we were to think along these lines?

@causalinf So the bacons decomp essentially would make the double demeaned indicator be written as a linear combination of sums of basis vectors right? For a 2*2 case, there is only one basis vector , so you'd have essentially nothing to worry about: No perverse comparisons if you will.

I’m going slowly through this continuous did paper by Callaway, bacon and Sant’Anna on my substack. I usually teach the parameters and the estimator but really downplay the TWFE decomposition. It sort of dawned on me that bacons decomp had made me interested in did tho.

@TahaSidiq @harikv73 @asad__tariq The matching was done on a subset so I don't know how helpful the eval would be

@rajdeep_c28 @harikv73 @asad__tariq I can provide api key to run it, it can serve as one of the evals since we already have human matched dataset?

Public Good Alert 🚨 If you’ve ever spent hours merging messy admin datasets reconciling districts/subdistricts splits,merges, or renaming, well, no more! we(I,@harikv73,@asad__tariq) have built adminlineage, a Py package that uses AI to solve this #AI pypi.org/project/adminl…

I have spent my career running field experiments. With my new book coming out I have decided to run a few on all of you! Here is the first one! I am giving away a bundle: 1 signed copy of my new book Experimental Economics: Theory and Practice + 1 free t-shirt of the book. But, there is a Dictator Game twist. In the comments please tag a friend. Before you know if you win, tell me your allocation: keep both, give one, or give it all away. If you are randomly selected, I'm holding you to what you said. Your choice is already public, and your friend can see it right now. Your chances of winning do not depend on your allocation. I will randomly select a winner this Saturday at midnight EST and report back. Extra credit giveaway: if the random person chosen also retweeted this message, then I will also send a signed copy of The Voltage Effect to both them and their partner. And, yes, your partner can sign up too to double your chances! Good luck!

Thankful to @AbhiroopMukho for having me on this project. If you are interested in intergenerational issues, LASI could be a rich dataset to generate insights, especially because it goes way back than any other dataset at present.

Are there any statistical procedures with names worse than ANOVA and ANOCOVA?

Not the jumpscares from @jmwooldridge when I am literally scratching my head over a TWFE model.

@Nyx_r6 Also you bringing up " why if fabian does it it's okay i do it it's not" is the literal definition of comparison jsuk

To clarify I would be idiotic to compare myself to GOATS of the game. All I was looking to do was started needed conversations. I acknowledge the mistakes I have made as a leader/player/teammate etc, they eat at me. If people wanna talk about anything specific my DMs are open.

@Nyx_r6 You cannot talk in the same way as Fabian if you are making mistakes that cost your team. It comes off as a blame-game. I am pretty sure Fabian makes mistakes, it's just that his achievements outnumber mistakes

I’ve seen mixed reactions on Fabians timeouts A big reason others don’t team with me is because I talk in the same way Fabian is here. Clearly calling out mistakes, not sugar coating anything. What do people think? Why is it wrong when I do it, put praised when Fabian does?

@_alice_evans If it is a highly relevant variable, should it not be in the regression?

Though, as I said in my substack, I do not understand why the regressions control for religion, since this is a highly relevant variable when it comes to gender segregation.

“Where men’s honour depends on women’s seclusion, cross-gender friendships are rare”. This also predicts low female employment. Great coverage of my favourite new paper in @TheEconomist

@HumanProgress If that was true, the ideal situation would be continuous procreation unless we can guarantee one genius per family, hence most number of geniuses. Extend that , your solution is infinite procreation within human limits. I don't see how that helps at all.

@HumanProgress What is important is growth, which should be attributed to expansion of knowledge. Saying things like "More people = more geniuses" is assuming a functional relationship between the two when the relation is stochastic.

David Attenborough famously said that "Anyone who thinks that you can have infinite growth in a finite environment is either a madman or an economist." The madmen and economists, however, have history on their side. humanprogress.org/growth-is-good…

Timeless advice

@jmwooldridge So should it be epsilon_it or u_it?

After reading my students' senior seminar projects, I need to find out which of my colleagues are using epsilon_it, rather than u_it, for the idiosyncratic errors in models for panel.🤔

@NPCI_NPCI Give refund for embarrassment caused

NPCI is currently facing intermittent technical issues, leading to partial UPI transaction declines. We are working to resolve the issue, and will keep you updated. We regret the inconvenience caused.

Thanks for the lovely dinner @AbhiroopMukho sir.

#Stata19 is here! 🔹 Machine learning via H2O: Ensemble decision trees 🔹 Conditional average treatment effects (CATE) 🔹 High-dimensional fixed effects (HDFE) 🔹 New and improved graphics features 🔹 And much more 🔗 Explore all the new features: stata.com/stata19