We started @malengo in 2021 to support Ugandan students on their journey to a Bachelor's degree in Germany.

That year, we admitted 10 students.

In 2022, we admitted 30.

Please meet the @malengo class of 2023:
ONE HUNDRED AND TWENTY STUDENTS!

Here's how we did it: 1/n We were fortunate to attract generous donors, including the Livelihood Impact Fund, Unorthodox Philanthropy, @GlobalDevInc, and many of you.

But it became clear quickly that while the returns to the program are massive, so is the upfront investment. So donations are not enough.

In July 2018, I first came across a paper called “The Sad Truth About Happiness Scales” by Tim Bond and Kevin Lang. My heart sank when I read it (as did that of others, possibly including @ImranRasul3).  1/n The paper starts with this problem: if we force people to tell us about their happiness using integers without knowing the distribution of the continuous underlying variable, we cannot easily use group averages to learn about the average level of happiness in groups of people.

Remember remotestudentexchange.org? It enables professors to open their online classes to students from low-income countries. It was a roaring success initially! But this past semester, interest waned. In this thread, I'll tell you what happened; and then ask for your advice!

1/n It started in January 2021, when some of my wonderful colleagues at @SU_Economics agreed to open their classes to guest students from @StrathU in Nairobi.

Storytime! Here is how and why I switched from working on cash transfers to working on international migration.

If you're still looking to donate this year, it's also a pitch! Donation link at malengo.org. US donations are tax-free.

1/n In May 2009, I sat on a bench in Harvard square with @PaulFNiehaus. Together with others, he was thinking about setting up a cash transfer charity, which they were calling "GiveDirect". I was interested in studying the effects of cash transfers on stress. It seemed like a match!

I'm sad and annoyed that my community is humiliating and ridiculing someone who has a deficient understanding of econometrics and was not modest about it. This is not the #EconTwitter I know and love.

A rope: 1/n I can understand that it's frustrating for @Jabaluck and colleagues to pull of an amazing study and then have to deal with confident but wrong takes. I also understand getting impatient with them and clapping back a little at some point. But: 2/n

#Econtwitter: In the context of this paper, a broader technical point I've been meaning to ask you about. When we quantify a treatment effect in percentage or standard deviation terms, it seems to me that the correct quantity to compare to is the *control group at endline*. 1/n

https://twitter.com/jhaushofer/status/1328259423475798017

The reason is that any other comparison would bake in either secular change, or treatment effects.

For treatment effects, we know how to get this right: E.g., suppose endline income is $10 in treatment, $5 in control. Baseline income was $1 in both.

Today @haralduhlig, lead editor of the @JPolEcon, a prominent journal in economics, called for @paulkrugman to step down as @nytimes columnist to make room for a POC. I think this is an important moment for the economics profession because of the context of this tweet. Thread: In June, Uhlig called BLM members who advocated defunding the police "flat-earthers and creatonists": "some... wish to go and protest... while you are still young and responsibility does not matter. Enjoy! Express yourself! Just don't break anything, ok? And be back by 8pm."

This is a non-apology apology @haralduhlig. You first apologize for having irritated people. Not for the substance of what you said, but for their reaction. Not a good start. [thread]

https://twitter.com/haralduhlig/status/1270475625237790727

Then you apologize for having chosen the wrong words and comparisons. Not the substance of your wholesale dismissal of particular goals of BLM, but how you put it. To be fair, your tone was a significant part of the problem. But then you double down:

I often run into randomized experiments (and am guilty myself) where treatment is much more expensive than control, but sample sizes in the two groups are equal. Has been said before but bears repeating: when c(T) > c(~T), the optimal sample size ratio T/~T is sqrt((c(~T)/c(T)). Implications: 1. With fixed budget B and sigma1=sigma0, the power-maximizing group sizes are n(~T)*=B/ (sqrt(c(~T)*c(T))+c(~T)) and n(T)*=B/(sqrt(c(~T)*c(T))+c(T)).