Jason Kerwin
@jkerwin.bsky.social
Development Economist @ UW. Hawaiʻi born, Stanford alum, Michigan PhD. Views my own. My website: http://jasonkerwin.com
Last Thursday I gave this quarter's Paul Heyne Seminar. My talk was about why USAID is great for America and also the world, why we should keep it, and how we can make it even better.
A key point is that it is the best brand in the industry, by far.
A key point is that it is the best brand in the industry, by far.
June 3, 2025 at 5:57 PM
Last Thursday I gave this quarter's Paul Heyne Seminar. My talk was about why USAID is great for America and also the world, why we should keep it, and how we can make it even better.
A key point is that it is the best brand in the industry, by far.
A key point is that it is the best brand in the industry, by far.
USAID is an incredible brand and source of soft power. Killing it off would be a disaster for US foreign policy and for millions of the world’s poorest people.
March 6, 2025 at 4:13 PM
USAID is an incredible brand and source of soft power. Killing it off would be a disaster for US foreign policy and for millions of the world’s poorest people.
Your slides should avoid having tons of extra junk on them, but the default Beamer themes make that hard. So I made my own .sty file that fixes all the standard issues and put it on my website for everyone to use:
February 27, 2025 at 11:11 PM
Your slides should avoid having tons of extra junk on them, but the default Beamer themes make that hard. So I made my own .sty file that fixes all the standard issues and put it on my website for everyone to use:
What should you do instead? Use design-based reasoning, plus the double lasso:
February 21, 2025 at 8:29 PM
What should you do instead? Use design-based reasoning, plus the double lasso:
Link to post: jasonkerwin.com/nonparibus/2...
February 21, 2025 at 8:29 PM
Link to post: jasonkerwin.com/nonparibus/2...
Are you picking controls for your RCT based on which variables have "significant" differences in your balance table? Don't do that.
February 21, 2025 at 8:29 PM
Are you picking controls for your RCT based on which variables have "significant" differences in your balance table? Don't do that.
Here’s one answer to this question—more sample size wouldn’t actually help much in Econ RCTs. I find this surprising!
January 27, 2025 at 3:43 PM
Here’s one answer to this question—more sample size wouldn’t actually help much in Econ RCTs. I find this surprising!
The calculation is identical if you make the MDE bigger. Doubling the sample size gets you from 80% to 99% power.
Since most interventions have d=0.1, powering studies to detect differences of half that size is reasonable.
Since most interventions have d=0.1, powering studies to detect differences of half that size is reasonable.
January 27, 2025 at 3:31 AM
The calculation is identical if you make the MDE bigger. Doubling the sample size gets you from 80% to 99% power.
Since most interventions have d=0.1, powering studies to detect differences of half that size is reasonable.
Since most interventions have d=0.1, powering studies to detect differences of half that size is reasonable.
Why don’t we just double the sample size for every study and get 99% power instead of 80%? Seems like it would solve many problems
#econtwitter #econsky
#econtwitter #econsky
January 26, 2025 at 5:44 PM
Why don’t we just double the sample size for every study and get 99% power instead of 80%? Seems like it would solve many problems
#econtwitter #econsky
#econtwitter #econsky
They randomized clusters within matched pairs, which can lead to overrejection problems. They allude to this as a potential issue but I do not see any actual corrections for it in the paper.
November 21, 2024 at 9:21 PM
They randomized clusters within matched pairs, which can lead to overrejection problems. They allude to this as a potential issue but I do not see any actual corrections for it in the paper.
I also find their justification for doing a clustered design strange. Shared characteristics within a cluster are a *problem* for clustered designs, not a reason for them. Contamination seems unlikely here.
November 21, 2024 at 9:18 PM
I also find their justification for doing a clustered design strange. Shared characteristics within a cluster are a *problem* for clustered designs, not a reason for them. Contamination seems unlikely here.
There are only 8 clusters here, unfortunately. The authors apply some corrections for clustering don’t seem to understand that this only matters for the CIs and not the point estimates.
November 21, 2024 at 9:16 PM
There are only 8 clusters here, unfortunately. The authors apply some corrections for clustering don’t seem to understand that this only matters for the CIs and not the point estimates.
I know ~nothing about poetry and these still seemed super obvious.
November 21, 2024 at 4:43 AM
I know ~nothing about poetry and these still seemed super obvious.
What's the best descriptive paper you've ever read? For me it's probably this one. Teachers in Uganda were missing from the classroom *half* the time! Stunning.
documents1.worldbank.org/curated/zh/8...
documents1.worldbank.org/curated/zh/8...
November 20, 2024 at 10:57 PM
What's the best descriptive paper you've ever read? For me it's probably this one. Teachers in Uganda were missing from the classroom *half* the time! Stunning.
documents1.worldbank.org/curated/zh/8...
documents1.worldbank.org/curated/zh/8...
Finally got to meet my coauthors Francis Annan & Alfredo Burlando in person! Only remembered to grab a selfie with Francis (and only just before I left).
November 18, 2024 at 10:45 PM
Finally got to meet my coauthors Francis Annan & Alfredo Burlando in person! Only remembered to grab a selfie with Francis (and only just before I left).
Thrilled to be in Berkeley for the CEGA Research Retreat! Stopped at In-n-Out to grab the best burgers on the planet
November 15, 2024 at 4:46 AM
Thrilled to be in Berkeley for the CEGA Research Retreat! Stopped at In-n-Out to grab the best burgers on the planet
Those at the Mokuluas, they’re real islands. Could be an edited photo though
November 15, 2024 at 4:32 AM
Those at the Mokuluas, they’re real islands. Could be an edited photo though
You might still have roadblocks to doing this process, but those roadblocks are going to be primarily the fact that you still need to do research. So go do it! Seek truth, and enjoy yourself. That's what this job is about. (7/N)
January 23, 2024 at 7:11 PM
You might still have roadblocks to doing this process, but those roadblocks are going to be primarily the fact that you still need to do research. So go do it! Seek truth, and enjoy yourself. That's what this job is about. (7/N)
Draw the rest of the horse. (5/N)
January 23, 2024 at 7:10 PM
Draw the rest of the horse. (5/N)
Then write your introduction. This part is hard, but there is a specific formula that good intros follow. (4/N)
January 23, 2024 at 7:09 PM
Then write your introduction. This part is hard, but there is a specific formula that good intros follow. (4/N)
Steps 1-3 are to do the actual research, which includes explaining things to people and making sure your argument is clear. (3/N)
January 23, 2024 at 7:09 PM
Steps 1-3 are to do the actual research, which includes explaining things to people and making sure your argument is clear. (3/N)
Tell it to draw you a picture in tikz and it will just do it:
January 6, 2024 at 10:06 PM
Tell it to draw you a picture in tikz and it will just do it: