Perhaps this is a reason for more investigator-based grant mechanisms where they discuss general research agendas rather than concrete plans for specific projects.
04.03.2026 15:30
👍 6
🔁 1
💬 0
📌 0
Perhaps this is a reason for more investigator-based grant mechanisms where they discuss general research agendas rather than concrete plans for specific projects.
But with the speed that methods can be developed, tested, and applied using things like Claude Code, 8 months is way too slow. By the time a reviewer sees a proposal, the work will probably have been done ages ago!
Then you submit the proposal, and during the 8 months that it's reviewed, you work out the kinks and apply the method to real data. Hopefully the grant is funded before you finish that work.
In this area, the traditional pattern is that you derive and mostly test a method before submitting the grant proposal (since you can't fund methods work without substantial evidence that it already works).
One concern that I don't see people talking about much is how AI will affect the rhythm of grant applications for stat gen work.
It's definitely not just a you problem (which is what prompted me to make the figure). When I get my Claude tokens back, I'll update the figure with more lead-in.
Disclaimer: I don't entirely vouch for the validity of these numbers, which were generated with 10 minutes of vibe coding with Claude Code and 5 minutes of spot checking to make sure it was all sensible. And of course, lots of limitations to my design. Just thought it was interesting.
There is also a drop in active projects in 2025. This could be because people have stopped using the RAP or because more people are piling onto a smaller number of applications.
It looks like there is a dip in preprints in 2025 when labs were required to make the switch, but the dip has largely recovered (unless you think the 2024 trajectory should have continued). The big drop at the end is just because the most recent quarter is only halfway over.
For this figure, I count the number of bioRxiv and medRxiv preprints that cite a UKB application number since January 2024 and plot the number by quarter. I also plot the number of open UKB applications over that same time period.
Over the course of 2025, UKB researchers were forced to move all of their data analysis to the cloud rather than working with local copies. What was the effect of this move on scientific output? Here is a simple first pass at answering that question.
Very exciting! Congratulations!
I'm now going to send all my referee reports for papers (or researchers) I don't like as a series png screenshots! What a great idea!
CSWEP strongly condemns Larry Summers’ behavior as revealed in the email correspondence with the late Jeffrey Epstein. While abuse of power in the economics profession is not new, rarely has the intent behind such abuse been so clearly stated.
I had to rescue Dad from Elko once while I was driving back Seattle because he had a major stomach flu and his carpool left him behind. Had to detour down a windy canyon in a blizzard with like 5 foot visibility and I'd start fish tailing any time I drove over 10 mph. Elko seemed nice enough though.
University of Colorado Boulder Department of Psychology & Neuroscience is searching for TWO tenure track assistant professors. This is open to all department areas, including Behavioral, Psychiatric, and Statistical Genetics🧬 jobs.colorado.edu/jobs/JobDeta...
This is exactly what I argued for months. It's not sustainable! But the airport considerations won me over in the end.
So annoying! This was ultimately the reason we hyphenated our kids' last names. Both their names have like 10 syllables now, but I'm less likely to get detained at an airport.
Excited to share our preprint on PGI portability across ancestries—now on bioRxiv. With co-authors @aysuo.bsky.social, @paturley.bsky.social, @alextisyoung.bsky.social, and @Dan_J_Benjamin. Preprint: doi.org/10.1101/2025... Thread below for details.
Essentially anything by Brandon Sanderson
Wish I had seen this earlier...
uncheck 'autoplay videos and GIFs'
Settings -> Content and media -> uncheck 'autoplay videos and GIFs'
Excited to present our new tool "PEStimate", an online app for predicting risk reductions after Polygenic Embryo Screening (PES).
Find the app here:
polygenicembryo.shinyapps.io/pestimate/
Preprint:
www.medrxiv.org/content/10.1...
I'm looking for a PhD student to do some trailblazing at the frontier of personality genomics and lifespan gene-environment transactions (start Fall 2026)! If you want to work with me: psychology.msu.edu/graduatestud...
(PS @drmeltemyucel.bsky.social is taking a student too, in moral psych/dev!)
Great thread! We make this point in the NEJM paper, but I think it flew under the radar for many.
I think I ultimately disagree too, but I thought it was thought provoking. Like the case where maybe a parent is asked to share their DNA and they explicitly say they don't consent. Genotyping a sibling and then imputing their DNA may be legal, but it feels icky, no?
New pre-print online: Imputing parental genotypes through Mendelian Imputation: Ethical and Legal Considerations
osf.io/preprints/ps...
The CCT paper has a discussion of why you should use the conventional bw even with the correction due to coverage maybe, but I'm lost. I'm getting pretty close to a 5% T1 error rate even when I use the larger bw. Has anyone smarter than me thought through this and can tell me what I'm missing?
I ran a few quick simulations with quartic trends and local linear regression, but used bw's at different multiples of the CCT bw. As expected, using uncorrected estimates, the MSE is minimized at the CCT bw, but the bias-corrected estimate's MSE is minimized at a 2-3 times larger bw.
RD optimal bw question: The CCT approach in rdrobust minimizes the MSE assuming you use local polynomial trends. Then it does a bias correction, which reduces the bias but increases the SEs. But then doesn't that mean that the new estimate no longer minimizes the MSE? #econsky