Josh Merfeld
@merfeld.bsky.social
Development, labor, and nerdy measurement stuff | Senior Lecturer, Uni Queensland | IZA | J-PAL | insulin addict | dad of an awesome girl
https://joshmerfeld.github.io/
https://joshmerfeld.github.io/
Winter time in Queensland
August 23, 2025 at 4:56 AM
Winter time in Queensland
This guy doesn’t have a name because he’s ugly and annoying
August 10, 2025 at 9:46 AM
This guy doesn’t have a name because he’s ugly and annoying
Just a horrible place
July 22, 2025 at 7:37 AM
Just a horrible place
The key takeaway is that ML methods -- XGBoost, in particular, does really well -- can outperform a traditional small area estimation technique (EBP). We see this improved performance in almost all statistics: accuracy is higher, coverage rates are higher, and CIs are smaller!
May 16, 2025 at 3:34 AM
The key takeaway is that ML methods -- XGBoost, in particular, does really well -- can outperform a traditional small area estimation technique (EBP). We see this improved performance in almost all statistics: accuracy is higher, coverage rates are higher, and CIs are smaller!
There’s this thing called the dawn phenomenon, where your blood glucose goes up right around the time you wake up. So I wake up and take a couple units of insulin. Then I go for a run, thinking it should look good when I get back.
My diabetes was like “fuck you josh”
My diabetes was like “fuck you josh”
April 13, 2025 at 12:36 AM
There’s this thing called the dawn phenomenon, where your blood glucose goes up right around the time you wake up. So I wake up and take a couple units of insulin. Then I go for a run, thinking it should look good when I get back.
My diabetes was like “fuck you josh”
My diabetes was like “fuck you josh”
In India, same pattern where there is a weird discontinuous jump RIGHT where the variable is dichotomized. Note that these were pre-specified.
February 24, 2025 at 3:56 AM
In India, same pattern where there is a weird discontinuous jump RIGHT where the variable is dichotomized. Note that these were pre-specified.
That SMS only group looking sketchier and sketchier.
February 24, 2025 at 3:56 AM
That SMS only group looking sketchier and sketchier.
I'm just going to put this screenshot here:
February 24, 2025 at 3:56 AM
I'm just going to put this screenshot here:
You can see that more clearly here. The ONLY jump is right where it's needed for results. Otherwise there is essentially no positive correlation.
February 24, 2025 at 3:56 AM
You can see that more clearly here. The ONLY jump is right where it's needed for results. Otherwise there is essentially no positive correlation.
Okay wow. Looking at the two endlines, the replicators compare within-respondent answers to some COVID-19 compliance outcomes. Basically ALL correlations perfectly map to the dotted line, where the original authors dichotomize the outcome.
February 24, 2025 at 3:56 AM
Okay wow. Looking at the two endlines, the replicators compare within-respondent answers to some COVID-19 compliance outcomes. Basically ALL correlations perfectly map to the dotted line, where the original authors dichotomize the outcome.
Reader, here is the data for the two studies. Note that they are not always IDENTICAL, but are clearly the same data, then manipulated/changed by a constant (consistently).
February 24, 2025 at 3:56 AM
Reader, here is the data for the two studies. Note that they are not always IDENTICAL, but are clearly the same data, then manipulated/changed by a constant (consistently).
Right so the re-use of earlier samples is pretty clear. Check out Table 1. Clearly no randomization in the new study. (FWIW it looks like previous studies may have also not been randomized as described.)
February 24, 2025 at 3:56 AM
Right so the re-use of earlier samples is pretty clear. Check out Table 1. Clearly no randomization in the new study. (FWIW it looks like previous studies may have also not been randomized as described.)
Unexpected start to my Wednesday!
January 22, 2025 at 1:28 PM
Unexpected start to my Wednesday!
GitHub copilot killing it
December 11, 2024 at 10:43 AM
GitHub copilot killing it
This honestly looks like a fake account
December 8, 2024 at 11:23 PM
This honestly looks like a fake account
God sushi (well sashimi in this case) is so good.
December 7, 2024 at 10:59 AM
God sushi (well sashimi in this case) is so good.
Well pivot_* with tidyverse is like
December 1, 2024 at 11:55 PM
Well pivot_* with tidyverse is like