Nathaniel B
LightNews LightNews
banner
bechhof.bsky.social
Nathaniel B
@bechhof.bsky.social
1.8K followers 710 following 46 posts
statistics obsessor, economics enjoyer, occasional engineer, founder at TrovBase
Posts Replies Media Videos
Reposted by Nathaniel B
vincentab.bsky.social
Whoa—my book is up for pre-order!

𝐌𝐨𝐝𝐞𝐥 𝐭𝐨 𝐌𝐞𝐚𝐧𝐢𝐧𝐠: 𝐇𝐨𝐰 𝐭𝐨 𝐈𝐧𝐭𝐞𝐫𝐩𝐫𝐞𝐭 𝐒𝐭𝐚𝐭 & 𝐌𝐋 𝐌𝐨𝐝𝐞𝐥𝐬 𝐢𝐧 #Rstats 𝐚𝐧𝐝 #PyData

The book presents an ultra-simple and powerful workflow to make sense of ± any model you fit

The web version will stay free forever and my proceeds go to charity.

tinyurl.com/4fk56fc8
September 17, 2025 at 7:49 PM
Reposted by Nathaniel B
lakens.bsky.social
You have not seen second generation p-values used anywhere because they were badly reinvented equivalence tests, and people should use equivalence tests instead. See open.lnu.se/index.php/me...
September 26, 2025 at 2:43 AM
Reposted by Nathaniel B
reveconstudies.bsky.social
"New study finds salary benchmarking cuts pay gaps by 25%. Pay dispersion partly arises from firms’ uncertainty about market rates, with key implications for pay transparency policy."

New paper from Perez-Truglia, Li & Cullen

www.restud.com/whats-my-emp...:

#econsky
#REStud
September 17, 2025 at 10:44 AM
Reposted by Nathaniel B
posit.co
Posit and AWS have signed a Strategic Collaboration Agreement!

This collaboration helps customers modernize their data infrastructure & accelerate their data science journey on the cloud, making it easier for teams to build and run critical data science outcomes.

Read now posit.co/blog/posit-s...
September 12, 2025 at 5:54 PM
Reposted by Nathaniel B
thomasp85.com
I am beyond excited to announce that ggplot2 4.0.0 has just landed on CRAN.

It's not every day we have a new major #ggplot2 release but it is a fitting 18 year birthday present for the package.

Get an overview of the release in this blog post and be on the lookout for more in-depth posts #rstats
ggplot2 4.0.0
A new major version of ggplot2 has been released on CRAN. Find out what is new here.
www.tidyverse.org
September 11, 2025 at 11:20 AM
Reposted by Nathaniel B
dingdingpeng.the100.ci
PS there’s a (mildly) updated version available now: osf.io/g4s2a_v2
OSF
osf.io
September 11, 2025 at 6:11 AM
Reposted by Nathaniel B
dingdingpeng.the100.ci
Ever stared at a table of regression coefficients & wondered what you're doing with your life?

Very excited to share this gentle introduction to another way of making sense of statistical models (w @vincentab.bsky.social)
Preprint: doi.org/10.31234/osf...
Website: j-rohrer.github.io/marginal-psy...
Models as Prediction Machines: How to Convert Confusing Coefficients into Clear Quantities Abstract Psychological researchers usually make sense of regression models by interpreting coefficient estimates directly. This works well enough for simple linear models, but is more challenging for more complex models with, for example, categorical variables, interactions, non-linearities, and hierarchical structures. Here, we introduce an alternative approach to making sense of statistical models. The central idea is to abstract away from the mechanics of estimation, and to treat models as “counterfactual prediction machines,” which are subsequently queried to estimate quantities and conduct tests that matter substantively. This workflow is model-agnostic; it can be applied in a consistent fashion to draw causal or descriptive inference from a wide range of models. We illustrate how to implement this workflow with the marginaleffects package, which supports over 100 different classes of models in R and Python, and present two worked examples. These examples show how the workflow can be applied across designs (e.g., observational study, randomized experiment) to answer different research questions (e.g., associations, causal effects, effect heterogeneity) while facing various challenges (e.g., controlling for confounders in a flexible manner, modelling ordinal outcomes, and interpreting non-linear models). Figure illustrating model predictions. On the X-axis the predictor, annual gross income in Euro. On the Y-axis the outcome, predicted life satisfaction. A solid line marks the curve of predictions on which individual data points are marked as model-implied outcomes at incomes of interest. Comparing two such predictions gives us a comparison. We can also fit a tangent to the line of predictions, which illustrates the slope at any given point of the curve. A figure illustrating various ways to include age as a predictor in a model. On the x-axis age (predictor), on the y-axis the outcome (model-implied importance of friends, including confidence intervals). Illustrated are 1. age as a categorical predictor, resultings in the predictions bouncing around a lot with wide confidence intervals 2. age as a linear predictor, which forces a straight line through the data points that has a very tight confidence band and 3. age splines, which lies somewhere in between as it smoothly follows the data but has more uncertainty than the straight line.
August 25, 2025 at 11:49 AM
Reposted by Nathaniel B
gaborbekes.bsky.social
If interested please test the notebooks.
Examples:
Describe
github.com/gabors-data-...
OLS
github.com/gabors-data-...
Random forest, SHAP
github.com/gabors-data-...
Panel FE, FD
github.com/gabors-data-...
August 15, 2025 at 9:26 PM
Reposted by Nathaniel B
scot-j-pol-econ.bsky.social
Very pleased to announce our special issue on Gender Norms, edited by @emmatominey.bsky.social and @nikkishure.bsky.social.

onlinelibrary.wiley.com/doi/toc/10.1...
June 18, 2025 at 10:20 AM
Reposted by Nathaniel B
economeager.bsky.social
this was excellent. a highly recommended way to spend 5 hours.
scientificdiscovery.dev Saloni @scientificdiscovery.dev · Jun 11
LAUNCH DAY 🚀

Today I’m launching a new podcast, Hard Drugs, with Jacob Trefethen (@jacobtref.bsky.social)

Our first episode is about lenacapavir — a new HIV drug that blocks infections with an efficacy rate of nearly 100%, and which could completely change the fight against HIV worldwide.
Lenacapavir: The miracle drug that could end AIDS
Hard Drugs · Episode
open.spotify.com
June 14, 2025 at 1:29 PM
bechhof.bsky.social
I indulged in a bit of fun doing a first pass of a regularized model for homicide declines with ChatGPT, using @jeffasher.bsky.social's data from jasher.substack.com/p/why-i-thin... to estimate "underlying" rates of homicide decline by city.
Scatter plot of 31 U.S. cities (excluding “Total”) showing raw vs. Bayesian MAP–regularized percent change in homicides from 2024 to 2025. Raw changes are on the x-axis, regularized changes on the y-axis. Each point is labeled by city, with small arrows connecting any labels placed more than 10 pixels from their marker. A caption below summarizes the log‐ratio transformation, hierarchical Normal(μ,τ²) model with an Inverse‐Gamma penalty on τ², and the posterior‐mean back-transform to percent change. Screenshot of Jeff Asher’s Substack titled “Murder is down a lot in big cities.” Below the title is a subtitle: “2025 vs 2024 YTD murders in top 30 most murderous cities.” A table lists four columns—City, YTD 2024, YTD 2025, and % Change—and shows, in descending order of decline, that Denver fell from 29 to 13 homicides (–55.2 %), Birmingham from 61 to 33 (–45.9 %), Cleveland 54→30 (–44.4 %), San Antonio 50→30 (–40.0 %), Las Vegas 54→35 (–35.2 %), Atlanta 50→33 (–34.0 %), Dallas 90→60 (–33.3 %), Indianapolis 70→49 (–30.0 %), and so on.
June 13, 2025 at 1:33 AM
Reposted by Nathaniel B
cghlewis.bsky.social
Thoughtfully planning your data collection efforts has a huge return on investment.

As someone who has cleaned data from projects both with and without data management planning, I can safely say it is worth the effort.

40 hours of planning > 80+ hours of cleaning messy data
May 13, 2025 at 8:13 PM
bechhof.bsky.social
There's just something special about running a proper Bayesian MCMC script that I can't replace with anything else after doing it for almost 10 years
A screenshot of a Bayesian statistical model training interface showing NUTS sampler progress. The display includes information about a model with 45 subgroup columns, training on 52,380 of 68,850 rows. The sampling is being performed using PyMC NUTS (JAX-compiled) with 4 chains running in parallel. The table shows the current progress of each chain with metrics including number of draws (ranging from 601-644), step size (0.01), gradient evaluations, sampling speed (approximately 1.14-1.22 draws/s), elapsed time (8:49), and remaining time estimates.
May 13, 2025 at 4:21 AM
Reposted by Nathaniel B
julianreif.bsky.social
“First, about half the time I reanalyze a study, I find that there are important bugs in the code, or that adding more data makes the mathematical finding go away, or that there’s a compelling alternative explanation for the results.”

davidroodman.com/blog/2025/05...
Appeal to Me: First Trial of a “Replication Opinion” - David Roodman
My employer, Open Philanthropy, strives to make grants in light of evidence. Of course, many uncertainties in our decision-making are irreducible. No amount of thumbing through peer-reviewed journals ...
davidroodman.com
May 10, 2025 at 3:02 AM
Reposted by Nathaniel B
johannarickne.bsky.social
7/ A series of errors create an illusion of statistically significant treatment effects of similar size in the five strategies. Examples include changing/adding/removing variables in regressions, using incorrect bandwidths, and changing clustering methods.
May 9, 2025 at 9:29 AM
Reposted by Nathaniel B
johannarickne.bsky.social
1/ The paper claimed that the reform caused a jump in rape by 50–60 percent. This seemed hard to reconcile with the flat time trend in reported rapes around the time of the reform.
May 9, 2025 at 9:29 AM
Reposted by Nathaniel B
lesja.bsky.social
I need a newsletter because I have so many words inside of me about the "they say they want to raise the birth rate and yet they are decimating maternal and infant health surveillance infrastructure [or whatever else], what a gotcha" thing
April 4, 2025 at 6:29 PM
Reposted by Nathaniel B
umpamdk.bsky.social
@icpsr.bsky.social @umich.edu is hosting DataLumos, which is a crowdsource repository of government data archive.icpsr.umich.edu/datalumos/home
A crowd-sourced repository for valuable government data<!-- --> | <!-- -->DATALUMOS
archive.icpsr.umich.edu
April 2, 2025 at 8:04 PM
Reposted by Nathaniel B
scientificdiscovery.dev
Seeing the "shingles vaccine protects from dementia" claim going around again.

Unfortunately the main study it's based on is just not convincing. @epiellie.bsky.social has written a great post about why:
Shingles vaccine & dementia prevention: too good to be true?
A public peer-review
medium.com
April 2, 2025 at 10:35 PM
Reposted by Nathaniel B
andrew.heiss.phd
hey hey my constant use of the palmerpenguins data is now relevant for my public policy classes
April 3, 2025 at 12:51 AM
Reposted by Nathaniel B
instrumenthull.bsky.social
📢New WP w/ @borusyak.bsky.social

We derive the most powerful recentered IVs for formula treatments & propose an algorithm for approximating them

This approach yields *huge* power gains, relative to conventional simulated IV, when estimating Medicaid crowdout effects

Check it out! t.co/UPg2XQBefN
March 28, 2025 at 1:54 PM
bechhof.bsky.social
Standardization and division of labor FTW
March 26, 2025 at 3:35 PM
Reposted by Nathaniel B
jorammutenge.com
Always remove whitespace before doing a unique count of values to get an accurate count.

In @pola.rs, there's a function to do just that!

#100DaysOfPolars
#datasky
Image showing polars code for data analysis.
March 24, 2025 at 5:03 AM
Reposted by Nathaniel B
nber.org
Healthcare employment in the US has skyrocketed since 1980. Healthcare is a middle-class jobs engine, but "manufacturing-to-meds" transitions are not saving the Rust Belt, from @gottliebecon.bsky.social, @nealemahoney.bsky.social, Kevin Rinz, and Victoria Udalova https://www.nber.org/papers/w33583
March 23, 2025 at 7:00 PM