Opens profile photo
Follow
Nathaniel Bechhofer
@bechhof
Economics PhD student at UCSD #Python/#rstats/#EconTwitter & everything social science; (some) opinions revised regularly
Joined June 2014

Nathaniel Bechhofer’s Tweets

Pinned Tweet
As a birthday present to myself, we are launching : a data management platform for researchers that makes best practices accessible and standardized. I’m psyched to be working with and to help build tooling for modern open science!
Quote Tweet
F it, we’re building in public. Introducing: @TrovBase vast-bird-dd8.notion.site/TrovBase-48151
Show this thread
35
Post 3/4 for dplyr 1.1.0 is out! #rstats 🎉 Today, we'll look at how the vctrs 📦 has upgraded dplyr's vector functions, like `case_when()` and `between()`. We'll also look at two powerful new helpers: `case_match()` and `consecutive_id()`!
173
This is a delightful notebook that showcases almost all the crunchy goodness in
Quote Tweet
How many times could the Lord of the Rings trilogy have been replayed in your lifetime? Want to magically reveal a disco dino? 🦖 Check out and tinker in our Taste of Observable notebook for this & other examples highlighting how Observable speeds up collaborative data work! 👇 twitter.com/observablehq/s…
Show this thread
1
9
Fascinating new paper - the stylized fact of regional divergence in incomes across states in the US is driven almost entirely by the top portion of the income distribution. At lower percentiles, there's actually been some weak convergence.
Quote Tweet
Prior studies often consider the relationship between mean state income in some base year and income growth over some subsequent period. Extending this approach across the distribution shows top percentiles actually *diverging* while the rest of the distribution converges slowly
Show this thread
Image
2
43
I don't think people fully grasp how much of Protestant Christianity is going to die off in the next 3 decades. 68% of Missouri Synod Lutherans have seen their 55th birthday. It's 57% of Southern Baptists. There's no major denomination where a majority are under 45 yrs old!
Image
Quote Tweet
@ryanburge I'm looking for a graph you previously posted showing age distribution profiles for a number of mainline protestant denominations. Could you pop it up again. I just remember that the PCUSA didn't have hardly anyone under the age of 65!
250
1,619
Exhibit A of inherent challenge in self-serve analytics. The data may be right, but inexperience in data leads to the wrong conclusion. People are sharing this and it’s absurdly wrong. “Traffic down 11% since ChatGPT launch.” Can you spot the problem?
6
15
Show this thread
Next post in the journey on features in the new {dplyr} release! Learn about .by in this post.
Quote Tweet
Post 2/4 for dplyr 1.1.0 is out! #rstats 🎉 Today, we'll look at `.by` - dplyr's new per-operation grouping syntax (inspired by data.table's own `by` argument)! With `.by`, you never need to remember that pesky `ungroup()` 😉 tidyverse.org/blog/2023/02/d
4
Thanks @EcSciAssoc for making replication materials available in one place! Likely great materials in here for students doing replications, those who are building off existing experiments etc. Extra thanks to data editor Eva Ranhill
Quote Tweet
The @EcScienceAssoc wants everyone to know about our new database of replication materials economicscience.org/replication Got questions? Contact our data editor Eva Ranehill data-editor@economicscience.org #ExperimentalEcon #EconTwitter
Image
10
SQL databases don't like to reshape, pivoting out is poorly supported. We still have options. One option is the PIVOT function. PIVOT needs an aggregation function to work. The trick is to use COUNT(column), which gives us one-hot encoded integers.
Image
Quote Tweet
Image
Day 4 of #pandas-in-DB v.s. #SQL week For #machinelearning, one-hot encoding is a common operation for transforming a categorical feature into a set of binary attributes. Now, how would you do this in your #datawarehouse?
Show this thread
2
32
I'm torn about this. I'm not sure what I hate the most: Overfitting or copyright law.
Quote Tweet
Models such as Stable Diffusion are trained on copyrighted, trademarked, private, and sensitive images. Yet, our new paper shows that diffusion models memorize images from their training data and emit them at generation time. Paper: arxiv.org/abs/2301.13188 👇[1/9]
Show this thread
Image
1
35