Imitating Established Scholars' Workflows

First, let me say that I’ve learned by imperfectly imitating other great scholars in economics, not limited to the ones I mention here. That said, caveat emptor still applies: AI is never perfect, and neither am I. This is just me sharing my first attempts at using Claude Code and other resources, like Quarto, in my research workflow.

For readers who (understandably) refuse to read past the abstract:
(1) Use Quarto instead of Word, (2) experiment with Claude Code for research assistance, and (3) automate your table generation.

What I can candidly endorse—regardless of your opinions about AI or how research should be done—are a few simple tools and habits. There are a few tools and processes that help keep my very ADHD brain from turning my research workflow into an unreplicable mess. Or, at the very least, mitigate the amount I (or you!) go off the rails.

So I’ll briefly tell you what I recommend using, plus a Github repo to try some of it out.

You can under- or over-engineer it to your heart’s content. Use it for:

• Lecture notes

• Research notes

• Making your CV in dual formatting (Not immaculate but simple, and cross-renders to LaTeX formats if you need it.)

• It integrates with your RStudio/R markdown workflow.

• If you don’t use RStudio for research, and insist on being Stata-native, don’t ignore this part. But there are work-arounds where you can basically get your Stata output and scripts pasted into the Quarto documents for your notes. It’s more upfront cost, sure, but when you wait 6 months for a reviewer and they ask for replication code, I'm fairly certain future you will thank present you for it.

• It can integrate into your website (I had Claude Code help with the setup, I won’t pretend to be an expert on that.)

(Just take incorporating it into your work at your own pace and trust levels)

I’m not going to elaborate much here. I’m just going to tell you my opinion upfront: I think academic research is increasingly going to make AI usage the norm. Just be a good economist and don’t just tell it to write your code and papers, or give it minimal oversight.

I’m not the first to say this. I’m not going to try and convince you that I’m not an AI fanboy, or that AI doesn’t have negative impacts interspersed with the positive ones, or whether it’s net good here. Just positive statements (in the economic sense) here on it.

I’m using it to manage and help write some of my code for research, plus giving me edits of my work so I can basically write

• Pedro Sant’Anna has his Claude Code workflow on Github, with documentation on his website. I made my own over-engineered workflow based on it.

• If you’re thinking already that I’m proselytizing about using AI for help on coding and research, I blame Scott Cunningham and his substack entries for it. I was hesitant, adopted some of his tips, and I believe it pays dividends, even if upfront fixed costs seem initially high.

This one is simple. I’ll die on this hill. It saves you the awkwardness of realizing you have old results mixed with new ones. Either in the same script as your regressions, or closely connected, just have it write it to your Overleaf or paper directory directly. Make the code idempotent so you don’t have a million outdated versions (save old versions as needed of course) and have only final results shoved in the directory. For Overleaf, I do this, then use \input{} commands to place the table TeX file in the main manuscript directly where I want it. Yes, I know journals want your tables inline. Save that for the final draft before submission. This is much less work upfront and keeps your workflow clean.

In short, implement as:

1. Code runs as usual

2. Whenever you run regression code, always run table code.

3. Make your cloud storage sync accordingly if using Overleaf.

4. If you trust the concept, try my template method.

This one might be obvious, but I was once an ignorant, more inefficient grad student who didn’t realize this was so easy (and so much better than doing it via Word) and would have done it sooner.

Don’t fret if the formatting isn’t perfect. Have Claude Code help on the formatting if you need it to, or just brute force it at end stage. Sometimes, you just need the current results visually represented in the paper, not the final pretty table. This is great for both, in my opinion.

I’m trying this on a whim, and so I can just start telling people where to look and so I can stop having to send an email to my friends and colleagues every time I want to share how I did something, or in some rare cases, answer a solicited question on how I did X research task. If you have suggestions for me on what I should do, try, or discuss, I’m all ears for comments. Just know that the caveat emptor principle goes both ways. But I’d rather hear criticism than ignore it.

Lastly, I created this blog for one purpose: to consolidate my unsolicited and sometimes helpful advice and resources for academics, particularly in economics. My penchant for giving friends tips like this is becoming so repetitive that, for their sakes and my own, I’m using this blog to start sharing those I find most useful. While it might not be all original resources, I want it to be an outlet for research resources I find helpful and actually use often. Some of the advice will be basic, and some economists might read it and think to themselves, “No shit, sherlock. I learned that as a first-year Ph.D. student.” — that’s fine. But if you were or ever had been like me and felt lost finding what works for you at any point, I might as well share what I do use. Determine for yourself whether or not you find my resources and opinions helpful on a case-by-case basis.