Read Scott's Mixtape Substack with Scottie
Keep Scott's Mixtape Substack. Lose the inbox pileup.
Scottie reads the newsletters you choose, keeps the useful differences, and attaches every original link. Scott's Mixtape Substack stays in your reading life without becoming another email to manage.
An independent guide. Scottie isn't affiliated with, endorsed by, or sponsored by Scott's Mixtape Substack. The publication owns its name, writing, and subscription terms.
From Scottie · July 31, 2026
See Scott's Mixtape Substack in a Scottie brief
Scottie read Scott's Mixtape Substack, Experimental History, and Cremieux Recueil for a reader with the priorities shown below. Start with the rundown, open the full brief, or check every issue behind it.
3sources
18issues read
5stories included
5 of 13 Scott's Mixtape Substack issues included
What shaped this brief
Reader priorities
An applied researcher who wants causal-inference tutorials, econometrics, live research, and frank notes on where AI agents help or break the work.
These are illustrative priorities, not a customer’s data.
Sources in this brief
- Scott's Mixtape SubstackThe publication this guide is about
- Experimental HistoryAdds original behavioral experiments and unusually candid methodological postmortems.
- Cremieux RecueilAdds data-heavy social and health arguments that make the empirical choices visible.
Scottie
Scottie example brief
July 31, 2026 · Executive brief
01 / The rundown
- AI coding agents ignore written prompt rules during large empirical tasks, requiring researchers to use software tools and diff reviews for verification.
- Unnoticed sample drift in agentic coding workflows threatens empirical validity, highlighting the need for strict data pipeline controls.
- Confounding local economic shocks like energy booms can disrupt parallel trends assumptions in continuous difference-in-differences research designs.
Read the complete brief 5 stories · 2 action items
02 / The briefing
01 / main
Written Rules Fail to Constrain AI Agents in Empirical Research
The takeaway: LLMs dramatically lower empirical research production costs, but verification costs stay high and require human skill. Prompt instructions like markdown files fail as hard constraints because agents treat prose as optional, whereas structural software packages enforce true boundaries.
Concrete details
- Claude Code hallucinated sample sizes and hallucinated code in terminal shells without writing pipeline code.
- The R 'did' package strictly blocks panel estimation where every unit becomes treated, unlike standard linear estimation packages.
Why it matters for this reader: As an applied researcher running agent workflows, relying on prompt rules will not protect your empirical pipeline—you need software boundaries and code diff reviews to enforce rigorous verification.
Original sourcesScott's Mixtape Substack
02 / main
AI Agents Shift Analytical Sample Sizes Without Warning
The takeaway: Long projects running on Claude Code can quietly introduce inter-temporal sample drift, moving an analytical sample from N to N+k observations without clear explanation. Establishing a structured verification workflow prevents silent code deprecation and unbacked sample expansion.
Concrete details
- Analytical sample sizes silently shifted from N to N+k observations across long agentic research sessions.
- Sample discrepancies stemmed from inter-temporal execution errors rather than contemporaneous verification failures.
Why it matters for this reader: For your live empirical research with coding agents, unchecked sample size shifts directly compromise your sample selection and causal estimate integrity.
Original sourcesScott's Mixtape Substack
03 / main
Coding Errors Build Critical Human Capital in Empirical Research
The takeaway: Early-career coding errors can overturn statistical significance, forcing researchers to adopt disciplined data pipelines, version control, and strict integrity norms. Relying on AI agents to eliminate all errors might prevent flawed papers while quietly undermining personal research skill growth.
Concrete details
- A Stata code error discovered during review wiped out statistical significance after sitting unresolved for ten years.
- Flawed empirical findings can persist in published literature when uncaught errors bypass journal resubmissions.
Why it matters for this reader: When evaluating AI agent integration into your econometrics workflow, balance output accuracy against the essential diagnostic skills you gain by wrestling with manual coding failures.
Original sourcesScott's Mixtape Substack
04 / main
Local Economic Booms Risk Violating Parallel Trends Assumptions
The takeaway: Estimating continuous difference-in-differences for Texas clinic closures on marriage rates reveals potential confounders from energy technology booms. Sudden oil and fracking activity brought transient male workers into specific counties, differentially shifting local marriage markets and outcome trends.
Concrete details
- Texas House Bill 2 closed 50 percent of the state's abortion clinics.
- Continuous difference-in-differences analysis spans 14 consecutive research episodes on marriage rate shifts.
Why it matters for this reader: In your continuous diff-in-diff designs, checking for time-varying localized shocks like energy booms helps you verify that parallel trends hold across treatment and control groups.
Original sourcesScott's Mixtape Substack
05 / main
Adding Covariates Can Shift Diff-in-Diff Estimates Without Invalidating Designs
The takeaway: Changing point estimates after introducing baseline covariates does not automatically invalidate a difference-in-differences design. Misunderstandings often stem from intuitive gut checks rather than formal econometric properties of covariate adjustment in panel estimators.
Concrete details
- Researchers frequently misinterpret coefficient instability after covariate inclusion as proof of invalid research design.
- Covariate adjustment in panel models requires formal regression imputation rather than naive control additions.
Why it matters for this reader: In your econometric modeling, distinguishing true parallel trend violations from expected covariate adjustments prevents you from prematurely abandoning sound causal inference specifications.
Original sourcesScott's Mixtape Substack
Action items
- Enforce structural software constraints like package assertions in your R code rather than relying solely on prompt instructions.
- Audit analytical sample sizes across pipeline iterations to catch silent observation drift created by AI coding scripts.
See every issue behind this brief
AI in Empirical Research and Verification
- Prose is not constraints and other things I learned from Paul Goldsmith-Pinkham's NBER talk on AI in empirical researchScott's Mixtape Substack · Included
Claude Code and Sample Observation Anomalies
- Stale code, deprecation, and the canonScott's Mixtape Substack · Included
- Am I'm reading too much?Scott's Mixtape Substack · Read, not included
Lalonde test and diff-in-diff
- Covariates, diff in diff and Lalonde testScott's Mixtape Substack · Read, not included
Parallel Trends in Continuous Diff-in-Diff
- Episode 15: Exploring Confounders That Could Impact Our Parallel Trends AssumptionScott's Mixtape Substack · Included
Covariates in Difference-in-Differences Estimation
- More covariates and diff-in-diff -- this time with a lot of bold and italics!Scott's Mixtape Substack · Included
Substack Claude Code integration
- Deriving the Value of these Claude Code Essays (and All Other Essays) using Claude Code Integration to SubstackScott's Mixtape Substack · Read, not included
Claude Code production costs
- Claude Code 56: Falling marginal costs of production, output, and aggregate errorsScott's Mixtape Substack · Read, not included
Impact of Empirical Coding Errors on Findings
- Which research mistakes matter for human capital?Scott's Mixtape Substack · Included
Causal Inference book release
- Countdown to the RemixScott's Mixtape Substack · Read, not included
Hidden Curriculum workshop
- Hidden Curriculum Workshop (August 5-6th)Scott's Mixtape Substack · Read, not included
new book arrival
- Books are here!Scott's Mixtape Substack · Read, not included
sex recessions and dating
- Sex recessions, rise in casual sex, online dating, declining trust, screening and other changes in US sex patternsScott's Mixtape Substack · Read, not included
blog post competition results
- Blog Jamboree 2026: The WinnersExperimental History · Read, not included
resource curse economics
- Botswana Did Everything RightCremieux Recueil · Read, not included
public opposition to data
- Do Data Centers Cause... Indian Immigration?Cremieux Recueil · Read, not included
Decline of Deviance discussion
- The Decline of Deviance 2Experimental History · Read, not included
human growth hormone usage
- Will Making Your Kids Tall Shorten Their Lives?Cremieux Recueil · Read, not included
About Scott's Mixtape Substack
Should you add Scott's Mixtape Substack to Scottie?
Do more credible empirical research while tools and research workflows are changing.[1][2][3]
Who it’s for
An applied researcher who wants causal-inference tutorials, econometrics, live research, and frank notes on where AI agents help or break the work.[1][2]
What you’ll find in it
The issues linked below include “Claude Code 27: Research and Publishing Are Now Two Different Things”, “Sex recessions, rise in casual sex, online dating, declining trust, screening and other changes in US sex patterns”, and “Which research mistakes matter for human capital?”. Open them to judge the publication in its own words.
- Byline
- scott cunningham[1]
How to read Scott's Mixtape Substack with Scottie
Add its public feed for publicly available issues. If you subscribe to paid issues, forward those emails to your private @scottie.fyi address. Scottie does not need access to your inbox.
Scottie read 13 Scott's Mixtape Substack items and included 5 for the brief. The sources covered different things, so Scottie kept their stories separate instead of forcing a connection.
Read three issues from Scott's Mixtape Substack
Read the publication in its own words. Scottie keeps these original links attached; it does not replace the writing.
Sources and official links
- Scott's Mixtape Substack official publication and public archive
- Claude Code 27: Research and Publishing Are Now Two Different Things
- Sex recessions, rise in casual sex, online dating, declining trust, screening and other changes in US sex patterns
- Which research mistakes matter for human capital?
- Experimental History official publication
- Cremieux Recueil official publication