Too Early To Say is Victoria Cholette's independent applied-economics lab. AI agents draft code, search records, and run candidate analyses. Victoria sets the economic question, identification strategy, checks, and interpretation.
The work is for applied economists, policy analysts, and research leaders who need to know whether faster production still produces defensible evidence.
The research asks whether AI shifts the real cost of policy work away from running the analysis and toward specifying the question, checking the answer, and judging what it means.
The question is an economic one. Which tasks move to the agent, where new errors enter, which checks catch them, and who stays accountable for the policy claim.
Evidence on time saved, errors introduced, and review work required.
In Difference-in-Differences in Python, the data-generating process plants an effect of 1.60. The estimator runs clean and still returns the wrong number when it targets the wrong comparison.
Agents: implement and rerun candidate estimators. Researcher: defines the estimand, answer key, and failure rule.
Policy findings with traceable data, methods, and limits.
In The Food Desert Myth, the published 4.7-fold comparison is under reconciliation across the article, data, and public materials until it matches a public record.
Agents: assemble and check source files. Researcher: defines access, defends the comparison, and interprets the finding.
Reusable checks that catch clean-looking failures.
In Instrumental Variables in Python, an F-statistic of 2,051 coexists with a causally invalid estimate.
Agents: run estimators and diagnostics. Researcher: specifies the simulated treatment and outcome equations and decides which claim the evidence supports.
A balanced covariate table can coexist with a missing counterfactual.
Why a clean TWFE coefficient can estimate a comparison that does not match the economic question.
A planted-truth test separates code that runs from an estimator that recovers the target.