Use when aer-identification has fixed the design, after methodology choice and before aer-robustness or aer-tables-figures, to run an AER-track analysis with StatsPAI — the agent-native Python engine and MCP server for causal inference, robustness, sensitivity, and publication-ready table export.
SkillScore77Maintained
Automatic
◆Data Analyst pack
Last verified 8/8/2026
What it changes
Passed 15 of 19 test cases with the skill, and 4 of 19 without it.
Without skill23%
With skill77%
Checks it passed that the model alone missed
Uses bartik to execute the shift-share estimation
Uses rddensity to execute the McCrary density test
Uses interactive_fe to fit interactive fixed effects models
and 14 more in the full report
12 fixed5 passed either way5 still failing0 broke
Pass-rate lift
+55 pp
Tokens
+82%
Agent turns
0%
Response time
-46%
Model gemini-3.6-flashMeasured 8/8/2026v117 of 22 cases passed with skill · 19 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 10
Prompt
In an administrative threshold design, we need to test for manipulation or sorting of running variable values near the cutoff. Which function in our econometric engine conducts this density discontinuity test?
What was checked, without and with the skill
Checked
Without skill
With skill
Uses rddensity to execute the McCrary density test
Missed
Passed
case-15 · gemini-3.6-flash · 8/8/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md227 linessingle file
AER StatsPAI
Overview
aer-statspai is the implementation engine option for this stack. Where
aer-identification and aer-robustness decide which estimator and which
diagnostics a referee will demand, this skill is about running them — through
StatsPAI, an open-source,
agent-native Python platform that exposes 1,000+ causal-inference and
econometrics functions behind one unified API, plus a machine-readable MCP
server an agent can drive directly.
It is one more choice, not a replacement. The Stata / R / Python templates in
templates/ remain the default for users who want drop-in, version-pinned
scripts. Reach for StatsPAI when you want a single Python surface that covers the
whole AER identification toolkit, self-describes its assumptions to an agent, and
exports publication-ready LaTeX / Word / Excel tables from the estimator object.
This skill does not override the methodology. The modern-default rules in
aer-identification (no TWFE on staggered data, Anderson-Rubin under weak IV,
local-linear RDD, placebo inference for SCM) still bind. StatsPAI executes those
rules; it does not relax them.
When to Use
You want to run the empirical analysis interactively from the agent, not
just receive template code to run later by hand
You want one Python dependency covering DiD, IV, RDD, SCM, matching, DML, and
causal forests instead of stitching together fixest, did, rdrobust,
linearmodels, and scdata
A StatsPAI MCP server is connected (tools prefixed statspai) and you can
chain detect_design → recommend → fit → audit_result → sensitivity → bibtex
You need publication-ready tables straight from the estimator object
(.to_latex(), .to_docx(), .to_excel()) to hand to aer-tables-figures
When NOT to Use
The deposited replication package must be Stata-only (some coauthors /
Data Editor workflows assume .do files) → use templates/stata/
You want exact, pinned, language-native scripts to commit verbatim into an