StatsPAI

Install

pip install statspai

Python 3.10 – 3.13 on Linux and macOS. The core install covers estimation, diagnostics, the bundled datasets, .xlsx / .docx / LaTeX export, and matplotlib figures (.plot(), sp.ggdid() and the rest). seaborn / plotly styles and heavier backends are optional extras:

The interactive figure editor additionally needs ipywidgets inside Jupyter.

import statspai as sp

print(sp.datasets.list_datasets()[["name", "design", "source"]])

sp is the convention used throughout these docs, but the alias is your choice. If sp is already taken in your code (for example by import scipy as sp), import statspai as si works just as well.

StatsPAI bundles 14 datasets that load offline. Most are real published extracts (source == "bundled CSV"): Card (1995) NLSYM schooling data, LaLonde/NSW with a PSID comparison group, the U.S. Senate RD data distributed with R's rdrobust, California Proposition 99, the castle-doctrine panel, and NHEFS, among others. A few are deterministic simulated replicas calibrated to a published design (source == "simulated"), including the Callaway–Sant'Anna mpdta panel used below; their numbers are not the numbers from the original data.

At a glance: 1,627 registered functions across 90 submodules; 729k LOC (core) + 496k LOC (tests). Run python scripts/registry_stats.py to reproduce these numbers.

---

Beginner Examples With Results

The outputs below were produced with StatsPAI 1.29.0 on the bundled datasets and are pinned by tests/test_readme_examples.py, so they cannot drift from the code silently. Example 6 uses the Stata vce() grammar and post-estimation commands introduced in 1.29.0; on an older release install from source with pip install "statspai @ git+https://github.com/brycewang-stanford/StatsPAI". Long summaries are abridged (... marks omitted lines); the numbers are pinned by tests/test_readme_examples.py and tests/test_synth_placebo_pvalue.py, so they cannot silently drift from the code again.

Everyday Workflow

import statspai as sp

card = sp.datasets.card_1995()
r1 = sp.regress(
    "lwage ~ educ + exper + expersq + black + south + smsa",
    data=card,
    robust="hc1",
)
r2 = sp.ivreg("lwage ~ (educ ~ nearc4) + exper + expersq + black + south + smsa", data=card)

print(r1.summary())                          # human-readable table
print(r1.tidy().head())                      # broom-style dataframe
print(r1.test("black = south"))              # Wald test, like Stata's `test`
print(r1.lincom("black - south"))            # like Stata's `lincom`
tbl = sp.regtable(r1, r2, model_labels=["OLS", "2SLS"])
tbl.to_word("table.docx")                    # Word table
tbl.to_excel("results.xlsx")                 # Excel table

Useful docs:

- Getting started and Cookbook - Choosing an estimator: DiD, IV, RD, matching, synthetic control - Migrating: from R, Stata/R command translators, shared argument grammar - Exporting regression tables - Agents: agent API, MCP workflow for economists - Evidence: stability and validation tiers, parity matrix, tier census

---

Compared With Other Python Packages

StatsPAI aims to be one broad Stata/R-style workbench. Several focused Python packages do one part of that job, often with a longer track record; if you only need that part, they are good choices.

Use StatsPAI when you want one package, one function registry, and one agent interface across the everyday Stata/R empirical workflow.

---

feols: bit-exact vs fixest::feols [py/R/Stata] (headline rel_est 5.2e-15 within rel_est<=1e-06, rel_se<=1e-06)

s["reference_versions"] # {'R': 'R version 4.5.2 (2025-10-31)', 'fixest': '0.14.0'}

sp.parity_summary() # coverage counts, including the unverified gap sp.parity_matrix(status="bit-exact")

Grades: `bit-exact` (headline relative error ≤ 1e-6 against a named R/Stata
reference), `aligned` (a documented, pre-registered looser tolerance),
`analytical-only` (recovers a known DGP truth or closed-form identity),
`external-replication` (reproduces published-paper numbers), and `unverified`
(registered but no parity evidence attached **yet** — the honest gap). The full,
auto-generated matrix is published at
[docs/parity.md](https://brycewang-stanford.github.io/StatsPAI/parity/).

For your own data, `sp.cross_validate` re-runs one estimand through every
independent engine installed locally and reports whether they agree: