brycewang-stanford/Auto-Empirical-Research-Skills / skills/40-py-econometrics-pyfixest
pyfixest-reference
VerifiedDense, machine-readable API reference for PyFixest — high-dimensional fixed-effects OLS/WLS/IV and Poisson (feols, fepois, feglm), clustered/robust standard errors, R-style formula syntax, and post-estimation. Use when writing or debugging Python fixed-effects regressions with the pyfixest package.
PyFixest LLM Skill Reference
Dense, machine-readable reference for LLMs. No prose padding. Version: matches latest PyFixest release.
Package Import
import pyfixest as pf
Core Estimation Functions
pf.feols() — OLS / WLS / IV with Fixed Effects
pf.feols(
fml: str, # Formula: "Y ~ X1 + X2 | fe1 + fe2" or IV: "Y ~ exog | fe | endog ~ inst"
data: pd.DataFrame,
vcov: str | dict = None, # "iid", "HC1"-"HC3", {"CRV1": "clust"}, {"CRV3": "clust"}, {"CRV1": "c1+c2"}
weights: str = None, # Column name for weights
ssc: dict = None, # Small sample correction, see pf.ssc()
fixef_rm: str = "singleton", # "none" or "singleton"
drop_intercept: bool = False,
split: str = None, # Column name to split sample by
fsplit: str = None, # Like split but also fits on full sample
weights_type: str = "aweights", # "aweights" or "fweights"
solver: str = "scipy.linalg.solve",
lean: bool = False,
) -> Feols | FixestMulti
Returns Feols for single model, FixestMulti for multiple estimation syntax.
pf.fepois() — Poisson Regression with Fixed Effects
pf.fepois(
fml: str, # "Y ~ X1 + X2 | fe1 + fe2"
data: pd.DataFrame,
vcov: str | dict = None,
ssc: dict = None,
fixef_rm: str = "singleton",
iwls_tol: float = 1e-08,
iwls_maxiter: int = 25,
separation_check: list[str] = None, # ["fe"] to check for separated FE
split: str = None,
fsplit: str = None,
) -> Fepois | FixestMulti
pf.feglm() — GLM (without FE demeaning, WIP)
pf.feglm(
fml: str,
data: pd.DataFrame,
family: str, # "gaussian", "logit", "probit"
vcov: str | dict = None,
separation_check: list[str] = None,
split: str = None,
fsplit: str = None,
) -> Feglm | FixestMulti
pf.quantreg() — Quantile Regression
pf.quantreg(
fml: str,
data: pd.DataFrame,
vcov: str | dict = "nid",
quantile: float | list[float] = 0.5, # Single or list of quantiles
method: str = "fn", # "fn" (Frisch-Newton)
split: str = None,
fsplit: str = None,
) -> Feols | FixestMulti
Formula Syntax
Basic
"Y ~ X1 + X2" # OLS
"Y ~ X1 + X2 | fe1" # OLS + one FE
"Y ~ X1 + X2 | fe1 + fe2" # OLS + two-way FE
"Y ~ X1 + C(categorical)" # Categorical variable as dummies
"Y ~ X1 + i(factor, ref=0)" # Factor variable for event studies
"Y ~ X1 + i(f1, X2)" # Interaction: factor × continuous
"Y ~ 1 | fe1 | X1 ~ Z1" # IV: depvar ~ exog | fe | endog ~ inst
"Y ~ 1 | X1 ~ Z1 + Z2" # IV without FE
"Y ~ X1:X2" # Interaction only
"Y ~ X1*X2" # X1 + X2 + X1:X2
"Y ~ X1 + I(X1**2)" # Polynomial term
Multiple Estimation Operators
"Y ~ X1 + sw(X2, X3)" # Stepwise: two models, X2 then X3
"Y ~ X1 + sw0(X2, X3)" # Stepwise with empty: three models
"Y ~ X1 + csw(X2, X3)" # Cumulative stepwise: X2, then X2+X3
"Y ~ X1 + csw0(X2, X3)" # Cumulative with empty: three models
"Y + Y2 ~ X1" # Multiple dependent variables
"Y ~ X1 | csw0(fe1, fe2)" # Stepwise fixed effects
Split Sample
pf.feols("Y ~ X1 | fe1", data=data, split="group_var") # Separate by group
pf.feols("Y ~ X1 | fe1", data=data, fsplit="group_var") # Separate + full sample
Post-Estimation Methods (Feols object)
Extracting Results
fit.summary() # Print summary
fit.tidy(alpha=0.05) # pd.DataFrame: Estimate, Std. Error, t value, Pr(>|t|), CI
fit.coef() # pd.Series of coefficients
fit.se() # pd.Series of standard errors
fit.tstat() # pd.Series of t-statistics
fit.pvalue() # pd.Series of p-values
fit.confint(alpha=0.05) # pd.DataFrame of confidence intervals
fit.confint(joint=True) # Simultaneous confidence bands (multiplier bootstrap)
Changing Inference
fit.vcov("iid") # IID standard errors
fit.vcov("HC1") # Heteroskedasticity-robust
fit.vcov({"CRV1": "cluster_var"}) # One-way cluster-robust
fit.vcov({"CRV3": "cluster_var"}) # CRV3 cluster-robust
fit.vcov({"CRV1": "c1 + c2"}) # Two-way clustering
Returns self — chainable: fit.vcov("HC1").summary().
Visualization
fit.coefplot() # Coefficient plot
pf.coefplot([fit1, fit2], keep="X1") # Compare models
pf.iplot([fit1, fit2], coord_flip=False) # Event study plot (for i() vars)
pf.qplot(fit_qr) # Quantile regression plot
Prediction
fit.predict() # In-sample predictions
fit.predict(newdata=df_new) # Out-of-sample
fit.predict(type="response") # Response scale (GLMs)
fit.predict(type="link") # Link scale (GLMs)
Inference Methods
# Wild cluster bootstrap
fit.wildboottest(param="X1", reps=999, cluster="clust_var")
# Randomization inference
fit.ritest(resampvar="X1=0", reps=1000, cluster="group_id")
# Causal cluster variance estimator (Abadie et al. 2023)
fit.ccv(treatment="treat_var", pk=0.05, n_splits=2, seed=42)
# Wald test: H0: beta = 0
fit.wald_test(R=np.eye(k))
# Wald test: H0: R @ beta = q
fit.wald_test(R=R_matrix, q=q_vector)
IV Diagnostics (Feiv objects)
fit_iv._model_1st_stage # First stage Feols object
fit_iv._f_stat_1st_stage # First stage F-statistic
fit_iv.IV_Diag() # Run IV diagnostics
fit_iv._eff_F # Effective F-stat (Olea & Pflueger 2013)
Online Learning
fit.update(X_new, y_new) # Sherman-Morrison coefficient update
Reporting Functions
pf.etable() — Regression Tables
pf.etable(
models, # Feols, list[Feols], or FixestMulti
type: str = "gt", # "gt" (Great Tables), "tex" (LaTeX), "md" (markdown), "df" (DataFrame)
signif_code: list = None, # e.g. [0.001, 0.01, 0.05]
coef_fmt: str = "b \n (se)", # Format: b=coef, se=SE, p=pval, t=tstat, ci_l, ci_u
keep: str | list = None, # Regex pattern(s) to keep
drop: str | list = None, # Regex pattern(s) to drop
labels: dict = None, # {"old_name": "New Label"}
felabels: dict = None, # {"fe_var": "FE Label"}
show_fe: bool = True,
show_se_type: bool = True,
notes: str = "",
model_heads: list = None, # Custom column headers
caption: str = None, # Via kwargs
file_name: str = None, # Save to file (.tex, .html)
)
pf.summary() — Print Results
pf.summary(models, digits=3) # models: Feols, list, or FixestMulti
pf.dtable() — Descriptive Statistics
pf.dtable(
df: pd.DataFrame,
vars: list, # Column names
stats: list = None, # ["count", "mean", "std", "min", "max", "median"]
bycol: list[str] = None, # Group columns (shown as separate column groups)
byrow: str = None, # Group variable (shown as row sections)
type: str = "gt", # "gt", "tex", "md", "df"
labels: dict = None,
digits: int = 2,
)
Multiple Testing Corrections
pf.bonferroni(models, param="X1") # Bonferroni adjusted p-values
pf.rwolf(models, param="X1", reps=9999, seed=42) # Romano-Wolf
pf.wyoung(models, param="X1", reps=9999, seed=42) # Westfall-Young
Difference-in-Differences
pf.event_study() — Unified Event Study API
pf.event_study(
data: pd.DataFrame,
yname: str, # Outcome column
idname: str, # Unit ID column
tname: str, # Time column
gname: str, # Group (first treatment period) column
xfml: str = None, # Additional covariates formula
cluster: str = None, # Cluster variable
estimator: str = "twfe", # "twfe" or "did2s"
att: bool = True,
)
pf.did2s() — Gardner's Two-Stage DID
pf.did2s(
data: pd.DataFrame,
yname: str,
first_stage: str, # "~ covariates | fe1 + fe2"
second_stage: str, # "~ i(rel_year, ref=-1.0)"
treatment: str, # Treatment indicator column
cluster: str,
weights: str = None,
)
pf.lpdid() — Local Projections DID
pf.lpdid(
data: pd.DataFrame,
yname: str,
idname: str,
tname: str,
gname: str,
vcov: str | dict = None,
pre_window: int = None,
post_window: int = None,
never_treated: int = 0,
att: bool = True,
xfml: str = None,
)
pf.panelview() — Treatment Visualization
pf.panelview(data, unit="unit_col", time="time_col", treat="treat_col")
Small Sample Correction
pf.ssc(
k_adj: bool = True, # Adjust for number of estimated parameters
k_fixef: str = "nonnested", # "nonnested" or "nested" FE adjustment
G_adj: bool = True, # Adjust for number of clusters
G_df: str = "min", # "min" or "conventional"
)
# Usage:
pf.feols("Y ~ X1 | fe1", data=data, ssc=pf.ssc(k_adj=True))
Data Generators
pf.get_data(N=1000, seed=1234, model="Feols") # Synthetic data: "Feols" or "Fepois"
pf.get_twin_data(N_pairs=500, seed=42) # Twin study data (returns to education)
pf.get_worker_panel(N_workers=500, N_firms=50, N_years=11, seed=42) # Worker-firm panel
Variance-Covariance Options
| vcov | Description |
|---|---|
"iid" | Spherical errors (homoskedastic, uncorrelated) |
"HC1" | Heteroskedasticity-robust (White) |
"HC2" | HC2 robust |
"HC3" | HC3 robust (jackknife-like) |
{"CRV1": "var"} | One-way cluster-robust |
{"CRV3": "var"} | CRV3 cluster-robust |
{"CRV1": "v1 + v2"} | Two-way clustering |
Default: CRV1 clustered by first FE variable (if FE present), else "iid".
Common Patterns
# Basic OLS with FE and clustering
fit = pf.feols("Y ~ X1 + X2 | fe1 + fe2", data=df, vcov={"CRV1": "fe1"})
# IV regression
fit_iv = pf.feols("Y ~ exog | fe1 | endog ~ instrument", data=df)
# Multiple specifications at once
fits = pf.feols("Y ~ X1 | csw0(fe1, fe2, fe3)", data=df)
fits.etable()
# Poisson with FE
fit_pois = pf.fepois("count ~ X1 + X2 | fe1", data=df)
# Event study
fit_es = pf.feols("Y ~ i(rel_time, ref=-1) | unit + time", data=df)
pf.iplot(fit_es)
# Publication table
pf.etable([fit1, fit2, fit3], type="tex", file_name="table1.tex",
labels={"X1": "Treatment"}, felabels={"fe1": "Unit FE"})
# Adjust SE after estimation
fit.vcov({"CRV1": "cluster"}).summary()
# Compare R fixest syntax → PyFixest
# R: feols(Y ~ X1 | fe1, data, cluster = ~fe1)
# Py: pf.feols("Y ~ X1 | fe1", data=data, vcov={"CRV1": "fe1"})
FixestMulti Methods
When multiple estimation syntax is used, returns FixestMulti:
multi = pf.feols("Y + Y2 ~ X1 | csw0(fe1, fe2)", data=df)
multi.etable() # Table of all models
multi.summary() # Print all summaries
multi.coefplot() # Plot all models
multi.vcov("HC1") # Update all models' inference
multi.fetch_model(0) # Get first Feols object
multi.all_fitted_models["Y~X1"] # Access by formula key
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-changelog-updatemorluto/reaPrepare or rewrite REA release changelogs and GitHub release notes from pinned Git history, with verified contributor thanks and Release Please synchronization. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| academy-guideanthropics/skillsStop and check this skill before finishing any reply to a question about how to use Claude or a Claude product — it recommends matching courses, tutorials, and use cases from Claude Academy (academy.claude.com), Anthropic's learning hub. Trigger on: "how do I", "how can I", "getting started with", "what can Claude do", "teach me", "learn to use"; questions about artifacts, projects, skills, plugins, connectors, MCP; requests about rolling Claude out to a team, class, or organization; and any ask for training materials, onboarding content, or learning resources. Use it when the user is learning how to use a feature or product — not when they are mid-task and just want the task done. This skill composes with other skills: after consulting product documentation to answer how a Claude feature works, also check here for a matching course or tutorial — a docs-grounded answer and an Academy recommendation belong together. Only recommend on a strong match; never invent Academy content. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| claude-apianthropics/skillsReference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file). | Stars: 198k | Updated | Claude CodeCodex | Unverified | Install |
| doc-coauthoringanthropics/skillsGuide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |