SkillAgentSearch skills...

pyfixest-reference

Dense, 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

Install / Use

npx skills add brycewang-stanford/Auto-Empirical-Research-Skills --skill 40-py-econometrics-pyfixest

Installs into whichever agent you are using.

About this skill
📄

SKILL.md

Installable skill definition

Quality Score

95/100

Supported Platforms

Universal

Our assessment of pyfixest-reference

pyfixest-reference scores 95/100 on our quality scale, 150th of 950 AI & Machine Learning skills we index (top 16%).

Its SKILL.md is 12 KB long, well organised into 50 sections with 27 code examples: a thorough specification that gives an agent plenty to work with.

With 4,360 GitHub stars, it is one of the more widely adopted skills in the catalogue.

Substance
30/30
Structure
20/20
Description
15/15
Adoption
15/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated about 2 months ago, so pyfixest-reference is actively maintained.
  • Our last check on 2026-11-15 found the source still online.
  • No license is declared. By default that means all rights are reserved: you can read it, but reusing or redistributing it is not clearly permitted. Ask the author before building on it commercially.
  • Its trust signals score 88/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

Safety scan

No issues found

Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.

AI review by kimi-k2.7-code on 2026-09-29. Automated pattern scan on 2026-09-28. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.

pyfixest-reference compared with similar skills

All 4 of these similar skills score higher than pyfixest-reference; compare them before choosing.

SkillScoreStarsUpdatedFormat
pyfixest-reference (this skill)by brycewang-stanford954.4k54d agoSKILL.md
claude-memby thedotmack100117.6ktodayCLAUDE.md
Agent-Reachby Panniantong100112.3k9d agoCLAUDE.md
Understand-Anythingby Egonex-AI10092.8k1d agoCLAUDE.md
headroomby headroomlabs-ai10079.8ktodayCLAUDE.md

Frequently asked questions

How do I install pyfixest-reference?
Run npx skills add brycewang-stanford/Auto-Empirical-Research-Skills --skill pyfixest-reference. The install tabs above show the steps for each supported agent.
Which AI agents does pyfixest-reference work with?
It is written for Universal, as a SKILL.md file. Other agents that read the same format can often use it too.
Is pyfixest-reference safe to use?
Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It declares no license and scores 88/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is pyfixest-reference still maintained?
The repository was last updated about 2 months ago, so pyfixest-reference is actively maintained.

name: pyfixest-reference description: "Dense, 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." allowed-tools: ["Read", "Write", "Edit", "Bash"]

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

Truncated for display — read the full file on GitHub.

Related Skills

View on GitHub
GitHub Stars4.4k
CategoryAI
Updated1mo ago
Forks527

Languages

Stata

Trust signals

88/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 medium