Statsmodels Cheat Sheet

Updated 2026-05-28

Next Topic: Streamlit Data Application Framework Cheat Sheet

🧠Study flashcards on this topic91 cards · spaced repetition→

Statsmodels is Python's inference-focused statistics and econometrics library, sitting on top of NumPy, SciPy, pandas, and Patsy. It matters because it combines model estimation with the outputs practitioners actually need: standard errors, hypothesis tests, diagnostics, confidence intervals, and interpretable summaries. A useful mental model is that statsmodels is less about black-box prediction and more about specifying the model carefully, checking assumptions, and defending the conclusions after fitting. As of version 0.14, major additions include treatment-effect estimation, hurdle and truncated count models, multiple seasonal decomposition (MSTL), and extended copula support.

What This Cheat Sheet Covers

This topic spans 11 focused tables and 147 indexed concepts, 91 flashcards. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.

Table 1: Core APIs and WorkflowTable 2: Linear and Robust Regression ModelsTable 3: GLM, Correlated Data, and Smooth ModelsTable 4: Discrete, Count, and Limited Dependent Variable ModelsTable 5: Hypothesis Tests, ANOVA, and Multiple ComparisonsTable 6: Regression Diagnostics and Inference ChecksTable 7: Time Series Diagnostics and DecompositionTable 8: Univariate Forecasting and Structural Time SeriesTable 9: Multivariate, State Space, and Regime-Switching ModelsTable 10: Nonparametric, Survival, Copulas, and Custom ModelingTable 11: Causal Inference and Missing Data

Quick IndexSubscribe to unlock

A jump-to index of every table row in this cheat sheet.

Mind MapSubscribe to unlock

An interactive map of every table and concept in this topic.

Table 1: Core APIs and Workflow

The two import surfaces — matrix-based sm and formula-based smf — determine how you build design matrices; almost every other call in the library flows from the results object returned by .fit().

Method	Example	Description
statsmodels.api	`import statsmodels.api as sm`	Main import surface for matrix-based model classes, datasets, graphics, and statistical tools.
statsmodels.formula.api	`import statsmodels.formula.api as smf`	Formula interface using R-style formulas and Patsy design matrices.
add_constant	`X = sm.add_constant(X)`	Adds an explicit intercept column for matrix-based models that do not include one by default.
from_formula	`sm.OLS.from_formula('y ~ x1 + x2', data=df)`	Alternate constructor that builds the design matrix from a formula and DataFrame.
fit	`res = sm.OLS(y, X).fit()`	Estimates the model and returns a results object with inference and diagnostics.
summary	`res.summary()`	Produces the standard text summary with coefficients, tests, fit metrics, and diagnostics.
params	`res.params`	Estimated coefficients in the fitted model.
predict	`res.predict(X_new)`	Returns fitted means or linear predictions for new observations.
get_prediction	`pred = res.get_prediction(X_new)`	Prediction wrapper that also tracks uncertainty for intervals and summaries.
summary_frame	`pred.summary_frame()`	Returns predictions and interval columns in a tidy DataFrame.
conf_int	`res.conf_int()`	Confidence intervals for fitted parameters.
info_criteria	`res.info_criteria('bic')`	Computes information criteria such as AIC, BIC, AICc, and HQIC.
load_pandas	`data = sm.datasets.longley.load_pandas()`	Loads a built-in statsmodels dataset as a pandas-backed Dataset object with `endog` and `exog`.
get_rdataset	`sm.datasets.get_rdataset('Duncan', 'carData')`	• Downloads and returns any dataset from the R Rdatasets collection • data is in the `.data` attribute

Table 2: Linear and Robust Regression Models

These models cover the classical linear and robust regression family. The matrix-based API requires add_constant; the formula API handles the intercept automatically. Covariance type (cov_type) controls standard error estimation without refitting.

Model	Example	Description
OLS	`res = sm.OLS(y, X).fit()`	Standard ordinary least squares for linear regression and classical inference.

The rest of this section is part of the full cheat sheet

Table 3: GLM, Correlated Data, and Smooth Models

GLMs extend linear regression to non-normal outcomes via a link function and variance structure; GEE and mixed models further generalize to correlated and clustered observations. Choosing the right family, link, and correlation structure is the central modelling decision in this table.

Model	Example	Description
GLM	`res = sm.GLM(y, X, family=sm.families.Poisson()).fit()`	Generalized linear model for exponential-family outcomes with a link function.

The rest of this section is part of the full cheat sheet

Table 4: Discrete, Count, and Limited Dependent Variable Models

This table covers the full range of models for binary, categorical, count, and zero-heavy outcomes — from standard Logit through zero-inflated and hurdle variants added in 0.14. Choosing between zero-inflated and hurdle models hinges on whether zeros arise from a separate process or from the same underlying count distribution.

Model	Example	Description
Logit	`res = sm.Logit(y, X).fit()`	Binary logistic regression estimated by maximum likelihood.

The rest of this section is part of the full cheat sheet

Table 1: Core APIs and Workflow

Method	Example	Description
statsmodels.api	`import statsmodels.api as sm`	Main import surface for matrix-based model classes, datasets, graphics, and statistical tools.
statsmodels.formula.api	`import statsmodels.formula.api as smf`	Formula interface using R-style formulas and Patsy design matrices.
add_constant	`X = sm.add_constant(X)`	Adds an explicit intercept column for matrix-based models that do not include one by default.
from_formula	`sm.OLS.from_formula('y ~ x1 + x2', data=df)`	Alternate constructor that builds the design matrix from a formula and DataFrame.
fit	`res = sm.OLS(y, X).fit()`	Estimates the model and returns a results object with inference and diagnostics.
summary	`res.summary()`	Produces the standard text summary with coefficients, tests, fit metrics, and diagnostics.
params	`res.params`	Estimated coefficients in the fitted model.
predict	`res.predict(X_new)`	Returns fitted means or linear predictions for new observations.
get_prediction	`pred = res.get_prediction(X_new)`	Prediction wrapper that also tracks uncertainty for intervals and summaries.
summary_frame	`pred.summary_frame()`	Returns predictions and interval columns in a tidy DataFrame.
conf_int	`res.conf_int()`	Confidence intervals for fitted parameters.
info_criteria	`res.info_criteria('bic')`	Computes information criteria such as AIC, BIC, AICc, and HQIC.
load_pandas	`data = sm.datasets.longley.load_pandas()`	Loads a built-in statsmodels dataset as a pandas-backed Dataset object with `endog` and `exog`.
get_rdataset	`sm.datasets.get_rdataset('Duncan', 'carData')`	• Downloads and returns any dataset from the R Rdatasets collection • data is in the `.data` attribute

Table 2: Linear and Robust Regression Models

Model	Example	Description
OLS	`res = sm.OLS(y, X).fit()`	Standard ordinary least squares for linear regression and classical inference.

Table 3: GLM, Correlated Data, and Smooth Models

Model	Example	Description
GLM	`res = sm.GLM(y, X, family=sm.families.Poisson()).fit()`	Generalized linear model for exponential-family outcomes with a link function.

Table 4: Discrete, Count, and Limited Dependent Variable Models

Model	Example	Description
Logit	`res = sm.Logit(y, X).fit()`	Binary logistic regression estimated by maximum likelihood.