statsmodels

0.14.6 · active · verified Sun Mar 29

statsmodels is a Python package offering a wide array of statistical models, hypothesis tests, and statistical data exploration tools. It provides classes and functions for the estimation of many different statistical models, including linear regression, generalized linear models, discrete choice models, and time series analysis. Currently at version 0.14.6, the library follows a loose, long time-based release cycle for its dependencies, typically updating minimal versions every one and a half to two years. [2, 3, 5, 7]

Warnings

breaking The `scikits` namespace was deprecated and eventually removed in versions prior to 0.5.0. Direct imports from `scikits.statsmodels` are no longer valid.
Fix: Always use `import statsmodels.api as sm` or direct imports from `statsmodels.<submodule>` (e.g., `statsmodels.regression.linear_model`). [29, 31]
breaking The signature of `model.predict` methods changed in versions prior to 0.5.0. It now explicitly requires the `params` argument (e.g., `model.predict(params, exog)`), rather than assuming the model has already been fit and omitting `params`.
Fix: Ensure `model.predict` calls explicitly pass the `params` argument from the fitted model, e.g., `results.predict(exog)` or `model.predict(results.params, exog)`. [29, 31]
deprecated The `statsmodels.tsa.arima_model.ARMA` and `statsmodels.tsa.arima_model.ARIMA` classes have been deprecated. Using them will raise a `FutureWarning`.
Fix: Migrate to `statsmodels.tsa.arima.model.ARIMA`. The new API provides more consistent handling and features. [34]
gotcha When using the direct `statsmodels.api.OLS(y, X)` interface (without formulas), an intercept term (constant) is NOT automatically added to the `X` (exog) design matrix. This differs from some other statistical software and can lead to incorrect models if an intercept is expected.
Fix: Explicitly add a constant term using `X = sm.add_constant(X)` from `statsmodels.api` before fitting the model, or use the `statsmodels.formula.api` interface which handles intercepts automatically. [19, 33]
breaking Pandas' `Panel` object and `pandas.stats.ols` (among others) were deprecated and removed in Pandas 0.20.1 and later. Users relying on these for panel data or OLS directly from Pandas will need to switch.
Fix: For OLS functionality, `statsmodels.api.OLS` is the recommended replacement. For panel data, Pandas recommends using a `MultiIndex` DataFrame or `xarray`, which can then be used with `statsmodels` models where appropriate (e.g., `MixedLM` for some panel-like structures). [37]
breaking Statsmodels 0.14.2 introduced compatibility with NumPy 2.0.0. While `statsmodels` itself may run on older NumPy versions, if you upgrade to NumPy 2.0, all other Python scientific stack dependencies (like SciPy and Pandas) *must also be NumPy 2.0 compatible* to avoid runtime issues. This release also increased the minimum Python version to 3.9 to match NumPy 2.0.
Fix: Ensure your entire scientific Python environment has compatible versions of all libraries when moving to NumPy 2.0. Check dependency release notes for NumPy 2.0 compatibility. [35]

Install

pip install statsmodels PyPI

Imports

statsmodels.api
```
import statsmodels.api as sm
```
Main entry point for most common models (e.g., OLS, GLM) when using NumPy arrays or pre-processed Pandas DataFrames. It's stable and recommended for direct model fitting. [18, 20]
statsmodels.formula.api
```
import statsmodels.formula.api as smf
```
Provides an R-style formula interface, highly recommended for exploratory data analysis and when working directly with Pandas DataFrames and categorical variables. [3, 18, 20]
Specific Submodule
```
from statsmodels.tsa.arima.model import ARIMA
```
Direct imports from submodules are used for specialized functionality (e.g., time series). Be aware that older submodule paths (like `statsmodels.tsa.arima_model`) might be deprecated or removed in newer versions, use the `statsmodels.tsa.arima.model` path instead. [20, 34]

Quickstart

This example demonstrates how to fit a simple Ordinary Least Squares (OLS) regression model using the R-style formula interface provided by `statsmodels.formula.api`. It shows creating sample data, defining the model with a formula, fitting it, and then printing a comprehensive summary of the results, including coefficients, R-squared, and various statistical tests. [3, 6, 33]

import statsmodels.formula.api as smf
import pandas as pd
import numpy as np

# 1. Create a sample DataFrame
np.random.seed(42)
data = {
    'y': 10 + 2 * np.random.rand(100) + 3 * np.random.randn(100),
    'x1': np.random.rand(100) * 10,
    'x2': np.random.randint(0, 2, 100) # categorical variable example
}
df = pd.DataFrame(data)

# 2. Fit OLS (Ordinary Least Squares) model using R-style formula
#    'y ~ x1 + C(x2)' means y is dependent on x1 and categorical x2
model = smf.ols('y ~ x1 + C(x2)', data=df)
results = model.fit()

# 3. Print the summary of the regression results
print(results.summary())

view raw JSON →