statsmodels.genmod.generalized_estimating_equations.GEE.fit

GEE.fit(maxiter=60, ctol=1e-06, start_params=None, params_niter=1, first_dep_update=0, cov_type='robust', ddof_scale=None, scaling_factor=1.0)

Fits a marginal regression model using generalized estimating equations (GEE).

Parameters:

maxiter : integer

The maximum number of iterations

ctol : float

The convergence criterion for stopping the Gauss-Seidel iterations

start_params : array-like

A vector of starting values for the regression coefficients. If None, a default is chosen.

params_niter : integer

The number of Gauss-Seidel updates of the mean structure parameters that take place prior to each update of the dependence structure.

first_dep_update : integer

No dependence structure updates occur before this iteration number.

cov_type : string

One of “robust”, “naive”, or “bias_reduced”.

ddof_scale : scalar or None

The scale parameter is estimated as the sum of squared Pearson residuals divided by N - ddof_scale, where N is the total sample size. If ddof_scale is None, the number of covariates (including an intercept if present) is used.

scaling_factor : scalar

The estimated covariance of the parameter estimates is scaled by this value. Default is 1, Stata uses N / (N - g), where N is the total sample size and g is the average group size.

Returns:

An instance of the GEEResults class or subclass

Notes

If convergence difficulties occur, increase the values of first_dep_update and/or params_niter. Setting first_dep_update to a greater value (e.g. ~10-20) causes the algorithm to move close to the GLM solution before attempting to identify the dependence structure.

For the Gaussian family, there is no benefit to setting params_niter to a value greater than 1, since the mean structure parameters converge in one step.