statsmodels.genmod.generalized_estimating_equations.GEE.fit_regularized

GEE.fit_regularized(pen_wt, scad_param=3.7, maxiter=100, ddof_scale=None, update_assoc=5, ctol=1e-05, ztol=0.001, eps=1e-06, scale=None)[source]

Regularized estimation for GEE.

Parameters:
pen_wtfloat

The penalty weight (a non-negative scalar).

scad_paramfloat

Non-negative scalar determining the shape of the Scad penalty.

maxiterint

The maximum number of iterations.

ddof_scaleint

Value to subtract from nobs when calculating the denominator degrees of freedom for t-statistics, defaults to the number of columns in exog.

update_assocint

The dependence parameters are updated every update_assoc iterations of the mean structure parameter updates.

ctolfloat

Convergence criterion, default is one order of magnitude smaller than proposed in section 3.1 of Wang et al.

ztolfloat

Coefficients smaller than this value are treated as being zero, default is based on section 5 of Wang et al.

epsnon-negative scalar

Numerical constant, see section 3.2 of Wang et al.

scalefloat or str

If a float, this value is used as the scale parameter. If “X2”, the scale parameter is always estimated using Pearson’s chi-square method (e.g. as in a quasi-Poisson analysis). If None, the default approach for the family is used to estimate the scale parameter.

Returns:
GEEResults instance. Note that not all methods of the results
class make sense when the model has been fit with regularization.

Notes

This implementation assumes that the link is canonical.

References

Wang L, Zhou J, Qu A. (2012). Penalized generalized estimating equations for high-dimensional longitudinal data analysis. Biometrics. 2012 Jun;68(2):353-60. doi: 10.1111/j.1541-0420.2011.01678.x. https://www.ncbi.nlm.nih.gov/pubmed/21955051 http://users.stat.umn.edu/~wangx346/research/GEE_selection.pdf