statsmodels.discrete.discrete_model.LogitResults.resid_dev¶

LogitResults.resid_dev()¶

Deviance residuals

Notes

Deviance residuals are defined

$d_j = \pm\left(2\left[Y_j\ln\left(\frac{Y_j}{M_jp_j}\right) + (M_j - Y_j\ln\left(\frac{M_j-Y_j}{M_j(1-p_j)} \right) \right] \right)^{1/2}$

where

$p_j = cdf(X\beta)$ and $M_j$ is the total number of observations sharing the covariate pattern $j$ .

For now $M_j$ is always set to 1.