Cox DFBETA influence diagnostics — morie_cox_dfbeta

The approximate change in each coefficient from deleting each subject, \(L_i I^{-1}\) where \(L_i\) is the score residual. dfbetas rescales by the coefficient's own standard error, so a value near 1 means that one subject moves beta by a full standard error.

Usage

morie_cox_dfbeta_influence(fit)

Arguments

fit: a Cox fit carrying information and se.

Value

list with dfbeta, dfbetas, score_residuals, max_influence, most_influential (0-based, to match the Python module).

Details

Influence is not outlyingness: a subject with an extreme covariate in a well-populated region may have none, and an ordinary-looking subject at the end of follow-up may have a great deal. Read dfbetas, not dfbeta.

References

Cain, K. C. and Lange, N. T. (1984). Approximate case influence for the proportional hazards regression model with censored data. Biometrics, 40(2), 493-499.

Examples

set.seed(1)
X <- matrix(rnorm(60), ncol = 1)
tt <- rexp(60, exp(X * 0.5))
f <- morie_efron_tie_correction(tt, rep(1, 60), X)
morie_cox_dfbeta_influence(f)$most_influential
#> [1] 5