Summarize a logistic_cal object
# S3 method for logistic_cal
summary(object, conf_level = 0.95, ...)estimates and conf_level*100 confidence intervals for calibration intercept and calibration slope.
The former is estimated from a glm (family = binomial("logit")) where the linear predictor (logit(p)) is included as an offset.
Results of the three likelihood ratio tests described by Miller et al. (2013) (see details).
a logistic_cal object
width of the confidence interval (0.95 gives 95% CI)
ignored
The likelihood ratio tests proposed by Miller et al. test the following: The first assesses weak calibration overall by testing the null hypothesis that the intercept (a) and slope (b) are equal to 0 and 1, respectively. The second assesses calibration in the large and tests the intercept against 0 with the slope fixed to 1. The third test assesses the calibration slope after correcting for calibration in the large (by estimating a new intercept term). Note the p-values from the calibration intercept and calibration slope estimates will typically agree with the p-values from the second and third likelihood ratio tests but will not always match perfectly as the former are based on z-statistics and the latter are based on log likelihood differences (chi-squared statistics).
Miller, M. E., Langefeld, C. D., Tierney, W. M., Hui, S. L., & McDonald, C. J. (1993). Validation of probabilistic predictions. Medical Decision Making, 13(1), 49-57.