Update Model LR Statistics After Multiple Imputation
LRupdate(fit, anova)new fit object like fit but with the substitutions made
an rms fit object
the result of processMI(..., 'anova')
Frank Harrell
For fits from orm, lrm, orm, cph, psm that were created using fit.mult.impute with lrt=TRUE or equivalent options and for which anova was obtained using processMI(fit, 'anova') to compute imputation-adjusted LR statistics. LRupdate uses the last line of the anova result (containing the overall model LR chi-square) to update Model L.R. in the fit stats component, and to adjust any of the new R-square measures in stats.
For models using Nagelkerke's R-squared, these are set to NA as they would need to be recomputed with a new intercept-only log-likelihood, which is not computed by anova. For ols models, R-squared is left alone as it is sample-size-independent and print.ols prints the correct adjusted R-squared due to fit.mult.impute correcting the residual d.f. in stacked fits.
processMI.fit.mult.impute(), Hmisc::R2Measures()
if (FALSE) {
a <- aregImpute(~ y + x1 + x2, n.impute=30, data=d)
f <- fit.mult.impute(y ~ x1 + x2, lrm, a, data=d, lrt=TRUE)
a <- processMI(f, 'anova')
f <- LRupdate(f, a)
print(f, r2=1:4) # print all imputation-corrected R2 measures
}
Run the code above in your browser using DataLab