r use of rms ols function with multiple objects and confidence intervals

Question

This example below uses the lm function and tidy(..., conf.int=TRUE) easily generates summary estimates and C.I for multiple model objects.

library(tidyverse)
library(broom)

mtcars %>%
  gather(predictor, measure, -mpg) %>%
  group_by(predictor) %>%
  do(tidy(lm(mpg ~ measure, .),conf.int=TRUE))

How do I generate similar output using rms::ols function ? tidy does not work on ols so is there a way to tweak the output from ols function to include C.I from multiple model objects ? Thanks in advance.

Answer 1

Here is a bare-bones/stripped-down version of tidy.ols. It has many limitations (see below), but should do what you want ...

tidy.ols <- function(x, ...) {
   se <- sqrt(diag(x$var)) 
   rdf <- x$stats["n"] - x$stats["d.f."] - 1
   cc <- coef(x)
   ci <- confint(x)
   tibble(term = names(cc),
          estimate = cc,
          std.error = se,
          statistic = cc/se,
          df = rdf,
          p.value = 2*pt(-abs(statistic), df = rdf),
          conf.low = ci[,1],
          conf.high = ci[,2])
}
mtcars %>%
  gather(predictor, measure, -mpg) %>%
  group_by(predictor) %>%
  do(tidy(ols(mpg ~ measure, .)))

limitations

• ignores additional args without checking
• haven't checked p-value calcs etc.
• doesn't allow choice of confint or not, or setting conf level
• doesn't consider other possible bells and whistles of ols ...

It seems like support for rms objects has been discussed a lot on the tidymodels/broom GitHub site but everything there is stale ... I opened an issue if anyone wants to chime in.