poma: Examine proportional odds and parallelism assumptions of `orm` and `lrm` model fits.

Description

Based on codes and strategies from Frank Harrell's canonical `Regression Modeling Strategies` text

Usage

poma(mod.orm, cutval, ...)

Arguments

mod.orm: Model fit of class `orm` or `lrm`. For `fit.mult.impute` objects, `poma` will refit model on a singly-imputed data-set
cutval: Numeric vector; sequence of observed values to cut outcome
...: parameters to pass to `impactPO` function such as `newdata`, `nonpo`, and `B`.

Author

Yong Hao Pua <puayonghao@gmail.com>

Details

Strategy 1: Compare PO model fit with models that relax the PO assumption (for discrete response variable)
Strategy 2: Apply different link functions to Prob of Binary Ys (defined by cutval). Regress transformed outcome on combined X and assess constancy of slopes (betas) across cut-points
Strategy 3: Generate score residual plot for each predictor (for response variable with <10 unique levels)
Strategy 4: Assess parallelism of link function transformed inverse CDFs curves for different XBeta levels (for response variables with >=10 unique levels)

Examples

Run this code


if (FALSE) {
## orm model (response variable has fewer than 10 unique levels)
mod.orm <- orm(carb ~ cyl + hp , x = TRUE, y = TRUE, data = mtcars)
poma(mod.orm)


## runs rms::impactPO when its args are supplied
## More examples: (https://yhpua.github.io/poma/)
d <- expand.grid(hp = c(90, 180), vs = c(0, 1))
mod.orm <- orm(cyl ~ vs + hp , x = TRUE, y = TRUE, data = mtcars)
poma(mod.orm, newdata = d)


## orm model (response variable has >=10 unique levels)
mod.orm <- orm(mpg ~ cyl + hp , x=TRUE, y=TRUE, data = mtcars)
poma(mod.orm)


## orm model using imputation
dat <- mtcars
## introduce NAs
dat[sample(rownames(dat), 10), "cyl"] <- NA
im <- aregImpute(~ cyl + wt + mpg + am, data = dat)
aa <- fit.mult.impute(mpg ~ cyl + wt , xtrans = im, data = dat, fitter = orm)
poma(aa)
}