multiRR: Inference on relative risk under multinomial logistic...
In youjin1207/logisticRR: Adjusted Relative Risk from Logistic Regression

Description Usage Arguments Value Author(s) Examples

View source: R/multiRR.R

Inference on relative risk under multinomial logistic regression

1 2	multiRR(formula, basecov = 0, fixcov = NULL, data, boot = FALSE, n.boot = 100)

`formula`	a formula term that is passed into `multinom()` where response should be a factor having `K` different levels. Every term appearing in the formula should be well-defined as a column of `data`. Reference response should be specified as the first level.
`basecov`	a baseline value of exposure variable. Defaults to `0`.
`fixcov`	a data frame of fixed value for each of adjusted confounders. If there is no confounder other than the exposure variable of interest, `fixcov` = `NULL`; if `fixcov` is missing for existing covariates, they are all set to `0` (for numerical covariates) or to the first level (for factor covariates).
`data`	a data frame containing response variable and all the terms used in `formula`.
`boot`	a logical value whether bootstrap samples are generated or not. Defaults to `FALSE`.
`n.boot`	if `boot = TRUE`, the number of bootstrap samples. Defaults to `100`.

`fit`	an object of class `multinom`.
`RRR`	(adjusted) relative risk ratio of `K` different responses compared to reference response under exposure at baseline (`basecov`) and `basecov + 1`.
`RR`	(adjusted) relative risk of `K` different responses under exposure at baseline (`basecov`) and `basecov + 1`.
`delta.var`	estimated variance of relative risk (`RR`) using Delta method.
`boot.rr`	if `boot = TRUE`, a vector of `RR`'s using bootstrap samples.
`boot.rrr`	if `boot = TRUE`, a vector of relative risk ratio (`RRR`)'s using bootstrap samples.
`boot.var`	estimated sampled variance using bootstraps if `boot = TRUE`.
`fix.cov`	a data frame of fixed value for each of adjsuted confounders.

Youjin Lee

n <- 500
set.seed(1234)
X <- rbinom(n, 1, 0.3)
W <- rbinom(n, 1, 0.3)
W[sample(1:n, n/3)] = 2
Y <- rbinom(n, 1, plogis(X - W))
dat <- as.data.frame(cbind(Y, X, W))
result <- multiRR(W ~ X + Y, basecov = 0, data = dat, boot = TRUE, n.boot = 100)
print(apply(result$boot.rr, 2, sd)) # estimated standard errors using Delta method
print(sqrt(result$delta.var)) # estimated standard errors using bootstrap