clusterSEs (version 1.1)

cluster.im.mlogit: Cluster-Adjusted Standard Errors and p-Values for mlogit

Description

Computes p-values and standard errors for multinomial logit models based on cluster-specific model estimation (Ibragimov and Muller 2010). A separate model is estimated in each cluster, and then p-values are computed based on a t/normal distribution of the cluster-specific estimates.

Usage

cluster.im.mlogit(mod, dat, cluster, report = TRUE, se = FALSE,
  truncate = FALSE)

Arguments

mod

A model estimated using mlogit.

dat

The data set used to estimate mod.

cluster

A formula of the clustering variable.

report

Should a table of results be printed to the console?

se

Should standard errors be returned?

truncate

Should outlying cluster-specific beta estimates be excluded?

Value

• A list with the elements
• p.values
  A matrix of the estimated p-values.
• se
  The estimated standard errors (if requested).

References

Ibragimov, Rustam, and Ulrich K. Muller. 2010. "t-Statistic Based Correlation and Heterogeneity Robust Inference." Journal of Business & Economic Statistics 28(4): 453-468.

Examples

# predict method of hospital admission
require(VGAMdata)
data(vtinpat)
vtinpat$hos.num <- as.numeric(vtinpat$hospital)
vtinpat$age <- as.numeric(vtinpat$age.group)
vtinpat.mlogit <- mlogit.data(vtinpat, choice = "admit", shape="wide")
vt.mod <- mlogit(admit ~ 0 | age + sex, data = vtinpat.mlogit)
summary(vt.mod)

# compute cluster-adjusted p-values (takes a while)
clust.p <- cluster.im.mlogit(vt.mod, dat=vtinpat.mlogit, cluster = ~ hos.num,
           report=TRUE, se=TRUE, truncate=TRUE)

# compute 95% confidence intervals
ci.lo <- coefficients(vt.mod) - qt(0.975, df=13)*clust.p$se
ci.hi <- coefficients(vt.mod) + qt(0.975, df=13)*clust.p$se
ci <- cbind(ci.lo, ci.hi)
colnames(ci) <- c("95% lower bound", "95% upper bound")
ci