multinomial function

Estimate probabilities in contingency table

multinomial(
  x,
  data = parent.frame(),
  marginal = FALSE,
  transform,
  vcov = TRUE,
  IC = TRUE,
  ...
)

Arguments

• x: Formula (or matrix or data.frame with observations, 1 or 2 columns)
• data: Optional data.frame
• marginal: If TRUE the marginals are estimated
• transform: Optional transformation of parameters (e.g., logit)
• vcov: Calculate asymptotic variance (default TRUE)
• IC: Return ic decomposition (default TRUE)
• ...: Additional arguments to lower-level functions

Examples

set.seed(1)
breaks <- c(-Inf,-1,0,Inf)
m <- lvm(); covariance(m,pairwise=TRUE) <- ~y1+y2+y3+y4
d <- transform(sim(m,5e2),
              z1=cut(y1,breaks=breaks),
              z2=cut(y2,breaks=breaks),
              z3=cut(y3,breaks=breaks),
              z4=cut(y4,breaks=breaks))

multinomial(d[,5])
(a1 <- multinomial(d[,5:6]))
(K1 <- kappa(a1)) ## Cohen's kappa

K2 <- kappa(d[,7:8])
## Testing difference K1-K2:
estimate(merge(K1,K2,id=TRUE),diff)

estimate(merge(K1,K2,id=FALSE),diff) ## Wrong std.err ignoring dependence
sqrt(vcov(K1)+vcov(K2))

## Average of the two kappas:
estimate(merge(K1,K2,id=TRUE),function(x) mean(x))
estimate(merge(K1,K2,id=FALSE),function(x) mean(x)) ## Independence
##'
## Goodman-Kruskal's gamma
m2 <- lvm(); covariance(m2) <- y1~y2
breaks1 <- c(-Inf,-1,0,Inf)
breaks2 <- c(-Inf,0,Inf)
d2 <- transform(sim(m2,5e2),
              z1=cut(y1,breaks=breaks1),
              z2=cut(y2,breaks=breaks2))

(g1 <- gkgamma(d2[,3:4]))
## same as
## Not run:

gkgamma(table(d2[,3:4]))
gkgamma(multinomial(d2[,3:4]))
## End(Not run)

##partial gamma
d2$x <- rbinom(nrow(d2),2,0.5)
gkgamma(z1~z2|x,data=d2)

Author(s)

Klaus K. Holst