Multinomial fits

[bgreenwell]

Jittering (at least on the response scale) does show some promise for multinomial fits:

library(nnet)
data(df1, package = "sure")
fit <- multinom(y ~ x, data = df1)
# fit <- multinom(y ~ x + I(x ^ 2), data = df1)
y <- as.integer(df1$y)
s <- runif(length(y), min = y, max = y + 1)
prob <- fit$fitted.values
j <- seq_len(ncol(prob))  # j = 1, 2, ..., J
jmat <- matrix(rep(j, times = nrow(prob)), ncol = ncol(prob),
               byrow = TRUE)  # 1, 2, ..., J; 1, 2, ..., J; ...
es <- rowSums((jmat + 0.5) * prob)
r <- s - es
scatter.smooth(df1$x, r, lpars = list(lwd = 2, col = "red"))