PlackettLuce: pltree-summaries – R documentation

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

pltree-summaries

Plackett-Luce Tree Summaries

Description

Obtain the coefficients, variance-covariance matrix, AIC, or predictions from a Plackett-Luce tree fitted by pltree().

Usage

## S3 method for class 'pltree'
coef(object, node = NULL, drop = TRUE, ...)

## S3 method for class 'pltree'
vcov(object, node = nodeids(object, terminal = TRUE), ...)

## S3 method for class 'pltree'
AIC(object, newdata = NULL, ...)

## S3 method for class 'pltree'
predict(
  object,
  newdata = NULL,
  type = c("itempar", "rank", "best", "node"),
  ...
)

Arguments

`object`	a fitted model object of class `"pltree"`.
`node`	a vector of node ids specifying the nodes to summarise, by default the ids of the terminal nodes.
`drop`	if `TRUE` return the coefficients as a vector when only one node is selected.
`...`	additional arguments passed to `itempar` by `predict`, and to `model.frame` by `AIC`.
`newdata`	an optional data frame to use instead of the original data. For `AIC` this must include the response variable.
`type`	the type of prediction to return for each group, one of: `"itempar"` to give the result of `itempar` (by default the fitted probability of each item being ranked first out of all objects), `"rank"` the corresponding rank, `"best"` the topped ranked item, or `"node"` the node of the tree the group belongs to.

Details

AIC computes -2 * L + 2 * p where L is the joint likelihood of the observed rankings under the tree model and p is the degrees of freedom used to fit the tree model.

Examples

data(beans)
# fit tree based on pairwise comparisons with variety B
pairB <- data.frame(Winner = ifelse(beans$var_b == "Worse",
                                    "Local", beans$variety_b),
                    Loser = ifelse(beans$var_b == "Worse",
                                   beans$variety_b, "Local"),
                    stringsAsFactors = FALSE, row.names = NULL)
beans$G <- as.rankings(pairB, input = "orderings",
                       index = rep(seq(nrow(beans)), 1))

mod <- pltree(G ~ ., data = beans[c("G", "maxTN")])

coef(mod, node = 3)
AIC(mod)

# treat first row from each year as new data
newdata <- beans[!duplicated(beans$year),]

## fitted probabilities
predict(mod, newdata)

## fitted log-abilities, with Local as reference
predict(mod, newdata, log = TRUE, ref = "Local")

## variety ranks
predict(mod, newdata, type = "rank")

## top ranked variety
predict(mod, newdata, type = "best")

## node the trial belongs to
predict(mod, newdata, type = "node")

PlackettLuce

Plackett-Luce Models for Rankings

v0.4.0

GPL-3

Authors

Heather Turner [aut, cre] (<https://orcid.org/0000-0002-1256-3375>), Ioannis Kosmidis [aut] (<https://orcid.org/0000-0003-1556-0302>), David Firth [aut] (<https://orcid.org/0000-0003-0302-2312>), Jacob van Etten [ctb] (<https://orcid.org/0000-0001-7554-2558>)

Initial release