llbt.worth: Function to calculate and print worth parameters from LLBT...

In prefmod: Utilities to Fit Paired Comparison Models for Preferences

Function to calculate and print worth parameters from LLBT model results

Description

Worth parameters are calculated from the results of an LLBT model fit, i.e., from llbtPC.fit or from a gnm-fit, respectively. For the latter, the function only works if the design matrix had been generated using llbt.design.

Usage

llbt.worth(fitobj, outmat = "worth")

Arguments

fitobj	result of an LLBT model fit using either llbtPC.fit or gnm (having used a design matrix generated by llbt.design).
outmat	a matrix of estimated worth parameters (outmat = "worth", the default) or LLBT model parameters (outmat = "lambda").

Details

If the LLBT model includes categorical subject covariates, the function provides estimates for all groups formed by the full crossclassification. Numerical subject covariates are not implemented (yet)(see Warning below).

Value

llbt.worth returns a matrix of worth or model parameters. If subject covariates have been specified, each column represents a group defined by the crossclassification of the subject covariates.

In case of object-specific covariates (gnm-fit only) the rows are collapsed to the number of different combinations of object-specific covariate values and labelled accordingly. Additionally, there is an attribute objtable containing a summary of original objects (items) and their reparameterisation with object-specific covariates. This is a list or a matrix.

The function plot gives a plot of the estimates.

Warning

If the LLBT model has been fitted including numerical subject covariates, they are ignored. However, estimates for the remaining predictors are calculated for convenience. Please note, that these cannot be interpreted as standard estimates but are intercepts of the regression model where the objects (or reparameterised objects) are explained by one or more numerical subject covariates.

Note

If a position effect has been fitted (for details see Dittrich, et. al., 1998), the corresponding variable must have been named pos.

Author(s)

Reinhold Hatzinger

See Also

llbtPC.fit, llbt.design, plot

Examples

# fit only first three objects with SEX effect
mod <- llbtPC.fit(cemspc, nitems = 3, formel = ~SEX, elim = ~SEX, undec = TRUE)

# calculate and print worth parameters
mw <- llbt.worth(mod)
mw

# the same using llbt.design and gnm
des <- llbt.design(cemspc, nitems = 3, cat.scovs = "SEX")
m2  <- gnm(y ~ o1+o2+o3 + SEX:(o1+o2+o3) + g1, elim = SEX:mu,
  data = des, family = poisson)

# calculate and plot worth parameters
w2 <- llbt.worth(m2)
plot(w2)


# model with object specific covariates
latin <- c(0, 1, 1, 0, 1, 0)     # object-specific covariate
LAT   <- data.frame(LAT = latin) # objcovs must be data frame with named columns

onames <- c("LO", "PA", "MI", "SG", "BA", "ST")

des <- llbt.design(cemspc, nitems = 6, objnames = onames, objcovs = LAT)
m3  <- gnm(y ~ LAT + g1, eliminate = mu, data = des, family = poisson)
w3  <- llbt.worth(m3)
w3
attr(w3, "objtable")

prefmod documentation built on Sept. 30, 2023, 5:06 p.m.