MConjoint (version 0.1)

MConjoint-package: Perform Conjoint Analysis using multiple designs

Description

The Multiple Conjoint Analysis package changes the meaning and use of traditional holdout cases. Rather than using the holdout cases to check a single design, the "holdouts" are used to create a large set of designs, each of which is analyzed. The average result is used

Details

Package:

MConjoint

Type:

Package

Version:

0.1

Date:

2013-05-14

License:

GPL-3

The use of the routines centers around something I call a "despack" a design package. A despack contains despack$cards: a list of the m cards for which ranks are obtained; despack$designs: a list of designs each with n cards drawn from the list of m cards; despack$samples, a list of samples of length n, drawn from 1:m, corresponding to the cards used in the design; despack$coeffs: a list of matrices of linear coefficients; despack$all.utils: a list of lists of utility values, on for each column of the coeffs matrices; despack$all.imps, a list of matrices of importances, one column for each utility; despack$utils: a list of utilities (average taken over first index of the list of lists; despack$imps: the average of the list of importance matices

Start with a data set, full.design, with all possible cards. (This may be the full factorial design (all combinations of levels)) or some combinations may be removed.

Obtain a "good" design of n cards (for information on what makes a design good see the documentation for mc.good.desgins). To this you add extra.cards cards in such a way that you maximize the number of subsets of the m=n + extra.cards of length n that lead to "good" designs.

Both operations can be done by calling

orig.design = mc.get.initial.design(full.design)

orig.design$design will be the m cards for which you will collect data

You then obtain your data, data, a matrix with each column corresponding the the ranks given to the cards by one subject. Then run

despack = good.designs(orig.design$design)

This will give an initial despack, with $cards, $samples, and $designs

Fill the other elements of despack by calling

despack=M.Conjoint(despack,data)

This will print a summary with the utilites and the importances averaged over the subjects (an operation that may or may not be useful)

Examples

#  A simple conjoint problem.  Managers can make hiring descisions
#  based on the factors
#  University: Prestige, Excellent, Good; Sex: Male, Female;
#  Dress: smart, messy; Hair: long, short.
#  We want to determine the importance of these factors.
#  We interview two managers.  The first picks first by
#  University, then by sex, male over female, then
#  by dress, smart over messy, and does not care about hair
#  length.  The second is like the first except that
#  this manager picks female over male.

# start with the full factorial design

data(hire.candidates)

#get a questionaire

hire.questionaire = mc.get.initial.design(hire.candidates,max.trials=10)

#collect the data

data(hire.data)

#get a design pack for the analyis

hire.despack=mc.good.designs(hire.questionaire$design, size=20)

#do the conjoint analysis

hire.despack=M.Conjoint(hire.despack,hire.data)

# (note this illustrates the danger of averaging utilities.
# The average utility for both Male and Female is small, but
# Sex is important to both managers)