86

Available online at www.jstage.jst.go.jp/

Agricultural Information Research 17(2), 2008. 8694
Topics
Design and Analysis of Choice Experiments Using R:
A Brief Introduction
Hideo Aizaki*
1)
and Kazushi Nishimura
2)
1) National Institute for Rural Engineering, National Agriculture and Food Research Organization, 2-1-6 Kannondai, Tsukuba-city,
Ibaraki, 305-8609, Japan
2) National Agricultural Research Center for Kyushu Okinawa Region, National Agriculture and Food Research Organization,
2421 Suya, Koshi-city, Kumamoto, 861-1192, Japan
Abstract
This paper briefly introduces choice experiments using R, which is a language and environment for statistical computing.
Choice experiments belong to the family of stated preference methods and are applied to numerous issues in research fields
such as marketing science, transportation economics, environmental economics, agricultural economics, or health economics.
We explain the process of creating choice sets using the AlgDesign package and the process of statistically analyzing
responses using the survival package. Since R is free software, readers of this document may find it easy to apply choice
experiments to their research projects.
Key words
choice-based conjoint analysis, design of experiments, factorial design, discrete choice model, conditional logit model, survival
package, AlgDesign package
Introduction
This paper, which is a briefly modified English edition of
Aizaki and Nishimura (2007), aims to explain the design and
analysis of choice experiments using R (R Development Core
Team 2006). Choice experiments, also called as choice-based
conjoint analysis, choice modeling, or stated choice method,
belong to the family of stated preference methods and are applied
to numerous issues in research fields such as marketing science,
transportation economics, environmental economics, agricultural
economics, or health economics (see Louviere et al. 2000, Ryan
and Farrar 2000, Bennett and Blamey 2001, Bateman et al. 2002,
Hensher et al. 2005). Choice experiments comprise seven steps:
characterizing the decision process, identifying and describing
the attributes, developing an experimental design, developing the
questionnaire, collecting data, estimating model, and interpreting
the results of policy analysis or decision support (Champ et al.
2003). However, this paper only describes a method for develop-
ing an experimental design (designing choice sets) and estimating
model (statistically analyzing responses) using R. Detailed
descriptions of the other steps and practical techniques for choice
experiments are referred to in the abovementioned books or
papers. R and its packages that are used in this paper are supposed
to be downloaded from the official web site of R (The R Project
for Statistical Computing <http://www.r-project.org/>) and
installed onto your personal computer. Readers who are not
familiar with R should refer to web sites, books, or documents
related to R (a list of publications is displayed on the web site).
Creating choice sets using the AlgDesign package
Attributes, levels, and question
Following the hypothetical example of consumers valuation
of milk, which is based on previous research (Sawada et al. 2002,
Iwamoto et al. 2003), we explain the method for designing the
choice sets and statistically analyzing the response data for choice
experiments using R. Fig. 1 shows an example of a question for
choice experiments, wherein respondents are assumed to be asked
to select one among three alternatives, including two choices of
* Corresponding Author
E-mail: [email protected]
Agricultural Information Research 17(2),2008
87
milk and none of the two (i.e., neither milk A nor milk B is pre-
ferred). Each type of milk is expressed by three attributes:
HACCP label, Eco label, and Price per liter. The HACCP
(hazard analysis of critical control points) label indicates that the
HACCP system has been used for food safety control during the
process of manufacturing milk in a factory. The Eco label shows
that the raw milk is produced by cows that are fed by ecological
dairy farmers who manage their dairy farms in harmony with the
environment. Each of these attributes has two levels: Yes or
No. The price per liter attribute comprises four levels: 145 yen,
150 yen, 155 yen, and 160 yen ($1=approximately 115 yen as of
October 2007). Each type of milk is supposed to be the same
except for these three attributes.
Respondents are assumed to be asked to answer eight ques-
tions, as shown in Fig. 1. Table 1 shows total choice sets, and
each line describes the two types of milk (alternatives). For exam-
ple, line 1 in Table 1 corresponds to question 1 (Fig. 1), and it
shows that the respondents can select one among two types of
milk in question 1milk A is characterized by costing 145 yen
without both the HACCP and Eco labels and milk B is character-
ized by costing 160 yen with the HACCP label and without the
Eco label.
How to create choice sets
Following the first of four methods explained by Louviere et
al. (2000, p. 114), the total choice sets can be created through five
steps using R (Fig. 2).
Step 1: The function gen.factorial( ) included in the AlgDe-
sign package (Wheeler 2004a) is used for creating a full
factorial design.
Step 2: The function optFederov( ) included in the AlgDesign
package (Wheeler 2004b) is used for generating a frac-
tional factorial design ((mixed) orthogonal array) from
the full factorial design.
Step 3: Making M-1 copies of the fractional factorial design.
Step 4: In order to create a choice set with M alternatives, ran-
domly select one of the alternatives (rows) from each of
the M sets of the fractional factorial design without
replacement. Repeat this step until all alternatives in each
of the M sets of the fractional factorial design are
assigned to P choice sets.
Step 5: Translating the design codes in the choice sets into codes
with a unique and corresponding level, and then adding
the none-of-these option to each choice set optionally.
Step 1: Creating a full factorial design
A full factorial design with two two-level factors (HACCP
label and Eco label) and one four-level factor (price attribute) can
be created using the function gen.factorial( ). In the full fac-
torial design, two or more attributes (independent variables) are
manipulated and all combinations of the levels of each attributes
are included. In our case, where there are two two-level attributes
and one four-level attribute, the full factorial design comprises
sixteen combinations of the levels of each attributes (=224).
After attaching the AlgDesign package by using the function
library( ), the full factorial design is generated and assigned to
an object ffd using the function gen.factorial( ) as follows.
> library(AlgDesign)
> ffd <- gen.factorial(c(2,2,4), varNames=c("HAC",
"ECO","PRI"), factors="all")
> ffd
HAC ECO PRI
1 1 1 1
2 2 1 1
3 1 2 1
4 2 2 1
5 1 1 2
6 2 1 2
7 1 2 2
8 2 2 2
9 1 1 3
10 2 1 3
11 1 2 3
12 2 2 3
13 1 1 4
14 2 1 4
15 1 2 4
16 2 2 4
Fig. 1 An example of choice experiment questions
Table 1 Total choice sets excluding the none-of-these option
Question
Milk A Milk B
HACCP Eco Price HACCP Eco Price
1 No No 145 yen Yes No 160 yen
2 Yes No 155 yen Yes Yes 145 yen
3 Yes Yes 150 yen No Yes 160 yen
4 No No 150 yen No No 145 yen
5 No Yes 155 yen Yes Yes 150 yen
6 Yes Yes 145 yen Yes No 155 yen
7 No Yes 160 yen No Yes 155 yen
8 Yes No 160 yen No No 150 yen
Design and Analysis of Choice Experiments using R: A Brief Introduction
88
The function gen.factorial( ) has three arguments: c( ),
varNames, and factors. The argument c( ) in the function is
used for setting the attributes and levels included in a full factorial
design. In order to construct a full factorial design having two
two-level attributes and one four-level attribute, the argument
c( ) is assigned values of 2, 2, and 4. If a full factorial design
with three two-level attributes, two four-level attributes, and one
six-level attribute needs to be created, the argument c( ) will be
c(2,2,2,4,4,6). The argument varNames is used for assign-
ing the names of the attributes to each column in the full factorial
design. Here, the first, second, and third columns are called HAC
(HACCP label attribute with two levels), ECO (Eco label attribute
with two levels), and PRI (price attribute with four levels),
respectively. The last argument factors=all indicates that all
attributes are factors. There is normally no need to modify this
argument as long as the steps described in this paper are followed.
Executing the abovementioned command creates the full facto-
rial design with two two-level attributes and one four-level
attribute, and the design is assigned to the object ffd. The name
of the object (ffd), which can be arbitrarily decided, lists the
elements of the object. We find that the full factorial design com-
prises sixteen rows.
Fig. 2 Steps followed in constructing choice sets
Agricultural Information Research 17(2),2008
89
Step 2: Generating a fractional factorial design
Consider that we have to create a half-fractional factorial
design where the number of rows of the full factorial design is
reduced from sixteen to eight. The function optFederov( ) gen-
erates the fractional factorial design from the full factorial design,
which is then assigned to an object des, after arbitrarily setting
the random seed by using the function set.seed( ) as follows.
> set.seed(54321)
> des <- optFederov(~.,ffd,8)
> des
$D
[1] 0.25
$A
[1] 6.333333
$Ge
[1] 1
$Dea
[1] 1
$design
HAC ECO PRI
1 1 1 1
4 2 2 1
5 1 1 2
8 2 2 2
10 2 1 3
11 1 2 3
14 2 1 4
15 1 2 4
$rows
[1] 1 4 5 8 10 11 14 15
The function optFederov( ) has three arguments: ~., ffd,
and 8. The argument ~. implies that all data variables are used lin-
early and their names are used in a model. The argument ffd indi-
cates the name of data containing the candidate list, which is the
same as the name of the object containing the full factorial design
created above. The argument 8 indicates the number of rows
(alternatives) in the fractional factorial design.
The object des contains some objects (elements) as a result of
the function optFederov( ). The fractional factorial design that
is required below and is called design in the object des is
assigned to an object alt1 from the object des by a notation $,
which is used to access an element of the object, as follows.
> alt1 <- des$design
> alt1
HAC ECO PRI
1 1 1 1
4 2 2 1
5 1 1 2
8 2 2 2
10 2 1 3
11 1 2 3
14 2 1 4
15 1 2 4
Step 3: Making copies of the fractional factorial design
M-1 copies of the fractional factorial design created in step 2
are made. Since two packs of milk are shown for each question in
the choice experiments (M=2), a copy of the fractional factorial
design must be made. The copy is created in an object alt2 using
the operator <- in the following manner.
> alt2 <- alt1
> alt2
HAC ECO PRI
1 1 1 1
4 2 2 1
5 1 1 2
8 2 2 2
10 2 1 3
11 1 2 3
14 2 1 4
15 1 2 4
If two or more copies of the fractional factorial design are
required, the names of the objects containing these copies must be
different from each other.
Step 4: Creating choice sets by using random selection with-
out replacement
We randomly select rows (alternatives) from each of the M
fractional factorial designs without replacement, and this selec-
tion is repeated until all rows in each of the M sets of the frac-
tional factorial design are assigned to P total choice sets. P is
equal to the number of rows in the designs. In R, a different and
new uniform random variable is added to every fractional facto-
rial design; then, each design is sorted on the basis of its corre-
sponding random variable.
First, a new uniform random number is added to the object
alt1 as follows.
> alt1 <- transform(alt1, r1=runif(8))
> alt1
HAC ECO PRI r1
1 1 1 1 0.05374091
4 2 2 1 0.75569340
5 1 1 2 0.15476521
8 2 2 2 0.12866907
Design and Analysis of Choice Experiments using R: A Brief Introduction
90
10 2 1 3 0.11170085
11 1 2 3 0.23430698
14 2 1 4 0.91222889
15 1 2 4 0.78156383
The command transform(alt1, r1=runif(8)) implies that
a uniform random variable r1 containing eight values is added to
the object alt1. The number of values contained in the random
variable is equal to the number of rows in the object alt1. If the
object alt1 comprises n rows, the argument runif(8) is changed
to runif(n). Similarly, the following command is implemented
in order to add another uniform random variable r2 to the object
alt2.
> alt2 <- transform(alt2, r2=runif(8))
> alt2
HAC ECO PRI r2
1 1 1 1 0.55358492
4 2 2 1 0.17772093
5 1 1 2 0.97593965
8 2 2 2 0.68168019
10 2 1 3 0.72350066
11 1 2 3 0.81742403
14 2 1 4 0.02218425
15 1 2 4 0.49822342
If the number of fractional factorial designs is M, the abovemen-
tioned command is repeated M times.
Next, each fractional factorial design is sorted on the basis of
its corresponding uniform random variable. The following com-
mand is executed on the object alt1 by using the function order
( ).
> alt1_sort <- alt1[order(alt1$r1),]
> alt1_sort
HAC ECO PRI r1
1 1 1 1 0.05374091
10 2 1 3 0.11170085
8 2 2 2 0.12866907
5 1 1 2 0.15476521
11 1 2 3 0.23430698
4 2 2 1 0.75569340
15 1 2 4 0.78156383
14 2 1 4 0.91222889
A similar command is executed on all fractional factorial
designs (only alt2 in our case).
> alt2_sort <- alt2[order(alt2$r2),]
> alt2_sort
HAC ECO PRI r2
14 2 1 4 0.02218425
4 2 2 1 0.17772093
15 1 2 4 0.49822342
1 1 1 1 0.55358492
8 2 2 2 0.68168019
10 2 1 3 0.72350066
11 1 2 3 0.81742403
5 1 1 2 0.97593965
Each line of the sorted fractional factorial designs (objects
alt1_sort and alt2_sort) corresponds to each question of the
choice experiments (the upper side of Fig. 3). For example, the
same line of the objects alt1_sort and alt2_sort shows the
combination of the attributes levels of milk A and milk B for the
same question, respectively. The first line of the object
alt1_sort shows the combination of the attributes levels of milk
A for question 1, whereas the eighth line of the object corre-
sponds to the same for question 8. Similarly, the first line of the
object alt2_sort shows the combination of the attributes levels
of milk B for question 1, while the eighth line corresponds to the
same for question 8. When two or more alternatives having the
same combination of attributes levels appear in one question, it is
necessary to repeat step 4.
Step 5: Translating the design codes
In step 5, let us translate each design code in the choice sets
into each unique and corresponding level (lower side of Fig. 3).
Following the relationship between the design codes and levels
shown in Table 2, fractional factorial designs shown in the upper
side of Fig. 3 are translated to the designs shown in the lower side
of Fig. 3 that are equivalent to Table 1.
Analyzing responses using the survival package
Representative component of utility
A conditional logit model, which has been applied in numerous
previous researches related to choice experiments, is assumed to
be used for the statistical analysis of the responses to the choice
experiments questions in our example. This model is based on the
random utility theory where a respondents utility is divided into
two components: a representative (systematic) component and a
random component. The representative component of utility for
the none-of-these option is normalized to be 0 and that for the two
types of milk is assumed to be as follows.
V
j
=ASC+
H
HAC
j
+
E
ECO
j
+
P
PRI
j
(1)
where V
j
denotes the representative component of utility for milk
j ( j=A, B). ASC is the alternative specific constant. HAC
j
takes a
Agricultural Information Research 17(2),2008
91
value of 1 if milk j was produced by using the HACCP system
and 0 otherwise. ECO
j
takes a value of 1 if milk j was obtained
from ecological dairy farmers and 0 otherwise. PRI
j
denotes the
price of milk j.
H
,
E
, and
P
are unknown parameters associated
with HAC
j
, ECO
j
, and PRI
j
, respectively.
The conditional logit model can be executed by using the func-
tion clogit( ) that is included in the survival package (Lumley
2006). Frequently, in order to consider the effects of individual
characteristics on the valuation of attributes, the interaction
between individual characteristics and attribute variables are
included in the model. The following two cases are shown: first,
independent variables comprising only attribute variables and
second, independent variables comprising attribute variables and
individual characteristics.
Creating and reading data set
Suppose that ten consumers responded to the eight questions
for the choice experiments (Table 1). For the purpose of simplic-
ity, the sample size is set to be smaller than that in the case of
practical choice experiments. The responses data listed in Table 3
have been randomly created.
Fig. 4 shows a part of a data set for the function clogit( ),
which is created from Table 1 and Table 3. Each respondent
answers eight questions and each question comprises three alter-
natives, that is, milk A, milk B, and the none-of-these option.
Each alternative comprises one row of the data set. Therefore, the
data set comprises 240 rows, with 10 blocks (10 individuals) hav-
ing eight sub-blocks (eight questions) and three rows (three alter-
natives). For example, the data for respondent 1 are listed from
row 2 to row 25, where rows 2, 3, and 4 denote data related to
milk A, milk B, and the none-of-these options, respectively, in
question 1.
Variable STR in column A is a stratification variable that is
used to identify each combination of respondents and questions.
The variable is assumed to be as follows.
Fig. 3 Creating total choice sets (excluding the none-of-these option) from sorted fractional factorial designs
Table 2 Relationship between the design codes and levels in each
attribute
Attribute
Design codes
1 2 3 4
HAC No Yes
ECO No Yes
PRI 145 yen 150 yen 155 yen 160 yen
Table 3 10 individuals responses to the eight questions for the
choice experiments
Respondent
Question
a
1 2 3 4 5 6 7 8
1 3 2 2 1 3 3 1 3
2 3 2 1 3 2 1 3 1
3 1 2 3 1 1 1 2 1
4 3 3 3 1 2 1 3 3
5 3 2 1 1 3 1 3 3
6 1 3 3 3 3 1 3 3
7 3 3 3 1 2 1 3 3
8 2 2 3 2 2 3 3 1
9 3 2 2 2 2 1 3 2
10 3 1 3 3 2 2 3 3
a
A value of 1 implies selecting milk A, 2 implies selecting milk B, and 3
implies selecting the none-of-these option.
Design and Analysis of Choice Experiments using R: A Brief Introduction
92
STR=100N+Q (2)
where variable N (=1, 2, 3, ..., n) is the respondents identification
number and variable Q (=1, 2, 3, ..., q) is a question number.
Since there are three alternatives to each question (choice situa-
tion), the stratification variable takes a different value every three
lines in our case. For example, the value of STR in the case of
question 1 for respondent 1 from row 2 to row 4 is 101
(=1001+1), while the value of STR in the case of question 8 for
respondent 10 from row 239 to row 241 is 1008 (=10010+8).
Variable RES in column B is a dummy variable that, at each
row, takes the value 1 if the alternative is selected and 0 other-
wise. For example, we observe that respondent 1 selected the
none-of-these option (third alternative) in the case of question 1
because the variable RES in row 4 takes a value of 1 and that in
both rows 2 and 3 takes a value of 0.
Columns C to F contain independent variables in the condi-
tional logit model. Variable ASC in column C is an alternative
specific constant for each type of milk, A and B. Variable HAC in
column D and variable ECO in column E represent dummy vari-
ables for the HACCP label and Eco label, respectively. Each
dummy variable takes a value of 1 if the milk has each label and
0 otherwise. Variable PRI in column F denotes the price of milk.
Each attribute variable of the HACCP label, Eco label, and price
corresponding to the none-of-these option takes a value of 0.
Because the function clogit( ) does not correspond to a ques-
tion (choice scenario) format including the none-of-these option,
it is necessary to normalize the representative component of util-
ity for the option to be 0.
Suppose the data set file data1.txt is saved as a tab delimited
file in the directly c:/test. The command read.delim( ) can be
used to read the data set file and create a data set object called
Fig. 4 A part of a data set for the function clogit ( )
Agricultural Information Research 17(2),2008
93
data1 in R as follows.
> data1 <- read.delim("c:/test/data1.txt")
Applying the conditional logit model
The conditional logit model can be applied by using the func-
tion clogit( ) included in the survival package in R. After
attaching the survival package, the function clogit( ) is exe-
cuted as follows.
> library(survival)
> clogout1 <-
clogit(RES~ASC+HAC+ECO+PRI+strata(STR),data=data1)
> clogout1
Call:
clogit(RES~ASC+HAC+ECO+PRI+strata(STR), data=data1)
coef exp(coef) se(coef) z p
ASC 12.6659 3.17e+05 5.2041 2.434 0.0150
HAC 0.8378 2.31e+00 0.4087 2.050 0.0400
ECO 0.3972 1.49e+00 0.4308 0.922 0.3600
PRI 0.0918 9.12e-01 0.0342 2.686 0.0072
Likelihood ratio test=22.9 on 4 df, p=0.00013 n=240
The function clogit( ) has two arguments the formula for
the estimation model and the name of the data set. The formula is
assumed to be RES~ASC+HAC+ECO+PRI+strata(STR), where
the left-hand side of the operator~shows the response variable
(RES) and the right-hand side shows independent variables (ASC,
HAC, ECO, and PRI) and the stratification variable (strata(STR)).
This model assumes that the representative component of utility
for milk is to be a linear combination of independent variables
(see equation (1)). The data set argument is data1. The output
from the function clogit( ) is assigned to the object clogout1
that includes some other objects (elements). For example, the out-
put of the function clogit( ) by default does not exhibit a log-
likelihood at zero and at convergence. These values are listed by
executing the following command.
> clogout1$loglik
[1] 87.88898 76.41796
The output shows that the values of the log-likelihood at zero
and that at convergence are 87.88898 and 76.41796, respec-
tively.
Considering the effects of individual characteristics on their
valuation
A method for reflecting the respondents individual character-
istics in the conditional logit model is to use a new variable that
multiplies the individual characteristic variable and attribute vari-
ables (including ASC). Suppose that the effect of the respondents
gender on their valuation of the price variable is statistically
tested, that is, a new variable PRI
j
FEM is introduced as one of
independent variables, where variable FEM takes a value of 1 if
the respondent is a female and 0 otherwise. The representative
component of utility for milk considering the effect of the vari-
able FEM on the variable PRI
j
is as follows.
V
j
=ASC+
H
HAC
j
+
E
ECO
j
+
P
PRI
j
+
PF
PRI
j
FEM (3)
where
PF
is the coefficient for the variable PRI
j
FEM. An inter-
action between two variables in formula is expressed by using
the operation :, that is, PRI:FEM. A data set (data2) for this
model is made from Table 1 and Table 4 that has been randomly
created. The conditional logit model can be estimated as follows.
> clogout2 <-
clogit(RES~ASC+HAC+ECO+PRI+PRI:FEM+strata(STR),data=d
ata2)
> clogout2
Call:
clogit(RES ~ ASC + HAC + ECO + PRI + PRI:FEM +
strata(STR), data = data2)
coef exp (coef) se (coef) z p
ASC 9.81506 1.83e+04 5.68911 1.725 0.084
HAC 0.84498 2.33e+00 0.43903 1.925 0.054
ECO 0.47206 1.60e+00 0.45666 1.034 0.300
PRI 0.07533 9.27e-01 0.03725 2.022 0.043
PRI:FEM
0.00217 9.98e-01 0.00311 0.697 0.490
Likelihood ratio test=34.5 on 5 df, p=1.87e-06 n= 240
The coefficient of the variable PRI:FEM has a negative sign but
Table 4 10 individuals responses to eight questions for the choice
experiments and their value for the variable FEM
Respondent
Question
a
FEM
b
1 2 3 4 5 6 7 8
1 3 2 2 1 3 3 1 3 1
2 3 2 1 3 2 1 3 1 0
3 3 2 3 1 1 3 3 3 1
4 3 3 3 1 2 1 3 3 0
5 3 3 3 3 3 1 3 3 1
6 1 3 3 3 3 1 3 3 0
7 3 3 3 3 2 1 3 3 1
8 2 2 3 2 2 3 3 1 0
9 3 3 2 2 2 1 3 2 1
10 3 1 3 3 2 2 3 3 0
a
A value of 1 implies selecting milk A, 2 implies selecting milk B, and 3
implies selecting the none-of-these alternative.
b
A value of 1 implies the respondent is a female and 0 otherwise.
Design and Analysis of Choice Experiments using R: A Brief Introduction
94
it is not significant (p=0.490), implying that the estimated coeffi-
cient of the variable PRI for females is not statistically different
from that for males.
Conclusion
This paper briefly illustrates a method for designing choice sets
and statistically analyzing the responses for choice experiments
using R. Readers who are interested in both choice experiments
and R can attempt other methods of design and analysis using R.
References
Aizaki, H. and K. Nishimura (2007) Introduction to Choice Experiments
Using R (Dta-Kaiseki-Kankyou R ni yoru Sentakugata-Konjyoin-
to-Bunseki Nyuumon), Technical Report of the National Institute
for Rural Engineering (Nouson Kougaku Kenkyu-Syo Gihou), 206:
151173 (in Japanese with English Summary), URL http://nkk.
naro.affrc.go.jp/library/publication/seika/giho/206/206-11.pdf.
Bateman, I., R. T. Carson, B. Day, M. Hanemann, N. Hanley, T. Hett, M.
Jones-Lee, G. Loomes, S. Mourato, E. Ozdemiroglu, D. W. Pearce
OBE, R. Sugden and J. Swanson (2002) Economic Valuation with
Stated Preference Techniques: A Manual, Edward Elgar, Chelten-
ham, UK.
Bennett, J. and R. Blamey (Eds.) (2001) The Choice Modelling Ap-
proach to Environmental Valuation, Edward Elgar, Cheltenham,
UK.
Champ, P. A., K. J. Boyle and T. C. Brown (2003) A Primer on Nonmar-
ket Valuation, Kluwer Academic Publishers, Dordrecht, Nether-
lands.
Hensher, D. A., J. M. Rose and W. H. Greene (2005) Applied Choice
Analysis, Cambridge University Press, Cambridge, UK.
Iwamoto, H., Y. Yamamoto, K. Sato, and M. Sawada (2003) Effects of
HACCP and Eco Labels on Japanese Consumers Choice of Milk,
Contributed paper presented for the 47th Annual Conference of the
Australian Agricultural and Resource Economics Society, Freman-
tle, The Esplanade Hotel, February 1114, <http://www.aares.info/
files/AARES/rest/Iwamoto.pdf>, browsed on 29 November 2007.
Louviere, J. J., D. A. Hensher and J. D. Swait (2000) Stated Choice
Methods, Cambridge University Press, Cambridge, UK.
Lumley, T. (2006) clogit. S original by T. Therneau, ported by T. Lum-
ley: The survival package version 2.29.
R Development Core Team (2006) R: A Language and Environment for
Statistical Computing. R Foundation for Statistical Computing, Vi-
enna, Austria. ISBN 3-900051-07-0. URL http://www.r-project.org/.
Ryan, M. and S. Farrar (2000) Using Conjoint Analysis to Elicit Prefer-
ences for Health Care, BMJ: British Medical Journal, 320: 1530
1533.
Sawada, M., H. Iwamoto and Y. Yamamoto (2002) Consumers' Valua-
tion of HACCP and ECO Labeled Milk (Syouhisya no HACCP oyo-
bi Kankyou-Taisaku-Ninshou-Hyouji Gyuunyu ni taisuru Kachi-
Hyouka), in Higuchi, A. and K. Awaji (Eds.), Changes in Agricul-
tural Environment and Policy (Nougyou no Yoken-Hendou to
Taiou-Saku), Association of Agricultural and Forestry Statistics,
Tokyo, Japan, 4658 (in Japanese).
Wheeler, R. E. (2004a) gen.factorial. AlgDesign. The R project for statis-
tical computing. URL http://www.r-project.org/
Wheeler, R. E. (2004b) optFederov. AlgDesign. The R project for statis-
tical computing. URL http://www.r-project.org/
Received December 3, 2007
Accepted March 20, 2008