Full text of DOI:10.1080/000368400322165

This article was downloaded by: [University of South Florida]
On: 09 October 2014, At: 11:05
Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer
House, 37-41 Mortimer Street, London W1T 3JH, UK
Applied Economics
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/raec20
Applying conjoint analysis in economic evaluations:
an application to menorrhagia
Fernando San Miguel , Mandy Ryan & Emma McIntosh
Published online: 04 Oct 2010.
To cite this article: Fernando San Miguel , Mandy Ryan & Emma McIntosh (2000) Applying conjoint analysis in economic
evaluations: an application to menorrhagia, Applied Economics, 32:7, 823-833, DOI: 10.1080/000368400322165
To link to this article: http://dx.doi.org/10.1080/000368400322165
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”)
contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors
make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any
purpose of the Content. Any opinions and views expressed in this publication are the opinions and views
of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources of information.
Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs,
expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in
connection with, in relation to or arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions

±
Applied Economics, 2000, 32, 823 833
Applying conjoint analysis in economic
evaluations: an application to menorrhagia
FER NA NDO SA N MIGUEL*, MA NDY R Y A N and
EMMA MCINTOSH
Health Economics Research Unit, Department of Public Health, University Medical
Buildings, Foresterhill, Aberdeen, AB25 2ZD, Scotland, UK
The increased demand for health care, coupled with limited resources, means that
decisions have to be made concerning the allocation of scarce health care resources.
This paper considers how conjoint analysis (CA) can be used to aid this decision
making process. It is shown how the technique can be used to estimate marginal rates
of substitution between attributes, willingness to pay (WTP) if cost is included as an
attribute and overall utility scores for di€ erent ways of providing a service. The
technique is applied to consider women’s preferences for two surgical procedures
in the treatment of menorrhagia: hysterectomy and conservative surgery. The results
suggest conservative surgery is preferred to hysterectomy, as indicated by higher
utility scores for the former and a marginal WTP of £7593 to have conservative
surgery rather than hysterectomy. The internal validity of CA was also shown. It is
concluded that CA is a potentially useful instrument for policy makers. However,
numerous methodological issues need addressing before the technique becomes an
established instrument within economic evaluations.
I. BA CKGR OUND
can be used to estimate marginal rates of substitution
(MRS), willingness to pay (WTP) and utilities for di€ erent
ways of providing a service. Methodological issues in the
application of the technique are also addressed, including
time to complete questionnaire, ease of completion, the
existence of dominant preferences and theoretical validity.
It is argued that the output from a CA study could poten-
tially be used within an economic evaluation framework.
However, methodological issues must ®rst be addressed.
The increased demand for health care, coupled with limited
resources, means that decisions have to be made concern-
ing the allocation of scarce health care resources. Health
economics has helped inform this debate by developing and
applying instruments such as Quality Adjusted Life Years
(QALYs) (Williams, 1985; Torrance, 1986; Kind, 1988)
and willingness to pay (WTP) (Donaldson et al., 1995;
Donaldson et al., 1997; Ryan et al., 1997; Donaldson and
Shackley, 1997). This paper considers the application of
conjoint analysis (CA) in economic evaluations. The aim
is to demonstrate how the application of CA can be of use
to policy makers concerned with the allocation of scarce
health care resources.
The following section provides background information
on CA. The technique is then applied to establish the ben-
e®ts of two alternative procedures for the treatment of
menorrhagia: hysterectomy and conservative surgery.
Information is ®rst provided on studies that have evaluated
these two treatments. Following this it is shown how CA
II. CONJOINT A NA LY SIS
Conjoint analysis (CA) refers to a family of stated prefer-
ences techniques used to measure individual preferences.
The technique originated in mathematical psychology
(Luce and Tukey, 1964) and has been widely used in mar-
ket research (Acito and Jain, 1980; Green and Srinivasan,
1990). Here ranking and rating techniques are most com-
monly used to elicit preferences. Use of the technique by
transport economists led to the development of discrete
* Author for correspondence: E-mail: sm.fernando@abdn.ac.uk
823
#
±
±
Applied Economics ISSN 0003 6846 print/ISSN 1466 4283 online
2000 Taylor & Francis Ltd
http://www.tandf.co.uk/journals

824
F. San Miguel et al.
ˆ
DV …H1 ¡ H0†
‡
A ‡
A2 ‡ ¢ ¢ ¢
-
-
-
2H1
choice models (Hensher, 1994; Louviere, 1988; Sheldon
and Steer, 1988). This approach has been developed in
the health economics literature (Propper, 1990; Ryan and
Farrar, 1995; Ryan, 1995; Propper, 1995; Ryan and
Hughes, 1997; Bryan et al., 1997; Vick and Scott, 1998;
van der Pol and Cairns, 1998).
0H1
1H1
1
H1g
‡ N A ‡ Y ‡ C ‡
-
"
H1
n
¡ B0
-
‡
A ‡
A2 ‡ ¢ ¢ ¢
-
-
2H0
1H0
1
H0
H0g
‡ N ‡ Y ‡ C ‡
…6†
…7†
"
H0
The discrete choice approach is rooted in random utility
theory (McFadden, 1973; Hanemann, 1984). Here the ana-
lyst does not have complete information about the indi-
vidual’s utility function. The utility derived from the use
or consumption of a good is therefore a function of both a
utility function that the researcher estimates and an error
term representing the unobservable factors in the indi-
vidual’s utility function:
which is equal to
ˆ
DV …H1 ¡ H0†
‡ å A ‡
¬
0
¬
³
i
i
DV …H1 ¡ H0† is the change in utility in moving from H0 to
H . are the parameters of the model to be estimated.
¬
¬
0
1
i
represents the constant term and shows the general prefer-
ence for H1 or H0 when all other attributes in the model are
the same. A positive value for the constant term means that
H1 is preferred to H0 when all attributes are the same, and
a negative value that H0 is preferred to H1. `Ai’ is the
di€ erence in the level of attribute i between H1 and H0
ˆ
V Hq…A† UHq…A† ‡
…1†
"
Hq
V Hq…A† represents the indirect utility function of individual
q for H with attributes A, UHq…A† represents the measur-
able component of utility estimated by the researcher, with
and
is the unobservable error term for the model.
³
Given that the discrete choice model involves estimating
a di€ erence model, and assuming a linear additive model,
terms common to both members of the left-hand side of
Equation 6 drop out of the measurable part of the utility
function. Thus, individual income …Y † and personal char-
acteristics …C† drop out of Equation 7.
H, q and A as de®ned above, and re¯ ects the unobserv-
"
able factors in the individual’s utility function. Individuals
are assumed to be utility maximizers, i.e. individual q will
choose health care intervention H1 over H0 if:
V H1q…A†
V
H0q…A†
…2†
>
There are ®ve main stages involved in estimating
Equation 7: identifying the attributes; assigning levels to
these attributes; presenting scenarios to individuals which
involve di€ erent levels of the attributes; obtaining prefer-
ences for these scenarios and analysing the responses.
Having estimated Equation 7 it is possible to estimate:
marginal rates of substitution (MRS) between the attri-
Following Equation 1, Equation 2 can be represented as:
…UH1q
‡
†
…U
‡
¡
†
†
…3†
"
>
"
"
H1q
H0q
H0q
H1q
or
…UH1q ¡ UH0q
†
…
…4†
> "H0q
butes (
); willingness to pay (WTP) if cost is included
¬i=¬j
Given that error terms are unknown, we do not know
whether H would be chosen over H (since may be
as an attribute, and total utility scores for di€ erent com-
binations of levels of the attributes. In what follows the ®ve
stages of a CA study, and estimation of MRS and utilities,
are demonstrated with reference to a study looking at
women’s preferences for alternative treatments for menor-
rhagia.
"
H0q
1
0
greater than H1q). Given this, a probability model is esti-
"
mated where:
ˆ
ˆ
…
j
P…H1q A H† PH1q
¡
†
…UH1q ¡ UH0q†
;
"
"
<
H0q
H1q
…5†
The probability of choosing alternative H1 (over H0) by
individual q, given the set of attributes A and the choice
set H, is given by the probability that the error di€ erence is
smaller than the di€ erence in the `observable’ utility com-
ponent between H1 and H0. For purposes of empirical
measurement, a probability distribution is assumed for
III. A PPLY ING CONJOINT A NA LY SIS TO
A SSESS PREFER ENCES FOR THE
TR EA TMENT OF MENOR R HA GIA
Background to conjoint analysis study
…
¡ H1q†. Usually logit and probit models are used.
"
"
H0q
When the discrete choice models are used, individuals
Menorrahagia (or excessive menstrual bleeding) a€ ects
approximately 22% of healthy women (Gath et al.,
1987). Currently 60% of referrals for menorrhagia are
treated by hysterectomy within ®ve years. However,
recently new minimal access surgical alternatives to hyster-
ectomy, known as conservative treatments, have been
developed (Coulter et al., 1995; Pinion et al., 1994).
Randomized trials have shown reduced post-operative
morbidity, reduced operative complications and shorter
are asked to choose between pairs or groups of alternatives.
These alternatives are di€ erent descriptions of the services
or interventions. Each individual provides a number of
answers (choices) which allow the researcher to estimate
the probabilistic models and the utility functions described
earlier.
Assuming a linear utility function, the utility to be esti-
mated in moving from H0 to H1 is:

Applying conjoint analysis in economic evaluations
825
stay in hospital following conservative surgery (Cameron et
al., 1986; Sculpher et al., 1993; Pinion et al., 1994).
However, hysterectomy shows greater e€ ectiveness in
terms of the relief of symptoms since two-thirds of
women who have conservative treatment will require re-
treatment (either hysterectomy or conservative) within 4
years. To date no work has established women’s prefer-
ences for the two treatment processes.
Stage 2 involves assigning these attributes level. These
should be realistic and capable of being traded. Levels
were based on results from previous studies, thus ensuring
realism of the scenarios presented to individuals. Including
cost as an attribute to estimate WTP indirectly raises ques-
tions concerning the range of levels. The issues raised here
are similar to those raised in devising the bid vector in a
closed ended WTP study (Ryan et al., 1997). To establish
such levels a direct WTP question was included in the pilot
study. This resulted in the levels of cost being changed from
£800, £1100, £1400, £1500 to £500, £1200, £1400, £2500
and £5000.
Sample and setting
A pilot self-completion questionnaire was given to a con-
venience sample of 24 women working in the Department
of Public Health at the University of Aberdeen. The aim
here was to establish whether respondents understood the
questions, whether trading was taking place, and to ascer-
tain levels for the cost attribute (see below). The main self-
completion questionnaire was given to a sample of 146
women working mainly at Aberdeen Maternity Hospital
and the Department of General Practice at the University
of Aberdeen. Due to time constraints no reminders were
sent.
Stage 3 involves de®ning the scenarios to be presented.
The attributes and levels selected gave rise to 1296 possible
scenarios (34 £ 42). This was reduced to 25 scenarios using
the SPEED computer package (Bradley, 1991). It was
assumed that the interactions between the attributes were
insigni®cant and therefore could be dropped from the
analysis.
In stage 4 a discrete choice model was used to obtain
preferences. The 25 scenarios produced by SPEED were
randomly split into two groups of 12 and 13 scenarios
and two versions of the CA questionnaire were produced.
In both questionnaires all scenarios were compared with
the current situation of hysterectomy. Women were ran-
domly allocated to the two questionnaires. An example
of one of the discrete choices is shown in Fig. 1. (For a
copy of the questionnaires see San Miguel et al., 1997.)
Given that multiple observations are obtained from indi-
viduals, in stage 5 responses were analysed using a random-
e€ ects probit model (Propper, 1995). A general to speci®c
approach was adopted, with explanatory variables being
excluded by backward elimination.
Stages in a conjoint analysis study
The ®rst stage of a CA study involves de®ning the attri-
butes of interest. A number of methods exist to do this.
These include literature reviews, group discussions, indi-
vidual interviews, and direct questioning of individuals.
Alternatively, it may be that the researcher is interested
in given attributes from a pre-de®ned policy question, or
that results from previous randomized controlled trials
indicate attributes of a service that need to be valued. In
this study the attributes were de®ned by reviewing previous
literature (Cameron et al., 1996; Pinion et al., 1994;
Sculpher et al., 1993, 1996). The attributes were: number
of nights in hospital after operation; time to return to nor-
mal activity; chance of complications following operation;
chance of re-treatment with conservative treatment; chance
of conservative re-treatment and chance of hysterectomy
re-treatment. In addition, cost was selected as an attribute
so that WTP could be estimated indirectly (Table 1).
Assuming a linear additive utility function, the model
estimated was:
ˆ
†
f
¬
C¡
H
0C
C
DV
‡
1CNightsC
¬
‡
Time ‡ 3CCompC
¬
2C
¬
…
‡
ReCon ‡
ReHys ‡ 6CCostC
¬
¬
¬
4C
C
5C
C
g
f
¬
‡ eC ‡ uC
¡
‡
1HNightsH
¬
‡
2HTimeH
¬
0H
‡
‡
3HCompH
‡
4HReConH
‡
5HReHysH
¬
¬
¬
g
6HCostHeh ‡ uH
¬
Table 1. Attributes and levels of the CA questionnaire
Attributes
Levels
0, 2, 4, and 7*
Nights in hospital after intervention
Time to return to normal activity
2, 3, 4 and 11*
Chance of complications following the operation
Probability of re-treatment with conservative surgery
Probability of re-treatment with hysterectomy
Cost
5%, 10%, 15%, 20% and 45%*
0%*, 10%, 20% and 30%
0%*, 10%, 20% and 30%
£500, £1200, £1400,* £2500 and £5000
* Levels de®ned for the ®xed scenario of hysterectomy

826
F. San Miguel et al.
Hysterectomy
Conservative
Number of nights in hospital after operation
Time to return to normal activity (weeks)
Chance of complications following operation
Chance of re-treatment with Conservative
Chance of re-treatment with Hysterectomy
Cost of the treatment
7
11
45%
0%
0%
1,400
0
2
20%
15%
30%
5,000
Prefer hysterectomy
Prefer conservative
Which treatment would you
prefer? (tick one box only)
Fig. 1. Example of discrete choice question in conjoint analysis questionnaire
which is equal to
are continuous. That is, there is always some level of
improvement in one attribute that can compensate an indi-
vidual for a deterioration in the level of another attribute.
Estimates of MRS using CA have to date adopted such a
model. However, when CA is applied in health economics
this may be too strong an assumption. For example, some
people may have strong preferences for the actual process
of care (hysterectomy versus conservative management),
health improvements, risk of side e€ ects or probability of
survival after a given intervention. The decision process
may then follow some type of non-compensatory process,
such as lexicographic preference, target setting or domi-
nance (Drakopoulos, 1991a, 1991b; Scott, 1998). There is
some evidence in the health economics CA literature of
dominant preferences with regard to given attributes and
the process of care (Ryan, 1995; Bryan et al., 1997). In this
study we investigated the existence of dominant preferences
by testing whether any respondent chose the same treat-
ment for all choices. Non-traders were dropped from the
analysis and MRS, WTP and utilities re-estimated for
traders only. The t-test was used to compare WTP esti-
mates for the entire sample with the sample of traders.
The results from the random-e€ ects probit regression
analysis were used to test the theoretical validity of CA,
i.e. the extent of which results are consistent with economic
theory, or sometimes, more generally, a priori expectations.
This was done in two ways. First, given that lower levels of
all attributes are to be preferred, all attribute coe cients
were expected to have a negative sign in the regression
equation. Second, to test for diminishing marginal utility
of income, an interaction variable was created between cost
and a dummy variable for individual net income. In the
ˆ
¡ †
DV
‡
Nights ‡ Time ‡ 3Comp
¬
¬
¬
2
¬
C
…
H
0
1
‡
ReCon ‡ ReHys ‡ Cost ‡
…8†
¬
4
¬
5
¬
6
³
DV
is the change in utility in moving from hysterect-
omy to conservative surgery,
C
H
¡ †
…
represents the constant
¬
0
term, Nights is the di€ erence in the nights stay in hospital
after the intervention for both types of treatment, Time is
the di€ erence in time to return to normal activities, Comp is
the di€ erence in complications following both types of
interventions, ReCon is the di€ erence in the chance of hav-
ing a re-treatment with conservative surgery, ReHys is the
di€ erence in the chance of having a re-treatment with hys-
terectomy and Cost is the di€ erence in cost between hys-
terectomy and conservative surgery. is the error term.1
³
Having estimated Equations 8, MRS, WTP and utilities
were estimated. For illustrative purposes the case of mov-
ing from the current to the mean level of the variables for
conservative treatment, as shown by the randomized trials,
is considered.
Methodological issues addressed
Given that CA is relatively new, information was collected
on whether respondents had di culty completing the ques-
tions and how long it took them to complete the question-
naire. Individuals were asked how di cult/easy they found
the choice questions. Possible responses were on a 1 to 5
scale, with 1 representing extremely easy and 5 extremely
di cult.
CA has its theoretical basis in classical economic demand
theory (where preferences satisfy certain axioms and a util-
ity function can therefore be de®ned), such as Lancaster’s
characteristics theory of value (Lancaster, 1966; 1971) and
ˆ
data set income had been de®ned as 1 less than £6800;
ˆ
ˆ
ˆ
±
±
±
2
£6800 £11 100;
3
£11 101 £17 100;
£25 700 and 5 more than £25 700. These groups were
rede®ned into low income (original groups 1, 2, 3, 4) and
4
£17 101
ˆ
random utility theory (McFadden, 1973).
assumption of classical economic theory is that preferences
A crucial
1 With data collected using CA multiple observations are obtained from each individual. Given this, random e€ ects models are used. The
error term ³ takes account of errors across both respondents and observations (Propper, 1995).

Applying conjoint analysis in economic evaluations
827

828
F. San Miguel et al.
Table 3. Marginal rates of substitution between the signi®cant attributes*
Chance of
complications (%)
Chance of re-treatment
with hysterectomy (%)
Signi®cant attributes
Cost
Chance of complications (%)
Chance or re-treatment with hysterectomy (%)
Cost (£)
1 (n/a)
0.67 (n/a)
132 (n/a)
1.47 (n/a)
1 (1)
195 (192)
0.007 (n/a)
0.005 (0.005)
1 (1)
* Figures in brackets represent MRS for the sample of traders.
high income (original group 5). We would expect the higher
income group to attach less weight to cost, resulting in a
lower cost coe cient. Con®dence intervals were estimated
for the cost coe cient for the low and high income groups,
and a t-test used to see if they were signi®cantly di€ erent.
and they should be compensated with £3907 for a 20%
increase in the chance of re-treatment with hysterectomy.
If there were no di€ erences in the attributes between hys-
terectomy and conservative surgery (which in this particu-
lar case is unrealistic) women are willing to pay £6768 to
have conservative surgery. Thus, total WTP for changing
from hysterectomy to conservative surgery is £6832, i.e.
women prefer conservative surgery to hysterectomy as indi-
cated also by the positive utility score obtained.
IV. R ESULTS
Of the 146 questionnaires given out in the main study, 75
were completed (51% response rate without a reminder).
The results from the general and speci®c models are pre-
sented in Table 2. If the entire sample is considered, three
of the six attributes are signi®cant at the 1% level. These
attributes all have the expected negative sign, indicating
that the lower the level of these attributes in the conserva-
tive treatment compared with hysterectomy, the more
likely the individual is to choose conservative treatment.
This is what we would expect, providing support for the
theoretical validity of the model. The `chance of re-treat-
ment with hysterectomy’ is the most important, as indi-
cated by the relative size of the coe cient. The positive
constant implies a general preference for conservative
treatment over hysterectomy when all attributes are the
same in the two treatment regimes.
MRS between signi®cant attributes are shown in Table
3. For the total sample, women were willing to pay £195 for
a 1% reduction in the chance of re-treatment with hyster-
ectomy and £132 for a 1% reduction in the change of
complications. They would be willing to accept a 1.47%
increase in the chance of complications for 1% reduction in
the chance of re-treatment with hysterectomy.
Overall utility scores and WTP measures (following a
change from conservative surgery to hysterectomy) are pre-
sented in Table 4. For the total sample, a 30% decrease in
the chance of complications following conservative surgery
increases utility by 1.340, a reduction in cost of £200
increases utility by 0.067 and an increase in the chance of
re-treatment with hysterectomy of 20% decreases utility by
1.318. The overall e€ ect of these changes (when moving
from hysterectomy to conservative surgery) is an increase
in utility of 2.37.
Methodological issues
Individuals appeared to ®nd the questionnaire not too dif-
®cult to complete, with 18.9% of respondents giving a
response of 1, 24.3% a response of 2, 25.7% a response
of 3, 20.3% a response of 4 and 10.8% a response of 5. The
average time to complete the questionnaire was 14 minutes.
There was some evidence of limited trading, with 36
(48%) respondents consistently having a preference for
±
one type of treatment over another
27 respondents
always preferred conservative treatment and nine always
preferred hysterectomy. If only `traders’ are considered,
only two of the six attributes are signi®cant, with these
coe cients again having the expected sign (Table 2). The
positive constant for all the traders again implies a general
preference for conservative treatment over hysterectomy
when all attributes are the same in the two types of treat-
ment. However, the higher constant term for the total
sample suggests that the non-traders have a stronger
preference. This is what we would expect. MRS between
signi®cant attributes for the traders are shown in Table 3.
Women who were trading were willing to pay £192 for a
1% reduction in the chance of re-treatment with hysterect-
omy compared to £195 for the entire sample. This di€ er-
ˆ
ence was not signi®cant at the 5% level (t 1 04;
:
ˆ
p
0 15). Following a change from hysterectomy to con-
:
servative, utility increases by 0.7593 for the sample of
traders, with an associated WTP of £2966.
Evidence of theoretical validity was found, with all coef-
®cients having the expected signs. In addition to this,
diminishing marginal utility of income was found, with
the cost coe cient being signi®cantly lower for the high
income group compared to the low income group
The results from the utility score estimates are supported
by the WTP estimates. Respondents were willing to pay
£3971 for a 30% reduction in the chance of complications
ˆ
ˆ
(t 3 27; p 0 01) (Table 5). This resulted in the low
income group being willing to pay £92 for a 1% reduction
:
:

Applying conjoint analysis in economic evaluations
829

830
F. San Miguel et al.
±
Table 5. Testing for theoretical validity results from the segmented Random E€ ects Probit model for the speci®c model
Speci®c model
Low-income
marginal
WTP (£)
5828
High income
marginal
WTP (£)
8829
Attribute*
Constant
Coe cient
2.5166
p-value
0.000
Unsegmented attributes
7
7
7
7
7
Chance of complications
0.039
0.069
0.020
0.000
92
140
243
7
Chance of re-treatment with hysterectomy
160
Segmentation by income
Cost-low income
(95% con®dence interval)
Cost-high income
7
7
7
7
0.00043
0.000
0.000
n/a
n/a
n/a
n/a
7
(
(
0.00055; 0.00031)
0.00028
7
0.00036 ; 0.00019)
(95% con®dence interval)
n
849
236.56
0.52
513.98
69.2%
7
Log-likelihood
McFadden R2
Chi-squared
Individual predictions
* Only the signi®cant attributes have been included.
in the chance of complications compared to £140 by the
pared and thereafter decisions made concerning the e -
cient allocation of scarce resources (Propper, 1990; 1995).
Estimated utility scores can be used within the framework
of a cost-utility framework whereby cost per util can be
estimated (Ryan and Hughes, 1997). However, before the
information provided from a CA study can be con®dently
used at the policy level, numerous issues need addressing.
In this study there was some evidence of limited trading.
This may have been a result of the study design. The hys-
terectomy scenario was held constant throughout at the
realistic level, resulting in all choices presented to respon-
dents having better levels for the `chance of re-treatment’
attributes and worse levels for the `chance of complica-
tions’ attribute. Given that these two attributes were sig-
ni®cant predictors of bene®ts (when all the sample was
considered), the questionnaire design may have resulted
in individuals always choosing one type of treatment over
another. That is, they may not have had dominant prefer-
ences. Future research should look at methods of testing
for the existence of non-continuous preferences, and deter-
mine whether they are lexicographic, target setting or
dominant preference. A crucial factor here is to ensure
that levels of attributes are set such that respondents, if
they are traders, are willing to trade. Any well designed
CA study can incorporate most preference structures.
While well designed CA studies can accommodate most
forms of decision making, an important issue that arises
here is the relevance of such studies for policy. It has been
ˆ
ˆ
high income group (t 12 4; p 0 001) and £160 for a 1%
reduction in the chance of re-treatment with hysterectomy
:
:
ˆ
compared to £243 for the high income group (t 35;
ˆ
p
0 001). If all attributes were the same across hysterect-
:
omy and conservative treatment, the high income group
would be willing to pay £8829 to have their preferred
option (conservative treatment) compared to £5828 for
the low income group being the di€ erence signi®cant
2
ˆ
ˆ
(t 25 1; p 0 001). These results all support the theor-
:
:
etical validity of CA. Due to sample size limitations, no
segmentation was carried out on the traders.
V. DISCUSSION A ND CONCLUSIONS
The aim of this paper was to demonstrate the application
of CA in the valuation of health care interventions. It was
shown how the technique can be used to estimate MRS
between attributes, WTP and utilities scores for di€ erent
ways of providing a service. Such information is likely to be
very useful to policy makers within an economic evaluation
framework. For example, estimated MRS will provide
guidance on the characteristics of services that decision
makers should focus on to maximize bene®ts within a lim-
ited budget (Ryan and Farrar, 1995). Estimated monetary
±
measures of bene®t can be used within a cost bene®t
framework, such that bene®ts and cost can be directly com-
2 To carry out signi®cance tests, standard errors are calculated from Taylor series approximation to the variance of a function of random
variables. Letting
2
2
ˆ
var …b1=b2† 1=b2‰var …b1† ¡ 2…b1=b2† cov …b1 ;b2† ‡ …b1=b2† var …b2†Š
ˆ
ˆ
ˆ
If b1 coefficient for attribute, 1;b2 coefficient for cost attribute, b1=b2 WTP for b1.

Applying conjoint analysis in economic evaluations
831
recommended in the literature that the levels chosen for
attributes should be realistic, plausible and capable of
being traded o€ (Bryan et al., 1997; Ryan and Hughes,
1997; van der Pol and Cairns, 1998). While it is theoretic-
ally possible, with well designed CA studies, to identify
areas in the indi€ erence map where individuals are trading,
such levels may not be realistic from a policy perspective.
If a subsample of the respondents are not willing to
trade, either because of the nature of their preferences, or
the level of attributes imposed by the ensuring realism for
the attribute levels, a crucial question is how to analyse
such responses. Most studies applying CA in health eco-
nomics have excluded non-traders (Propper, 1995; Ryan
and Hughes, 1997). In this study signi®cantly di€ erent
results were found when such an approach was adopted.
When all the respondents were considered (including those
with dominant preferences) three out of six attributes were
signi®cant and the goodness of ®t (given by the log-
likelihood estimate) was higher. It is important to consider
the policy implications of excluding non-traders. If the
non-traders are excluded from the analysis in this study,
the bene®ts from changing from hysterectomy to conserva-
tive treatment are underestimated (Table 4). This result is
clearly due to the e€ ect of the high proportion of respon-
dents having a strong preference for conservative treat-
ment. Thus, rather than excluding from the analysis
future work should focus on possible ways of analysing
such responses with the aim of providing more accurate
and useful information to policy makers.
Using cost to estimate WTP indirectly raises questions
concerning the de®nition of the cost attribute in a collec-
tively funded health care system. The following statement
was included when de®ning the cost attribute: `The aim of
asking you about cost is to measure how much you value
these treatments. This is done by asking you how much you
would be willing to pay for each of the treatments. The ques-
tions are purely hypothetical and there is no way you would
have to pay this amount’. This statement was included in an
attempt to reduce `protest bids’, i.e. respondents not
answering the question because they have a moral objec-
tion to being asked about WTP for health care. It is recog-
nized that such a statement may invite respondents to
overstate their true valuation of the service. Two studies
within the WTP literature have directly tested the overbid-
ding hypothesis, i.e. that respondents overstate their true
value when they are told they will not have to pay for the
proposed change (Bohm, 1972; Posavac, 1998). While
Bohm (1972) found no evidence of the overbidding hypoth-
esis, Posavac (1998) did. Future work should address the
over-pledging hypothesis within the context of CA studies.
An alternative approach to dealing with the de®nition of the
cost attribute would be to move away from asking patients
WTP at the point of consumption to an ex ante insurance
based or taxation based questions. Such an approach is cur-
rently being considered in the direct WTP literature (O’Brien
and Gafni, 1996; Olsen and Donaldson, 1998).
Output of CA studies can also be used within a cost-
utility framework, where cost per unit of bene®t can be
estimated (Farrar and Ryan, 1997). An important issue
to investigate here is whether the utility scores produced
in a CA study are cardinal. That is, if a particular health
care intervention has a utility score of 3, and another of 6,
can we conclude that the latter is twice as good as the
former? This is particularly important when comparing
utilities with cost. If a health care intervention costs twice
as much then for e cient decisions to be made we must
address the question of whether bene®ts are at least twice
as much. CA has its origins in mathematical psychology
where it has been proved that cardinal data can be
obtained from ordinal data (if preferences obey certain
axioms) (Luce and Tukey, 1964). These proofs apply to
data obtained from ranking and rating type exercises.
Within the context of the discrete choice approach,
research is required on the axioms that underlie random
utility theory, whether individuals obey these axioms, and,
whether the data produced from such studies can be treated
as cardinal.
Any technique that is used to assess bene®ts should be
internally consistent, reliable and valid. Within health eco-
nomics CA has been shown to be internally consistent and
theoretically valid (Bryan et al., 1997; Propper, 1990, 1995;
Ryan, 1995; Ryan and Hughes, 1997; van der Pol and
Cairns, 1998; Vick and Scott, 1998). One study has demon-
strated the convergent validity of CA with the closed-ended
WTP method (Ryan, 1996). Evidence of theoretical validity
has also been shown in this study.3 There is no work to
date looking at the reliability or external validity of CA
within health care and future work must address this.
In conclusion, CA is potentially a very useful tool for
measuring the bene®ts of health care interventions. It can
be used to estimate MRS, WTP and utility scores for dif-
ferent ways of providing a service. However, as with the
development of any new instrument, methodological work
is needed before CA becomes an established instrument in
economic evaluations.
A CKNOWLEDGEMENTS
Financial support from the Medical Research Council
(MRC), the Chief Scientist O ce of the Scottish O ce
3 CA studies in health care have traditionally used dominance tests to test for internal consistency. Given that the process of treatment
was signi®cantly di€ erent across the two types of treatment, such a test could not be employed here. Future work should explore tests of
internal consistency in CA experiments.

832
F. San Miguel et al.
Hensher, D. A. (1994) Stated preference analysis of travel choices:
Department of Health (SODOH) and the Government of
Navarre is acknowledged. The views expressed in the paper
are those of the authors.
±
the state of practice, Transportation, 21, 107 33.
Kind, P. (1988) The design and construction of quality of life
measures, Discussion Paper No 43, University of York,
York.
Lancaster, K. (1966) A new approach to consumer theory,
±
Journal of Political Economy, 74, 134 57.
R EFER ENCES
Lancaster, K. (1971) Consumer Demand, Columbia University
Acito, F. (1977) An investigation of some data collection issues in
conjoint measurement, in Contemporary Marketing Thought,
Proceedings of American Marketing Association.
Press, New York.
Louviere, J. L. (1988) Conjoint analysis modelling of stated pre-
ferences. A review of theory, methods, recent developments
and external validity, Journal of Transport Economics and
Acito, F. and Jain, A. K. (1980) Evaluation of conjoint analysis
results: a comparison of methods, Journal of Marketing
±
Policy, 22, 93 119.
±
Research, 17, 106 12.
Luce, R. D. and Tukey, J. W. (1964) Simultaneous conjoint meas-
Bohm, P. (1972) Estimating demand for public goods: an experi-
urement: A new type of fundamental measurement, Journal
±
±
of Mathematical Psychology, 1, 1 27.
ment, European Economic Review, 3, 111 30.
Bradley, M. (1991) User’s manual for the speed version 2.1 stated
preference experiment editor and designer, Hague Consulting
Group, Hague.
Bryan, B., Buxton, M., Sheldon, R. and Grant, A. (1997) The use
of magnetic resonance imaging (MRI) for the investigation
of knee injuries: a discrete choice conjoint analysis exercise.
Presented at the Health Economists’ Study Group Meeting,
University of York.
McFadden, D. (1973) Conditional L ogit Analysis of Qualitative
Choice Behaviour, University of California at Berkeley,
California.
O’Brien, B. and Gafni, A. (1996) When do the `dollars’ make
sense? Towards a conceptual framework for contingent
valuation studies in health care, Medical Decision Making,
±
16, 288 99.
Olsen, J. A. and Donaldson, C. (1998) Helicopters, hearts and hips:
Cameron, I. M., Mollison, J., Pinion, S. B., Atherton-Naji, A.,
Buckingham, K. and Torgerson, D. (1996) A cost compar-
ison of hysterectomy and hysteroscopic surgery for the treat-
ment of menorrhagia, European Journal of Obstetrics &
using willingness to pay to set priorities for public sector
±
health care programme, Social Science and Medicine, 46, 1 12.
Pinion, S. B., Parkin, D. E., Abramovich, D. R., Naki, A.,
Alexander, D., Russell, I. T. and Kitchener, H. C. (1994)
Randomised trial of hysterectomy, endometrial laser abla-
tion and transcervical endometrial resection for dysfunc-
±
Gynaecology and Reproductive Biology, 70, 87 92.
Coulter, A., Kelland, J., Long, A., Melville, A., O’Meara, S.,
Schulper, M., Sheldon, T. and Song, F. (1995) The manage-
ment of menorrhagia. What are the e€ ective ways of treating
excessive regular menstrual blood loss in primary and sec-
±
tional uterine bleeding, British Medical Journal, 309, 979 83.
Posavac, S. (1998) Strategic overbidding in contingent valuation:
stated economic value of public goods varies according to
consumers expectations of funding source, Journal of
±
ondary care? E€ ective Health Care, 9, 2 14.
±
Donaldson, C. and Shackley, P. (1997) Does process utility exist?
Economic Psychology, 19, 205 14.
A case study of willingness to pay for laparoscopic cholecys-
Propper, C. (1990) Contingent valuation of time spent on NHS
±
±
tectomy, Social Science and Medicine, 44, 699 707.
waiting lists, The Economic Journal, 100, 193 99.
Donaldson, C., Shackley, P., Abdalla, M. and Miedzybrodzka, Z.
(1995) Willingness to pay for antenatal carrier screening for
cystic ®brosis, Health Economics, 4, 439 52.
Donaldson, C., Shackley, P. and Abdalla, M. (1997) Using will-
Propper, C. (1995) The disutility of time spent on the United
Kingdom’s National Health Service waiting lists, The
±
±
Journal of Human Resources, 30, 677 700.
Ryan, M. and Farrar, S. (1995) Conjoint analysis: a new tool for
eliciting patients’ preferences. Brie®ng paper for the NHS
in Scotland No. 6, Health Economics Research Unit,
University of Aberdeen.
ingness to pay to value close substitutes: carrier screening for
±
cystic ®brosis revisited, Health Economics, 6, 145 59.
Drakopoulos, S. A. (1991) Hierarchical behaviour in economics,
±
discussion paper 91 07, Department of Economics,
Ryan, M. (1995) Economics and the patient’s utility function: an
application to assisted reproductive techniques, PhD Thesis,
Aberdeen, University of Aberdeen.
Ryan, M. (1996) Establishing the convergent validity of willing-
ness to pay and conjoint analysis for eliciting preferences,
paper presented to the Health Economists’ Study Group
meeting, University of Brunel.
University of Aberdeen.
Drakopoulos, S. A. (1991b) Threshold based demand and prod-
±
uct market clearing, discussion paper 91 10, Department of
Economics, University of Aberdeen.
Farrar, S. and Ryan, M. (1997) Prioritisation of clinical service
developments in a teaching hospital: an application of con-
joint analysis, Health Economics Research Unit Discussion
Paper No. 08/97, University of Aberdeen, Aberdeen.
Fowkes, T. and Wardman, M. (1988) The design of stated
preference travel choice experiments. With special reference
to interpersonal taste variation, Journal of Transport
Ryan, M. and Hughes, J. (1997) Using conjoint analysis to assess
women’s preferences for miscarriage management, Health
±
Economics, 6, 261 73.
Ryan, M., Ratcli€ e, J. and Tucker, J. (1997) Using willingness to
pay to value alternative models of antenatal care, Social
±
±
Science and Medicine, 44, 371 80.
Economics and Policy, 22, 27 44.
Gath, D., Osborn, M. and Bungay, G. (1987) Psychiatric disorder
San Miguel, F., Ryan, M. and McIntosh, E. (1997) Establishing
women’s preferences for the treatment of menorrhagia using
the technique of conjoint analysis, Health Economics
Research Unit Discussion Paper No. 06/97, University of
Aberdeen, Aberdeen.
Scott, A. (1998) Giving things up to have more of others. The
implications of limited substitutability for eliciting preferences
in health and health care, presented at the Health Economists’
Study Group Meeting, University of She eld, January.
and gynaecological symptoms in middle aged women: a com-
±
munity survey, British Medical Journal, 294, 213 18.
Green, P. E. and Srinivasan, V. (1990) Conjoint analysis in mar-
keting: new developments with implications for research and
±
practice, Journal of Marketing Research, 54, 3 19.
Hanemann, W. (1984) Welfare evaluations in contingent valua-
tion experiments with discrete responses: reply, American
±
Journal of Agricultural Economics, 2 341.

Applying conjoint analysis in economic evaluations
833
Sculpher, M. J., Bryan, S., Dwyer, N., Hutton, J. and Stirrat,
G. M. (1993) An economic evaluation of transcervical endo-
metrial resection versus abdominal hysterectomy for the
treatment of menorrhagia, British Journal of Obstetrics and
van der Pol, M. and Cairns, J. (1998) Establishing patient prefer-
ences for blood transfusion support: an application of con-
joint analysis, Journal Health Services Research and
Management, 3, 70 6.
Vick, S. and Scott, A. (1998) Agency in health care: examining
±
±
Gynaecology, 100, 244 52.
Sculpher, M. J., Dwyer, N. and Stirrat, G. M. (1996) Randomised
trial comparing hysterectomy and transcervical endometrial
resection: e€ ect on health related quality of life and costs two
years after surgery, British Journal of Obstetrics and
patients’ preferences for attributes of the doctor-patient
±
relationship, Journal of Health Economics, 17, 587
606.
Wardman, M. (1988) A comparison of revealed preference and
±
Gynaecology, 103, 142 9.
stated preference models, Journal of Transport Economics and
Sheldon, R. J. and Steer, J. K. (1982) the use of conjoint analysis
±
Policy, 22, 71 91.
±
in transport research, Transport Research, 145 58.
Williams, A. (1985) Economics of coronary artery bypass graft-
Torrance, G. (1986) Measurement of health state utilities for eco-
±
ing, British Medical Journal, 1, 326 29.
±
nomic appraisal: a review, Journal of Health Economics, 5, 1 30.