Professional Documents
Culture Documents
SSRN Id1818631
SSRN Id1818631
SSRN Id1818631
net/publication/227377190
CITATIONS READS
11 385
2 authors:
Some of the authors of this publication are also working on these related projects:
Theoretical and Empirical Studies on the Dynamic Diffusion of Low Carbon Innovations: the Perspective of Socio-technical System Evolution View project
All content following this page was uploaded by José María Grisolía on 06 November 2017.
HETEROGENEITY IN WILLINGNESS-
TO-PAY FOR THEATRE PRODUCTIONS:
INDIVIDUAL SPECIFIC WILLINGNESS-
TO-PAY ESTIMATES FOR THEATRES,
SHOWS AND THEIR ATTRIBUTES
Abstract
This paper investigates people’s preferences for live theatre, and heterogeneity in
willingness-to-pay (WTP). A stated preference discrete choice model estimates
utility and WTP for different attributes of theatre productions. Previous studies
assumed fixed coefficients across all theatre-goers. This study allows coefficients of
attributes to vary across individuals in the population, providing information on the
heterogeneity of tastes. Theatre-goers’ choices are used to estimate individual-
based parameters for a person’s tastes in theatre productions. The analysis reveals
the heterogeneity of tastes for different types of plays and ticket price; and the
significance of ‘reviews’ by critics and ‘word of mouth’ opinions as important
variables determining choice.
I Introduction
Early research on the demand for the performing arts considered demand mainly a
function of price. Subsequently more sophisticated models and empirical
applications included not only own price, but also the price of substitutes, as
well as the quality of the theatrical production in the demand function (see
Throsby, 1994; Corning and Levey, 2002; Akdede and King, 2006; among others).
These studies used revealed preference (RP) data to assess the effect of the
attributes of the production, and the effect of variations in socio-economic and
demographic characteristics, on demand. The study reported here, in contrast,
uses stated preference (SP) data. SP data overcomes many of the problems
associated with RP data, and also allows heterogeneity in tastes and preferences
to be more fully explored.
The aim of this study is to assess the effect of the characteristics of theatre
productions on demand, and in particular the effect of information about
n
Universidad de Las Palmas de Gran Canaria
nn
University of Newcastle
378
H E T E R O G E NE I T Y I N W I L L I N G N E S S - T O - P A Y 379
the IPs estimation; this is followed by a section on estimates of WTP for various
attributes of productions; and finally some conclusions are drawn.
II Theoretical Background
Discrete choice models are based on the random utility model (see, for instance
Ortúzar and Willumsen, 2001) which postulates that each good provides utility
that can be measured in terms of the characteristics of the good, subject to some
error term. It is assumed that individual n can choose between J alternatives (in
this case theatre shows) from j 5 1 to J. Since the study adopts a SP choice
experiment, the respondent will face T choice situations from t 5 1 to T. Thus,
the utility of alternative j for individual n in the choice situation t can be
expressed in terms of the following expression:
Unjt ¼ bXnjt þ enjt ; ð1Þ
where Xnjt is a vector of observed variables and b is a vector of coefficients to be
estimated. The second term is stochastic and represents the part of the utility
which is not observed by the researcher. Assumptions made about the
distribution of the random error part lead to different types of model. The
simplest model is the MNL which is derived by placing restrictive assumptions
on the random component of the utility: error disturbances are assumed to be
independently and identically distributed (IID) according to a Type 1 extreme
value distribution with the cumulative distribution function. For linear indirect
utility, it can be proven (McFadden, 1973) that the probability of the individual
utility of alternative j in the choice situation t is the largest among those in the
choice set C facing individual n is
expðb0 xjtK Þ
Pnjt ¼ P 0 8j; k 2 C: ð2Þ
k expðb xjtK Þ
This simulated Pn can then be inserted into the likelihood which is the
simulated log likelihood.
The IPs are by obtained conditioning the population distribution of
parameters b, from the MXL model, on the real choices made by the individual.
This information works as a prior in Bayes’ rule. Considering individual choices
Yn it is possible to derive the conditional density hðbn jyn Þ of any bn taking the
Bayes’ rule:
X P
Pðyn jbn Þgðbn jb; Þ
hðbn jyn ; b; Þ¼ P : ð6Þ
Pðyn jb; Þ
This ratio shows that the conditional distribution hðbn jyn Þ of the IPs can be
obtained as the probability of observing the individual choices given bn
multiplied by the range of betas in the population, divided by the probability of
observing these choices in the population of betas. This last expression is the
mixed logit represented in equation (4) and it is constant.
The mean of the estimated IPs is calculated as the mean of b in the
subpopulation of individuals who chose yn, using the integral:
Z X
¼ bhðb jyn ; b;
b Þdb: ð7Þ
n
Pn ðyn jbrn Þ
PR r
: ð8Þ
r¼1 Pn ðyn jbn Þ
1
For a discussion about advantages of SP over RP see Louviere et al. (2000).
Table 1
Attributes and levels
Variable Levels
2
Orthogonality occurs when all estimable effects are uncorrelated.
3
Balance occurs when each level occurs equally often within each factor or alternative.
4
The experimental design was developed using macros in Excel following Rose et al. (2008).
bundles, is that combination which produces most information for the model. This
produced 24 different choice alternatives which were paired to form 12 choice sets
from which respondents indicated their most preferred alternative.
In the current study, each respondent was presented with 12 different choice
sets. Research has shown that the precision of choice experiment estimates is
highest in the range of 6 to 13 choice situations (Caussade et al., 2005). Twelve
choice sets were thus chosen as a reasonable limit for the choice experiment.
A labelled experiment was chosen; that is, a choice set where each alternative was
labelled. This was done for purposes of realism, to set the theatrical production in
the context of one of the five most important theatres in the local area.
The 12 SP choice sets formed the core of the questionnaire. However, the
questionnaire also included some additional questions to ascertain the cultural
habits of the respondent; and also the socio-economic and demographic
characteristics of the respondent, such as age, gender, income and educational
level, that might be hypothesised to affect choice.
IV Data
A self-completion questionnaire was distributed to people attending the Northern
Stage theatre. It is often argued that SP surveys should be administered by
personal interview, because respondents are not usually familiar with discrete
choice experiment type questionnaires (Arrow et al., 1993). However, interviewer
surveys are expensive. A simple mail questionnaire, on the other hand, is the least
expensive survey method, but mail surveys typically have a lower response rate
(McDaniel and Gates, 2004). As a compromise it was decided to use a mixed
methodology with an interviewer handing questionnaires and motivating the
respondent by briefly explaining the purposes of the research and the question-
naire, which the respondent then self-completed and mailed back to the researcher.
In addition, a cash incentive5 was offered for completed questionnaires returned,
the terms and conditions of which were set out in a signed cover letter to the
questionnaire. This method, where an interviewer talks to a potential res-
pondent and convinces him or her to complete the questionnaire, results in a much
higher response rate than a passive mail survey at a notably lower cost than a face
to face interview. Moreover, the respondent can take the questionnaire home to
complete, allowing more time to think about responses. Giving respondents time to
think has been shown to produce more accurate and reliable results (Whittington
et al., 1992). The survey achieved a 55% response rate.
The survey commenced in November 2008, and covered a number of different
theatre productions, before ending with the Royal Shakespeare Company play
(Othello) in February 2009. Six hundred questionnaires were handed out to randomly
selected people attending performances at Northern Stage over this period, and 332
questionnaires were returned. Of the respondents, 70% were women, 77% graduates
or postgraduates, and 33% of respondents had a monthly income over d3000.
5
This was the chance to participate in a cash draw, with the first prize being d50 and two
other prizes of d25 each.
V Modelling
This section presents results from three models: first a MNL with socioeconomic
variables; second a MXL; and finally an IP estimation.
A MXL model provides two parameters: a mean and standard deviation for
each estimated parameter for the population. This type of model generates
random parameters that distribute among the population according to a
specified distribution, whereas IP estimates are obtained by conditioning these
random parameters on the individual’s observed choices. Thus, the former is
called an ‘unconditional’ distribution and the later a ‘conditional’ distribution
(Train, 2003; Hess, 2007).
Table 2
MNL and MXL models
Table 2 (Continued )
Basic MNL Mixed Logit
As expected the MXL model performs much better than MNL, as can been
seen from their log-likelihood values, in spite of the fact that the MNL model
has 46 parameters; whereas the MXL model has only 23 parameters. In
addition, Table 3 shows several tests that confirm this. These are Akaike and
Bayesian information criterion. In all cases, these statistics use the log-likelihood
weighted by number of individuals and parameters to check which model fits
better, which will be the one with the lowest value.
In the MXL model, the variation (standard deviation) of the population
parameters were statistically significant for comedy, drama, price, and reviews3
(‘must-see’) and for most of the levels of WOM particularly the first level
(‘poor’) which accounts for the largest variance. Normal distributions were set
for all these parameters. This result points to the fact that most of the
heterogeneity is associated with these variables.
Various reasons might be postulated for heterogeneity in these variables.
People may use WOM information in different ways. This is consistent with the
idea that, although WOM is usually more trusted information, some individuals
may have tastes different from ‘the average’ and therefore may not be so
influenced by this assessment; while some others may not be able to so readily
identify with this source of information. In contrast Reviews do not follow a
population distribution pattern except for the highest ‘must-see’ level. This
might reflect that the fact that attitude towards reviews varies less: the source of
information is clearly defined, the critic is supposedly an expert and everybody
knows where to find these reviews. Sometimes reviews by critics differ from
popular tastes and this may be perceived in the statistical significance of the
standard deviation of the ‘must-see’. In general it can be seen (in Table 2) that
for intermediate levels (Review 1 and Review 2, which represent ‘average’ and
‘very good’) and WOM2 (‘average’) there is less heterogeneity associated with
these situations: the coefficient of the standard deviation of these variables is not
statistically significant (and hence is not reported). This is in contrast to the
extreme of the scale (i.e. a critic’s review 3: ‘must see’) which might polarise
Scottish Journal of Political Economy
r 2011 The Authors. Scottish Journal of Political Economy r 2011 Scottish Economic Society.
388 J. M. GRISOLIA AND K. WILLIS
Table 3
Tests to discriminate between models
attitudes between those who therefore would see the play and those who might
not, thus leading to heterogeneity in tastes.
The other significant random parameters were the type of show (drama and
comedy) and the price. It is worth noting the high degree of statistical
significance and also the large coefficient associated with the standard deviation
for comedy reflects the heterogeneity of tastes for this type of show, compared
with drama which has a much smaller standard deviation.
An important issue in most of models is the presence and frequency of
counterintuitive signs in some parameters, particularly for price. Some solutions
have been suggested to overcome this, such as specifying a lognormal
distribution to ensure the coefficient has the same sign for all individuals
(Train, 2003) but this option usually leads to estimation problems. Another
approach could be to use a triangular distribution which has become a more
common method (Campbell et al., 2010). In this study a negative sign is expected
for price and Wom1 (poor); and a positive sign for Wom3 and Wom4. It is not
clear, however, what the expected sign for drama and comedy should be, since
these attributes are dependent on subjective tastes. In case of ‘reviews’ we should
expect positive signs for ‘very good’ and ‘must see’.
Table 4
Comparson of results from the population means and the individual parameters models
Wom3 Wom4
Wom1 very must
Comedy Drama Reviews3 Price poor good see
Unconditional mean
Mean 0.112 0.304 1.48 0.0297 1.67 1.41 1.6
SD 0.727 0.404 0.299 0.0365 0.857 0.28 0.261
Prob (o0)% 45.5 22.1 0 20.10 98.00 0 0
Prob (40)% 54.5 77.9 100 79.90 2.00 100 100
Lower 1.329 0.473 0.926 3.288 0.848 1.103 0.103
Upper 1.520 1.101 2.076 0.026 1.964 2.104 0.042
Conditional means
Mean 0.125 0.325 1.430 0.029 1.588 1.413 1.608
SD 0.485 0.103 0.043 0.025 0.294 0.031 0.045
Min 0.9257 0.09223 1.307 0.1026 1.945 1.326 1.498
Max 1.771 0.5942 1.541 0.02122 0.3121 1.5 1.771
Perc (o0)% 59.00 0 0 10.42 100 100 100
Perc (40)% 41.00 100 100 89.58 0 0 0
Lower 0.826 0.123 1.346 2.165 1.353 1.519 0.078
Upper 1.076 0.527 1.515 1.012 1.473 1.697 0.019
8
Individual-based parameters where estimated using Winbugs (Lunn et al., 2000).
VI WTP
In its simplest formulation WTP for changes in one unit of an attribute (i.e. a
level of an attribute) is obtained as the ratio of the coefficient of this attribute, or
attribute level (l), and the price coefficient (p). This result is equivalent to the
compensating variation (Small and Rosen, 1981). Thus, the WTP for one unit of
attribute k, for a linear function, is
When interaction terms are included this formula is more complex. Formula
(10) provides a simple point estimate that is assumed to represent the whole
population, which may not be realistic. One way to overcome this issue is to
calculate confidence intervals. Unfortunately this task is not easy, as it is known
that the parameters in ratio (10) distribute asymptotically normal. The ratio of
two random normal variables is another random variable that distributes Cauchy
(Arnold and Brockett, 1992) which is unstable with undefined variance and
unknown mean. Thus, estimating a confidence interval can be problematic if only
a standard procedure is applied. There is, however, an alternative, namely to use
a formula suggested by Armstrong et al. (2001) which is easy to apply for point
estimations in the MNL since it is not necessary to know the WTP distribution.
In using MXL, calculations of WTP imply the added difficulty of working
with two parameters, the mean and the standard deviation. WTP can however
be estimated using simulation. It is known, however, that this will produce WTP
with an excessive variance (Hensher et al., 2006).
Finally, IPs estimates can be used to derive an individual’s WTP simply
taking the individual ratio of conditional mean of the parameter of interest and
the conditional mean of the price. This last method is not only simpler but also
provides accurate results, and it is the one adopted in this article.
Scottish Journal of Political Economy
r 2011 The Authors. Scottish Journal of Political Economy r 2011 Scottish Economic Society.
H E T E R O G E NE I T Y I N W I L L I N G N E S S - T O - P A Y 391
Comedy Drama
40
60
50 30
Frequency
Frequency
40
30 20
20
10
10
0 0
0
0
00
50
00
50
00
00
10
20
30
40
50
60
.0
.5
0.
0.
1.
1.
2.
0.
0.
0.
0.
0.
0.
0.
–1
–0
Frequency
40
30 20
20
10
10
0 0
40
50
60
70
80
–0 0
–0 8
–0 6
–0 4
0. 2
00
02
.1
.1
.0
.0
.0
.0
1.
1.
1.
1.
1.
0.
–0
–0
30
Frequency
60
20
40
10 20
0 0
30
35
40
45
50
00
00
00
.5
.5
1.
1.
1.
1.
1.
0.
–2
–1
–1
–0
30
Frequency
20
10
0
30
35
40
45
50
55
1.
1.
1.
1.
1.
1.
Table 5
Willingness to pay for all theatre play attributes based on individual parametersa and MNL models
IP estimates
Mean (d per play) 4.83 8.33 38.56 43.15 38.05 43.33
Upper (d per play) 36.151 18.208 73.330 1.167 72.213 82.275
Lower (d per play) 26.491 1.548 3.790 85.133 3.887 4.385
MNL estimates
Mean (d per play) 17.79 12.40 45.86 32.87 38.95 47.51
Upper (d per play) 25.90 21.40 58.24 22.09 51.24 61.72
Lower (d per play) 10.89 4.05 37.01 46.31 29.18 36.33
a
Confidence interval for 95%.
Ortúzar (2005) who suggest considering whether or not these individuals give
due attention to the price attribute. Analysis of the data revealed that these
extreme cases corresponded with those observations that have the lowest
statistical significance for the price parameter, in some cases close to zero.9
To avoid the problem of a price coefficient close to zero generating infinite
moments, and to obtain a more reliable distribution, only those individuals were
selected who had a price coefficient which exceeded a statistically significance
threshold. It was decided to set the minimum as 75% significance; that is, a t-ratio
of 0.647. Proceeding in this way meant trimming the sample from 332 to 216
individuals. The WTP results are shown in Table 5, for both the IP and MNL
models. The IP model produces slightly lower estimates of mean WTP for comedy
and drama productions compared with the MNL model, although the results for
the effect of information (reviews and word of mouth) on WTP are quite similar.
Finally it is instructive to estimate the WTP for different types of production.
In doing so, several different attributes were selected to calculate this complex
ratio, since it is not only the type of production but also other attributes such as
whether the writer is known or unknown, the type of reviews and WOM about the
play, that affect WTP. A realistic assumption might be that theatre plays usually
have good reviews, from ‘very good’ to ‘must see’; that ‘word of mouth’ is not
usually available, and that ‘known’ authors are more frequently represented.
Therefore the scenario in Table 6 is for a ‘must-see’ play with a ‘known’ author
showing at the Theatre Royal. The WTP results presented in Table 6 are for the
three basic types of play, and are shown as histograms in Figure 2.
This type of analysis provides a better understanding of consumer tastes
which could be exploited for marketing purposes. For instance, the model can
clearly identify those shows which would command a higher price on the basis of
9
Daly et al. (2009) argue that estimating WTP from random parameters will produce infinite
moments. This is the result of values of cost coefficients being close to zero. They demonstrate
that the usual practise of applying triangular, lognormal, uniform or normal distributions will
not solve the problem. On the other hand they discard simulation (Espino et al., 2006) as a way
to overcome the problem and reject the ratio of individual parameters as another way to obtain
individual WTP. Nevertheless this later practise has been common in those few papers that
estimate individual WTP such as Hensher et al. (2006) and Hess (2007) and is adopted here.
Table 6
Willingness to pay for a play in Theatre Royal with a must-see reviews and known author
Frequency
40
30
30
20
20
10 10
0 0
–1 00
–5 0
00
00
15 0
00
00
00
00
.0
.0
.0
.0
.
0.
0.
0.
0.
0.
0.
0.
50
00
50
00
00
10
10
20
–1
–2
–1
Experimental Theatre in
Theatre Royal
50 Mean =60.17
Std. Dev. =27.781
40 N =216
Frequency
30
20
10
0
–1 00
–5 0
00
00
15 0
00
.0
.0
0
.
0.
0.
0.
0.
50
00
50
10
–1
Figure 2. WTP distributions for some types of shows at the Theatre Royal.
their attributes, including reviews and WOM where the show had been running
say in London before going on tour. Drama can, on average, command a higher
ticket price than comedy. Also the price of a ticket could be increased by d38.56
on average (i.e. dividing the ‘must see’ coefficient by the price coefficient) for a
‘must-see’ play, compared with a play without a ‘review’. In general the impact
of ‘reviews’ is impressive, but even this information is subject to heterogeneity
(see Figure 1), with some individuals willing to pay more than the average and
some individuals willing to pay much less than the average.
Scottish Journal of Political Economy
r 2011 The Authors. Scottish Journal of Political Economy r 2011 Scottish Economic Society.
394 J. M. GRISOLIA AND K. WILLIS
VII Conclusions
IP estimation from discrete choice models applied to theatre production provides a
method which can be used to accurately estimate the value of different plays. The
fact that these estimations come from IPs allows the researcher to address
heterogeneity in the sample, overcoming the problems of interval estimation.
Unlike the use of systematic variation, the IPs estimation provides a specific
structure of tastes even in cases where the socioeconomic information is missing,
which can occur for a number of observations in questionnaire surveys where
respondents are reluctant to provide such information. In addition since the
conditional estimation is more precise and simpler than the one from mixed logit
it is also possible to select ex-post the observations based on their significance in
order to obtain a more realistic distributions of WTP for theatre. However, both
the IP and MNL models produce similar estimates of WTP for theatre
attributes, when a strict comparison is made between the two models.
The results of both models (IP and MNL) reveal that there is considerable
heterogeneity in preferences for the attributes of theatre productions, and in
theatre-goers WTP for tickets based on these attributes. These are factors which
theatre managers and producers of plays should consider in deciding which
plays, with attribute bundles, to stage; and in any strategy to maximise revenue.
This type of model has an enormous potential to detect differences and segments
among theatre-goers, that might justify greater differential pricing strategies; and to
assist theatre management to capture more theatre-goer consumer surplus. In
addition, the approach can assist in a better understanding the social value of
performing arts and how this varies by different types of show. A SP survey with IPs
estimation appears to be more precise and flexible than RP methods.
Acknowledgements
This work is supported by the Arts and Humanities Research Council; and the
Arts Council England [Grant: AHRC Fellowship in Economic Impact
Assessment of Arts and Humanities]. The authors would like to thank Mauricio
Sillano for his help with the WinBUGS code; and Juan de Dios Ortúzar, and an
anonymous referee, for their helpful advice and comments.
References
AKDEDE, S. and KING, J. (2006). Demand for and productivity analysis of Turkish public
theatre. Journal of Cultural Economics, 30, pp. 219–31.
ARMSTRONG, P., GARRIDO, R. A. and ORTUZAR, J. de D. (2001). Confidence intervals to
bound the value of time. Transportation Research, 37, E, pp. 143–61.
ARNOLD, B. and BROCKETT, P. (1992). On the distribution whose components ratios are
Cauchy. The American Statistician, 46, pp. 25–6.
ARROW, K. J., SOLLOW, R., PORTNEY, P. R., LEAMER, E. E., RADNER, R. and SCHUMAN, H.
(1993). Report of the national oceanic atmospheric administration. Federal Register, 58,
10, pp. 4016–614.
CAMPBELL, D., HESS, S., SCARPA, R. and ROSE, J. M. (2010). Accommodating coefficient
outliers in discrete choice modeling: a comparison of discrete and continuous mixing
approaches. In S. Hess and A. Daly (eds.), Choice Modelling: The State-of-the-Art and the
State-of-Practice. Bingley: Emerald Publishing, pp. 331–52.
CAUSSADE, S., ORTUZAR, J. de D., RIZZI, L. I. and HENSHER, D. A. (2005). Assessing the
influence of design dimensions on stated choice estimates. Transportation Research B, 39,
pp. 621–40.
CORNING, J. and LEVEY, A. (2002). Demand for live theatre with market segmentation and
seasonality. Journal of Cultural Economics, 26, pp. 217–35.
DALY, A., HESS, S. and TRAIN, K. (2009). Assuring Finite Moments for Willingness To Pay in
Random Coefficient Models. Working Paper. Institute for Transport Studies, Leeds
University, Leeds.
ESPINO, R., ORTÚZAR, J. de D. and ROMÁN, C. (2006). Confidence intervals for willingness-
to-pay measures in mode choice models. Networks and Spatial Economics, 6, 2, pp. 81–96.
FORREST, D., GRIMES, K. and WOODS, R. (2000). Is it worth subsidizing regional repertory
theatre? Oxford Economic Papers, 52, pp. 381–97.
GRISOLÍA, J. M. and WILLIS, K. G. (2010). An evening at the theatre: using choice experiments
to model preferences for theatres and theatrical productions. Applied Economics, 6
December 2010 [Epub ahead of print. 6 December 2010].
HENSHER, D. A., GREENE, W. H. and ROSE, J. M. (2006). Deriving willingness-to-pay estimates
of travel-time savings from individual-based parameters. Environment and Planning A, 38,
12, pp. 2365–76.
HESS, S. (2007). Posterior analysis of random taste coefficients in air travel behaviour. Journal
of Air Transport Management, 13, pp. 203–12.
LANCASTER, K. J. (1966). A new approach to consumer theory. The Journal of Political
Economy, 74, pp. 132–57.
LOUVIERE, J. J., HENSHER, D. A. and SWAIT, J. D. (2000). Stated Choice Methods: Analysis and
Application. Cambridge: Cambridge University Press.
LUNN, D. J., THOMAS, A., BEST, N. and SPIEGELHALTER, D. (2000). WinBUGS a Bayesian
modelling framework: concepts, structure, and extensibility. Statistics and Computing, 10,
pp. 325–37.
MCDANIEL, K. and GATES, R. (2004). Marketing Research Essentials. New York: John Wiley &
Sons.
MCFADDEN, D. (1973). Conditional logit analysis of qualitative choice behaviour. In
P. Zarembka (ed.), Frontiers of Econometrics. New York: Academic Press.
MCFADDEN, D. and TRAIN, K. (2000). Mixed MNL models for discrete response. Journal of
Applied Econometrics, 15, 5, pp. 447–70.
ORTúZAR, J. de D. and WILLUMSEN, L. G. (2001). Modelling Transport. Chichester: John Wiley.
ROSE, J. M. and BLIEMER, M. J. C. (2004). The Design of Stated Choice Experiments: the State
of Practice. Working Paper. University of Sydney. Institute of Transport and Logistics
Studies, Sydney.
ROSE, J. M., BLIEMER, M. J. C., HENSHER, D. A. and COLLINS, A. T. (2008). Designing
efficient stated choice experiments in the presence of reference alternatives. Transportation
Research Part B: Methodological, 42, 4, pp. 395–406.
SCARPA, R., WILLIS, K. G. and ACUTT, M. (2004). Individual specific welfare measures for
public goods: a latent class approach to residential customers of Yorkshire Water. In
P. Koundouri (ed.), Econometrics Informing Natural Resources Management. Cheltenham:
Edward Elgar.
SILLANO, M. and ORTÚZAR, J. de D. (2005). Willingness-to-pay estimation with mixed logit
models: some new evidence. Environment and Planning A, 37, 3, pp. 525–50.
SMALL, K. and ROSEN, H. (1981). Applied welfare economics of discrete choice models.
Econometrica, 49, pp. 105–30.
THROSBY, D. (1994). The production and consumption of the arts: a view of cultural economics.
Journal of Economic Literature, 32, pp. 1–29.
TRAIN, K. (1998). Recreation demand models with taste differences over people. Land
Economics, 74, 2, pp. 230–9.
TRAIN, K. (2003). Discrete Choice Models with simulation. Cambridge: Cambridge University Press.
URRUTIAGUER, D. (2002). Quality judgements and demand for French public theatre. Journal
of Cultural Economics, 22, pp. 185–202.
WHITTINGTON, D., SMITH, V. K., OKORAFOR, A., OKORE, A., LIU, J. L. and MCPHAIL, A.
(1992). Giving respondents time to think in contingent valuation studies: a developing country
application. Journal of Environmental Economics and Management, 22, pp. 205–25.