Evaluating Animal Welfare with Choice Experiments: An Application to Swedish Pig Production

Evaluating Animal Welfare with Choice Experiments:
An Application to Swedish Pig Production
Carolina Liljenstolpe
Department of Economics, SLU, Uppsala, Sweden.
E-mail: carolina.liljenstolpe@ekon.slu.se

In this study, the demand for animal welfare attributes when buying pork fillet is investigated
among Swedish respondents. The issue is of importance in order to ensure an economically
viable pig industry while applying an increasing number of animal friendly practices. In order
to obtain information about consumer demand, an indirect utility function and willingness to
pay (WTP) for animal welfare attributes are estimated. The attributes are solely associated
with animal friendly practices. An investigation of numerous housing and managerial
practices of pig production has not yet been performed. The indirect utility function is
estimated using a random parameter logit model. A realistic approach when modeling
consumer choice is to allow for heterogeneity in preferences. The relevance of assuming
randomness of some of the parameters is evaluated by using a specification test developed by
McFadden and Train (2000). The WTP is also estimated at the individual level. The results
indicate that WTP for animal welfare attributes may be negative or positive. The preferences
are also heterogeneous among respondents, which may be explained by a segmentation of
preferences. Finally, the WTP estimates for animal welfare practices are compared with cost
estimates for such production systems. [Econlit subject codes: C010, C500, Q100] r 2008
Wiley-Liss, Inc.


Pig production in Organisation for Economic Co-operation and Development
(OECD) countries during the recent decades has been undergoing an industrializa-
tion process, characterized in part by fewer, larger, and more efficient production
operations (Lundeheim & Holmgren, 1994; Paarlberg, Boehlje, Foster, Doering, &
Wallace, 1999). Along with a substantial demand for inexpensive meat products, and
yet a consideration for sensory-specific qualities and food safety, the public debate
has focused on animal rights and the handling of animals in the industrialized sector
(see for example American Veterinary Medical Association [AVMA], 2006 and
Lindgren & Forslund, 1990). Integration of more animal friendly practices into the
production system by imposing stricter animal welfare regulations1 is a problematic

   Animal welfare regulation in this regard implies the rules that are stipulated in order to improve the
wellbeing of livestock and affect practices in production in terms of housing conditions, veterinary care,
feed and transportation.

Agribusiness, Vol. 24 (1) 67–84 (2008)                               r 2008 Wiley Periodicals, Inc.
Published online in Wiley InterScience (www.interscience.wiley.com).     DOI: 10.1002/AGR.20147


issue as it probably incurs additional costs for producers which, in turn, increases the
price paid by consumers (Henson & Traill, 2000; McInerney, 1991; Stott et al., 2005).
Consequently, it is vital that there exists a demand for animal friendly practices in
order to ensure an economically viable industry. If that is the case there ought to
exist a willingness to pay (WTP) for animal welfare attributes, i.e.,, for procedures
that promote improvements in animal well-being. Such demand analysis could be of
value for the conventional as well as for the organic pig production because it could
enhance the competitiveness of the industry. Public demand in Sweden for animal
welfare attributes could also be of international interest, as Sweden has already
adopted relatively strict regulations for pig production (e.g., ‘‘The Swedish model’’2).
Also, national demand should be taken into account prior to a harmonization of
animal welfare legislation (see European Commission, 2006).
   The credence character of animal friendly practices raises a complex problem
when evaluating the demand for products with specific qualities. As actual sales
data for hypothetical products are not available, a postulated demand for animal
welfare attributes is required in order to evaluate animal friendly production.
Previous surveys examining hypothetical markets have mostly used the contingent
valuation method (CVM)—a few them include Anderson and Frykblom
(1999), Bennett (1997), Bennett and Larson (1996), Drake and Holm (1989), and
Rolfe (1999). The drawback of the CVM approach is that estimates are obtained as
absolute values and therefore comparisons are not meaningful. Due to the diversified
character of preferences, it can be useful to adopt a multiple choice approach for
the evaluation of agricultural production systems. Den Ouden (1996) used conjoint
analysis to evaluate 12 attributes in pig production, using a small sample of
consumers and experts on pig welfare issues. It is common practice in choice
experiments to assume that preferences are heterogeneous across respondents.
Often a random parameter logit model (RPL) is used to estimate WTP. Extensive
surveys of the evaluation of preferences have been performed to consider preference
heterogeneity for animal welfare. More recent studies include Andersen (2003),
Carlsson, Frykblom, and Lagerkvist (2004), Enneking (2004), Lagerkvist, Carlsson,
and Viske (2006), and Larue, West, Gendron, and Lambert (2004). Previous studies
of animal welfare attributes use a mixture of animal welfare attributes and food
safety attributes. However, these studies reveal a relatively high evaluation of
food safety related attributes compared to animal welfare attributes, which indicates
that preferences may be segmented. A choice experiment study of WTP
for animal welfare attributes adhering to the Swedish model have not been
performed hitherto.
   The objective of this study is to evaluate animal welfare attributes and animal
friendly production standards in the Swedish pig production. These may be
attributable to the voluntary rules stipulated in ‘‘the Swedish model’’ or are practiced
experimentally. The WTP for animal welfare attributes among costumers that buy

   The ‘‘Swedish model’’ refers to Swedish Animal Welfare Acts of 1988 and additional directives of 1989
and 1993 (Swedish Gov. Offices, 1988, 1989, and 1993). Some additional voluntary action programmes are
normally also included in this definition. The features of the ‘‘Swedish model’’ that are analyzed in this
survey are rules concerning bedding straw, air partition, castration, transportation, feed, housing and

Agribusiness     DOI 10.1002/agr

pork fillet is estimated by applying a RPL model. This model also allows
for individual ranking of WTP, which makes it possible to estimate the distribution
of WTP and hence detect its diversity. This study provides insights into
consumer behavior: why results from similar studies may differ and why a
substantial segmentation of preferences may exist. Furthermore, in order to
motivate the use of the animal welfare attributes, a comparison between
WTP values and the corresponding costs for such production systems is made. A
comparison as such has not been done in previous choice experiments of pig
production. Most animal welfare attributes are regarded as welfare improving by the
respondents, which to some extent supports the results of previous studies. Also,
even if a hypothetical bias exists, the WTP values exceed approximated costs of
implementation. These results motivate the use of the animal friendly practices into
production systems.
  The article is organized as follows. First, the econometric model is presented with a
definition of the indirect utility function and WTP and derivation of WTP estimates
on individual level. Next, the design of the choice experiment is described, followed
by presentation of the econometric results. At the end of the article, a discussion of
the findings and some concluding remarks are presented.


The relative utility of an individual in a discrete choice model is represented by linear
utility function. Utility is assumed to either increase or decrease according to price
and animal welfare attributes, depending on how the respondent regards animal
friendly practices. For an individual n choosing alternative j, the indirect utility is
assumed to take the following form:
                                Unj ¼ anj þ gj Sn þ b0n xnj þ Enj                     ð2:1Þ
The obtained indirect utility may vary between choice j and individuals n (the total
number of individuals is n 5 1,y, N). Indirect utility is assumed to consist of a
deterministic part Vnj 5 anj1cjsn1bn’xnj and a stochastic part enj. The deterministic
component of the utility function consists of anj which is the option specific intercept
that corresponds to individual n’s intrinsic preference for alternative j. The
socioeconomic and demographic characteristics of the individual, sn, and the
coefficient vector cj correspond to the systematic preference heterogeneity among the
individuals in the sample. Altogether, 12 animal welfare attribute coefficients are
estimated and, with the price coefficient bn1 , there are 13 coefficients;
(bn ¼ ½bn1 ; . . . ; bn13 ’). These coefficients are assumed to be generic (i.e., the
coefficients of the explanatory variables do not vary across the options). Hence,
an assumption of stable preferences is made.
   The logit choice probability is derived by assuming independence and
a distribution of the error terms in the utility function (see McFadden, 1974).
This causes the multinomial logit model (MNL) to overestimate the joint probability
of close substitutes because of its independence of irrelevant alternatives (IIA)
property and because it does not allow for random taste variation as the unknown
utility terms enj are assumed to be independent and identically distributed
(i.i.d.). The IIA property is restrictive. In order to test whether certain parameters
exhibit randomness, McFadden and Train (2000) propose a Lagrange Multiplier
                                                             Agribusiness   DOI 10.1002/agr

test.3 The artificial variables zni capture heterogeneity in terms of a correlation
between the chosen and nonchosen alternatives. The hypothesis that the coefficients
of the variables in zni are equal to zero is tested.
  The RPL model relaxes the IIA and iid by assuming heterogeneity among the
sample of individuals (for a derivation of the model, see for example Ben-Akiva &
Lerman, 1985 and Revelt & Train, 1998). The extent to which an individual bn differs
from the population mean b constitutes an unobserved taste variation in the sample.
One cannot observe enough replications to obtain estimates of bn. Instead, the
expected value across the population is used and distributional assumptions are
made about each coefficient in bn 5 [bn1,y,bn13]’ (Ben-Akiva & Lerman). The
distributional assumptions of the parameters may have a considerable impact on the
results. Hensher and Greene (2003) developed an empirical procedure to identify the
true distribution. A procedure is suggested that estimates (n1) models, at each step
removing one individual observation. The parameter estimates are plotted in order
to establish the empirical profile of the unobserved heterogeneity.
  The RPL model is estimated by simulating maximum likelihood with Nlogit 3.0
(Greene, 2002). Several authors suggest that the fit and computation time of RPL
models would be improved with fewer and more even draws from the distribution,
so-called Halton draws (Bhat, 2000; Hensher, 2001; Revelt & Train, 1998). Train
(2000) suggests that several hundred replications are needed to obtain an unbiased
  The WTP of individual n is most often defined as the net income change that
equates to a change in quality or quantity of a particular good (Freeman, 1993; Just,
Hueth, & Schmitz, 2004). This makes the interpretation of a marginal WTP from the
utility function straightforward; marginal WTP is the marginal change in the price
parameter required to keep utility constant after a marginal change in attribute
parameter. Based on the estimated parameters, the marginal WTP for animal welfare
attributes are calculated. The standard errors can be approximated by using the
Delta method (Greene, 2000).
  The RPL model makes it possible to retrieve individual-level parameters from the
estimated model by the use of Bayes Theorem. Thus, it is possible to obtain an
estimate of the location of single respondent and a sample distribution of WTP
(Train, 2003).
                                            b bn Pni f ðbn Þdbn
                                  E½bn  ¼ Rn                                   ð2:2Þ
                                               b Pni f ðdbn Þn

   The product of the estimated logit choice probabilities Pni in a choice set and the choice vector, xni: is
summed over each choice set in order to obtain xni:
                                                 xni ¼       xni  Pni
The artificial variables in zni are created as:

                                                zni ¼ ðxnj  xni Þ2
The vector zni is linearly independent of the vector of chosen alternatives xnj and the MNL model is re-
estimated including the artificial variables. A likelihood ratio test is performed in order to test for the
hypothesis where artificial variables are to be omitted from the MNL model or if mixing of certain
parameters is needed. For a proof of the specification test, see McFadden and Train (2000).

Agribusiness      DOI 10.1002/agr

TABLE 1.      Animal Welfare Attributes in Swedish Pig Production

Attribute           Level

Transport           1. Transports according to existing regulations and limited by time
                    2. Mobile abattoirs
                    3. Transports according to existing regulations and limited by distance
Castration          1. Castration of piglet without anesthesia
                    2. Castration of piglet with anesthesia
                    3. No castration
Housing system      1. Reared in a pen holding 8 pigs (size 5 4.92 square feet/cwt pig)
                    2. Reared in deep litter holding 50 pigs (size 5 7.14 square feet/cwt pig)
                    3. Reared in a pen holding 8 pigs with a possibility to stay inside or outside
                       (size 5 10.9 square feet/cwt pig). During summertime, pasture is provided
                       with an opportunity for mud bathing and grazing
Feed                1. No restrictions on feed, or minimum limit of on-farm produced feed
                    2. No minimum level of on-farm produced feed but all feed must
                       be Swedish
                    3. All feed must be Swedish and at least half of it has to be produced at
                       the farm
Mixing of pigs      1. Mixing of unfamiliar pigs allowed
                    2. Mixing of unfamiliar pigs forbidden
Stock size          1. A maximum of 400 pigs in one section
                    2. A maximum of 200 pigs in one section
                    3. A maximum of 100 pigs in one section
Bedding straw       1. No minimum restriction of bedding straw
                    2. Minimum amount of bedding straw
Price (US$/lb)      0.18, 0.45, 0.73, 0.91, 1.30, 1.50 and 1.90

The sample distribution may illustrate potential problems of unobserved hetero-
geneity; if the parameters change sign due to large standard deviations and if there
are segmentations in preferences. The integral in 2.2 is complex and must be
estimated by numerical simulations.


Based on the literature,4 current regulation of organic pig rearing, two focus groups5
and interviews with representatives from consumer associations—the Swedish
Farmers Federation (LRF), Swedish Meats, and the Swedish University of
Agricultural Sciences—, seven welfare attributes in pig production are defined.
The welfare attributes (cf. Table 1) concern transportation, housing systems, stock
density, supply of bedding straw, castration, mixing pigs from different litters, and
types of feed. The choice experiment data was collected exclusively from Swedish

   Due to space limitation, the literature review is not included in the paper but can be provided by the
author upon request.
  Due to space limitation, the notes from the focus group discussion are not included in the paper but
may be provided by the author upon request.

                                                                      Agribusiness     DOI 10.1002/agr

consumers using conjoint choice modeling technique (Louviere, Hensher, & Swait,
2000). There are 972 possible combinations of animal welfare attributes (or utility
levels). With the OPTEX procedure in SAS, a linear D-optimal design procedure
(Kuhfeld, 2001), 32 orthogonal combinations were created for the survey. These
were blocked into four different survey versions, each containing four choice sets.
The product chosen for the survey was pork fillet, which constitutes about 1.4% of
the pig carcass. The price of pork fillet depends on market demand and is
characterized by a complex relationship with the carcass price (Swedish Meats,
  Each respondent evaluated four choice sets, choosing among three options. The
first option always referred to a base scenario level with no extra charge. Alternatives
2 and 3 included an extra charge due to the enhanced level of animal well-being. The
prices of the attribute vector were chosen in order to be realistic in terms of the
associated costs of production systems and to be acceptable to the respondent with
respect to their budget constraints. It was not necessary to have all four choice-sets
completed in order to be included in the data set. The questionnaire consisted of
two parts. In addition to multiple choice questions, the respondents provided
socioeconomic information on income, education, and age. Each questionnaire was
accompanied by an information sheet explaining the different stages and attributes
of the Swedish pig production chain. This was necessary to minimize potential
information bias due to the limited knowledge of respondents regarding agricultural

A sample of Swedish respondents aged between 18 and 75 was obtained from
SPAR.6 A total of 3,000 individuals in Sweden received the questionnaire in May
2002. After 2 weeks, a reminder was sent to those who had not responded.
Altogether, 1,400 (45%) of the questionnaires were returned and out of these 1,250
(43%) were available for an empirical analysis. Table 2 presents some demographic
and socioeconomic statistics of the sample.
   Where national statistics were available, comparisons with the descriptive statistics
of the sample were performed to assess the representativeness of the sample. The
mean age in the sample was slightly higher: women had a higher response rate and the
average ratio of children/respondent and the average number of persons per
household were higher. Vegetarians were excluded from the sample. The socio-
economic variables that were significant at the 10% level were included in the
analysis. The specification of the RPL model (Table 3) includes the variables gender
and income. The generic form facilitates the interaction with the option specific
intercept anj and be included in the two alternatives that imply changes in animal
friendly practices. Men in the sample were found to derive greater utility from the use
of animal friendly practices. A negative sign on ‘Income’ indicates that respondents
with high income are less concerned with improved animal well-being.
   Randomness/taste variation among the respondents is confirmed through the
estimation of a MNL model including artificial variables. McFadden and Train
(2000) use t-statistics to test the hypothesis of the coefficient vector being statistically

      ’’Swedish census registry’’

Agribusiness        DOI 10.1002/agr

TABLE 2.       Demographic and Socioeconomic Statistics of the Respondents

Variables     Description                                          Mean        Std deva Min        Max

Age           Average age of respondent                            46.10       15.23       18      75
Male          Proportion of men in sample                          0.4463      0.4971      0       1
Child         Proportion households with children                  0.3434      0.4749      0       1
Dummy         Proportion of questions concerning pork              0.1014      0.3018      0       1
                chops or pork fillet
Prhh          Average number of persons in household               2.6120      1.2820      1       8
Inc           Average household income/month after tax             2,1450      994         1028    4625
Rel           Proportion of persons who consider                   0.5203      0.4996      0       1
                themselves to have relation to the
                agricultural sector
Sass          Proportion of members of a ‘‘socially                0.1308      0.3372      0       1
                oriented’’ association
Eass          Proportion of members of an                          0.1252      0.3310      0       1
                ‘‘environmentally oriented’’ association
Shop          Proportion of respondents doing the                  0.8386      0.3679      0       1
                household shopping
NonVeg        Proportion of non-vegetarians in sample              0.9834      0.1278      0       1
Samh          Proportion living in a village (1,000–9,999          0.1841      0.3875      0       1
Minc          Proportion living in a minor city                    0.1595      0.3661      0       1
                (10,000–39,999 inhabitants)
Medc          Proportion living in a medium sized city             0.2094      0.4069      0       1
                (440,000 inhabitants)
Stad          Proportion living in a big city (Stockholm,          0.2707      0.4443      0       1
                Gothenburg or Malmö)
Note. According to Statistics Sweden (2003) the average age was 44.8 years in 2002. There were 49.74%
women and 50.26% men between 18–75 years in the population in 2002. Proportion of households with
children was 38.26 %. Average number of persons in household was 2.01. The average disposable income
(net of all taxes and social transfers) for all households in Sweden 2002 was 1,708 US$/month. Proportions
of individuals living in a small village were 6%, in a minor city 33%, in a medium sized city 32% and in a
big city 29%.
 Std dev: standard deviation.

different from zero. The decision rule applied in order to reject the null of no
randomness requires the absolute t-value to be greater than one. The artificial
variables ‘‘Mobile slaughter’ (t 5 2.619), ‘‘Big Box’’ (t 5 1.117), ‘‘No castration’’
(t 5 1.501), ‘‘Stock limit: 200 pigs’’ (t 5 2.004), ‘‘Stock limit: 100 pigs’’ (t 5 2.066),
‘‘No mixing of pigs’’ (t 5 1.801), and ‘‘Minimum amount of bedding straw’’
(t 5 1.082) all have an absolute t-value in excess of one. Accordingly, they are
considered as random parameters.
   The empirical distributions of the random parameters were generated from 800
repeated estimations of a model for all but one respondent; i.e., removing one individual
each time. For ‘‘Mobile slaughter’’ the estimated value from the full-sample MNL model
is 0.34007 and is placed in the middle of the histogram in Figure 1 below.

   The results from these MNL estimations are available from the author upon request.

                                                                      Agribusiness      DOI 10.1002/agr

        350           mtsp







               0.332 0.333 0.334 0.335 0.336 0.337 0.338 0.339 0.340 0.341 0.342 0.343 0.344 0.345 0.346 0.347

                    Figure 1       The empirical distribution of ‘‘Mobile slaughter.’’

   The histogram plot suggests that individual error terms enj are drawn from a
normal distribution. Depending on the form of the histogram, a triangular
distribution could be more suitable. However, estimation imposing a triangular
distribution did not considerably change the results compared to a normal
distribution. The assumption that the random parameters follow a normal
distribution implies that both negative and positive values for this parameter may
exist if preferences are very heterogeneous.
   The RPL model was simulated using 300 replications, with a maximum of 100
iterations through Halton draws. As random attributes ‘‘Minimum amount of
bedding straw’’ and ‘‘Big box’’ caused nonconvergence, the parameters were
modeled as fixed in the random model specification. Their random status was
questionable in light of their absolute t-values which were close to unity in the
specification test. Furthermore, the price parameter was kept fixed to ensure that the
distribution of WTP matches the distribution of animal welfare attributes (Ruud,
1996). The results of the RPL estimation are presented in Table 3. Not surprisingly,
the price coefficient of the utility function is negative and significantly different from
zero, which means that a price increase lessens the probability that a respondent
would choose an improved quality alternative.
   Based on the estimated parameters, the marginal WTP values are calculated and
the results are reported in Table 4. The standard errors are obtained by the Delta
   Most of the attributes are perceived by respondents as likely to improve animal
well-being. Among the random variables, the attributes ‘‘Mobile slaughter’’ and
‘‘Stock limit: 100 pigs’’ yield the highest WTP. The ‘‘Mobile slaughter’’ attribute
yields a mean WTP about 19% higher than base. The individual level parameters of
marginal WTP are obtained using Bayes rule. The results of these calculations are
presented in Table 5.
Agribusiness        DOI 10.1002/agr

            TABLE 3.         Random Parameter Logit Model

            Attribute                                    Coefficient        SEa

            Fixed effects
            Transports decided by distance, b1           0.0045            0.1630
            Castration with anesthesia, b4               0.4592            0.1700
            Big box, b5                                  0.3115            0.1782
            In-out box, b6                               0.8734            0.3153
            Swedish feed, b7                             0.4486            0.2340
            Farm feed, b8                                0.6256            0.1521
            Minimum amount of bedding straw, b12         0.2499            0.1492
            Intercept, ánj                              0.6511            0.3099
            Price, b1                                    0.0168           0.0096
            Gender, g1                                   0.3506            0.1225
            Inc, g2                                      0.0342           0.0143
            Random effects
            Mobile slaughter, b2                         0.5091            0.1531
            No castration, b3                            0.4030           0.2113
            Stock limit: 200 pigs, b9                    0.3876            0.1907
            Stock limit: 100 pigs, b10                   0.5404            0.2437
            No mixing of unfamiliar pigs, b11            0.3640            0.1367
            Log-likelihood                               3529
            Pseudo-R2                                    0.1455
                SE: standard error.

TABLE 4.        Mean Willingness to Pay (WTP) for Random Parameters

                                       WTP             90% confidence       from base scenario
Attribute                              (US$/lb) SEa    interval            price (%)

Mobile slaughter, b2                   1.41     0.68   0.29–2.53           19
No castration, b3                      1.13    1.00   2.78–0.52          15
Stock limit: 200 pigs, b9              1.09     0.79   0.21–2.39          15
Stock limit: 100 pigs, b10             1.50     1.02   0.18–3.18          20
No mixing of unfamiliar pigs, b11      1.00     0.68   0.12–2.12          13
SE: standard error.

  The mean individual WTP values are similar to the estimated population means
reported in Table 4. The probability of a sign reversal is reported in the last column
in Table 5. The probabilities of a sign reversal are high, which indicates that there
may exist heterogeneity in preferences. The highest degree of heterogeneity is found
for the ‘‘Mobile slaughter,’’ ‘‘No castration’’ and ‘‘No mixing’’ attributes. Figure 2
shows the actual frequency distribution of WTP for ‘‘Mobile slaughter.’’
  This distribution is bi-modal. Hence one segment of the population perceives
benefit from the attribute, another segment sees drawbacks with the attribute.
                                                            Agribusiness      DOI 10.1002/agr

TABLE 5.                Estimated Individual WTP

Attribute                                                    Mean        Std deva       Min         Max        reversal (%)

Mobile slaughter, b2                                         1.54        2.44           5.69       6.77       0.36
No castration, b3                                            1.21       1.61           4.92       3.19       0.18
Stock limit: 200 pigs, b9                                    1.21        0.83           0.34       3.03       0.08
Stock limit: 100 pigs, b10                                   1.57        1.00           1.13       3.23       0.11
No mixing of unfamiliar pigs, b11                            1.01        1.78           3.45       3.91       0.32
    Std dev: standard deviation.










                  -7      -6      -5     -4   -3   -2   -1     0     1    2     3   4    5      6   7      8

             Figure 2                  Individual marginal WTP for the attribute ‘‘Mobile slaughter.’’


6.1. Heterogeneous Preferences
We can draw several conclusions from our model measuring consumer utility derived
from animal welfare attributes in Swedish pig production. First, preferences for such
attributes are heterogeneous among the respondents. The estimated distributions of
individual WTP indicate that the probabilities of a sign reversal are high. Sometimes,
this is considered as a problem8 (Hensher & Greene, 2003). However, for the

   Lognormal distribution is often applied in order to reduce a probability of sign reversal. However, this
distribution has a long right-hand tail which can be a disadvantage for WTP estimates that result in a large
proportion of unreasonable values. Further, the lognormal distribution is known to cause problems with
convergence in the model estimation (Hensher & Greene, 2003). The convergence problem is also
experienced in this study.

Agribusiness                DOI 10.1002/agr

evaluation of animal welfare it may be reasonable that individual parameter
estimates take different signs as one animal welfare attribute may be perceived to
enhance the well-being of pigs by one individual while another individual might be
concerned about the attribute’s impact on for example food safety. The results of the
pre-investigatory focus group discussions give some further insights into this issue.
The interviews reveal a definite concern over three issues: food prices, food safety,
and animal well-being. On the one hand, animal welfare attributes concerning
housing systems, like deep litter boxes and indoor-outdoor boxes, were considered
very important, along with the importance of handling animals humanely, from
birth to slaughter. Also, ‘‘Mobile slaughter’’ was regarded as a good alternative to
transportation. On the other hand, the importance of high quality feed and concerns
about the use of antibiotics and traces of pesticides or medicines in food were
evident. Contradictory opinions which possibly interact and affect the results were
also observed by Ngapo et al. (2003), where fear of BSE spread via animal feed was
expressed and long-distance transport in confined spaces were considered as
deleterious. Verbeke (2000) and Verbeke and Ward (2001) also found conflicts in
consumer perceptions and behavior between food safety and animal well-being. If
different motives underlie individual choices, e.g., food safety, or environmental and
animal welfare concerns, the utility parameters may have a discrete support, which
cannot be handled within the RPL modeling framework. According to Hanemann
and Kanninen (1998), heterogeneous preferences among respondents may require
different probability models. Thus, too many aims within the choice-set may be
further exacerbated by combining attributes related to animal friendly practices and
attributes closely associated to food safety matters within the same choice-set. The
attributes in this study are solely related to animal friendly practices. Nevertheless,
the attribute ‘‘No castration’’ may be considered as a ‘‘food safety’’ oriented
attribute due to the increased risk of boar taint resulting from not castrating piglets.
The attribute ‘‘Mobile slaughter’’ may be ‘‘environmentally’’ oriented as the use of
mobile abattoirs implies that the pigs do not have to be transported over long
distances. However, these attributes also have an important animal welfare
dimension. Therefore heterogeneity of preferences and discrete supports for
attributes may be the reason why choice experiment studies concerning attributes
in pig production may give contrasting and sometimes contradictory results. In
Table 6, the results from this study are compared with results from Carlsson et al.
(2004) and Lagerkvist et al. (2006). In Carlsson et al. (2004), the mean WTP for

TABLE 6. Comparison Between Choice Experiment Studies of Animal Welfare in Pig

Attribute         This study    Carlsson et al. (2004)   Lagerkvist et al. (2006)

Mobile abattoir   119%          110% for beef 18%        n.a.
                                  for pork
No castration     15%          n.a.                     21%
Outdoor pigs      132%          167%                     34% (interacted with shopping

                                                           Agribusiness    DOI 10.1002/agr

‘‘Mobile abattoir’’ is about 10% higher than the base scenario option for beef and
8% higher for pork. In Lagerkvist et al. (2006), ‘‘No castration’’ was intensely
disliked; 21% lower than the base scenario price. Carlsson et al. (2004) found the
attribute ‘‘Outdoors summertime’’ to be particularly important, with a WTP about
67% above the base scenario price. Lagerkvist et al. (2006) found varying support
for the outdoor attribute: a negative WTP of 34% from the base scenario when
shopping experience was interacting with the attribute and an ordinary WTP of 58%
above the base scenario.

6.2. Cost of Implementation
Is it economically feasible to incorporate animal friendly practices into production?
Some of the attributes (e.g., ‘‘Mobile slaughter’’ and ‘‘No castration’’) are
only practiced on small scale basis, which complicates the assessment of differences
in earnings. However, calculations may still be performed to provide ‘‘rough’’
benefits-costs estimates for each attribute. These results are presented in Table 7.
As the fillet constitutes 2% of the carcass weight, the fillet price is not representative
of the average price of the carcass. Instead the average retail price of
pork is approximated. For that purpose, it is assumed that the average retail
price and is directly related to the wholesale price and production cost. The
wholesale price in 2002 averaged 0.54 US$/lb (Swedish Meats, 2006). Assuming

TABLE 7.       Cost-Benefit Approximations of Animal Friendly Practices

                                                           Cost change
                                                           for marketing
                              WTP for      Cost change at VAT and
                     WTP      attribute    wholesale level inflationcd             Revenue
Attribute            (%)      (US$/lb)a    (US$/lb)b       (US$l/lb)              (US$/lb)e     Ratiof

Mobile abattoir:
South of Sweden 19            0.172        0.018               0.032              0.141         0.814
North of Sweden                            0.064              0.113             0.286         0.343
No castration    15          0.136       0.031               0.050              0.186        0.632
Air partition:
Limit 200 pigs       15       0.136        0.001               0.002              1.134         0.985
Limit 100 pigs       20       0.181                                               1.179         0.989
No mixing            13       0.118        0.002               0.003              0.115         0.975
  Willingness to Pay (WTP) for attribute(US$/lb): wtp*0,91.
  Cost change: 0,54*change in cost. From Andersson et al. (1997), Botermans (2003), and Helgesson
  The inflation rate is derived from consumer price index (CPI).
  Adjusted cost change: cost change*(adjustment level depends on inflation, the VAT, and retail margin).
  Revenue: wtp for attribute-adjusted cost change.
 The ratio is calculated as (1-adjusted change in cost*[wtp for attribute]1) and can be interpreted as the
maximum allowed size of a hypothetical bias.

Agribusiness     DOI 10.1002/agr

a margin of 50%9 and value added adjustment,10 the average retail price of pork
would be 0.91 US$/lb.
   In a Swedish study that evaluated the costs of a mobile slaughter system for pigs
(Helgesson, 2000), large cost differences were found between different regions in the
country. The north, with smaller-sized abattoirs and considerable transport distances,
would gain 0.064 US$/lb by adopting mobile systems, whereas the south would incur
an increase in the slaughter cost of 0.018 US$/lb pork. Adjusting the associated costs
and for retail margin, inflation11 and the value added tax, the revenues are 0.14 and
0.29 US$/lb pork, respectively. Hence, it can be concluded that the increase in
slaughter costs in both southern and northern Sweden may be compensated by a high
WTP for mobile abattoirs. The attribute ‘‘No castration’’ is regarded negatively by
the respondents as indicated by a WTP 15% below the base scenario price. In Britain
and Denmark, un-castrated male piglets are produced as fattening pigs, which means
that some 5–9% of the herd has to be eliminated by slaughter due to boar taint. Using
the model developed by Botermans (2003),12 the cost of not castrating male piglets
would increase overall cost by 6% or 0.031 US$/lb. An important observation is that
this cost measure accounts for rejection rate of 9% due to boar taint. Alternative uses
for this meat are not included in this approximation. The adjusted values for ‘‘No
castration’’ yield a return of 0.19 US$/lb. It may therefore be justified to retain the
practice of castrating piglets. Botermans’ model13 yields a 0.3% cost increase
(0.0014 US$/lb) with air partitioning. Thus, taking into account the attributes ‘‘Stock
limit: 200 pigs’’ and ‘‘Stock limit: 100 pigs’’ (assuming the same increase in costs when
producing the pig barn) yield adjusted profit of 1.14 and 1.18 US$/lb pork. According
to Andersson and Jonasson (1997), the increasing costs as due to reduced capacity
utilization of farm buildings may be compensated by a reduced need for antibiotics in
feed. If this is the case, the benefits of applying air partition would be further
improved. Andersson, Campos, and Jonasson (2000) estimated the cost of not
separating litters to 0.40 US$/pig. For a live slaughter weight of 50 lb, a cost increase
of 0.002 US$/lb is expected. The adjusted value of WTP is 0.003 and yield an
approximated economic benefit of 0.115 US$/lb, which motivates the practice of not
mixing litters. In order to assess the maximum allowed size of a hypothetical bias, the
ratio between the WTP value and cost approximate is calculated. These ratios are
overall high. This indicates that the practices can be economically motivated, even if a
large hypothetical bias exists.

6.3. Results in an International Perspective
As the results of this study are obtained from data from Swedish respondents, it is of
interest to learn whether these results may be generalized to consumers in other

   Normally, the retail margin for fresh meat is assessed to be about 30–50% (Supermarket, 2005).
      In 2002 the value added tax for food commodities reached 12%.
   The inflation rate is derived from the consumer price index (CPI).
   For this particular experiment the elimination of castration is assumed to promote growth rate (1
5%), improve feed conversion (16%), increase classification rate (11.5%), reduce the time/sow and year
(10%) while the rejection rate is assumed to be high for this scenario (19%) (Botermans, 2003).
   The number of slaughter pig places and sow places increases by 10%, growth rate increases 10%, feed
conversion improves by 2%, contagion rate decreases by 10%, mortality rate decreases by 2% and use of
medicines decreases by 5% (Botermans, 2003).

                                                                      Agribusiness   DOI 10.1002/agr

European markets? Ngapo et al. (2003) compiled results from focus groups in four
different EU countries: France, Britain, Sweden, and Denmark. The results of this
study indicate that preferences are heterogeneous both within and between countries.
National differences in perception of meat quality were found. While the Swedish
group was concerned about ‘‘Slaughtered on farm,’’ ‘‘Raised nearby’’ and ‘‘From a
small abattoir,’’ the French respondents had clear preferences for an appetizing
visual appearance, the English respondents for disease-free pigs, and the Danish for
cooking qualities. Thus, preferences may reveal national segmentation patterns.
Hence, the relative importance of WTP vis-à-vis mobile abattoirs in this study may
be considered typical of Swedish consumers.
   The level of WTP observed in this study raises the question: Why is the share of
organic products in the market so small? Organic, KRAV-labeled meat constitutes
only about 1% of the meat market in Sweden. In the focus group discussions,
organic production was associated with environmentally friendly production. One
problem may be that the KRAV label is not considered a true ‘‘animal welfare’’
designation. The KRAV label includes additional environmental regulations that
presumably affect the pricing. This is supported by Andersen (2003), who found that
consumers perceive organic eggs to be more ‘‘environmentally friendly.’’ However,
‘‘conventional eggs’’ ‘‘free-range eggs,’’ and ‘‘barn eggs’’ are believed to be produced
with a greater degree of animal well-being. Thus, there exist information
asymmetries regarding food safety products, products produced with animal friendly
practices and environmentally safe products. It is therefore not possible to conclude
with certainty that a small market share of KRAV labeled products implies a weak
demand for, or disinterest, in animal friendly practices.


Animal welfare attributes of pig production associated with ‘‘the Swedish model’’ or
practiced on an experimental basis have been analyzed. The attributes pertain to
transportation, housing systems, feed, castration, stock size, mixing of pigs, and
improved environment. As far as we know, this kind of large-scale choice
experiment14 for animal welfare attributes in primary pig production had not been
undertaken before. However, some important issues should be taken into
consideration when interpreting the results.
  In order to be able to calculate WTP from the estimated parameters, exogeneity of
price in the utility function has to be assumed. If the respondents have well defined
preferences an inclusion of a price attribute should not affect the valuation of animal
welfare attributes. However, it should be taken into consideration that inclusion of
prices in choice experiment can lead to different preference ranking. Inclusion of a
price attribute may decrease the estimated values of marginal rate of substitution
between attributes (Tversky & Thaler, 1990). Another important issue is that the
WTP values for different attributes are not additive; one cannot obtain a total value
of WTP by summing all values (see Nilsson, Foster, & Lusk, 2006 for a derivation).
The degree of WTP depends on the experimental design. Therefore, the WTP
obtained should be interpreted with caution. One may still suspect that the WTP

    Den Ouden (1996) performed a conjoint analysis of 12 welfare attributes in pig production chain in a
small sample of respondents.

Agribusiness     DOI 10.1002/agr

values are overstated (see for example Bennett, 1997; Cummings & Taylor, 1999;
Harrison & Rutstrom, 1995; Frykblom, 1997; or Johannesson, 1997). Some authors
have tackled the problem of hypothetical bias by calibrating factors. The size of
commodity-specific calibration factors has been estimated by Alfnes (2003), Bennett
(1997), Fox, Shogren, Hayes, and Kliebenstein (1998) and List and Shogren (1998).
The size of the factor is assumed to be commodity specific (List & Shogren). A wide
range of calibration factors is suggested for meat products.15 In order to ensure that
the absolute values of WTP are realistic, one has to take into account the
hypothetical bias. In order to provide viable policy recommendations from the
estimated costs and WTP for animal friendly practices presented in Table 7, one
must also deal with possibly inflated values of WTP.
   The RPL model approach is an appealing tool to account for heterogeneous
preferences. The model allows a range of attitudes towards animal welfare attributes
and identifies a mean and a spread of values around the mean. However, a well-
known shortcoming of the model is that it requires strong distributional
assumptions. Some heterogeneities can be treated within the RPL model formula-
tion; but, if one suspects that the evaluation of attributes is affected by strong
underlying preferences, i.e., similar preferences within groups and considerable
intergroup heterogeneity, the estimation might be improved by using a latent model
formulation (see for example Ben-Akiva et al., 2002). To identify the right
segmentation criteria is an important topic for further research. As advocated by
Greene and Hensher (2003), it should be worthwhile to compare and contrast the
RPL model and the latent class model. The use of a latent class formulation may
elicit further information regarding market asymmetries, the importance of
socioeconomic characteristics, and the segmentation of preferences, which could
be useful for policy and marketing analysis.


  This study has been conducted as a part of the research theme project ’Animal
Welfare for Quality in Food Production’ at the Swedish University of Agricultural
Sciences. The research program funded the study. This article has benefited from
comments by professor Hans Andersson, professor Yves Surry, associate professor
Lotta Rydhmer and two anonymous reviewers. The responsibility of any remaining
errors lies with the author.

Alfnes, F. (2003). Willingness to pay for quality in experimental auction markets and stated
   choice surveys. Doctoral dissertation, Agricultural University of Norway, Ås.
Andersen, L. (2003). Consumer evaluation of environmental and animal welfare labelling:
   An econometric analysis of panel data using mixed multinomial logit model (Working
   Paper 6). AKF, Denmark.
Anderson, J., & Frykblom, P. (1999). Exploring nonmarket values for the social impact of
   farm animal welfare (Working Paper 99/2). Department of Economics, SLU, Sweden.

    List and Shogren (1998) proposed a factor of 0.6 in an irradiated/radiated meat survey. Alfnes (2003)
reports a calibration factor of 0.1 in a hypothetical survey of hormone treated US-beef. Bennett (1997)
finds that the factor is nearly 0.1 for a legislation of battery cages.

                                                                      Agribusiness     DOI 10.1002/agr

Andersson, H., & Jonasson, L. (1997). Den svenska modellen–hävstång eller ok för svensk
  grisproduktion? [The Swedish model-lever or yoke for Swedish pig production.] Optimering
  av den svenska modellen–Delprojekt I. Ett samarbetsprojekt mellan Lantmännen,
  Slakteriförbundets FoU-grupp Svin, Sveriges svinproducenter och Svenska djurhälsovården.
Andersson, H., Campos, M., & Jonasson, L. (2000). Kostnadspåverkande faktorer i Svensk
  grisköttproduktion. [Cost affecting factors of Swedish pig production.] Optimering av den
  Svenska modellen–Delprojekt II. Ett samarbetsprojekt mellan Lantmännen, Slakteriför-
  bundets FoU-grupp Svin, Sveriges svinproducenter och Svenska djurhälsovården.
American Veterinary Medical Association. (2006). American Veterinary Policy Statements
  AVMA Animal Welfare principles. www.AVMA.org.
Ben Akiva, M., & Lerman, S. (1985). Discrete choice analysis. Cambridge, MA: The MIT
Ben-Akiva, M., Walker, J., Bernardino, A.T., Gopinath, D.A., Morikawa, T., &
  Polydoropoulou, A. (2002). Integration of choice and latent variable models. In:
  Mahmassani H. (Ed.), Perpetual motion: Travel behaviour research opportunities and
  application challenges (Vol. 21, pp. 431–470). Oxford: Elsevier Science.
Bennett, R. (1997). Farm animal and welfare and food policy. Food Policy, 22, 281–288.
Bennett, R., & Larson D. (1996). Contingent valuation of the perceived benefits of farm
  animal welfare legislation: An exploratory survey 1. Journal of Agricultural Economics, 47,
Bhat, C.R. (2000). Flexible model structures for discrete choice analysis. In: Hensher D.A. &
  Button K.J. (Eds.), Handbook of transport modelling (Vol. 1, pp. 71–90). Oxford, England:
  Pergamon Press.
Botermans, J. (2003). A model for economical evaluation of different production systems and
  animal welfare measures for pigs (Report 143). JBT, Alnarp, Sweden.
Carlsson, F., Frykblom, P., & Lagerkvist, C. (2004). Consumer willingness to pay for
  animal welfare–transportation of farm animals to slaughter versus the use of mobile
  abattoirs (Working Paper 149). Department of Economics, Gothenburg University,
Cummings, R.G., Harrison, G.W., & Rutstrom, E.E. (1995). Homegrown values and
  hypothetical surveys - Is the dichotomous choice approach incentive-compatible? American
  Economic Review, 85, 260–266.
Cummings, R.G., & Taylor, L.O. (1999). Unbiased value estimates for environmental goods:
  A cheap talk design for the contingent valuation method. American Economic Review, 89,
Den Ouden, M. (1996). Economic modelling of pork production-marketing chains.
  Unpublished doctoral dissertation. Wageningen Agricultural University, The Netherlands.
DFS 2006:4. Swedish Code of Statuses.
Drake, L., & Holm, H. (1989). Konsumenters attityder till alternativt producerat kött
  (Report 21). [Consumer attitudes for organic pork.] Department of Economics, SLU,
Enneking, U. (2004). Willingness-to-pay for safety improvements in the German meat
  sector: The case of the Q&S label. European Review of Agricultural Economics, 31,
European Commission (2006). EU Action plan. IP/06/64.
Fox, J.A., Shogren, J.F., Hayes, D.J., & Kliebenstein, J.B. (1998). CVM-X: Calibrating
  contingent values with experimental auction markets. Journal of Agricultural Economics,
  80, 455–465.
Freeman, A. (1993). The measurement and environmental and resource values: Theory and
  methods. Washington D.C.: Resources for the Future.
Frykblom, P. (1997). Hypothetical question modes and real willingness to pay. Journal of
  Environmental Economics and Management, 34, 275–287.
Greene, W.H. (2000). Econometric analysis. (4th ed.). New Jersey: Prentice-Hall.
Greene, W.H. (2002). Nlogit version 3.0 reference guide. New York: Econometric Software.
Greene, W.H. & Hensher, D.A. (2003). A latent class model for discrete choice analysis:
  Contrast with the mixed logit. Transportation Research Part B, 37, 681–698.

Agribusiness   DOI 10.1002/agr

Hanemann, W., & Kanninen, B. (1998). The statistical analysis of discrete response contingent
   valuation data (Working paper 798). Californian Agricultural Experiment Station:
   Giannini Foundation of Agricultural Economics.
Helgesson, A. (2000). Avoiding transports of live animals–evaluation of a mobile slaughter
   system for pigs (Report 242). Department of Economics. Swedish University of
   Agricultural Sciences, Uppsala, Sweden.
Hensher, D.A. (2001). Measurement of the valuation of travel time savings. Journal of
   Transport Economics and Policy, 35, 71–98.
Hensher, D.A., & Greene, W.H. (2003). The mixed logit model: The state of practice and
   warnings for the unwary. Transportation, 30, 133–176.
Henson, S., & Traill, W. (2000). Measuring perceived performance of the food system and
   consumer food-related welfare. Journal of Agricultural Economics, 51, 388–404.
Johannesson, M. (1997). Some further experimental results on hypothetical versus real
   willingness to pay. Applied Economic Letters, 4, 535–536.
Just, R., Hueth, D., & Schmitz, A. (2004). The welfare economics of public policy: A practical
   approach to project and policy evaluation. Cheltenham, U.K.: Edwar Elgar.
Kuhfeld, W. (2001). Multinomial logit, discrete choice modelling. An introduction to
   designing choice experiments, and collecting, processing and analysing choice data with
   SAS. SAS Institute TS-643.
Lagerkvist, C, Carlsson, F., & Viske, D. (2006). Swedish consumer preferences for animal
   welfare and biotech: A choice experiment. AgBioForum, 9, 51–58.
Larue, B., West, G., Gendron, C., & Lambert, R. (2004). Consumer response to functional
   foods produced by conventional, organic or genetic manipulation. Agribusiness, 20,
Lindgren, A., & Forslund, K. (1990). Min ko vill ha roligt: inhopp i djurskyddsdebatten - hur
   och varför det blev som det blev. [My cow wants to have fun.] Rabén & Sjögren,
   Stockholm, Sweden.
List, J.A., & Shogren, J.F. (1998). Calibration of the difference between actual and
   hypothetical valuations in field experiments. Journal of Economic Behaviour &
   Organization, 37, 193–205.
Louviere, J., Hensher, D., & Swait, J. (2000). Stated choice methods: Analysis and application.
   Cambridge, U.K.: Cambridge University Press.
Lundeheim, N., & Holmgren, N. (1994). Utveckling av uppfödningsformer och hälsa hos
   slaktsvin. [Improvements in breeding and animal welfare in pig production.] Svensk
   Veterinärtidning, 54, 469–474.
McFadden, D. (1974). Conditional logit analysis of qualitative choice behaviour. In:
   Zarembka P. (Ed.), Frontiers in econometrics (pp. 105–142). New York: Academic Press.
McFadden, D., & Train, K. (2000). Mixed mnl models for discrete response. Journal of
   Applied Econometrics, 15, 447–470.
McInerney, J. (1991). A socioeconomic perspective on animal-welfare. 1. Outlook on
   Agriculture, 20, 51–56.
Ngapo, T.M., Dransfield, E., Martin, J.F., Magnusson, M., Bredahl, L., & Nute, G.R. (2003).
   Consumers perceptions: Pork and pig productions. Insights from France, England Sweden
   and Denmark. Meat Science, 66, 125–134.
Nilsson, T., Foster, K., & Lusk, J.L. (2006). Marketing opportunities for certified pork chops.
   Canadian Journal of Agricultural Economics, 54, 567–583.
Paarlberg, P., Boehlje, M., Foster, K., Doering, O., & Wallace, T. (1999). Structural change
   and Market performance in agriculture: Critical issues and concerns in the pork industry
   (Staff Paper 99-14). Department of Economics, Purdue University, West Lafayette, USA.
Revelt, D., & Train, K. (1998). Mixed logit with repeated choices: Households’ choices of
   appliance efficiency level. Review of Economics and Statistics, 80, 1–11.
Rolfe, J. (1999). Ethical rules and the demand for free range eggs. Economic Analysis &
   Policy, 29, 187–206.
Ruud, P. (1996). Simulation of the multinomial probit model: An analysis of covariance
   matrix estimation (Working Paper). Department of Economics. University of California,
Statistics Sweden (2003). Yearbook of Agricultural Statistics. Swedish Board of Agriculture.

                                                               Agribusiness    DOI 10.1002/agr

Stott, A.W., Milne, C.E., Goddard, P.J., & Waterhouse, A. (2005). Projected effect of
   alternative management strategies on profit and animal welfare in extensive sheep
   production systems in Great Britain. Livestock Production Science, 97, 161–171.
Supermarket (2005). www.market.se.
Swedish Government Offices. (1989). The NFA code of statutes–Animal welfare act (No.
   SJVFS 1989:20). Stockholm, Sweden: Author.
Swedish Government Offices. (1993). The NFA code of statutes–Animal welfare act (No.
   SJVFS 1993:129). Stockholm, Sweden: Author.
Swedish Government Offices. (1988). The Swedish code of statutes–Animal welfare act (No.
   SFS 1988:534, 539). Stockholm, Sweden: Author.
Swedish Meats (2006). www.swedishmeats.se.
Train, K. (2000). Halton sequences for mixed logit (Working Paper E00-278). Economics
   Department, University of California, Berkeley, USA.
Train, K. (2003). Discrete choice models with simulation. Cambridge, U.K.: Cambridge
   University Press.
Tversky, A., & Thaler, R.H. (1990). Preference reversals. Journal of Economic Perspectives, 5,
Verbeke, W. (2000). Influences on the consumer decision making process towards fresh meats.
   Insights from Belgium and implications. British Food journal, 102, 522–538.
Verbeke, W., & Ward, R.W. (2001). A fresh meat almost ideal demand system incorporating
   negative TV press and advertising impact. Agricultural Economics, 25, 359–374.

Carolina Liljenstolpe holds a Master of Agricultural Economics from the Swedish University of
Agricultural Sciences, and will defend her PhD thesis in 2008. Her research interests include
choice experiments and sectoral modeling.

Agribusiness   DOI 10.1002/agr
You can also read
Next slide ... Cancel