Abstract
This paper explores the demand and willingness to pay (WTP) for value-added services to chicken. Since the demand for such services are likely to be highly segmented and often applies only to a market niche, models based on assumptions of homogeneity among consumers are likely to be inappropriate. For this reason, this paper combines discrete and continuous mixing distributions to concurrently identify the size of the niche market and the heterogeneity among consumers within the market niche. Failing to account for the niche market nature of value-added services is shown to have implications for predictions of WTP, demand and total revenue.
1. Introduction
Food safety legislation within the European Union has focused on ensuring safety and traceability of food from ‘farm-to-fork’ and has prompted the use of various labelling policies as a method of ensuring integrity in the food chain and enhanced consumer confidence. In addition to the obvious health benefits of ensuring food safety, there are numerous economic benefits that can be achieved. The most obvious is the aversion of economic losses associated with food scares, which can cause substantial volatility in the agricultural sector (Serra, 2011). A further important economic benefit is the value-added services to food products through additional features such as country of origin labelling and safety assurance labels (cf. Hu et al., 2011a, b; Martínez Michel, Anders and Wismer, 2011, for recent stated preference studies).
In this study, we employ the choice experiment methodology to explore preferences for services that improve the safety and quality of food products over that which exists. According to Louviere, Hensher and Swait (2000), choice experiments are advantageous because they closely simulate real-world purchasing decisions where a respondent has to select a product from a set of options. This method is aimed at determining how individuals trade-off among the value-added services and their levels and is intended to uncover willingness to pay (WTP) estimates for the services and their relative values.
This study adds to a growing literature that has examined preferences for features that are perceived to affect food quality and safety (e.g. Lusk, Roosen and Fox, 2003; Alfnes et al., 2006; Lusk, Nilsson and Foster, 2007; Ubilava and Foster, 2009; Olynk, Tonsor and Wolf, 2010; Ortega et al., 2011). The focus of these studies range from examining preferences for reduction in particular factors such as avoidance of pathogens in food or avoidance of pesticides or GM in food production (e.g. Carlsson, Frykblom and Lagerkvist, 2007a), or for a range of features that are perceived to enhance food safety and quality, such as traceability (Loureiro and Umberger, 2007), region of origin or origin labelling (Loureiro and Umberger, 2003; Enneking, 2004; Scarpa, Philippidis and Spalatro, 2005b; Loureiro, Gracia and Nayga, 2006; Bougherara and Combris, 2009; Ortega et al., 2011) and animal welfare (Lagerkvist, Carlsson and Viske, 2006; Carlsson, Frykblom and Lagerkvist, 2007b; Lagerkvist and Hess, 2011).
While food safety and quality features are shown to add value to food products, the demand is often segmented among different types of consumers (e.g. Hu et al., 2004; Scarpa and Del Giudice, 2004; Rigby and Burton, 2005; Mørkbak and Nordström, 2009) and may, in fact, represent only a niche market (e.g. Latacz-Lohmann and Foster, 1997; Loureiro and Hine, 2002; James, Rickard and Rossman, 2009). Indeed, as shown in Rigby and Burton (2006) and Balcombe, Fraser and Di Falco (2010), a proportion of consumers are likely to be ‘disinterested’ or ‘indifferent’, meaning that only a subset of consumers are willing to pay a price premium for the value-added services. With this recognition, models based on the assumptions of homogeneity are likely to be inappropriate and failing to realise the possibility of market niches could lead to biased results. For instance, consider the scenario where the market niche for a value-added service represents 40 per cent of consumers and that, on average, they are willing to pay a price premium of GBP 2.00 for the service, and that the remaining 60 per cent of consumers are not willing to pay anything extra over the standard product. In this case, erroneously deriving a single value would lead to a WTP estimate of GBP 0.80 and would, furthermore, give the (misleading) impression that all consumers are willing to pay this amount. Importantly, pricing decisions made on the basis of this model are likely to be suboptimal (i.e. setting the price premium at GBP 0.80 would generate an additional GBP 32.00 per 100 consumers, in contrast to GBP 80.00 per 100 consumers that would be achieved if the price premium was set at GBP 2.00).
This paper aims at exploring the demand for value-added services to chicken. But unlike previous studies, we are mindful of the fact that the markets for such services are likely to be tightly niched. For this reason, we implement a mixed logit model in WTP space that combines discrete and continuous mixing distributions, which is aimed at simultaneously identifying the size of the niche market and the heterogeneity in WTP within the market niche. We compare this specification against those which do not facilitate the market niche. To explore the role that failing to account for this has on welfare and demand predictions, we extend our analysis to explore the market shares at different price premiums and also to identify the optimal pricing of value-added services to ensure maximum revenue among food producers.
To test our methodology, we use a stated choice experiment dataset collected from a representative sample of individuals residing in Great Britain to establish the demand and WTP for a number of value-added services to chicken. These services are associated with scientific improvements to better ensure the integrity and safety of food products, as well as their region of origin. Results from our analysis reveal that, while the value-added services are preferred by respondents, they apply to only niche markets (possibly as low as one-fifth of the market) and that within these niches there is considerable heterogeneity in the WTP estimates. Accounting for the market niches is also found to be highly relevant for estimation outcomes as goodness-of-fit is significantly improved. Important implications for the WTP, demand and total revenue predictions of not allowing for the possibility of niche markets are also discovered. Crucially, naïve specifications are found to poorly predict the upper range in the WTP distributions and lead to erroneous market share predictions, overstated total revenue estimates and potentially lower profit margins.
The rest of the paper is organised as follows: Section 2 describes our econometric approach and Section 3 presents our empirical context. Section 4 reports the model estimation results and implications for WTP as well as the findings arising from the market share and an optimal pricing analysis. Section 5 concludes.
2. Method
We derive marginal WTP estimates using the framework of discrete choice models based on random utility maximisation (RUM). We start this section by outlining the multinomial logit (MNL) model, mainly to introduce the required notation and to provide a base model for which comparisons can be made. We then present a random parameters logit (RPL) model with an error component (EC) to address the heterogeneous nature of respondent's WTP for value-added services to chicken as well as the potential substitution effects between some of the alternatives. In an attempt to account for the fact that only a subset of respondents may be willing to pay for value-added food products, our final model uses a combined discrete mixtures (DMs) and RPL/EC model specification to simultaneously uncover the size of the niche market and the heterogeneity in WTP within these market segments as well as the substitution patterns. We finish this section by detailing our means of model estimation.
2.1. WTP for value-added food products
2.2. Accommodating heterogeneous WTP for value-added food products
While the assumption of homogeneity in the values that respondents are willing to pay for value-added services to food may hold in some cases, for a variety reasons one may postulate the hypothesis that it is more likely that the values will be heterogeneous across respondents (e.g. see Alfnes, 2004; Scarpa et al., 2005b; Rigby and Burton, 2006; Rigby, Balcombe and Burton, 2009; Ortega et al., 2011). Consequently, we are interested in capturing the heterogeneity in respondents' marginal WTP for value-added services to food products. For this reason, we treat α and w as random terms entering the utility function. As discussed by Scarpa, Ferrini and Willis (2005a), there may also be substitution patterns between the non-SQ alternatives. We accommodate this by introducing an error component, labelled η.
In this RPL/EC specification parameters of the continuous distributions (i.e. Ω) and the SQ constant are obtained. This generally leads to significant gains in model performance and, importantly, greater insights into choice behaviours and preferences for the choice experiment attributes.
2.3. Identifying niche markets for value-added food products
Given the widespread evidence suggesting that the demand for value-added products is highly segmented among different types of consumers (e.g. Hu et al., 2004; Scarpa and Del Giudice, 2004; Rigby and Burton, 2005) and that they are often niche markets (e.g. Latacz-Lohmann and Foster, 1997; Loureiro and Hine, 2002; James et al., 2009), it is likely that only a subset of respondents are willing to pay for the value-added services, while those remaining are not willing to pay anything extra (in which case there will be a spike at zero). We acknowledge, however, that this may not be the case in all empirical settings but should be evaluated on a case-by-case basis.
Despite the appeal of continuous representations, and previous work on ‘disinterest’ and ‘indifference’ by Rigby and Burton (2006) and Balcombe et al. (2010), unfortunately they do not adequately accommodate situations where there is a spike in the distribution. Moreover, if the distribution does contain a spike, the likely outcome of erroneously fitting a continuous distribution will be an under-estimation of the share who are not willing to pay anything extra and an over-estimation of probability between zero and the typical WTP value among the subset who are willing to pay for the value-added service. For this reason, we also consider the use of discrete representations of WTP.
We note the parallels of this modelling approach with the work on attribute non-attendance (e.g. Hensher, Rose, and Greene, 2005; Campbell, Hutchinson, and Scarpa, 2008), and more specifically to the studies that have used latent class specifications to investigate the issue (e.g. Scarpa et al., 2009; Campbell, Hensher, and Scarpa, 2011, 2012). While both result in zero (or negligible) marginal WTP estimations, they have a somewhat different meaning. In this paper, we prefer to think that respondents who are identified as having zero marginal WTP for a value-added service have no preference for it (or are indifferent to it), which in contrast to the view that it is because they ignored it when making their choice. While difficult to disentangle, the issue remains the same, namely that of uncovering the distribution of marginal WTP for the subset of respondents where the value-added service had a bearing on their choice.
2.4. Model estimation
The models above are estimated using Biogeme (Bierlaire, 2003). The MNL model (equation (2)) is estimated using maximum likelihood estimation, whereas, since the choice probabilities in the RPL/EC and DM-RPL/EC models (equations (4) and (5), respectively) cannot be calculated exactly (because the integrals do not have a closed form), they are estimated by simulating the log-likelihood with 500 quasi-random draws via Halton sampling. In the case of the DM-RPL/EC model, we are also mindful of the fact that models of this form are subject to local maxima. However, in an attempt to reduce the likelihood of reaching a local rather than a global maximum, we employ the CFSQP algorithm developed by Lawrence, Zhou and Tits (1997) and use a variety of random starting points. Specifically, we do this by estimating the model several times, but each time using a different vector of starting values, which are chosen randomly.
A key element with the specification of random taste variation is the assumption regarding the distribution of each of the random parameters (Hensher and Greene, 2003; Hess, Bierlaire and Polak, 2005; Rigby et al., 2009). Random parameters can take a number of predefined functional forms. While this affords the analyst with some control and flexibility, the random parameters are not observed and there is typically little a priori information about the shape of its distribution except possibly a sign constraint (Fosgerau and Hess, 2009). Consequently, the chosen distribution is essentially an arbitrary approximation (Hensher and Greene, 2003) requiring some possibly strong or unwarranted distributional assumptions about individual heterogeneity (Greene and Hensher, 2003). In this regard, specification testing and assessing the suitability of different distributional assumptions is warranted (see, Fosgerau and Bierlaire, 2007; Fosgerau, 2008 for an overview on such tests). Given the theoretical expectations of disutility for price and the widespread practice in WTP space models (e.g. Scarpa, Thiene and Train, 2008; Thiene and Scarpa, 2009; Balcombe et al. 2010), we specify α as having a log-normal distribution to ensure strictly negative values for the price coefficient as follows: , where υ is a standard normal deviate and μ and σ are the parameters to be estimated.2 After evaluating the results from various specifications and distributional assumptions, for the distributions of w we opt for a flexible distribution denoted by the following:
, where a represents the lower boundary, b provides the range and υ is again a standard normal deviate.3 We found that our data were most suitably characterised by this distribution given consideration to the plausible signs on the coefficients and with regard to the evaluation of the log-likelihood values using different distributions. The distribution is similar in nature to the normal and triangular distributions in the sense that it satisfies the need for a central tendency, but unlike the normal distribution it has the advantage of finite symmetric upper and lower limits, but eliminates the abrupt change at the mid-point and does not exhibit unrealistic linear probability slopes that are associated with the triangular distribution. We note that we do not specify the values a and b, but rather that separate values are estimated for each of the WTP distributions. We also place the restriction
to ensure that the distribution of WTP for the value-added services is always non-negative. For η, we specify it as a zero-mean error component as follows:
, where υ is once more a standard normal deviate.
3. Case study: demand for assured, safe and traceable food
This section starts with a general description of the empirical case study, including an overview of the value-added food attributes focused on in this study. Then we discuss the experimental design procedure used to generate the choice tasks followed with an outline of the survey implementation and data collection.
3.1. Stated choice experiment for value-added food products
In this paper, we report results from a case study exploring the WTP for value-added services to chicken meat, specifically, two uncooked chicken breasts. To identify the relevant food safety attributes and levels associated with this chicken product, the study design was informed by expert opinion from food scientists involved in developing methods to verify the safety and authenticity of food. Although these discussions helped establish the attributes of interest, we gathered information from food stores and undertook a series of focus group discussions with members of the general public and pilot surveying to further ensure that the attributes and levels were understandable and relevant to the general public. Following this, three main food safety attributes were identified: food testing standards, traceability standards and animal health/welfare standards. All three of these value-added services were defined as having two standards: an enhanced standard and a current standard. For food testing, the enhanced standard represented the use of additional testing to ensure safer food. For traceability, the enhanced standard consisted of the use of technology to verify the exact origins of the meat so that labelling fraud could not occur. For the animal health/welfare attribute, respondents were informed that the enhanced standard tested the animals for the presence of any drugs or diseases, whilst the current standard tested only for the presence of drugs. A region of origin attribute was also included to decipher preferences for chicken produced within the British Isles versus chicken products that came from outside this area. A final attribute was included to represent the price of the chicken product. The price attribute, which was reflective of supermarket prices, was depicted with six levels, ranging between GBP 2.00 and 4.50 in GBP 0.50 increments.
3.2. Experimental design
Having established the attributes and their levels, in an attempt to maximise sampling efficiency and account for the uncertainty with regard to the assumed parameter values, a Bayesian efficient experimental design was generated, based on the minimisation of the -error criterion (for a general overview of efficient experimental design literature, see e.g. Scarpa and Rose, 2008 and references cited therein). Our prior parameter estimates were informed on the basis of initial estimations produced from an MNL model on our pilot study of 400 observations (gathered from 50 respondents). We also used the feedback from the expert opinions, focus group discussions as well as evidence from the literature to help define the random priors and to establish the appropriate number of tasks that should be used in the final design.
The final design comprised of 16 choice tasks, which were blocked into two smaller designs, such that each respondent completed a panel of eight choice tasks. For each task, respondents were asked to choose between two experimentally designed alternatives and a ‘buy none’ (or SQ) option. When making their choices, respondents were asked to consider only the information presented in the choice task and to treat each task separately.
3.3. Survey implementation and data collection
The choice data were collected during September 2010 via an on-line survey. We recruited 622 respondents residing in Great Britain resulting in 4,976 observations for model estimation. The sample consisted of approximately equal numbers of male (49 per cent) and female (51 per cent) respondents, which is comparable with the national population statistics. In accordance with the regional breakdown, 86, 9 and 5 per cent of respondents resided in England, Scotland and Wales, respectively. Also in line with the breakdown of the adult population of Great Britain, the majority of respondents were younger than 45 years (69 per cent) and worked, either on a full-time or part-time basis (63 per cent). Of the 51 per cent of respondents who disclosed their income, the average annual gross income was almost GBP 28,000, which is also comparable with national statistics.
4. Results
We begin this section with the results obtained from the discrete choice models outlined in Section 2. Following this, we report the findings from a sensitivity analysis exploring the demand and total revenue generated associated with the value-added services at different price premiums.
4.1. Estimation results
Estimation results are presented in Table 1. As a point of reference, our analysis starts with the MNL model, with a marginal utility parameter for the price attribute (denoted by α); marginal WTP parameters for the value-added services (represented with ws);4 and, a constant for the ‘buy none’ option (indicated by C), whose coefficient can be interpreted as the marginal (dis-)utility for having neither types of chicken breasts. Analysing the results obtained from this model, we note, as anticipated, that the price coefficient and ‘buy none’ constant are both significantly estimated as having negative signs – implying that, ceteris paribus, respondents (i) prefer less-expensive chicken breasts and (ii) dislike the situation of not having any chicken breasts.
Estimation results
| MNL | RPL/ECa | DM-RPL/ECab | |||||
|---|---|---|---|---|---|---|---|
| Est. | Est. | Est. | |||||
| Price | α | −0.668 | 24.70 | – | – | – | – |
| μ | – | – | −0.163 | 3.17 | 0.025 | 0.49 | |
| σ | – | – | 0.605 | 13.10 | 0.593 | 9.96 | |
| Food testing | w | 1.146 | 19.81 | – | – | – | – |
| m | – | – | 1.365 | 21.20 | 2.041 | 17.93 | |
| s | – | – | 1.365 | 21.20 | 2.041 | 17.93 | |
| π1 | – | – | – | – | 0.572 | 11.15 | |
| Traceability | w | 0.666 | 11.97 | – | – | – | – |
| m | – | – | 0.662 | 12.84 | 0.626 | 12.85 | |
| s | – | – | 0.652 | 3.04 | 0.626 | 12.85 | |
| π1 | – | – | – | – | 1.000 | <0.01 | |
| Animal health/welfare | w | 1.069 | 17.74 | – | – | – | – |
| m | – | – | 1.227 | 18.06 | 1.904 | 15.51 | |
| s | – | – | 1.227 | 18.06 | 1.904 | 15.51 | |
| π1 | – | – | – | – | 0.567 | 9.60 | |
| Region of origin | w | 0.282 | 4.06 | – | – | – | – |
| m | – | – | 0.294 | 4.76 | 2.700 | 5.68 | |
| s | – | – | 0.294 | 4.76 | 2.700 | 5.68 | |
| π1 | – | – | – | – | 0.151 | 22.39 | |
| ‘Buy none’ constant | C | −2.317 | 24.83 | −5.043 | 16.53 | −5.838 | 14.92 |
| Error component | σ | – | – | 3.287 | 13.58 | 3.887 | 11.75 |
| LL | −4,168.53 | −3,601.29 | −3,494.37 | ||||
| Kc | 6 | 12 | 16 | ||||
| 0.237 | 0.339 | 0.339 | |||||
| AIC | 8,349.06 | 7,226.59 | 7,020.74 | ||||
| BIC | 8,388.14 | 7,304.74 | 7,124.94 | ||||
| MNL | RPL/ECa | DM-RPL/ECab | |||||
|---|---|---|---|---|---|---|---|
| Est. | Est. | Est. | |||||
| Price | α | −0.668 | 24.70 | – | – | – | – |
| μ | – | – | −0.163 | 3.17 | 0.025 | 0.49 | |
| σ | – | – | 0.605 | 13.10 | 0.593 | 9.96 | |
| Food testing | w | 1.146 | 19.81 | – | – | – | – |
| m | – | – | 1.365 | 21.20 | 2.041 | 17.93 | |
| s | – | – | 1.365 | 21.20 | 2.041 | 17.93 | |
| π1 | – | – | – | – | 0.572 | 11.15 | |
| Traceability | w | 0.666 | 11.97 | – | – | – | – |
| m | – | – | 0.662 | 12.84 | 0.626 | 12.85 | |
| s | – | – | 0.652 | 3.04 | 0.626 | 12.85 | |
| π1 | – | – | – | – | 1.000 | <0.01 | |
| Animal health/welfare | w | 1.069 | 17.74 | – | – | – | – |
| m | – | – | 1.227 | 18.06 | 1.904 | 15.51 | |
| s | – | – | 1.227 | 18.06 | 1.904 | 15.51 | |
| π1 | – | – | – | – | 0.567 | 9.60 | |
| Region of origin | w | 0.282 | 4.06 | – | – | – | – |
| m | – | – | 0.294 | 4.76 | 2.700 | 5.68 | |
| s | – | – | 0.294 | 4.76 | 2.700 | 5.68 | |
| π1 | – | – | – | – | 0.151 | 22.39 | |
| ‘Buy none’ constant | C | −2.317 | 24.83 | −5.043 | 16.53 | −5.838 | 14.92 |
| Error component | σ | – | – | 3.287 | 13.58 | 3.887 | 11.75 |
| LL | −4,168.53 | −3,601.29 | −3,494.37 | ||||
| Kc | 6 | 12 | 16 | ||||
| 0.237 | 0.339 | 0.339 | |||||
| AIC | 8,349.06 | 7,226.59 | 7,020.74 | ||||
| BIC | 8,388.14 | 7,304.74 | 7,124.94 | ||||
aTo facilitate comparison, we report the mid-point (m) and spread (s) estimates of the marginal WTP distributions, which can be derived using and s = b/2, respectively.
bSince , for the sake of brevity we report only
. We further note that the associated t-ratios for
test
cTotal number of estimated parameters.
dThe is an adjustment of the
statistic, penalising for the number of parameters K. It is defined by:
, where
and LL(0) are the log-likelihoods for the estimated and the model in which all parameters are set to zero, respectively.
Estimation results
| MNL | RPL/ECa | DM-RPL/ECab | |||||
|---|---|---|---|---|---|---|---|
| Est. | Est. | Est. | |||||
| Price | α | −0.668 | 24.70 | – | – | – | – |
| μ | – | – | −0.163 | 3.17 | 0.025 | 0.49 | |
| σ | – | – | 0.605 | 13.10 | 0.593 | 9.96 | |
| Food testing | w | 1.146 | 19.81 | – | – | – | – |
| m | – | – | 1.365 | 21.20 | 2.041 | 17.93 | |
| s | – | – | 1.365 | 21.20 | 2.041 | 17.93 | |
| π1 | – | – | – | – | 0.572 | 11.15 | |
| Traceability | w | 0.666 | 11.97 | – | – | – | – |
| m | – | – | 0.662 | 12.84 | 0.626 | 12.85 | |
| s | – | – | 0.652 | 3.04 | 0.626 | 12.85 | |
| π1 | – | – | – | – | 1.000 | <0.01 | |
| Animal health/welfare | w | 1.069 | 17.74 | – | – | – | – |
| m | – | – | 1.227 | 18.06 | 1.904 | 15.51 | |
| s | – | – | 1.227 | 18.06 | 1.904 | 15.51 | |
| π1 | – | – | – | – | 0.567 | 9.60 | |
| Region of origin | w | 0.282 | 4.06 | – | – | – | – |
| m | – | – | 0.294 | 4.76 | 2.700 | 5.68 | |
| s | – | – | 0.294 | 4.76 | 2.700 | 5.68 | |
| π1 | – | – | – | – | 0.151 | 22.39 | |
| ‘Buy none’ constant | C | −2.317 | 24.83 | −5.043 | 16.53 | −5.838 | 14.92 |
| Error component | σ | – | – | 3.287 | 13.58 | 3.887 | 11.75 |
| LL | −4,168.53 | −3,601.29 | −3,494.37 | ||||
| Kc | 6 | 12 | 16 | ||||
| 0.237 | 0.339 | 0.339 | |||||
| AIC | 8,349.06 | 7,226.59 | 7,020.74 | ||||
| BIC | 8,388.14 | 7,304.74 | 7,124.94 | ||||
| MNL | RPL/ECa | DM-RPL/ECab | |||||
|---|---|---|---|---|---|---|---|
| Est. | Est. | Est. | |||||
| Price | α | −0.668 | 24.70 | – | – | – | – |
| μ | – | – | −0.163 | 3.17 | 0.025 | 0.49 | |
| σ | – | – | 0.605 | 13.10 | 0.593 | 9.96 | |
| Food testing | w | 1.146 | 19.81 | – | – | – | – |
| m | – | – | 1.365 | 21.20 | 2.041 | 17.93 | |
| s | – | – | 1.365 | 21.20 | 2.041 | 17.93 | |
| π1 | – | – | – | – | 0.572 | 11.15 | |
| Traceability | w | 0.666 | 11.97 | – | – | – | – |
| m | – | – | 0.662 | 12.84 | 0.626 | 12.85 | |
| s | – | – | 0.652 | 3.04 | 0.626 | 12.85 | |
| π1 | – | – | – | – | 1.000 | <0.01 | |
| Animal health/welfare | w | 1.069 | 17.74 | – | – | – | – |
| m | – | – | 1.227 | 18.06 | 1.904 | 15.51 | |
| s | – | – | 1.227 | 18.06 | 1.904 | 15.51 | |
| π1 | – | – | – | – | 0.567 | 9.60 | |
| Region of origin | w | 0.282 | 4.06 | – | – | – | – |
| m | – | – | 0.294 | 4.76 | 2.700 | 5.68 | |
| s | – | – | 0.294 | 4.76 | 2.700 | 5.68 | |
| π1 | – | – | – | – | 0.151 | 22.39 | |
| ‘Buy none’ constant | C | −2.317 | 24.83 | −5.043 | 16.53 | −5.838 | 14.92 |
| Error component | σ | – | – | 3.287 | 13.58 | 3.887 | 11.75 |
| LL | −4,168.53 | −3,601.29 | −3,494.37 | ||||
| Kc | 6 | 12 | 16 | ||||
| 0.237 | 0.339 | 0.339 | |||||
| AIC | 8,349.06 | 7,226.59 | 7,020.74 | ||||
| BIC | 8,388.14 | 7,304.74 | 7,124.94 | ||||
aTo facilitate comparison, we report the mid-point (m) and spread (s) estimates of the marginal WTP distributions, which can be derived using and s = b/2, respectively.
bSince , for the sake of brevity we report only
. We further note that the associated t-ratios for
test
cTotal number of estimated parameters.
dThe is an adjustment of the
statistic, penalising for the number of parameters K. It is defined by:
, where
and LL(0) are the log-likelihoods for the estimated and the model in which all parameters are set to zero, respectively.
In line with a priori expectations, we find that the marginal WTP coefficients for the three food safety attributes are all significantly estimated as having positive signs – implying that respondents are willing to pay a price premium for these value-added services. Comparing the magnitudes of these coefficients suggests that respondents place the highest value on chicken that has undergone enhanced food testing to ensure food safety (GBP 1.15) and that the chicken was produced under enhanced animal health/welfare standards (GBP 1.07), whereas the ability to fully trace the chicken is predicted as having a considerably lower price premium (GBP 0.67). In accordance with prior expectations, the marginal WTP coefficient for the locally produced value-added service is also found to be positive, and significant – revealing that respondents are more likely to purchase chicken breasts that are produced in the British Isles, compared with chicken produced elsewhere. Nevertheless, we find that the price premium respondents would be willing to pay for locally produced chicken breasts (GBP 0.28) is considerably lower than all of the food safety value-added services.
Moving to the RPL/EC model, which is aimed at accommodating heterogeneity across respondents and substitution patterns of the non-‘buy none’ alternatives, we note that, for the price coefficient (which is negatively log-normally distributed), the location parameter (given by μ) and the spread parameter (denoted by σ) are both significant. Although difficult to interpret directly, the results do nevertheless indicate heterogeneous sensitivities to price, which is not surprising given the prevalence of differences in marginal utility of money that is likely to exist across respondents. Once again, the ‘buy none’ constant is negative, and significant. Moving to the estimated distributions of marginal WTP, which, unlike the other coefficients, are independent of the scale factor that is inversely related to the variance of the error term, reveals that the respondents are willing to pay a price premium for the value-added services. To enable more straightforward comparison of the implied marginal WTP distributions against the MNL estimates, we report the mid-point (m) and spread (s) parameters relating to each of the marginal WTP distributions. Looking first at the mid-point of these distributions, we highlight that they are in line with the marginal WTP estimates produced under the MNL model, which is an expected finding, given our distributional assumptions. The significant spread parameters for these marginal WTP distributions leads to the rejection of the null hypothesis of homogeneity in the price premium respondents are willing to pay for value-added services to food. With the possible exception of the traceability value-added service, we find that the spreads are equivalent to the mid-point, which, we admit, is an artefact of our distributional assumptions. Notwithstanding this, it implies that the distributions of the marginal WTP estimates range between zero and twice their mid-point. The fact that the spread of the error component is significant indicates that there is a degree of substitutability between the non-‘buy none’ alternatives. As is typically experienced with RPL/EC specifications, the overall level-of-fit is substantially improved when heterogeneity is facilitated – with an improvement of almost 570 log-likelihood units at the expense of fitting five additional parameters, equating to a likelihood ratio statistic of 1134.48 against the χ2 critical value of 12.59 . We recognise that this result stems from allowing preference heterogeneity and substitution patterns between the non-‘buy none’ alternatives, but also, to a large extent, because the correlation across choices made by the same respondent is captured.
Notwithstanding the behavioural insights afforded by the RPL/EC model over the MNL model, the RPL/EC model does not accommodate the situation where the market for value-added services is niched. Our final model (DM-RPL/EC) shown in Table 1 attempts to address this. In this model, the distribution of the price premiums that respondents are willing to pay incorporates a spike at zero as well as a portion facilitating the heterogeneity in WTP within the market niche. Looking first at the coefficients that explain the price attribute, ‘buy none’ constant and error component, we note that the inferences reached are in line with those obtained under the RPL/EC model. Of greater interest in this model is the impact on the estimated distributions of marginal WTP. Inspection of the parameters explaining these distributions reveal that, with the sole exception of the traceability value-added service, the predicted distributions are markedly different from those implied from the RPL/EC model. Crucially, for three of the value-added services (food testing, animal health/welfare and region of origin), we find that the value of are significantly different from one. This provides strong evidence that they represent only niche markets since they are not valued by all respondents. In fact, for the region of origin value-added service, we observe that the market niche represents 15 per cent, meaning that 85 per cent of respondents are not willing to pay a price premium for chicken breasts that are produced within the British Isles. The market niches for the food testing and animal health/welfare value-added services are in the order of 55–60 per cent. In conjunction with the strong evidence that the value-added services are valued by only a subset of respondents, we find that the predicted distributions of marginal WTP are of a higher magnitude among these subsets vis-à-vis those predicted under the RPL/EC and MNL models, which do not accommodate the fact that a subset are not willing to pay a price premium. This evidence calls into question some of the inferences derived from the RPL/EC model. Moreover, with a likelihood ratio statistic of 213.84 against the χ2 critical value of 9.49
for the DM-RPL/EC model versus the RPL/EC model, we can reject the null hypothesis that the more flexible specification, which incorporates market niches, does not lead to a better model fit. We note that this improvement in fit is supported by the
, AIC and BIC statistics, i.e. even after penalising for the additional parameters.
For a more detailed examination and sensitivity of the estimated marginal WTP distributions, in Figure 1 we graphically contrast the cumulative distribution functions (CDFs) of the unconditional distributions (based on 10,000 random draws) for the four value-added services across the three model specifications. The MNL model is based on the assumption of homogeneous WTP and, thus, their respective CDFs (represented using square characters) are constant leading to vertical lines. Turning our attention to the distributions of marginal WTP estimated under the RPL/EC model (denoted with circle characters) illustrates the extent of heterogeneity in the price premiums that respondents are willing to pay for value-added services to food. When these services concern enhanced food testing and higher animal health/welfare, the marginal WTP estimates range between zero and over GBP 2.00, whereas for traceable and locally produced chicken breasts, marginal WTP estimates are not found to exceed GBP 1.30 and 0.60, respectively.
Notwithstanding the fact that the medians of the marginal WTP distributions derived from the RPL/EC model are not markedly different for the MNL predictions, which is in accordance with our earlier remarks, the stark differences between the marginal WTP distributions derived from the two models reinforces the inappropriateness of the MNL model in predicting marginal WTP among our sample of respondents. Despite this, as one moves to the marginal WTP distributions produced from the DM-RPL/EC (signified with triangle character), further significant changes are apparent. First, the marginal WTP distributions have both a discrete and a continuous part, with the exception of the traceability in value-added services, which is virtually equivalent to that obtained from the RPL/EC model. For the remaining value-added services, there is a sizeable mass at zero reflecting a subset of respondents who are predicted as not willing to pay a price premium for the service. We also draw attention to the relative dispersion in marginal WTP within these three niche markets. For these value-added services, we find that some respondents would be willing to pay up to GBP 4.00 (and beyond). We remark that this is markedly higher than that predicted under the RPL/EC model and, in fact, in all three cases almost one-fifth of respondents are predicted as having higher marginal WTP values under the DM-RPL/EC model than the highest value attained from the RPL/EC model. This is an important finding as it highlights the implications of negating to account for the fact that the market is segmented and that the market for these value-added services to food are niched rather than mainstream. Both the MNL and the RPL/EC models give a misleading impression of marginal WTP, as the distributions are distorted by the subset of respondents not willing to pay for the value-added services.
4.2. Demand for value-added food products at different price premiums
While the values of in Table 1 give the ceteris paribus estimates of the size of the niche market for the respective value-added service, it is also of interest to explore how sensitive the market segments might be at different price premiums. As part of our analysis, we are, therefore, interested in exploring the impact that different price premiums for the value-added services have on the likely demand for the valued-added chicken versus standard chicken as well as its role on consumers deciding not to buy any chicken.
Similar to the approach by Rigby and Burton (2005), we consider a hypothetical sample of 10,000 purchasing decisions made by separate consumers and that their choice is restricted to either: (i) buying ‘value-added chicken’, with enhanced food testing, traceability and animal health/welfare, which is produced within the British Isles and entails a price premium ranging between GBP 0.00 and 4.00; (ii) buying ‘standard chicken’, with standard food testing, traceability and animal health/welfare, which is produced outside of the British Isles and priced at GBP 2.00 or (iii) buying none. Based on the model results, we simulate coefficients for each consumer within our hypothetical sample and use equation (3) to obtain the probability of each of these options being chosen by each consumer. These probabilities are calculated at price premiums that range between GBP 0.00 and 4.00, at every GBP 0.05 interval. In Figure 2 we plot the means of probabilities (i.e. market shares) for each choice outcome at each of the price premium intervals. We further compare this for the three model specifications.
As would be expected, if the price premium of the value-added chicken is zero, its market share is considerably higher (approximately 85 per cent) than that predicted for the standard chicken (approximately 10 per cent). Nevertheless, in line with theoretical expectations, as the price of the value-added chicken increases the proportion of consumers predicted to purchase it decreases. Indeed, by the time the price premium reaches GBP 4.00 (i.e. a total price of GBP 6.00 for the value-added chicken), the average probability (i.e. market share) falls to around 30 per cent. This is associated with an increase in the market share for standard chicken, which is at almost 50 per cent at this price premium. Increasing the price premium for the value-added chicken also leads to a growing share of the consumers buying neither the value-added nor standard chicken (as the price premium increases from GBP 0.00 to 4.00, the average proportion of consumers buying none increases from around 5 per cent to almost 20 per cent). We remark that these findings hold irrespective of the model specification.
While the means of the probabilities at the different price premiums are relatively comparable across the model specifications, especially in the case of the ‘buy none’ option, we recognise that the means do not fully represent the distribution and that, therefore, a closer examination of these unconditional distributions would provide a useful comparison. Given the superior model fit and intuitive appeal of the DM-RPL/EC model, in Figure 2 we also present histograms of the choice probabilities at various price premiums derived from this model. This clearly demonstrates the complete range in choice probabilities and that, irrespective of the price premium (at least within the GBP 0.00–4.00 range), there is likely to be a subset of consumers who will buy the value-added chicken.
4.3. Optimal price premium for value-added food products
As illustrated in Figure 2, the exploration of market shares at various price premiums can yield important information. However, given that increasing the price premium of value-added chicken simultaneously led to a lowering in the proportion of consumers who would buy the value-added chicken and an increasing quantity of consumers who would buy the (cheaper) standard chicken or buy no chicken, it raises questions relating to the appropriate (i.e. optimal) price premium for the value-added chicken. Specifically, what is the premium price level associated with the maximum profit or maximum revenue? Despite the fact that this type of information can be ascertained relatively straightforwardly from discrete choice model estimates and that it has considerable practical appeal and importance for marketers engaged in pricing decisions, this type of analysis is, for the most part, overlooked. In this section we therefore extend the demand analysis to arrive at the price premium for the value-added chicken that will lead to the maximum total revenue.5
Under the MNL model we find that the price premium of GBP 2.55 leads to the highest expected total revenue of GBP 30,726. In contrast, the price premium associated with the maximum total revenue of GBP 32,939 for the hypothetical sample generated from the RPL/EC model results is found to be GBP 2.75 (i.e. a total price of GBP 4.75 for the value-added chicken). At this price premium for the value-added chicken, this model predicts, on average, that 57 per cent of the hypothetical consumers would buy the value-added chicken, 30 per cent would buy the standard chicken and the remaining 13 per cent would not buy any of the chicken breasts. Moving to the optimal price premium established from the DM-RPL/EC model, we find that it is somewhat higher at GBP 2.95 (i.e. a total price of GBP 4.95 for the value-added chicken), and, importantly, that the expected total revenue of GBP 30,853 is lower than that predicted under the RPL/EC model. At this price premium, on average, 47 per cent of consumers are predicted to buy the value-added chicken, 37 per cent would buy the standard chicken and the remaining 15 per cent of the hypothetical sample would not buy any of the chicken breasts.
Although the RPL/EC model predicts a higher maximum expected total revenue, we draw attention to the fact that it is based on the premise that all consumers are willing to pay for the value-added services. From the results presented in Table 1, we find that this may be an erroneous assumption and that not accounting for the market segments is likely to result in misleading inferences (including inferences relating to expected total revenue). Indeed, if, as implied by the model diagnostics, the DM-RPL/EC is a more accurate model, Figure 3 clearly highlights that reaching pricing decisions on the basis of the RPL/EC model would result in an over prediction of total revenue. Moreover, considering the optimal price premium to be GBP 2.75 has implications for managing the product portfolio. If decisions are made on the basis of the RPL/EC model, the demand would be in the region of 5,700 units of value-added chicken and 3,000 units of standard chicken, whereas based on the predictions derived from the more reliable DM-RPL/EC model, the actual demand at this price premium is in the region of 5,000 units for the value-added chicken and around 3,500 units for the standard chicken. Importantly, setting the price based on the RPL/EC predictions could also lead to lower profits, and this difference will become more pronounced as the unit (variable) cost associated with producing the chicken increases (notwithstanding the additional costs of disposal of excess value-added chicken that are likely to be incurred if the optimal price premium implied by the RPL/EC model is used as well as the revenue foregone by under supplying standard chicken).
5. Conclusions
While choice experiments have been widely applied to establish the demand and the value consumers place on food attributes, they typically do not accommodate the fact that the markets for such attributes may be highly niched. In this paper, we develop a novel discrete choice model in WTP space that combines discrete and continuous mixing distributions to simultaneously uncover estimates of the size of the niche market for value-added services to chicken and the heterogeneity in WTP within this niche. Our findings suggest that: (i) the demand for value-added services to chicken does not apply to all consumers, but instead to niche markets; (ii) within the niche markets there is considerable heterogeneity; (iii) accounting for this is relevant for estimation outcomes as it significantly improves goodness-of-fit. In addition, comparison of marginal WTP estimates and results obtained from a sensitivity analysis of demand and total revenue predictions at different price premiums draw attention to the important implications for the marginal WTP, demand and total revenue predictions. In particular, we find that the naïve specifications are unable to detect the subset of respondents who are predicted as not willing to pay a price premium for the service. This is shown to create a misleading impression of marginal WTP. Importantly, these naïve specifications also appear to poorly predict the upper range in the marginal WTP distributions and lead to potentially erroneous market share predictions. Crucially, failing to account for the fact that the markets for the value-added services are niched rather than mainstream leads to an overstated total revenue estimate and is likely to lead to lower profit margins.
Methodological issues aside, our results indicate that a sizeable proportion of consumers are not willing to pay anything extra for value-added services to food. In particular, we find that just over half of consumers are willing to pay a price premium for chicken that meets higher food testing and animal health/welfare standards and that the respective figure for the locally produced value-added service is just over 15 per cent. Irrespective of model specification, we find that, on average, consumers are willing to pay the highest price premiums for chicken that meets higher standards of food testing and animal health/welfare standards, the least for locally produced chicken, with traceability ranking in-between. Our most flexible, and best fitting, model suggests that consumers would be willing to pay a price premium of up to around GBP 4.00 for the food testing, animal health/welfare standards and locally produced value-added services. In contrast, the marginal WTP for the traceability value-added service peaks at approximately GBP 1.30. Our analysis further shows that the maximum expected revenue per purchase decision is GBP 3.09, whereby almost half of consumers are predicted to buy value-added chicken at GBP 4.95 and 37 per cent are predicted to buy standard chicken at GBP 2.00.
In closing, we note a potential limitation. The discrete mixtures specification is vulnerable to some identification issues. While aimed at uncovering respondents with zero marginal WTP, the discrete mixtures may possibly also serve as a proxy for taste heterogeneity in general and may also be an artefact of attribute non-attendance. Notwithstanding this, the method outlined in this paper does represent a step forward in predicting the demand and value for value-added services to food. Despite the appeal, flexibility and widespread application of random parameters models, the results in this paper call into question the suitability of continuous distributions in situations where it is believed that the market is niched. As demonstrated in this paper and the widespread evidence of niche markets for value-added services to food, analysts engaged in exploring preferences for food attributes from stated choice data should benefit from using specifications capable of accommodating and identifying the niche markets to establish the robustness of their predictions of marginal WTP, demand and total revenue. The ability to predict the potential extent of the niche markets at different price premiums is likely to be of considerable appeal and usefulness to marketers. Knowledge of such information should be particularly useful when deciding on the appropriate marketing-mix. Furthermore, the approach promoted in this paper also facilitates heterogeneity in values within the market niche that consumers are willing to pay. This adds additional insight for marketers to fine-tune their marketing effort.
Given the sensitivity of the marginal WTP, demand and total revenue predictions when the possibility of niche markets are ignored, this type of analysis should become a recommended course of action in practice as well as an area for future research. Extending this analysis to include socio-demographic information as covariates would be a useful area for further research. Indeed, even a rudimentary examination of the conditional distributions of the marginal WTP estimates alongside the conditional probabilities of being within the market niche against the socio-demographic profile of respondents could offer good insight and provide additional information for marketers to tailor their marketing-mix to different consumer segments. Further research on alternative econometric specifications and distributional assumptions would also seem justified. Non-parametric approaches, or making use of mixtures of continuous distributions and mixed latent class models (cf. Greene and Hensher, 2013) with a view to accommodating multiple modes, are potentially important avenues for future research in this context.
Acknowledgements
This work was supported by funding from the Department for Employment and Learning (Northern Ireland) under the ASsured, SafE and Traceable (ASSET) food research project.
References
For further discussion on WTP space models, see Train and Weeks (2005) and Scarpa et al. (2008).
While we could have also specified a mass at zero for the price attribute to accommodate potential non-attendance, doing so could complicate the estimation of marginal WTP estimates and, moreover, would go beyond the primary focus of this paper, namely the identification of niche markets for the value-added service attributes.
In our analysis, we assume that the WTP distributions are independent. While this may be somewhat restrictive, it does not preclude us from exploring the issue which is at hand, namely the identification of niche markets.
The value-added service attributes are entered as dummy variables, with the value of one representing the enhanced standard for the three food safety attributes and chicken that has been produced within the British Isles in the case of the region of origin attribute.
We note that it would be preferable to calculate the price premium that would lead to maximum profit. Since this would require the establishment of additional cost (variable and fixed) assumptions, which we feared could trivialise this sensitivity analysis, we focus only on total revenue generated as it does not require any further assumptions from those used in Section 4.2.
