Sustainability trade-offs in animal husbandry: Consumers’ choice when they can't have it all

Abstract

1. Introduction

The agricultural sector within the European Union (EU) has experienced transformative developments in recent years. The ‘Farm to Fork strategy’, a key component of the European Green Deal, was developed to intend a change towards a sustainable food system by 2030 (European Commission 2020). Additionally, the recent reform of the Common Agricultural Policy also emphasizes the need for sustainable practices and the management of natural resources within the EU (Cap strategic plans 2022). While agriculture in Germany has evolved into an efficient and innovative economic industry, that is well integrated into international markets (Agethen 2019), the associated industrialization of the sector has led to a reduction or loss of social acceptance. Citizens are increasingly demanding a change in the area of livestock production (Bundesministerium für Ernährung und Landwirtschaft 2015; European Commission 2016; Weible et al. 2016; Heise 2017; Kantar Emnid 2017; Simons et al. 2018; Schulze et al. 2023). Although pork is the most consumed meat in Germany (Statista 2022), pig farming has come under public scrutiny due to concerns about the husbandry conditions within this sector (European Commission 2016; Christoph-Schulz and Rovers 2020; Birkle et al. 2022). While negative societal attitudes about animal husbandry are widespread, they are often built with limited knowledge (Simons et al. 2018). Adding complexity, public discussions in general fail to consider that measures successful in mitigating one shortcoming can aggravate others. When multiple goals cannot be pursued simultaneously, a potential for conflict arises (Boudreaux and Ozer 2013).

In livestock farming, conflicting goals can arise, such as balancing animal welfare with environmental pollution and public health (e.g. Tuyttens, 2005; Scott et al. 2006; Bracke et al. 2013; Bundesministerium für Ernährung und Landwirtschaft, 2015; Economou and Gousia 2015). For example, animal-friendly practices like outdoor access can lead to greater pollution and health risks (Spiller et al. 2015). Additionally, reducing antibiotic use in livestock production is important for limiting antibiotic resistance and protecting public health (World Health Organization et al. 2015; Woolhouse et al. 2015; Spellberg et al. 2016), but it can also compromise animal health (D'Angeli et al. 2016). Furthermore, stricter standards in animal farming can lead to increased production costs and efficiency loss, which can have implications for sectoral competitiveness (Belay and Jensen 2022a; 2022b). Failure to compensate for increased costs with higher market prices may accelerate the structural change in livestock farming, which will further induce undesirable social effects for family farms and rural communities while also affecting the international competitiveness of the sector (Bundesministerium für Ernährung und Landwirtschaft 2015). When higher requirements and standards result in higher market prices, they are linked to unwanted distributional effects. In summary, society faces conflicts between various sustainability goals in animal husbandry, including economic, social, environmental, human health, and animal welfare objectives. Measures aimed at addressing one goal may inadvertently worsen others, indicating a degree of incompatibility among multiple sustainability objectives (Sieben and Schildbach 1975).

Hence, gaining insights into consumers’ preferences for conflicting sustainability goals in livestock farming is relevant for developing effective marketing strategies and policy initiatives. In this regard, selecting an appropriate communication strategy is of great importance. Given that conflicting sustainability goals are rarely discussed publicly and considering consumers’ limited knowledge about animal husbandry, it is essential to initially inform consumers about the various dimensions of sustainability and the related trade-offs. Furthermore, it is important to acknowledge that consumers are not homogeneous and that different consumer segments may prioritize different sustainability dimensions.

In the present study, we focus on examining three main conflicts between different sustainability dimensions: first, the conflict between animal welfare and environmental protection; second, between human health and animal health; and third, between human health and animal welfare. Thus, the present study seeks to address the following three research questions: What are consumers’ preferences regarding different (conflicting) sustainability goals? How does positive information and trade-off information affect consumers’ preferences? And, what is the role of attitudinal and sociodemographic factors in influencing consumers’ preferences in the case of sustainability trade-offs? Based on these three research questions, nine research hypotheses are derived and tested in the subsequent chapters of this study.

Our study expands on previous research in several ways. Firstly, it examines three conflicts simultaneously within the context of pig farming, considering their interconnectedness and trade-offs. Previous studies either focused on the conflict between two sustainability goals (e.g. Van Asselt et al. 2019) or investigated a citizen instead of a consumer perspective (cf. Schütz, Busch and Sonntag 2023). Our study goes beyond previous research by examining consumers’ preferences and their Willingness to Pay (WTP) in relation to a number of conflicting sustainability goals in pig farming. Secondly, it investigates the effects of positive information and trade-off information on consumers’ preferences regarding these sustainability goals. By examining the influence of information on consumers’ preferences, we aim to contribute to a better understanding of how communication strategies and information provision can shape consumer decision-making in the context of sustainability trade-offs. Thirdly, we highlight the role of attitudinal and sociodemographic factors in predicting consumers’ preferences, thereby providing a deeper understanding of consumer decision-making in the context of sustainability trade-offs. Our findings have implications for the development of effective marketing strategies and policy initiatives that aim at aligning with consumers’ preferences and contribute to more sustainable livestock farming.

2. Literature overview and hypotheses development

2.1 Consumers’ preferences for sustainability goals in animal husbandry

Many studies have shown that consumers tend to prioritize animal husbandry sustainability attributes that benefit their own health over those who promote animal welfare or environmental benefits (Heid and Hamm 2013; Viegas et al. 2014; Dahlhausen et al. 2018; Grunert et al. 2018; Paudel et al. 2022). For example, Viegas et al. (2014) observed that consumers’ WTP is highest for the safety of meat products compared to other attributes. This is consistent with the results from Grunert et al. (2018), who showed that consumers prefer meat attributes providing individual health benefits, such as reduced antibiotic use, rather than those contributing to animal welfare or societal benefits. Similarly, according to Frewer et al. (2005) attributes such as food safety, no medicine residues and healthy product claims are considered by consumers as more important than animal welfare-oriented or environmentally friendly pig production. Along similar lines, Paudel et al. (2022) demonstrated that WTP for antibiotic-free pork chops is higher compared to both pork chops where only curative use of antibiotics is allowed and pork chops where antibiotics can also be used to promote growth. The results of a discrete choice experiment conducted by Dahlhausen et al. (2018) show that antibiotic-free claims elicit a higher WTP than improved animal welfare attributes for pork minute steaks. Furthermore, different studies on piglet castration revealed that citizens oppose piglet castration without anaesthesia because of the negative effects on animal welfare (Fredriksen et al. 2011; Heid and Hamm 2013); however, immunocastration as an alternative is opposed by a large share of respondents due to concerns about residues and negative effects on human health (Fredriksen, Johnsen and Skuterud 2011; Heid and Hamm 2013; Bergstra et al. 2017). Although the majority of literature in this field suggests that consumers/citizens value health benefits more than animal welfare or environmental benefits, other literature suggests that consumers can also be indecisive (Van Asselt et al. 2019). Moreover, when consumers’/citizens’ preference for animal welfare is contrasted with environmental benefits, the literature suggests that animal welfare is given priority (Viegas et al. 2014; Sonntag 2018; Kühl et al. 2019). Accordingly, we formulate the following two hypotheses about consumer preferences for different sustainability goals:

Hypothesis 1:Human health is a sustainability goal preferred by consumers over both (a) animal welfare and (b) the environment.

Hypothesis 2:Animal welfare is a sustainability goal preferred by consumers over the environment.

2.2 The role of information on consumers’ preferences for sustainability goals

As consumers/citizens often lack knowledge about animal husbandry systems (Simons et al. 2018), marketing or policy-induced information strategies may help in bridging this gap. Several studies demonstrate that informing consumers about different animal husbandry practices can influence consumers’ preferences (Risius and Hamm 2017; Wille et al. 2017; Altmann et al. 2022; Weingarten and Hartmann 2023). Furthermore, providing information about different sustainability goals can affect consumers’ evaluation of meat production. Palomo-Vélez et al. (2018) propose that animal welfare-related information is the most effective, while Cordts et al. (2014) showed that health information is more effective than environmental information. However, information can also have adverse effects, as it can, for example, induce confusion (Grunert 2002; Sonntag et al. 2019). Moreover, when information is two-sided (i.e. including information about drawbacks and benefits), it is likely to be less effective compared to one-sided information (Velde et al. 2018; Cornelis et al., 2020), although the decrease in effectiveness is found to be message and context-specific (O'Keefe 1999; Eisend 2006). With respect to the effect of (two-sided) information, we formulate the following hypotheses:

Hypothesis 3:Information regarding the animal welfare benefits of ‘open-barn’ husbandry practices and regarding the human health benefits of ‘antibiotic-free’ husbandry practices positively influences consumers’ preference for the (a) ‘open-barn’ labelled product and the ‘antibiotic-free’ labelled product.

Hypothesis 4:Information that displays both benefits and trade-offs linked to ‘open-barn’ husbandry practices and to ‘antibiotic-free’ husbandry practices, does not influence consumers’ preference for the (a) ‘open barn’ labelled product and the (b) ‘antibiotic-free’ labelled product.

2.3 The influence of attitudinal and socio-demographic factors on consumers’ preferences for sustainability goals

In several studies, attitudinal variables were considered determining factors for consumers’ preferences or WTP for products bearing different sustainability attributes. Rahmani et al. (2019) revealed that consumers who had, for example, a pro-environmental attitude were more likely to choose eggs characterized by higher animal welfare standards. Similar results were found by Katt and Meixner (2020); here, attitudes toward the environment played a significant role in influencing WTP for organic food. Focusing on animal welfare consciousness, Denver et al. (2021) showed that concerns regarding animal welfare affected WTP for animal welfare products positively. In a study by Bradford et al. (2022), behavioural beliefs around animal welfare were one of the main predictors of WTP to purchase pork labelled ‘raised without antibiotics.’ Furthermore, consumers who perceived a personal risk from antimicrobial resistance were willing to buy pork labelled ‘raised without antibiotics’ (Bradford et al. 2022). Attitudes toward consumers’ own health also significantly influenced WTP for organic food (Katt and Meixner 2020). Finally, general label perception might also influence consumers’ WTP for products bearing sustainability labels. If consumers perceive a label positively, they are more likely to purchase products carrying this label (Calvo Porral and Levy-Mangin 2016; Wang et al. 2020). Therefore, based on the literature, we derive the following hypotheses concerning the influence of attitudinal factors:

Hypothesis 5:Higher environmental consciousness leads to a stronger preference for a product supporting the environmental sustainability goal.

Hypothesis 6:Higher animal welfare consciousness leads to a stronger preference for a product supporting an animal welfare sustainability goal.

Hypothesis 7:Higher health consciousness leads to a stronger preference for a product supporting the health sustainability goal.

Hypothesis 8:Higher label perception leads to a stronger preference for (a) product supporting the environmental sustainability goal and (b) a product supporting the health sustainability goal.

With respect to sociodemographic factors, the findings of the meta-analysis by Clark et al. (2017) show that WTP for animal welfare is higher for women, and decreases with age, but increases with income and higher education. While Rahnama et al. (2017) found no effect of sociodemographic factors such as gender, age, education and income on choosing antibiotic-free chicken, Bradford et al. (2022) found a higher WTP among women, but not in relation to socio-economic status, age or education. In contrast, Denver et al. (2022) observed that high WTP for antibiotic-free pork tended to be more frequent among women and older respondents, while Denver et al. (2021) observed the same among well-educated people. Thus, we derive one final hypothesis concerning the effect of sociodemographic factors on the ‘open-barn’ labelled product, but do not make any directional predictions due to the heterogeneous results of prior literature.

Hypothesis 9:Sociodemographic factors such as (_i) gender, (_ii) age and (_iii) education influence preferences for (a) the ‘open-barn’ labelled product and (b) the ‘antibiotic-free’ labelled product.

3. Methods

3.1 Experimental design and procedure

Consumer data were collected in December 2020 through an online survey conducted by a market research agency that was compliant with the Declaration of Helsinki and the EU's General Data Protection Regulation (GDPR). Recruitment was designed to be representative of the overall population in Germany based on age (>18 years old), gender, education, and residence in Germany. In our study, we refrained from asking about income due to ethical concerns, which could have induced participants to drop out. The exclusion criteria were: following a vegetarian or vegan diet, bearing no responsibility for household food shopping, and never consuming meat or purchasing salami.

The study followed a 3 × 3 mixed experimental design comprising of within and between-subject factors. The first part of the survey collected sociodemographic information. The between-subjects factor, aimed at investigating the role of information provision on consumer choices, was introduced next. Respondents were randomly assigned to one of the three groups: the control group or one of the two experimental groups. Control group (‘control’) participants were provided with text about the local university, thus receiving information unrelated to the conducted experiment. The first experimental group (‘label info’) received a text about the beneficial characteristics of pig husbandry practices associated with the two fictitious labels—the ‘open-barn’ label contributing to improved animal welfare and the ‘antibiotic-free’ label contributing to the protection of human health. To match the length of texts received by all groups, an additional paragraph containing neutral information about the University of Bonn was included. The second experimental group (‘label + conflict’) received identical positive information about the two fictitious labels as the first experimental group. Furthermore, this text contained an additional paragraph describing the negative effects of production practices behind fictitious labels, namely the negative environmental effects of ‘open-barn’ production as well as the negative effects of an ‘antibiotic-free’ system on animal health. The three texts contained were of equal length (‘control’ = 328 words, ‘label info’ = 329 words, ‘label + control’ = 329 words) and were pretested with 20 persons (with an agricultural background as well as citizens with no affiliation to the agricultural sector). Information provision was followed by an attention check question, which was later used as a criterion to exclude inattentive participants. Translated information texts are enclosed in the appendix (Table A.1–A.3).

After being exposed to information treatments, respondents completed choice experiments based on the hypothetical contingent valuation method (CVM), which corresponds to the within-subject part of the experiment. Each respondent was presented with three choice situations, each comparing two salami products, that represented one of the following conflicts between different sustainability goals: (1)ʻopen barn’ labelled salami (vs. no label): ‘open-barn’ salami that improves animal welfare, but reduces environmental protection versus unlabelled salami as an alternative; (2) ‘antibiotic-free’ labelled salami (vs. no label): ‘antibiotic-free’ salami that supports human health, but reduces animal health versus unlabelled salami as an alternative; and (3) ‘antibiotic-free’ labelled salami (vs. ‘open barn’ labelled salami): a choice option that captures all aforementioned sustainability trade-offs implicitly and the conflict between animal welfare and human health explicitly. To minimize bias from consumers’ preferences for label design, both labels used in the experiment had identical colour schemes, arrangement of the textual elements, and positioning (Fig. 1). In addition, fictitious labels were used in order to avoid pre-existing associations. Furthermore, labels were chosen as the first communication means on the package due to their ease of comprehension and increasing use in communicating sustainability characteristics at the Point-of Sale (POS) (Hawley et al. 2012; Sonntag et al. 2023). Figure 1 shows the salami pictures used in the choice experiments. After the CVM assessment, a captcha check was used to identify and exclude bots. Finally, the attitudinal factors were measured and participants were debriefed. Materials can be accessed at: https://osf.io/etbkr/?view_only=52fc564512cb4e66a602c6377776b7d9.

Figure 1.

Example of the salami pictures with labels used in choice experiments. Description on salami packaging in German from left to right: ‘Feinschmecker Salami aus Schweinefleisch nach traditionellem Rezept, edel im Geschmack ʼ(English: ‘gourmet salami made from pork according to traditional recipe noble taste’), ‘Ohne Antibiotika—Bereits ab Geburt’ (English: ‘antibiotic-free—already from birth’), ‘Offenstall—Mehr Platz & Frischluft’ (English: ‘open barn—more space & fresh air’).

Open in new tabDownload slide

3.2 Measures

3.2.1 Contingent valuation method

The CVM was used to determine participants’ preferences for different (labelled) salami products in three choice situations as described in the previous section. CVM is a widely accepted and frequently employed technique in economics and environmental research to elicit individuals’ preferences and monetary valuations (Carson and Hanemann 2005; Boyle 2017). This method is particularly useful in eliciting WTP for products not yet available on the market, and in instances where the emphasis is on experimental, attitudinal or socio-demographic predictors of consumer choices (Zhang et al. 2020; Giotis and Drichoutis 2021; Yormirzoev et al. 2021; Schuermann and Woo 2022; Just and Goddard 2023). More precisely, we used a triple-bounded contingent valuation (CV) design, expanding the range of potential intervals. This approach allowed for a more refined and precise determination of WTP values, drawing on the methodologies established by Langford et al. (1996), Bateman et al. (2001).

To reduce a potential hypothetical bias, a short ‘cheap talk’ script was included (Lusk and Shogren 2007; Silva et al. 2011) that urged participants to imagine a daily shopping situation in the supermarket. Here, participants were asked to imagine standing in front of the charcuterie shelf in the supermarket where they would normally shop and wanting to buy a packaged pork salami and that all salami were the ones of their preferred brand. Furthermore, they were encouraged to consider the budget they generally spend at the supermarket, and that this purchase would reduce the money available for other purchases.

Next, participants were confronted with three choice situations with conflicting sustainability goals, corresponding to the within-subject part of the experiment described above. In each of the choice situations, a multinomial-choice question was asked first, followed by two dichotomous choice questions based on a Double-Bounded Contingent Valuation Method (DB-CVM) (Hanemann et al. 1991; Kealy and Turner 1993) (see Fig. 2). In the multinomial question, for each choice situation, two salami products linked to competing sustainability goals were offered at the same reference price of €1.49. The price of €1.49 for the salami was determined through a small market study involving 54 pork salami products from three different retailers. Based on this analysis, the gram amount was set at 80 g, which is the most common quantity found. This price and gram amount align with the average price and weight of similar conventional salami products in supermarkets.

Figure 2.

Between-subject choice situations and CV procedure. The multinomial choice question was translated into a dichotomous bid question |${B}_D$| with a fixed bid amount of €1.49 to allow for the application of the triple-bounded model. The notation |$B\_$| refers to the Bid variable, while the subscript D refers to the dichotomous choice. The first superscript h|$(l)$| is used to indicate whether the bid refers to the higher (lower) price in the first round. The second superscript |$h\,\, (l)$| is used to denote the higher (lower) value in the second round. At the first level, one of nine higher (lower) prices is randomly assigned.

Open in new tabDownload slide

Analogous to Tversky and Shafir (1992) and Jin (2014), the multinomial question had the following response options: (I) ‘I would choose salami A’; (II) ‘I would choose salami B’; (III) ‘I think both salamis are equally good and I am undecided’; and (IV) ‘I would not choose any of the salamis,’ with ‘A’ corresponding to the first-choice option and ‘B’ to the alternative, respectively. Options III and IV approximate the real choice situations of respondents in the supermarket, as they allow them to express indifference or opt out from the purchase.

Figure 2 illustrates the three choice situations that each respondent faced, with the exemplarily highlighting of the choice between ‘open barn (respectively A) and ‘no label’ (respectively B) salami.

Depending on respondents’ answers to the multinomial choice question, they were presented with two follow-up dichotomous bids, as shown in Fig. 2. By exemplarily highlighting the choice between ‘open barn’ and ‘no label’ salami (see Fig. 2), it means that consumers who selected either option (I) or (III), indicating a preference for ‘open barn’ salami over or equal to ‘no label’ salami in the multinomial question |$({B}_D = Yes)$|⁠, were randomly assigned to one of nine possible higher bids |$(B_D^h > $|€1.49)for the ‘open barn’ salami in the subsequent dichotomous choice question. The first higher bid price ranged from €1.55 to €2.03 in 6-cent intervals above €1.49. Those who chose option (II), thus stating that they prefer alternative ‘no label’ salami over ‘open barn’ salami in the multinomial question |$({B}_D = No)$|⁠, were randomly allocated to one of nine possible lower bids (⁠|$B_D^l < $|€1.49) for salami ‘A’ in the subsequent dichotomous choice question. In this context, the initial lower bid price ranged from €1.55 to €2.03, with intervals of 6 cents below €1.49. The follow-up bid ranges for option ‘open barn’ salami used in the DB-CVM were designed following Jin (2014), while the price of alternative ‘no label’ salami was kept constant at €1.49. The corresponding question was: ʻAssume that the “no label” salami is sold at a price of €1.49 Now, if the price of the “open barn” salami is €xx, would you choose the “open barn” salami?.’ Participants who selected option (IV) in the multinomial question were not considered for the follow-up dichotomous choice questions. The price level of the next bit was contingent upon the respondent's answer to the prior answer. If the earlier question was answered with ‘Yes’ (‘No’), the previously made surcharge doubled (halved) in the second query (Jin 2014). This contingent design resulted in eight possible response outcome combinations presented in Fig. 2. An overview of all possible prices is presented in the Appendix A2.1 and A2.2. The order of the choice situations was randomized.

3.2.2 Attitudinal factors

We measured four latent attitudinal factors: label perception, environmental consciousness, animal welfare consciousness, and health consciousness. The label perception scale was based on five items originally developed by Loureiro et al. (2006) and adapted by Gracia and de-Magistris (2016). Environmental consciousness was measured with five statements, based on Cordano et al. (2003). For measuring animal welfare consciousness, we used five statements from the scale of Cembalo et al. (2016). Health consciousness was measured with seven statements based on Dutta-Bergman (2004). All items were measured on a 7-point Likert scale (with 1 = strongly disagree and 7 = strongly agree).

3.3 Estimation procedure

In analyzing the data, we adhered to the methodology outlined by Meng et al. (2022) and estimated a multilevel-probit model. Consequently, the recoded multinomial responses were combined with the subsequent dichotomous choice questions. We assumed that the WTP amounts underlying the responses at each bid level were correlated (Alberini et al. 2003). Following the procedure outlined above, the data were structured hierarchically. For each respondent (Level 2), three bids/responses were obtained (Level 1), with responses coded as 1 for ‘Yes’ and 0 for ‘No’, as illustrated in Fig. 3. For instance, individual 1 did not select product ‘A’ in the first bid (multinomial choice question), was then offered a lower price (dichotomous choice 1) that was accepted, followed by a price ranging between the two prior bids (dichotomous choice 2), which was also accepted.

Figure 3.

Hierarchical structure of the data. Responses, |$1 = \text{`}Yes$|’, |$0 = \text{`}No$|’.

Open in new tabDownload slide

We analyzed the data by pooling the three bids/responses of each individual into a two-level mixed-effects probit model. The probit model assumes that there is an unobservable latent variable related to the actual binary response (⁠|$1 = \text{`}Yes$|’, |$0 = \text{`}No$|’) through a threshold concept (Bock, 1975). The response |${y}_i = 1$| occurs when |$y_{i\ }^*$| exceeds a threshold Y; here, |$P( {{y}_i = 1} ) = P( {y_{i\ }^* > 0} )$|⁠.Thus, the true WTP (⁠|$y_i^*$|⁠) of the participant is assumed to be latent and has the following linear response function:

where |${x}_i$| is a vector of the covariates with regression coefficients |$\beta $| for the fixed effects of each individual i. There are three responses per individual i. The parameter |${u}_i$| refers to individual-specific random effects. The random effects are not directly estimated as model parameters, but are summarized as variance components in the model. Finally, |${\epsilon }_i$| refers to the residuals and is assumed to be normally distributed with a mean of 0 |$(N( {0,1} )).$|

The triple-bounded design allows the restriction of the WTP within intervals. Based on the eight outcomes presented in Fig. 2 and the assumption of normal disturbance of |${\epsilon }_i$| the probabilities |$\pi $| of lying within an interval can be modelled for each outcome using cumulative density function |$F()$| as follows:

Thus, for example, the probability of saying ‘no’ to the first two questions followed by ‘yes’ to the third question means that the respondent's WTP is between the second bid and the lower third bid offered.

Consequently, the log-likelihood function including the eight probabilities can be represented as:

where |$\ln L\ $| is the natural logarithm of the likelihood and |$d_i^{*\ }$|are binary variables of the associated outcome (⁠|$d_i^{*\ } = 1$| denoting occurrence).

A two-level mixed-effects probit model was estimated for each choice situation. In the empirical application of equation (1), ten covariates|$\ ( {{x}_i} )$| were included. The variable Bid records the randomized price of the salami product offered to respondents. To investigate the effect of the information provision, dummy variables GroupLabel and GroupConflict were created, indicating whether the respondent was allocated to the ‘label info group’ or ‘label + conflict info group’, respectively. The control text served as the reference category. Furthermore, the four latent attitudinal constructs, as well as the influence of demographic covariates gender, age, and education, were included as predictors of consumer choice.

The estimation was conducted using meprobit and meprobit postestimation (margins and estat icc) commands in StataBE17 (StataCorp 17). Meprobit fits mixed-effect models for binominal or binary responses and allows many levels of nested clusters of random effects. The conditional distribution of the response given the random effects was assumed to be Bernoulli. The success probability was determined by the standard normal cumulative distribution. The postestimation command margins, dydx(*) atmeans estimates the average margins at the means of covariates and estat icc allows the estimation of the residual intraclass correlation between the latent responses of the same individual.

4. Results

4.1 Sample properties

In total, 1,570 participants completed the survey. A total of 226 participants were excluded as they answered the attention check question incorrectly (n_control = 46, n_{label info} = 45, n_{label + conflict} = 135). Another 13 participants (n_control = 4, n_{label info} = 7, n_{label+ conflict} = 2) had captcha values below 50 (Qualtrics 2023). In addition, 249 participants (n_control = 94, n_{label info} = 69, n_{label + conflict} = 86) were excluded as they displayed intransitive preferences. Thus, the final sample consists of 1,082 participants, with 392 participants assigned to the control group, 399 to the label info group, and 291 to the label + conflict group (Table 1).

Chi-square tests showed no significant associations between experimental groups and gender (χ²(2) = 0.9284, P = 0.629), age (χ²(10) = 8.5052, P = 0.580), or education (χ²(12) = 3.6469, P = 0.989). In terms of gender, age, and education level, the sample roughly corresponded to the German population (Destatis 2019, 2020a, 2020b).

The means and standard deviations for the constructs label perception, environmental consciousness, animal welfare consciousness, and health consciousness are shown in Table 2. Environmental consciousness (F(2, 1079) = 0.09, P =0.9152), animal welfare consciousness (F(2, 1079) = 0.35, P = 0.7081) and health consciousness (F(2, 1079) = 0.06, P = 0.9404) did not differ significantly between experimental conditions. Label perception (F(2, 1079) = 7.72, P < 0.001) differed significantly between groups. Bonferroni-adjusted post-hoc analysis revealed significantly higher perception scores in the label info group (P < .001).

4.2 Confirmatory factor analysis

Confirmatory factor analysis (CFA) was calculated to check the validity of the attitudinal factors used (Table 3). The CFA yielded a good fit, with |${\chi }^2$| = 849.59, df = 203, P < 0.001, |${{{\chi }^2} \!/ \!{{df}}} = 4.19$|⁠, Root Mean Squared Error of Approximation (RMSEA) = 0.054, Standardized Root Mean Square Residual (SRMR) = 0.050, Tucker–Lewis Index (TLI) = 0.925, and Comparative Fit Index (CFI) = 0.934.¹

All constructs exceeded the recommended threshold values for composite reliability (CR > 0.6; Bagozzi and Yi 1988) and average variance extracted (AVE > 0.5, Fornell and Larcker 1981). The Fornell and Larcker (1981) criterion shows good discriminant validity as the square root of AVE in every latent variable was higher than the correlation of the respective constructs with any other construct (Fornell and Larcker 1981). As CFA showed good reliability, all latent constructs were retained and the average value of scale items was used in the ensuing analyses.

4.3 Determinants of consumers’ choice

Overall, three two-level mixed-effect probit models, one corresponding to each choice situation, were estimated (Table 4). The variance of the random intercept at the individual level for the three models was: Model 1: |$\sigma _{M1}^2$|= 3.59, SE = 1.25; Model 2: |$\sigma _{M2}^2$|= 3.21, SE = 1.11; Model 3: |$\sigma _{M3}^2$| = 2.44, SE = 0.52. The likelihood-ratio test for all three models showed that there was enough variability between individuals to favour a mixed-effects probit regression over an ordinary probit model |$(\chi _{M1}^2$| = 182.41, P < 0.001; |$\chi _{M2}^2$| = 167.31, P < 0.001; |$\chi _{M3}^2$| = 192.52, P < 0.001). The interclass correlation coefficient (ICC) shows that in all three models more than 70% of the error variance is explained by the unobserved respondents’ heterogeneity (ICC_{Model 1} = 0.78, ICC_{Model 2} = 0.76, ICC_{Model 3} = 0.71).

Focusing on the effects of information on consumer choices, we observed that compared to participants in the control group those who received positive information about the two labels (GroupLabel) were more likely to select the ‘open barn’ salami than the ‘no label’ salami (β = 0.38, P < 0.05), and less likely to select the ‘antibiotic-free’ salami than the ‘open barn’ salami (β = −0.52, P < 0.001). However, there was no significant effect in the second-choice situation (antibiotic-free vs. no label). Providing consumers with information about benefits and trade-offs related to fictitious labels (GroupConflict) did not have a statistically significant effect on consumers’ choices in any of the three choice situations.

The attitudinal constructs label perception and environmental consciousness did not significantly influence any of the choices. However, animal welfare consciousness had a significant positive effect on the probability of choosing the ‘open barn’ (β = 0.41, P < 0.001), and the ‘antibiotic-free’ salami (β = 0.46, P < 0.001), when the alternative was a ‘no label’ salami. Similar findings hold for health consciousness, where the coefficient is positive and significant, indicating that health-conscious participants are more likely to choose the ‘open barn’ (β = 0.34, P < 0.05) or the ‘antibiotic-free’ (β = 0.24, P < 0.05) labelled salami.

Demographic variables indicate that gender had no influence on any of the three choices. A significant negative effect of age in the choice situation for antibiotic-free salami when contrasted with ‘no label’ revealed that the likelihood of purchasing the former decreased with increasing age (β = −0.12, P < 0.05). Lastly, our findings reveal that higher levels of education corresponded with a higher likelihood of a positive response in all three choice situations (Model 1: β = 0.16, P < 0.001; Model 2: β = 0.14, P < 0.05; Model 3: β = 0.12, P < 0.05).

Furthermore, the results reveal that Bid has a statistically significant and negative effect on the probability of choosing a salami bearing the respective label in all choice situations. Thus, increasing the price significantly decreases the probability of selecting the ‘open barn’ salami in Model 1 (β = −4.85, P < 0.001), the ‘antibiotic-free’ salami in Model 2 (β = −4.33, P < 0.001) (both vs. no label, respectively), and the ‘antibiotic-free’ (β = −2.91, p < 0.001) (vs. ‘open barn’) salami in Model 3.

In order to determine the nature and significance of the information treatment in our mixed-effects probit regression, the average marginal effects at their means were calculated using the coefficients generated from the mixed-effects probit regression. Holding all control variables constant at their means, information provision increases the likelihood of being consumers of the ‘open barn’ versus ‘no label’ by 5% (∆y/∆x = 0.05), and for ‘antibiotic-free’ versus ‘no label’ by 11% (∆y/∆x= −0.11).

4.4 Estimation of consumers’ willingness to pay

The mean WTP (M_WTP) for the three choice situations was calculated using the STATA nlcom command (Table 5). On average, consumers were willing to pay €2.06 and €2.12 for the ‘open barn’ and the ‘antibiotic-free’ salami, respectively. This implies a 57 cents premium for the ‘open barn’ salami (model 1; ‘open barn’ vs.’ no label’) and 63 cents premium for the ‘antibiotic-free’ salami (model 2; ‘antibiotic free’ vs. ‘no label’), relative to the ‘no label’ product offered at the reference price of €1.49. These results reveal higher consumer preferences for the labelled products and higher relative preference for the ‘antibiotic-free’ than for the ‘open barn’ label. Finally, the M_WTP from the last choice situation (model 3; ‘antibiotic free’ vs. ‘open barn’) shows that, on average, consumers are willing to pay €1.81 for the ‘antibiotic-free’ salami, when the ‘open barn’ is offered for €1.49, which corresponds to a 32-cent premium. This is consistent with the hierarchy of preferences established based on the first two choice situations—‘antibiotic-free’ salami commands a higher WTP than the ‘open barn’ salami.

Furthermore, for each model (choice situation), we visualized the probability of selecting a targeted product contingent on price while keeping the value of other predictors at average. The corresponding demand curves (Fig. 4) reflect choices made by respondents when they actually make a purchase decision between the presented options. Therefore, the demand curve is conditional.

Figure 4.

Estimated probability of selecting yes contingent on the bidding price. Curves estimated based on the results of multilevel mixed-effects probit regressions presented in Table 4 by varying the bidding price and using the average values of the covariates. The red line (€1.49) indicates the baseline price of the alternative products in each choice situation (shown in legend, in the brackets).

Open in new tabDownload slide

For models 1 and 2 where choice was made between labelled and unlabelled product, the shape of the demand curves shows that the majority of respondents would select a labelled option at the same price. However, the probability declines when the premium for the labelled product increases. When the price of the labelled alternative increases over €1.70, it is possible to see a divergence between the demand curves of model 1 and model 2 as consumers are less sensitive to a price increase for ‘antibiotic-free’ salami.

The demand curve for model 3 suggests that if both the ‘antibiotic-free’ and ‘open barn’ salami were offered at the average market price of €1.49, about three-quarters of consumers would choose ‘antibiotic-free’ salami. Furthermore, the slope of this curve differs from the previous two showing a faster loss of consumers first, but then a considerable WTP persistence of some consumers even at higher ends of the bidding price range, suggesting a clear preference of those for ‘antibiotic-free’ over ‘open barn’ salami.

5. Discussion and conclusion

In the ongoing social discourse regarding potential changes to current animal husbandry practices, it is necessary to acknowledge and address the inherent trade-offs between competing sustainability goals. Securing protected goods beyond the minimum legal requirements often is associated with higher production, processing, and distribution costs, inducing significant financial risk to suppliers if these additional sustainability attributes are not adequately appreciated and valued by consumers in the marketplace. Thus, understanding consumers’ priorities when faced with (conflicting) sustainability goals with respect to farm animal husbandry is crucial for both policymakers and private sector stakeholders. Therefore, it is necessary to investigate the impact of (conflicting) sustainability goals on consumers’ WTP. Equally important is to examine the role of information provision and sociodemographic and attitudinal factors in consumer decision-making processes within the context of animal husbandry practices.

In the forthcoming sections, we will assess whether our formulated hypotheses can be rejected in accordance with the results presented in Tables 4 and 5, as well as Fig. 4. Our investigation demonstrates that the majority of respondents are willing to pay a premium for a labelled salami compared to an unlabelled products, with a median WTP for ‘antibiotic-free’ salami being higher than for the ‘open barn’ salami. Moreover, a direct comparison between the two attributes confirms that, on average, consumers have a higher WTP for ‘antibiotic-free’ products. These findings provide support for our first and second hypotheses. These results suggest that, within the hierarchy of sustainability attributes, considerations related to personal health take precedence over animal welfare concerns, albeit animal welfare considerations still hold greater weight than environment-related sustainability attributes. This aligns with prior literature, which suggests a similar hierarchy of relative preferences for different sustainability goals (Sparks and Shepherd 1992; Grunert 1993; Grunert and Juhl 1995; Harper and Makatouni 2002; Berndsen and van der Pligt 2004; Jonge and van Trijp 2013; Cordts et al. 2014; Palomo-Vélez et al. 2018; Paudel et al. 2022).

We further observe the influence of pricing strategy on market share. Increasing the mark-up for a labelled product relative to both an unlabelled or other labelled product leads to a reduction in the market share of the said product. This finding confirms the negative price elasticity of demand. Thus, to the extent that it is socially desirable to enhance the market share of products promoting specific sustainability goals, though the WTP for such products does not cover the associated costs, additional market strategies or even governmental intervention could be considered.

One avenue to influence consumer behaviour is through informative labelling. In line with the literature (O'Keefe 1999; Eisend 2006), our study suggests that this strategy can lead to ambiguous effects as the impact of information provision depend on the nature of the sustainability conflict investigated and the type of information conveyed. Our findings are aligned with Hypothesis 3 (a) but not with Hypothesis 3 (b). Specifically, providing only positive information increased consumers’ WTP for the ‘open barn’ salami, but not for the ‘antibiotic-free’ salami, compared to the unlabelled salami. Moreover, this type of information reduces the WTP for the ‘antibiotic-free’ salami compared to the ‘open barn’ salami, although consumers still valued the former more than the latter. These findings imply that consumers perceive ‘antibiotic-free’ animal production as beneficial to their own health, with this salience being sufficiently captured by the label itself, rendering additional information ineffective in influencing consumer preferences. In contrast, consumers might be less aware of the positive animal welfare effects of ‘open barn’ production. Here, positive information serves to increase knowledge and, thereby, WTP for the ‘open barn’ salami

To achieve full transparency, consumers should be informed not only about the benefits, but also about the potential drawbacks of a production method. Our findings, however, suggest that such a strategy comes at a price. We argue that two-sided information has a neutralizing effect, as we could not detect any influence of the ‘label + conflict’ information on consumers’ WTP relative to the control group. Thus, we find support for Hypothesis 4 (a) and (b). This finding is in line with prior literature, indicating that two-sided information can be less effective than one-sided information (e.g. Velde et al. 2018) unless presented in a refutational manner, including an explanation of how drawbacks can be outweighed by benefits (Cornelis et al. 2020).

Integrating attitudinal factors into our analysis allows for additional valuable insights. Consistent with our sixth hypothesis, animal welfare-conscious participants exhibit a higher WTP for the ‘open barn’ salami (vs. ‘no label’). Surprisingly, this holds true for health-conscious consumers as well. Thus, consumers appear to associate improved animal welfare with enhanced product healthiness, a finding also observed in other studies (Katt and Meixner 2020; Denver et al. 2021; Bradford et al. 2022). Additionally, health-conscious consumers are more likely to have a higher WTP for the ‘antibiotic-free’ salami (vs. ‘no label’), which supports our seventh hypothesis. Surprisingly, this effect also holds for animal welfare-conscious consumers, hinting at potential misconception regarding the effects of antibiotic reduction on animal health and welfare. Contrary to our fifth hypothesis, environmental consciousness does not significantly impact consumers’ decisions in any of the three choice situations. The lack of effect could be attributed to the fact that none of the labels directly address environmental sustainability aspects. More specifically, the negative environmental effects of ‘open barn’ production were not apparent from the label and only explicitly stated in the ‘label + conflict’ information text. Thus, there was likely a lack of awareness among those not exposed to the text, as consumers typically have limited knowledge of animal husbandry (Simons et al. 2018). Similarly, label perception did not influence consumer preferences, which contrasts the eight hypotheses. In summary, our study finds support for Hypotheses 6 and 7, while this is not the case for the fifth and eighth hypotheses.

Furthermore, regarding socio-demographics, our results are in line with the meta-analysis of Clark et al. (2017) in that WTP for the animal welfare product ‘open barn’ is higher among younger and better-educated respondents. Additionally, our analysis confirms the findings of Denver et al. (2021) and Denver et al. (2022), showing that better-educated individuals have a higher WTP for the ‘antibiotic-free’ product. Surprisingly, contrary to Denver et al. (2021), younger individuals, not older ones, exhibit a higher WTP for these products. Moreover, gender does not emerge as a significant factor influencing consumer preferences in our analysis.

Some limitations of the study must be mentioned. Hypothetical bias may be present due to the nature of the CV experiment. Moreover, social desirability (e.g. Van de Mortel and Murgo 2006) could influence consumer responses, potentially leading to an overstatement of their WTP. Thus, we acknowledge that our study may not capture consumers’ WTP with absolute precision, but it offers valuable insights into the relative importance of different sustainability attributes and the role of information on consumers’ WTP. Building on previous research, we addressed multiple conflicting goals in pig farming identified through expert interviews and a literature review. Nevertheless, there are other sustainability conflicts not considered in this research. Lastly, the number of respondents assigned to each condition was imbalanced, as about three times as many participants in the ‘label + conflict’ group gave a wrong answer to the attention check question in comparison to the two other groups and had to be removed. A plausible explanation is that the attention checks were not equally difficult across groups, or that the conflict information induced confusion (Sonntag et al. 2019; Grunert 2002). Although the groups did not differ significantly with respect to their level of education, we cannot rule out that the groups differed in other variables, such as prior knowledge or conscientiousness.

The present study offers a number of implications for the agricultural sector and suggests avenues for future research. First, it is important to recognize the sustainability preferences of consumers. When confronted with conflicting sustainability goals, a hierarchy of preferences can be identified. Public health emerges as the top priority, followed by animal welfare considerations, with environmentally sustainability ranked as the least important. Second, the study underscores that consumer differ in their preferences for different protected goods, influenced by psychographic and socio-demographic factors. Policymakers and stakeholders in the agricultural sector should consider these variations when formulating strategies and policies; for example, by balancing the promotion of different protected goods when developing comprehensive livestock strategies. Third, the study emphasizes the importance of adequate communication in discussions about sustainability goals. Information provision significantly influences consumer choices and their willingness to pay. While one-sided information may effectively convey sustainability dimensions, it is essential to acknowledge that its impact can be neutralized when accompanied by information about the associated costs to other sustainability dimensions. These insights provide valuable guidance for the implementation of more effective marketing strategies, the development of informed policy initiatives, and for the establishment of a policy framework for the future of animal husbandry.

Future research avenues should explore diverse preference elicitation methods to strengthen the validity and generalizability of our findings. Moreover, examining effective strategies for informing consumers about existing and potential sustainability trade-offs is essential. We emphasize that such strategies must carefully balance transparency and effectiveness, as our study suggests that increased transparency can sometimes reduce effectiveness. Exploring the impact of refutational information in contexts involving conflicting sustainability goals holds promise.

Footnotes

Acknowledgments

We would like to thank Lena Große Streine for her important contribution in conducting the study and Johannes Simons for his constructive input. In addition, we would like to thank Yan Meng, Philip M. Clarke, and Ilias Goranitis for providing insights into their analysis and coding scheme in Stata.

Funding

This research is funded by the Ministry for Environment, Agriculture, Conservation and Consumer Protection of the State of North Rhine-Westphalia (MULNV) within the work of the teaching and research focus ‘Umweltverträgliche und Standortgerechte Landwirtschaft’ (USL) (17-02.04.01-07/2018). Publication of this work was supported by the Open Access Publication Fund of the University of Bonn.

Conflict of interest

Authors declare no conflict of interest.

Data Availability

The data underlying this article will be shared on reasonable request to the corresponding author.

References

Appendix

Figure A2.1.

Price levels:|${\boldsymbol{B}}_{\boldsymbol{D}}^{\boldsymbol{h}}$|as well as follow-up|${\boldsymbol{B}}_{\boldsymbol{D}}^{{\boldsymbol{hh}}}$|and|${\boldsymbol{B}}_{\boldsymbol{D}}^{{\boldsymbol{hl}}}$|⁠.

Open in new tabDownload slide

Figure A2.2.

Price levels: |${\boldsymbol{B}}_{\boldsymbol{D}}^{\boldsymbol{l}}$|as well as follow-up|${\boldsymbol{B}}_{\boldsymbol{D}}^{{\boldsymbol{lh}}}$|and|${\boldsymbol{B}}_{\boldsymbol{D}}^{{\boldsymbol{lh}}}$|⁠.

Open in new tabDownload slide