Willingness to Pay for Cheese and Yogurt from Production Systems with Climate Change Mitigation

By 2022, the population aged 30 to 60 years in the Greater Metropolitan Area (GAM ⁽²⁾, by its acronym in Spanish) was 1,599,969 (INEC, 2022b). This population was selected as the target since GAM has the highest population growth and concentrates the largest economic activity in the country (Arias and Sánchez, 2012). Based on the researchers' criteria, a quota sampling method was applied (Parra and Vázquez, 2017). Equation (1) was used to estimate the sample size from the total population, which exceeded 10,000 observations, and equation (2) was subsequently used to determine the representative quotas for the obtained sample. (1) n=(Z2)⋅(p)(q)(EE2) n = {{({Z^2}) \cdot (p)(q)} \over {({EE}^2)}} (2) ni=nNiN {n_i} = n{{{N_i}} \over N} Where: n – is the sample size to be determined; Z – is the confidence level coefficient; p, is the probability in favor; q – is the probability against (1 – p); and EE, is the estimated margin of error; n_i – is the quota size proportional to the group; N_i – refers to the size of stratum i; and N – is the number of elements in the population.

For this research, the proportion to be detected was unknown, so a value of 50% was used. A confidence level of 90% and an allowed error of 5% were also applied. By substituting these values into equation (1), a sample of 270 observations was obtained, distributed as shown in the following table.

Data collection was conducted through visits to public parks and shopping centers within the study area, as well as by distributing online forms to individuals with heterogeneous characteristics residing in the study area.

A structured questionnaire comprising 39 items was developed and divided into three main sections (Table 2).

Data collection was conducted in public spaces (parks and shopping centers), as well as online within the study area. The collected responses were subsequently tabulated to enable descriptive analysis and estimation of WTP.

To calculate the WTP for fresh cheese and yogurt with distinguishing attributes – such as originating from production systems that mitigate environmental impact – two methods were employed.

According to Hernandez et al. (2019), contingent valuation (CV) involves applying surveys to determine the value that individuals assign to a good or service and the factors influencing that value. In this study, CV was used to elicit the maximum WTP for organic dairy products. A two-stage question format was applied: an initial binary (yes/no) WTP question, followed by an open-ended question to capture the maximum amount respondents were willing to pay. Product quantities were standardized to 500 g of fresh cheese and 1000 ml of yogurt.

Qualitative response models allow estimation of the probability of a specific event occurring (Hossain et al., 2022; Valencia et al., 2015). To assess the population's willingness to pay a premium for organic fresh cheese and yogurt, logit and probit models were applied.

Logit and probit models estimate the likelihood that respondents would pay a premium for organic dairy products. These binary choice models allow the probability of a positive WTP response to be modeled as a function of explanatory variables (Table 3).

According to (Gujarati and Porter, 2010), logit models are given by the logarithm of the ratio of the probabilities in favor of an event occurring (equation 4), based on a cumulative logistic distribution function (equation 3). (3) 1−Pi=11+eZi 1 - {P_i} = {1 \over {1 + {e^{Zi}}}} (4) Li= ln (Pi1−Pi)=Zi=β1+β2Xi {L_i} = \ln \left({{{{P_i}} \over {1 - {P_i}}}} \right) = {Z_i} = {\beta_1} + {\beta_2}{X_i} Where, the expected value ⁽³⁾ is given by E(y = 1) = p; P_i – is the probability that individual i is WTP; Z_i – is the linear combination of explanatory variables that affect the WTP; L_i – is the log-odds (logit) of the probability that individual i is WTP; β₁ – is the intercept term in the model; β₂ – is the coefficient of the explanatory variable X_i; X_i – is the value of the explanatory variable for individual i; and the variance is expressed as var(y) = p(1 – p) (Hill et al., 2011).

The slope coefficient of a variable represents the change in the logarithm of the odds in favor of an event occurring associated with a one-unit change in that variable, while holding other variables constant. Therefore, equation 5 is used to calculate the rate of change in probability at a given point (Gujarati and Porter, 2010). (5) ∂Pi∂Xki=βkPi(1−Pi) {{\partial {P_i}} \over {\partial {X_{ki}}}} = {\beta_k}{P_i}(1 - {P_i})

On the other hand, probit models follow a cumulative normal distribution (CND), represented by equation 6, while equation 7 represents the inverse of the CND (Glantz and Kissell, 2014). (6) F(X)=∫−∞X012σ2πe−(x−μ)2/2σ2 F(X) = \int_{- \infty}^{{X_0}} {{1 \over {\sqrt {2{\sigma^2}\pi}}}{e^{- {{(x - \mu)}^2}/2{\sigma^2}}}} (7) Ii=F−1(Pi)=β1+β2Xi {I_i} = {F^{- 1}}({P_i}) = {\beta_1} + {\beta_2}{X_i} Where: μ – is the mean and σ² – the variance.

The calculation of the rate of change of the probability in probit models requires the density function of the standardized normal variable (equation 8). (8) ∂Pi∂Xki=βkf(Zi)yZi=β0+β1Xi+…+βkXki {{\partial {P_i}} \over {\partial {X_{ki}}}} = {\beta_k}f({Z_i})y{Z_i} = {\beta_0} + {\beta_1}{X_i} + \ldots + {\beta_k}{X_{ki}}

According to Orrego et al. (1997), the consumer can maximize utility by incorporating the demand for environmental services. Therefore, based on duality theory, the following equality can be proposed to define a relationship C between an individual's income and expenditure (Solórzano et al., 2020). (9) v(p,y−C,s,q1)+ε1=v(p,y,s,q0)+ε0 v(p,y - C,s,{q_1}) + {\varepsilon_1} = v(p,y,s,{q_0}) + {\varepsilon_0} Where: C – maximum willingness to pay for the offered change (going from q₀ to q₁); y – represents income; p – is a price vector; s – is the socioeconomic characteristics vector.

Finally, the linear equations (equations 10 and 11) and their transformation (equation 13) were derived. Equations 10 and 11 result from differentiating the utility function according to duality theory (Solórzano et al., 2020). (10) v0=α0+βy+ε0 {v_0} = {\alpha_0} + {\beta_y} + {\varepsilon_0} (11) v1=α1+β(γ−C)+ε1 {v_1} = {\alpha_1} + \beta (\gamma - C) + {\varepsilon_1} (12) Δv=α1+β(γ−C)+ε0−ε0−α0−βy−ε0Δv=α+βC+η \matrix{{\Delta v = {\alpha_1} + \beta (\gamma - C) + {\varepsilon_0} - {\varepsilon_0} - {\alpha_0} - \beta y - {\varepsilon_0}} \cr {\Delta v = \alpha + \beta C + \eta} \cr} (13) C=α+ηβ C = {{\alpha + \eta} \over \beta} Where: v₀, represents the initial utility of the consumer before paying for the improved product; ɛ₀ – is the random error term; β – is the sensitivity to income; y, is the consumer income; v₁ – is the utility after paying the cost of the improved product (C); C – is the average willingness to pay; α₁ – is the new utility intercept; ɛ₁ – is the error term in the new scenario; and η – is the combined error term after differentiation.

Data analysis was carried out using specialized software such as Microsoft Excel from Office 365, and econometric estimates were performed with Gretl 2023a.

A total of 282 observations were collected in the GAM: 43% in San José, 26% in Alajuela, 16% in Cartago and 15% in Heredia. Of these, 57% of respondents were women and 43% were men, and 169 respondents held a university degree.

Figure 1 shows a positive asymmetry for the age variable, ranging from 19 to 78 years. This is evident both in Figure 2 and by the fact that the mode for this variable does not coincide with the mean or the median (Díaz and Pita, 2001).

Fig. 1.

Distribution of survey participants by age

Source: Own elaboration.

Fig. 2.

Distribution of responses related to household income in US dollars (USD)

Source: own elaboration based.

The average family size was three persons, and 29.08% of respondents reported a family income ⁽⁴⁾ ranging between USD 997.97 and USD 1995.93. 26.24% reported income between USD 1995.94 and USD 3991.86, while 13.38% reported income between USD 498.98–USD 997.96 and USD 3991.87–USD 5987.78 (Fig. 3). These values coincide with the III and IV quintiles of average household income reported by INEC (INEC, 2023). In addition, the average household food expenditure was USD 516.22, consistent with the aforementioned quintiles.

Table 4 shows that 95% of respondents consume dairy products, with 92% consuming fresh cheese and 81% yogurt. Likewise, 64% indicated willingness to pay (WTP) a premium for fresh cheese, and 63% for organic yogurt. Further details on WTP are discussed below.

Respondents who consume fresh cheese and yogurt reported weekly average intakes of 533.45 g and 732.88 ml, respectively. The preferred sizes were 500 g for fresh cheese and 750 ml for yogurt (Table 4). The initial WTP was USD 6.08 for 500 g of fresh cheese and USD 5.54 for 1,000 ml of yogurt. When asked about products from production systems implementing environmental mitigation strategies, respondents reported higher WTP values: USD 6.43 and USD 10.48 for 500 g and 1000 g of organic fresh cheese, respectively, and USD 6.65 and USD 2.40 for 1,000 ml and 200 ml of organic yogurt, respectively.

A total of 94% of respondents expressed a moderate to high preference for fresh cheese, while 86% reported a similar preference for yogurt. The characteristics of fresh cheese rated as “very” or “very important” by at least 55% of consumers were: absence of intense smell, light flavor, product availability, variety of presentations, clear labeling, information on health benefits, and price. For yogurt, characteristics that reached the 55% threshold included animal welfare, protein content, product availability, clear labeling, information on health benefits, and price.

It is also noteworthy that consumers place high importance on sustainability in dairy production. Specifically, between 73% and 81% of respondents rated the following attributes as “very” or “very important”: products from farms that implement efficient water use (73%), are carbon neutral (69%), are deforestation-free (79%), and guarantee animal welfare (81%).

46% of respondents associate the term “organic” with the category “environment,” which includes responses such as water, air, environmental protection, forests, green, global warming, and environmental mitigation strategies. The next most frequent categories are “other” and “health,” each with 11% of responses. The “other” category includes words such as help, change, lie, decrease, effort, exclusive, easy to use, fresh, free, clean, and saltier, while the “health” category includes words such as health, light, clean, healthier, diet, nutritious, and vitamins. Table 6 presents responses in the remaining categories: “packaging”, “sustainability”, “price”, without additives”, “wellness”, “artisanal”, “quality”, “commitment”, “agricultural” and “don't know.”

To model the WTP for organic fresh cheese and yogurt, the binary qualitative response models – probit and logit – were used, with variables described in Appendix A. These models help control for potential response bias, as the predicted probabilities are always constrained between 0 and 1, with the partial effect varying depending on the parameters (Solórzano et al., 2020).

Table 4 shows that the willingness to pay a premium for organic fresh cheese is primarily influenced by the following variables: lactose-free, absence of odor, light flavor, household food expenses per capita, price, and the implementation of environmental impact mitigation strategies at the farm level.

Except for the price variable, all coefficients in both the logit and probit models are statistically significant and have the expected signs. Price was retained because it improves the goodness-of-fit measures. Both models achieve correct case prediction percentages above 70%; however, the probit model was selected as the preferred specification. Although the McFadden R2, which is intended to demonstrate that the independent variables have sufficient explanatory power given that the restricted model is not the same as the unrestricted model, is relatively low (0.1819), the likelihood ratio ⁽⁵⁾ has a p-value of 0.0000, indicating that the null hypothesis is rejected and the slope coefficients are simultaneously different from zero (Gujarati and Porter, 2010). The probit model also shows lower values for the Schwarz Criterion, and Akaike and Hannan Quinn criteria, and its predictive capacity (74.81%) is supported by an area under the curve of 0.762 for the binary probit, which ranges between 0 and 1 – the closer it is to the upper limit, the greater the model's ability to distinguish between the two classes (Fawcett, 2006; Martínez Pérez and Pérez Martin, 2023).

Based on the mean of the marginal changes for all observations, the probability that a consumer is willing to pay a premium for organic fresh cheese is 67%. This probability could decrease by 0.068 percentage points (pp) if the variable DESL increases by one category; increase by 0.110 pp if AO increases by one category; increase by 0.0690 pp if the variable SL increases by one category; increase by 0.1421 pp if the variable SP increases by one category; decrease by 0.0529 pp if P increases by one category; and finally, increase by 0.0001 pp if AHPC000 increases by one category. On the other hand, if interpreted in a generalized way from the odds ratio, the probit model presents an odds ratio of 2.04, indicating that the willingness to pay a premium for organic fresh cheese is 2.04 times more likely to be accepted by the consumer than not.

The significant coefficients with the expected signs for willingness to pay a premium for organic yogurt are light odor, light flavor, information on the production process, and the implementation of environmental impact mitigation strategies at the farm level. Both the logit and probit models achieved correct case prediction percentages above 70%; however, the logit model was selected as the one that best fits the data. Both models show a low McFadden R2 (0.1224), but the likelihood ratio has a p-value of 0.0000, rejecting the null hypothesis and indicating that the slope coefficients are simultaneously different from zero (Gujarati and Porter, 2010). The model's predictive capacity (71.24%) can be appreciated in Figure 6, which shows an area under the curve of 0.732 for the binary Logit.

Based on the mean of the marginal changes for all observations, the probability that a consumer is willing to pay a premium for organic yogurt is 63%. This probability could increase by 0.010 pp if OL increases by one category; increase by 0.066 pp if SLY increases by one category; decrease by 0.080 pp if IPRDY increases by one category; and increase by 0.131 pp if SP increases by one category. On the other hand, if interpreted in a generalized way from the probability ratio, the probit model shown above presents an odds ratio of 1.70, indicating that the willingness to pay a premium for organic yogurt is 1.70 times more likely to be accepted by the consumer than not.

Based on the coefficients from the models presented in the previous sections and using equation 19, the average willingness to pay was estimated as USD 6.35 for 0.5 g organic fresh cheese and USD 7.51 for 1,000 ml organic yogurt.