5.1. Survey design and implementation

The survey was organized into four parts that collected information about households: (1) socioeconomic demographic characteristics; (2) consumption of beef products; (3) knowledge, opinions, and experiences about beef quality; and (4) responses to two sequential stated choice experiments to assess their preferences for intrinsic and credence attributes (see Table 1). Questions in part 3 of the survey were designed to test hypothesis 1. The choice experiments in part 4 were developed to test hypotheses 2 and 3.

Table 1. Beef attributes selected for the study (intra-attributes in parentheses)

^a In this case, they all refer to credence attributes.

The survey was administered in person to 547 randomly selected households in the cities of Guayaquil and Santo Domingo during 2016. Guayaquil is the largest city in the country and is located in the coastal region. This city also accounts for 20% of total national beef expenditures, according to data from the 2011–2012 national survey on income and expenditures of rural and urban households (ENIGHUR). Santo Domingo is a middle-sized city located in the foothills of the Andes. It has special participation in the beef supply chain of Ecuador because it has the largest cattle market in the country.

The survey was directed to middle-income households,Footnote ⁴ and the interviewed person was the household member in charge of making beef purchases. Because the quality of beef experienced at consumption can also be altered by the way the meat is prepared (Grunert, Bredahl, and Brunso, Reference Grunert, Bredahl and Brunso2004), this study concentrates only on fresh meat purchased at the market for cooking at home.

5.2. Choice experiments’ design

The survey instruments included two sequential choice experiments for all consumers. Each choice experiment included multiple choice scenarios. The first choice experiment mirrors the current situation where consumers only observe and use intrinsic attributes and price for the selection of beef products. The second choice experiment also included information about credence attributes, in addition to price and intrinsic attributes, to reflect the potential scenario where consumers receive this additional information. Moreover, the second experiment included two levels of information provided to two randomly selected consumer groups: basic and more advanced information about the credence attributes.

The first choice experiment included only two intrinsic attributes (color and texture) and the price of “top round” beef cuts.Footnote ⁵ Beef consumers were asked to select between two product profiles (choice scenario) that differ across the intrinsic characteristics and price. Consumers could also choose a neither option. A total of six choice scenarios were given to each consumer in the first choice experiment. The color attribute included three levels (intra-attributes): dark red, bright red, and pale. The texture attribute included two levels: few muscle fibers and visible muscle fibers. The price attribute included four potential prices: $2.86/lb., $3.60/lb., $4.45/lb., and $5.31/lb.Footnote ⁶ These price levels were defined based on a sample of prices from 12 retail locations (supermarkets, meat supermarkets, municipal markets, and convenience stores) collected in Guayaquil and Santo Domingo between March and August 2016. The overall average value of prices during the period was $4.05. Average prices in Santo Domingo were about 8% lower. Therefore, price levels in the experiment were set by using price values about 10% and 30% above and below the average observed prices in both cities ($4.05).

The second experiment included the same intrinsic attributes as in the first experiment, the price levels, as well as the credence attributes (sanitary control, meat maturation, animal welfare, and traceability), which were selected based on the results of the focus groups and expert interviews. The levels for each credence attribute were the presence or absence of the attribute. As in the case of the first choice experiment, consumers were asked to select between two products that differ across the intrinsic and credence characteristics and price. Consumers could also choose a neither option. They were given a total of eight choice scenarios in this choice experiment. Figure 1 shows an example of one of the choice scenarios presented to the consumers.Footnote ⁷ With this second experiment, we expected to see how consumers value information on credence attributes (i.e., information that signals better-quality beef, hypothesis 2) and if consumer preferences for these intrinsic attributes change when being exposed to this type of information.

Figure 1. A scenario of the second choice experiment.

We also introduced a variation to the second choice experiment to evaluate the effect of education on preferences for the credence attributes. One-half of the sample of consumers was shown the credence attributes with only basic information about such attributes. This was done for the purpose of reflecting the situation of places where they have appropriate labels, but where consumers do not necessarily understand what that information means. The other half of the sample was provided educational information related to the credence attributes. The purpose of this variation is to test the effect of educational information on consumers’ WTP for credence attributes (hypothesis 3). The educational information included a brief explanation of each credence attribute and the links between the attribute and the effect it might have on beef quality. Details on this information are in the Appendix (Figure A1).

SAS software was used to create the experimental question design. The combination of all attributes and levels for the first experiment resulted in a total of 24 (4 × 3 × 2) possible product profiles and 276 possible choice scenarios (C²₂₄), where $C_n^r$ denotes the number of unordered subsets (i.e., combinations) of n objects taken r at a time (Hogg, Tanis, and Zimmermann, Reference Hogg, Tanis and Zimmermann2001). The number of product profiles and choice scenarios for the second experiment was 384 (4 × 3 × 2 × 2 × 2 × 2 × 2) and 73,536 (C²₃₈₄), respectively. Hence, fractional factorial designs were applied to choose 18 choice scenarios for the first experiment and 24 for the second experiment by comparing the D-efficiency for each combination of choice scenarios. Finally, the designs were blocked into three different versions of the questionnaires where each respondent was offered only six choice scenarios in the first experiment and eight choice scenarios in the second experiment.

5.3. Data analysis and models

Choices made by survey respondents were modeled assuming a random utility model. The utility of each choice depends on the identified product attributes (price and intrinsic and credence beef attributes). For consumer i choosing between C alternatives in choice occasion t, the utility of choice c (U_ict) is the following:Footnote ⁸

$${U_{ict}} = {V_{ict}} + {e_{ict}},$$(1)

where i = 1, …, I; c = 1, …, C; t = 1, …, T. V _ict is the portion of utility related to the identified product attributes, and e _ict reflects the effect of factors not included in V _ict (e.g., consumers’ knowledge, perceptions, etc.). It is further assumed that e _ict is i.i.d. (independent and identically distributed), and distributed extreme value and the variance of e _ict can differ across consumers, so $Var\left( {{e_{ict}}} \right) = k_i^2\left( {{{{\pi ^2}} \over 6}} \right),$ where K _i is a scale parameter corresponding to consumer i. Intuitively, this consumer-specific scale parameter reflects the variability of utility across choice situations for each consumer because, for example, unobserved factors can differ across choice situations (Train and Weeks, Reference Train, Weeks, Scarpa and Alberini2005).

Assuming a linear functional form for the parameters in the deterministic component of utility, and specifying utility as separable in price, p _ict , and nonprice attributes, X _ict , equation (1) can be rewritten as follows:

$${U_{ict}} = - {\alpha _i}{p_{ict}} + {{\bf{\beta '}}_i}{{\bf{X}}_{{\rm{ict}}}}{\rm{ + }}{{\rm{e}}_{{\rm{ict}}}},$$(2)

where β _i and α _i are individual-specific utility parameters corresponding to the nonprice and price attributes, respectively. Dividing equation (1) by the scale parameter k _i results in a new error term (ϵ _ict) with constant variance $({{{\pi ^2}} \over 6})$ that corresponds to the traditional indirect utility model in preference space (Train and Weeks, Reference Train, Weeks, Scarpa and Alberini2005):

$${U_{ict}} = - ({\alpha _i}/{k_i}){p_{ict}} + {\rm{ (}}{\bf \beta} /{k_i}{\rm{)}},{X_{ict}} + ({e_{ict}}/{k_i}) = - {\gamma _i}{p_{ict}} + {\rm{ }}{\tau '_i}{X_{ict}} + {\varepsilon _{ict}}.$$(3)

Because the WTP for an attribute is the ratio of the attribute’s coefficient to the price coefficient, W_i = τ_i /γ_i , equation (3) can be reparameterized if we multiply and divide τ_i by ${\gamma _i}({\rm{i}}.{\rm{e}}.,{{{\gamma _i}} \over {{\gamma _i}}}{\tau _i} = {\gamma _i}{w_i})$ :

$${U_{ict}} = - {\gamma _i}{p_{ict}} + ({\gamma _i}{W_i})'{X_{ict}} + {\varepsilon _{ict}},$$(4)

which is called the model in WTP space, and W_i is the vector of WTP values for all the attributes (Train and Weeks, Reference Train, Weeks, Scarpa and Alberini2005). We adopted the model in WTP space instead of the model in preference space because this approach tends to fit the data better and results in more plausible estimates of the WTP values for the nonprice attributes (Scarpa, Thiene, and Train, Reference Scarpa, Thiene and Train2008).Footnote ⁹ Given the assumption that each ϵ_ict is i.i.d. extreme value, the probability that consumer i chooses alternative c in choice occasion t, conditional on the coefficient vector ${\theta _i} = [{\gamma _i}{W'_i}]'$ , is (Revelt and Train, Reference Revelt and Train1998):

$${P_{ict}}\left( {{\theta _i}} \right) = {\rm{ }}{{{e^{{{\rm{V}}_{ict}}({\theta _i})}}} \over {\sum\nolimits_c {{e^{{V_{ict}}({\theta _i}),}}} }}$$(5)

where ${V_{ict}}({\theta _i}) = - {\gamma _i}{p_{ict}} + ({\gamma _i}{W_i})'{X_{ict}}$ Furthermore, conditional on θ_i , the probability of consumer i’s observed sequence of T choices is then (Train, Reference Train1998):

$${S_i}\left( {{\theta _i}} \right) = {\prod _t}{\rm{ }}{P_{ic(i,t)t}}\left( {{\theta _i}} \right),$$(6)

where c(i,t) denotes the specific alternative c that consumer i selects in choice occasion t. The coefficient vector θ_i is unobserved for each consumer i and varies in the population with density f( θ_i |Γ) where the parameters of the distribution of θ_i are Γ. Thus, the unconditional probability of the observed choice sequence (i.e., the mixed logit choice probability) is as follows:

$${P_i}\left( \Gamma \right) = \smallint {\rm{ }}{S_i}({\theta _i})f({\theta _i}|\Gamma )d{\theta _i}.$$(7)

The log-likelihood function for all n consumers is LL(Γ) = ∑ _i ln P_i (Γ). Estimation was carried out using simulated maximum likelihood procedures using STATA software (Rigby and Burton, Reference Rigby and Burton2006; Train, Reference Train1998, Reference Train2003). Regarding the distribution of the coefficients in θ_i , the price coefficient was specified to be lognormal, and the WTP distributions for all nonprice attributes were assumed to be normal.

5.4. Testing differences in willingness-to-pay coefficients across experiments and groups of respondents with varying information levels

Estimation of model coefficients was carried out combining respondents’ data from the first and second experiments. This approach allow us to test for differences in WTP coefficients both across the sequential choice experiments (as the credence attributes are included in the product profiles in addition to the intrinsic attributes) and across groups of consumers with different information about the credence attributes. Denoting the vector of intrinsic attributes as a_ict and the vector of credence attributes as Z_ict, the complete vector of nonprice attributes used for model estimation is ${X_{ict}} = {\rm{ }}{\left[ {{{a'}_{ict}}{{a'}_{ict}}x{d_{exp2}}{{z'}_{ict}}x{d_{exp2}}{{z'}_{ict}}x{d_{exp2}}x{d_{information}}} \right]^\prime }$ . In addition to the vector of intrinsic attributes a_ict , X_ict includes an interaction between the intrinsic attributes and a dummy variable for the second experiment (d _exp2) ${a'_{ict}}x{d_{exp2}}$ , and an interaction between credence attributes and the second experiment dummy ${z'_{ict}}x{d_{exp2}}.\,{X_{ict}}$ . also includes an interaction between the credence attributes, the second experiment dummy and a dummy variable (d _information ) to differentiate households that received additional information regarding the credence attributes: ${z'_{ict}}x{d_{exp2}}x{d_{information}}$ . The vector of intrinsic attributes a_ict , is included by itself and interacted with the dummy for the second experiment because these attributes were present in both experiments. The vector of credence attributes z_ict is only included interacted with the dummy for the second experiment because these attributes were not included in the first experiment. Similarly, the information dummy is only interacted with ${z'_{ict}}$ and d _exp2 as the information was only presented after the first experiment and focused on the credence attributes.

The corresponding complete vector of WTP coefficients in this specification, ${W_i} = [w{a'_i}w{a'_i}_{_,\,exp2}w{z'_i}_{,\,exp2}wzin{f'_i}_{_{,\,}exp{\bf{2}}}]'$ , includes WTP coefficients for credence attributes in the first experiment ${a'_{ict}}$ $(wa'_{i})$ , the interaction terms ${a'_{ict}}x{d_{exp2}}(w{a'_i}_{,exp{\bf{2}}}),{z'_{ict}}x{d_{exp2}}(w{z_{i,exp{\bf{2}}}})'$ and ${z'_{ict}}x{d_{exp2}}$ $x{d_{information}}(wzin{f'_{i,exp{\bf{2}}}});w{a'_i}$ represents WTP values for the intrinsic attributes in the first experiment, ${w'_{i, exp2}}$ reflects differences in WTP values for the intrinsic attributes in the second experiment relative to the first experiment, $w{z'_{i, exp2}}$ is the WTP values for the credence attributes which were only presented in the second choice experiment, and $wzin{f'_{i,exp2}}$ represents differences in WTP for the credence attributes among the group of respondents that received additional information relative to the group that only received basic information in the second experiment.

The use of the combined data from both experiments generates three options for the specification of the price coefficient (γ_i in equation 4). The alternative specifications are considered to evaluate changes in the price coefficient across experiments, as well as the sensitivity of the results to different assumptions regarding these coefficients. The first option assumes a constant price coefficient (γ_i = α_i/k_i ) for the first and second choice experiment:Footnote ^10, Footnote ¹¹

$${U_{ict}} = - {\gamma _i}({p_{ict}}x{d_{exp1}} + {p_{ict}}x{d_{exp2}}) + {\rm{ }}{\left( {{\gamma _i}{W_i}} \right)^\prime }{{\bf{X}}_{ict}} + {\varepsilon _{ict}},$$(8)

where the dummy d _exp1 differentiates data from first experiment. The price coefficient in this specification is estimated using information from both experiments. The other two specifications include, in addition to the price term p_ict , interactions between the price and dummies for the second (equation 9) and first experiment (equation 10):

$${U_{ict}} = - {\gamma _{i,exp1}}{p_{ict}} + ({\gamma _{i,exp1}}{r_{i,exp2}}){p_{ict}}x{d_{exp2}} + {\rm{ (}}{\gamma _{i,exp1}}{W_{iexp1}})'{{\bf{X}}_{ict}} + {\varepsilon _{ict}},$$(9)

$${U_{ict}} = - {\gamma _{i,exp2}}{p_{ict}} + ({\gamma _{i,exp2}}{r_{i,exp1}}){p_{ict}}x{d_{exp1}} + ({\gamma _{i,exp2}}{W_{iexp2}})'{{\bf{X}}_{ict}} + {\varepsilon _{ict}},$$(10)

where the coefficient γ_i,expk denotes the price coefficient for the kth experiment (k = 1, 2); r_i,expk = γ_i,expj / γ_i,expk ∀j≠k, the ratio between the price coefficient for the jth and kth experiment (j = 1, 2); and W _iexpk denotes WTP coefficients based on the kth experiment price coefficient. Note that whereas equation (8) assumes and imposes a constant price coefficient for both experiments, equations (9) and (10) allow for the estimation and testing for differences in price coefficients across models but use only one price coefficient for the estimation of WTP coefficients: equation (9) uses the price coefficient from the first experiment, and equation (10) uses the price coefficient from the second experiment.

5.5. Factors affecting consumers’ willingness to pay for beef attributes

The analyses of the relationship between consumers’ WTP values for the beef attributes and consumers’ characteristics were carried out using a two-step approach. In the first step, we estimated individual consumers’ WTP values for each beef attribute. In the second step, we used regression analyses to explore the association of consumer-related factors and beef attributes’ WTP values. Estimation of individual consumers’ WTP values is based on an application of Bayes’s rule. More specifically, the density of each θ_i (which includes the WTP coefficients) conditional on the individual’s sequence of choices and the population parameters is given by (Hess, Reference Hess2007; Revelt and Train, Reference Revelt and Train1999):

$$h\left( {{\theta _i}|\Gamma } \right) = {{{S_i}\left( {{\theta _i}} \right)f\left( {{\theta _i}|\Gamma } \right)} \over {{P_i}\left( {{\rm{ }}\Gamma } \right)}},$$(11)

so the expected value of θ_i is given by

$$E({\theta _i}|\Gamma ) = \smallint {\rm{ }}{\theta _i}h({\theta _i}|\Gamma ),$$(12)

and the simulated approximation to the individual vector of values for the attributes is therefore

$$\hat E({\theta _i}|\Gamma ) = {{\sum\nolimits_r {{\theta ^r}} {S_i}\left( {{\theta ^r}} \right)} \over {\sum\nolimits_r {{S_i}} ({\theta ^r})}},$$(13)

where θ^r is the rth draw from the population density f( θ_i |Γ), and S_i θ^r is the probability of individual i’s sequence of choices. Estimated parameters $\hat \Gamma $ were used instead of the parameters Γ (Hess, Reference Hess2007). The stability of the estimated Ê( θ_i |Γ) values was verified using various sizes for the number of sample draws. The empirical results are based on 1,000 draws.

Regarding the regression analyses, we used the following random effect regression model for the credence attributes (Campbell, Reference Campbell2007):

$$WT{P_{ia}} = {c_a} + {\bf{z}}_{\bf{i}}^{'}{\bf{b}} + {u_n} + {e_{ia}},$$(14)

where WTP_ia is the ith household WTP for attribute a (i.e., a = sanitary control, meat maturation, animal welfare, traceability). c_a and b are coefficients, Z_i is a vector of household-related factors, u_n is a household-specific random error, and e_ia is an idiosyncratic error term. This approach accounts for the panel nature of the data and provides estimates of (average) marginal effects of factors on the WTP values for a group of attributes. Households’ factors in X_i included households’ characteristics as well as variables related to knowledge, perceptions, and experiences regarding beef production and consumption (Mennecke et al., Reference Mennecke, Townsend, Hayes and Lonergan2007).

6.1. Summary of sociodemographic characteristics

A female was the person in charge of food purchases in 78% of surveyed households, although this percentage is larger in the case of Santo Domingo (see Table 2). The level of education of the females in the sample was in most cases high school (61%), followed by a bachelor’s degree (32%). In the case of men, the majority had a bachelor’s degree (56%), followed by 34% who had only a high school education. Furthermore, and consistent with census data (Instituto Nacional de Estadísticas y Censos [INEC], 2010), the level of education is higher among respondents from Guayaquil than among those from Santo Domingo (Table 2). Average household size was four members in both cities (Table 2), similar to the 2010 census average (3.80 for Guayaquil and 3.86 for Santo Domingo).

Table 2. Description and summary statistics of respondents’ characteristics and knowledge, experience, and perceptions

6.2. Consumption of beef products

Summary statistics show that 80% of the households consume beef at least once a week (Table 2). When asked about the cuts of beef purchased most frequently, 44% of respondents gave among their answers, cuts such as “soft meat,” “filleted meat,” and “meat with bone,” among others that do not correspond to actual beef cuts. This percentage is much larger in Santo Domingo (64%) than in Guayaquil (24%), in spite of the fact that cattle farming and slaughtering are much more important components of Santo Domingo’s economy.Footnote ¹² This reveals the lack of a culture on beef cuts even among those closer to the beef cattle sector. Moreover, the smaller percentage values in Guayaquil may be because of the higher levels of education and the presence of larger supermarkets and high-end meat stores in this city relative to Santo Domingo. For those who did answer actual beef cut names, the most popular beef cuts are outside round and tenderloin (both 19%), followed by top round (16%), rump (13%), ribs (12%), and loin (10%). However, there are differences between the two cities. In Guayaquil, top round is the most popular cut (26%), whereas in Santo Domingo it is both tenderloin and outside round (16% each).

Average monthly beef expenditure per household is $38 for the whole sample, but this average is larger in Guayaquil ($42/month) than in Santo Domingo ($33/month) attributable in part to higher meat prices in Guayaquil. The most visited places of purchase are supermarkets in Guayaquil (33% of the sample), followed by meat supermarkets (23%), and then by municipal markets (20%). On the other hand, municipal markets are the preferred outlet in Santo Domingo (47%), followed, although not closely, by supermarkets (23%) and then by small meat markets (18%; Table 2).

6.3. Knowledge, opinions, and experiences about beef quality

Only 7% of respondents indicated that they would like to increase beef consumption, but they do not consume more because of concerns about the health effects of beef consumption (50%) and high price (24%; see Table 2 for results differentiated by location). With respect to the group of households that would not like to increase beef consumption (93%), the main reasons for this were health concerns about beef consumption (58%), satisfaction with current levels of consumption (37%), and current health problems (21%). For both groups of individuals (those wanting and not wanting to increase beef consumption), concerns for animal welfare, the environmental impact of the cattle business, and dissatisfaction with local beef attributes were not among the most commonly cited reasons limiting beef consumption.

Regarding expectations for future consumption, 72% of the sample think their beef consumption will remain the same in the future, 24% think it will decrease, and only 3% expect it to increase. For those who expect beef consumption to decrease in the future, the most popular reasons were again concerns about the health effects of beef consumption (53%) and current health problems (49%).

The majority of survey respondents perceive the quality of the beef they consume as good or very good (about 79% of consumers), about 20% perceive its quality as fair, and only 1% of the respondents perceive its quality as bad or very bad. This is so, despite the fact that a large proportion of households (52%) have experienced disappointments when purchasing beef (either sometimes or oftentimes have ended up purchasing tough meat), and that 62% have avoided purchasing beef because they did not like the beef available at the first place they visited. These results are very similar across cities (Table 2).

Overall, these results do not support hypothesis 1 and a likely explanation for this is that, as noted earlier, Ecuadorian consumers do not have much knowledge about good-quality beef. These results also suggest that the relatively low per capita beef consumption in Ecuador does not seem to be driven by a poor overall quality perception of beef, but rather by a perception of beef as not very healthy: 61% of our sample expressed health concerns about beef consumption as reasons limiting either their current or future consumption.

In terms of households’ knowledge about beef credence attributes, the majority of survey respondents (at least 60%) were not very familiar with any of them. Traceability is the concept respondents seem to be the least familiar with (71% do not know anything about it), followed by meat maturation (61%). On the other hand, sanitary control shows the largest percentage of familiarity because 72% are familiar with this concept (Table 2).

About 90% of the respondents agreed that a quality certification (e.g., of sanitary control or animal welfare) would stimulate producers and wholesalers to improve the quality of beef. With respect to the most trusted institutions to provide quality certifications, a majority of households indicated they are more confident in international private organizations (about 53% indicated they are very confident about these institutions) compared with local private organizations (45%) and national government organizations (34%). Results are similar between the two cities, except for trust in a governmental institution, which is much lower in the case of Guayaquil (24% vs. 44% in Santo Domingo).

6.4. Mixed logit model results

The results of our baseline mixed logit estimation results (model 1), based on equation (9) and that uses the price coefficient of experiment 1, are shown in Table 3.Footnote ¹³ This model was selected as it resulted in the highest log-likelihood value among the three models considered and also generated the most plausible and conservative estimates of the mean of the WTP distributions. The model with the combined price coefficient (equation 8) resulted in WTP mean coefficients about twice as large (in absolute value) as those obtained with equation (9). The model using the second experiment price coefficient (equation 10) resulted in WTP mean coefficients that were about five times larger (in absolute value) than those obtained using equation (9). The overall relative magnitude of the nonprice coefficients was nonetheless consistent across models.Footnote ¹⁴

Table 3. Results of the mixed logit model in willingness-to-pay (WTP) space

Notes: Asterisks (***, **, and *) indicate significance the 1%, 5%, and 10% level, respectively. ASC, alternative specific constant.

^a dexp2 is a dummy variable denoting the second experiment.

^b dinformation is a dummy variable identifying households that received additional information about credence attributes.

^c All the coefficients of nonprice attributes correspond to parameters (mean and standard deviation) describing the distribution of WTP values for the attributes.

Table 3 also includes a restricted version of model specification (model 1) without interactions between all the credence attributes included in experiment 2 and the dummy variable to differentiate households that received additional information regarding the credence attributes (sanitary control, meat maturation, animal welfare, and traceability). A log-likelihood ratio test was used to compare the restricted model (model 2) with the unrestricted model (model 1), where the restricted model hypothesizes no effect of additional information on consumers’ WTP for the credence attributes. The null hypothesis that the restrictions are valid was rejected ; thus, this provides evidence that information efforts can affect WTP values for credence attributes.

Focusing on the parameters of the interactions between intrinsic attributes and the experiment 2 dummy (d _expe2 ) (in models 1 and 2), all the parameters are significant at the 0.05 level. Moreover, the magnitude of the coefficients suggests a reduction in the absolute value of the WTP mean estimates for the intrinsic attributes once information about the extrinsic attributes is added to the experiments. This is strong evidence that WTP values for product attributes differ depending on the attributes considered in an experiment, even if a constant price coefficient is assumed for the estimation of WTP values. The significance of the coefficient related to the price term interacted with the experiment 2 dummy in models 1 and 2 also provides evidence of differences of price coefficients across experiments.

6.5. Willingness-to-pay values

Table 4 summarizes WTP mean values across experiments estimated using the results of model 2 in Table 3. When considered alone (in experiment 1), mean WTP premiums and discounts for intrinsic attributes are very high. Relative to beef of pale color, consumers are willing to pay, on average, $4.87/lb. more for bright red beef and about $4.20/lb. less for dark red beef. Consumers are also willing to pay, on average, $3.53/lb. more dollars for beef with few muscle fibers than for beef with visible muscle fibers. Given the observed average market prices of $4.05/lb., these values suggest a strong preference for products with high-quality intrinsic attributes (bright red beef and few muscle fibers).

Table 4. Estimated mean willingness-to-pay (WTP) values ($/lb.)

Note: Standard errors of the mean WTP values are in parentheses.

WTP premium and discount values for the intrinsic attributes are reduced significantly once the credence attributes are included in the second choice experiment; however, in reference to current market prices, they are still important. For example, the mean WTP premium for bright red beef relative to pale color beef gets reduced to about $2.20/lb., less than half the WTP premium found in the first experiment. The estimated WTP values for the credence attributes also suggest that consumers are willing to pay premiums for these attributes. For consumers with no additional information regarding credence attributes, sanitary control was the credence attribute with the highest estimated mean WTP premium ($2.69/lb.), followed by animal welfare ($2.31/lb.), beef maturation ($0.62/lb.), and traceability ($0.36/lb.). WTP values for sanitary control, maturation, and traceability increase when additional information is provided. On the other hand, the mean WTP premium for animal welfare decreases with additional information. The result of beef maturation more than duplicates with additional information, making it now the credence attribute with the second-highest mean WTP. The final relative importance of WTP values for sanitary control and maturation in the second experiment could be revealing that consumers have a higher preference for beef products with attributes more closely (or easily) associated with the sensorial characteristics of beef and with human health. Overall, these results provide evidence of Ecuadorian consumers’ strong preferences for high-quality intrinsic and credence beef attributes.

6.6. Random effects regression results

Table 5 presents the results of the regression analysis for the relationship between WTP values for the credence attributes and consumers’ characteristics (see Table 5 for descriptive statistics). Dummy variables were included for sanitary control, meat maturation, and animal welfare; therefore, traceability is the baseline attribute.

Table 5. Credence attributes willingness-to-pay regression results

Note: Asterisks (***, **, and *) indicate significance the 1%, 5%, and 10% level, respectively.

We find that even after controlling for a large set of household-related factors, there are large differences in households’ WTP for sanitary control, meat maturation, and animal welfare relative to the WTP for the traceability attribute. Households’ characteristics like age, college education, and income were found to have a statistically significant association with households’ WTP for the credence attributes included; however, only education was found to be economically important: college-educated respondents are willing to pay $0.22 more for the credence attributes. On the other hand, each additional year of age is only found to be associated with a $0.004 decrease in the WTP values. Similarly, at the average income level ($1,165) an additional $100 of income is found to be associated with an increase in the WTP for the credence attributes of only $0.014. It is also important to point out that we did not find evidence of a statistically significant association between WTP for the credence attributes and location of the household.

We also found that negative perceptions about the healthiness of beef products and dissatisfaction with the beef quality available in the market have statistically and economic significant associations with households’ WTP for credence attributes. Households that perceive beef products as unhealthy are willing to pay $0.12 less for the credence attributes relative to households that do not have this perception. The WTP for credence attributes of households that are not satisfied with the quality of local beef (only 5% of the sample) is $0.40 lower than the WTP of the rest of households. These results are contrary to expected and can be understood as a paradox reflecting consumers’ rejection of meat products (Latvala, Reference Latvala2010). In other words, when perceptions are very negative, consumers may tend to reject the product, and this discourages their WTP.