Measuring Eco-Consumption Tendency as a Conjoint-Based Latent Variable

• Published in: Advances in Consumer Research Vol. 48; pp. 374 - 376
• Main Authors: Gorissen, Karen, Weijters, Bert, Baumgartner, Hans
• Format: Conference Proceeding
• Language: English
• Published: Urbana Association for Consumer Research 01.01.2020
• Subjects: Climate change; Consumer behavior; Consumers; Greenhouse gases; Sustainability
• ISSN: 0098-9258

Scientists agree that human activities are the primary driver of climate change, and shifting consumer choices toward more sustainable options can significantly reduce greenhouse gas emissions (Beckage et al., 2018). Against this backdrop, marketing researchers often want to measure eco-consumption tendencies (ECT), defined as consumers' inclination to select eco-friendly options from a choice set. Lange, Steinke, and Dewitte (2018) point out that most ECT studies rely exclusively on self-report measures of behavior or proposed psychological antecedents (e.g., intentions or attitudes). However, it is well-established in the methodological literature that responses to self-report items contain substantial individual variation that is not content-related but stylistic in nature (Baumgartner & Steenkamp, 2001). For instance, some respondents tend to agree with items irrespective of what is being measured, while others do not. Furthermore, self-report scales are susceptible to social desirability, especially if the content relates to morally charged issues (Steenkamp, De Jong, & Baumgartner, 2010). Finally, an essential limitation of self-report scales is the limited variation in the presented stimuli. That is, traditional self-report scales typically ask consumers to rate how likely they are to prefer one specific attribute, making the task very transparent and different from real life decisions. Conjoint analysis is a procedure that can help avoid some of these issues (Green & Srinivasan, 1990). In rating-based conjoint studies, consumers rate stimuli consisting of several experimentally manipulated attribute levels. By quantifying the relation between the attribute levels of the stimuli and the profile ratings for each respondent, a respondent-specific weight (or part-worth utility) can be computed that captures the relative contribution of each attribute level to overall ratings. Although conjoint analysis is an extensively validated method, it has mostly been used to address applied issues (Green & Srinivasan, 1990). In the current paper, we propose that conjoint designs that experimentally vary the eco-friendliness of products can be used to measure consumers' ECT, and we show how this can be integrated into a nomological network of antecedents, mediators, and outcomes using Structural Equation Modeling for Within-Subject Experiments (SEMWISE), a novel approach for analyzing conjoint data (Weijters & Baumgartner, 2019) that so far has not been applied in consumer research. Study 1 When identifying drivers of organic purchasing behavior, one difficulty is that many times the organic attribute is confounded with other attributes like taste and price (Nadricka, Millet, & Verlegh, 2019; van Doorn & Verhoef, 2011). In the first study, we illustrate how conjoint analysis can disentangle preferences for different attributes (and their potential interactions) by orthogonally manipulating them in a simple within-subject experiment. We then correlate the conjoint-based latent ECT variable with traditional self-report measures of ECT and related constructs in order to cross-validate the results, using the SEMWISE procedure. For a brief introduction to SEMWISE; for details, refer to Weijters and Baumgartner (2019). Method Participants (N = 560 coffee drinkers; Mage = 37.4 , SD = 8.6; 48.9% women) imagined ordering coffee and rated the attractiveness of eight coffee profiles using a continuous scale from 0 to 5 stars. The coffee profiles correspond to a full-profile complete conjoint design created by experimentally manipulating the (binary) attributes organic label (organic vs. regular), taste (consumer ratings of 3/5 'average taste' vs 5/5 'excellent taste'), and price (high euro2.40 vs. low euro2.00). Green consumption values were measured with an adapted translation (Weijters & Baumgartner, 2019) of the GREEN scale (Haws et al., 2013) consisting of six items, three are reverse-worded (e.g., T would describe myself as environmentally responsible', 'I'm not particularly bothered by worries about our environment' [reversed]). Health consciousness was measured with six items, half were reverse-worded (Chen, 2009) (e.g., T consider myself very health conscious,' 'I have the impression that other people pay more attention to their health than I do' [reversed]). Results and discussion First, we specify an unconditional SEMWISE model without environmentalism and health consciousness, and conduct a series of model comparisons. Throughout, indicator intercepts are fixed to zero, while indicator residual variances are freely estimated. For each set of effects (intercept, main effects, two-way interactions, three-way interaction), we consider three possibilities: (a) a model with freely estimated means and variances for the weight factors, implying that the average effect (part-worth utility) across respondents is non-zero and that there are individual differences in the strength of the effect; (b) a model with freely estimated means but zero variances for the weight factors, implying that the average effect is nonzero but the effect is the same for all respondents; and (c) a model in which the means and variances of the weight factors are zero, implying that there are no (significant) effects of the manipulated factors. Setting the variance of the three-way interaction weight factor to zero results in a just-identified model, which is used as the starting point for the model comparisons. We gradually impose additional restrictions on this model and evaluate them based on model fit indices (preferring lower BIC values, RMSEA < .08 [preferably < .05], SRMR < .08 [< .05], CFI > .90 [> .95], and TEI > .90 [> .95]). The model that includes a random intercept factor and random main effect weight factors, as well as fixed weight factors for the two-way interactions results as the preferred model, as it is the most parsimonious model that meets all fit index cutoff criteria and has the lowest BIC value (χ2(64.3) = 11555.4, p < .001, RMSEA = .065, CFI = .982, TEI = .973, SRMR =.044). The parameter estimates based on this model are reported in Table 1. The findings show that, on average, people like better taste and dislike higher prices, hence the positive weight factor mean for good taste and the negative weight factor mean for high price. The relatively small variance estimates for these weight factors suggest that these preferences are rather consistent across individuals. The results also indicate that, on average, people prefer regular over organic coffee (controlling for price and taste). This may be related to the fact that, in vice products, consumers often associate organic claims with lower quality (van Doorn and Verhoef 2011) and that an organic label tends to increase perceived taste and attractiveness for healthy food (Nadricka et al., 2019). The organic weight factor has a much larger variance than the weight factors for price and taste, indicating substantial heterogeneity in the part-worth utilities of organic claims. In addition to the main effects, there are three two-way interaction effects. First, the small negative weight factor mean for high price x good taste (Table 1) implies that consumers are slightly less demanding about taste for higher priced coffees (Plassmann, O'Doherty, Shiv, & Rangel, 2008) regardless of the review information. Second, consumers are less price sensitive for organic coffee (i.e., high price has a negative mean, but the interaction of organic x high price has a positive mean; Table 1). Finally, for organic coffee, good taste is slightly less important on average (i.e., good taste has a positive mean, but the interaction of organic x good taste has a negative weight factor mean; Table 1). Next, we add the covariates environmentalism and health consciousness, each modeled as a latent factor with six indicators. In addition, since for both scales half of the items are reverse-worded, we model a method factor on which all items have unit loadings (Maydeu-Olivares & Coffman, 2006) and is uncorrelated with the substantive factors in the model (note that items were not reversecoded for analysis). The intercept and main effect weight factors are modeled as dependent variables with environmentalism and health consciousness as antecedents (the two-way interaction weight factors have zero variance and cannot be modeled as dependent variables). The resulting model shows acceptable fit to the data (χ2(155) = 310.768, p < .001, RMSEA = .042, CFI = .971, TLI = .964, SRMR = .039). As expected, the organic label is valued more positively by consumers who score higher on environmentalism (B = .300, 95% CI = [.208, .392]) and (to a lesser extent) by consumers who score higher on health consciousness (B =.133, 95% CI = [.035, .230]). People higher in environmentalism also show a more positive attitude toward all the coffees (B = .145, 95% CI = [.049, .242]), possibly due to overlap in consumer segments visiting coffee bars and engaging in green consumption (Gilg, Barr, & Ford, 2005). No other regression weights were significant. In sum, we analyzed conjoint data for coffee in which taste, price and organic label were varied experimentally, and found strong main effects of these attributes on preferences. The attribute weights (part-worth utilities) also showed significant individual variation. This was particularly true for the weight factor of organic labeling, and the variation in part-worth utilities was associated with individual differences in self-reported environmental concern and (to a lesser extent) health concern. We analyzed the data using the recently proposed SEMWISE approach, which enables the estimation of part-worth utilities as latent weight factors that can be integrated into a structural network consisting of antecedents, mediators, and outcomes. Study 2 Consumers often rely on eco-information schemes (e.g., ecolabels) to assess products' environmental sustainability. A recent type of eco-