A Comparison of Three Psychometric Values Measures for Modeling Ecologically Conscious Consumption Behavior ()
1. Introduction
With the advent of reference dependent preferences [1] , “arbitrarily coherent” demand [2] , and laboratory experiments revealing limited support for the neo-classical, preference-based model of consumer choice [3] , researchers might welcome an alternative, stable construct on which to base models of consumer behavior. Values―“desirable transsituational goals… that serve as guiding principles in the life of a person…” ( [4] , p. 21)―offer one such alternative. Values “can motivate action…” ( [4] , p. 21) and have long been regarded as influencing individual attitudes, beliefs, and behaviors [4] [5] [6] .
Environmentally-linked purchase behavior offers an attractive test case for linking values and behaviors. Environmental psychologists have found that people who “hold collective, society-directed values are more likely to engage in environmentally and socially responsible behaviors” ( [7] , p. 559). Moreover, values may play a larger role in explaining environmental awareness as socio-demographics tend to be less useful in this context [8] .
One measure of environmentally-linked behavior is Ecologically Conscious Consumption Behavior (ECCB), which occurs when consumers “purchase products and services which they perceive to have a positive impact on the environment” ( [9] , p. 222). This note reports the construction and analysis of a consumer survey incorporating ECCB and three values measures: Schwartz’s Self-Transcendence Values [4] , [6] , Kahle’s List of Values (LOV) [10] , and Richins and Dawson’s Materialism [11] . We find that Schwartz’s constructs, both separately and as a group, add explanatory power to a model of ECCB compared to a model based solely on socio-demographics. Neither LOV nor Materialism improves model fit, though one element of Materialism, acquisition centrality, does. These findings: 1) suggest psychometrics can enhance models of environmentally-linked consumer behavior and 2) emphasize the importance of construct selection in a values-based approach to modeling.
Section 2 below develops the constructs we explore and our research hypotheses. Section 3 describes our sample and survey, while Section 4 describes the resulting data. Section 5 reports the statistical analysis. Section 6 concludes.
2. Theoretical Background
2.1. Ecologically Conscious Consumption Behavior
Roberts [9] , [12] suggests separating socially responsive consumption into one category focusing on environmental awareness and another focusing on other aspects of social concern, e.g., donations to charities, hiring of minorities, and other socially conscious activities. We follow this convention and report in this note analysis of environmentally-linked behaviors (ECCB); other socially-orient- ed behaviors are reported elsewhere.
2.2. Self-Transcendence Values
Schwartz [4] , [6] posits ten values: self-direction, stimulation, conformity, tradition, security, achievement, power, universalism, benevolence, and hedonism. As other-regarding constructs, benevolence and universalism offer the most promise for enhancing our understanding of ECCB. Schwartz [6] defines benevolence to focus on the welfare of people with whom one frequently interacts―one’s ingroup. The motivational goal behind benevolence is the “preservation and enhancement of the welfare of people with whom one is in frequent personal contact” ( [6] , p. 39). Schwartz ( [6] , p. 12) defines universalism based on the motivational goal behind its manifestation: “understanding, appreciation, tolerance, and protection for the welfare of all people and for nature.”
Schultz and Zelezny [13] examined the relationship between environmental attitudes and Schwartz’s measures, finding that environmental attitudes concerned with the physical environment are positively related to universalism, but negatively related to benevolence, reflecting the priority benevolent individuals attach to the ingroup’s welfare over that of society at large.
Consequently, we hypothesize:
H1: Individuals with high levels of universalism engage in more ECCB, and
H2: Individuals with high levels of benevolence engage in less ECCB.
2.3. The LOV
Kahle [10] developed a nine item List of Values (LOV)―sense of belonging; excitement; warm relationships; self-fulfillment; being well-respected; fun and enjoyment of life; security; self-respect; and a sense of accomplishment. Items are measured on a nine point scale anchored by Very Unimportant (1) to Very Important (9). Kamakura and Novak [14] classified the LOV into five motivational domains: self-direction (LOV’s self-respect and self-fulfillment), achievement (accomplishment, well-respected), enjoyment (fun and enjoyment, excitement, warm relationships), maturity (belonging, warm relationships), and security (security).
As Schwartz [4] , [6] merged his maturity value with other values dealing with the welfare of outgroups and nature, we postulate LOV’s maturity measure bears a similar relationship:
H3: Individuals with high levels of maturity engage in more ECCB.
The literature suggests that security is related to anthropocentric attitudes [13] . Thus,
H4: Individuals with high levels of security engage in less ECCB.
Schwartz and Bilsky ( [5] , p. 552) describe individuals high in self-direction as imaginative, independent, intellectual and logical who “refer to reliance on and gratification from one’s independent capacities for decision making, creativity, and action.” The more a consumer’s perception of an organization (or product) corresponds to the individual’s self-definition, the stronger the consumer-or- ganization identification [15] . Since consumers’ comparisons are likely based on human traits such as norms and values, we postulate:
H5: Individuals with high levels of self-direction engage in more ECCB.
Kamakura and Novak’s [14] achievement and enjoyment serve individual interests; consequently,
H6: Individuals with high levels of achievement engage in less ECCB, and
H7: Individuals with high levels of enjoyment engage in less ECCB.
2.4. Materialism
Materialism consists of success, acquisition centrality, and pursuit of happiness [11] . Success addresses the notion that people often judge their own and others’ success by their possessions. Acquisition centrality refers to the central position material objects hold in the lives of materialistic people. Pursuing happiness through acquiring material objects is a hallmark of materialistic individuals.
Richins and Dawson [11] support the idea that materialists are selfish with money and possessions. Consequently, we hypothesize:
H8: Individuals with high levels of materialism on the success dimension engage in less ECCB,
H9: Individuals with high levels of materialism on the acquisition centrality dimension engage in less ECCB, and
H10: Individuals with high levels of materialism on the pursuit of happiness dimension engage in less ECCB.
3. Survey Development
3.1. Sample
A random sample was drawn from the telephone directory of a large metropolitan area in the Southeastern United States. Potential respondents were pre-called and asked to participate in a study of “Consumer Purchase Behavior.” The addresses of those agreeing to participate were verified and a survey and self-addressed, stamped envelope mailed to them. We mailed 290 surveys; 149 were returned, yielding a 51.38 percent response rate. We deleted surveys with missing variables, leaving 141 observations.
3.2. ECCB
While Roberts [9] , [12] , [16] and his colleagues [8] examined a range of environmentally and socially-oriented activities using 40 survey items, we focus on environmentally-related purchase behavior. Items that are repetitive, outdated, or reflect behavior motivated by a desire to save money [9] were excluded, leaving 20 items. These were measured on a 7-point scale anchored by “Never True” (1) and “Always True” (7).
3.3. Control Variables
Social desirability was measured by 10 items from the Marlowe-Crowne Social Desirability Scale [17] . These are “true” or “false” items which we summed, with one point per item for the response indicating social desirability. We also measured gender, household income, age, and education.
3.4. Values
The three values constructs were included in the survey. Schwartz’s [6] instrument contains 56 items measuring 10 values. Due to the length of his instrument we measured only universalism and benevolence, those values likely to be associated with ECCB. We used the nine items Schwartz specified for each measure in the order they appeared in the complete instrument [6] . Respondents were instructed to read the entire list of 18 values and to “RATE EACH VALUE according to its DEGREE OF IMPORTANCE TO YOU AS A GUIDING PRINCIPLE IN YOUR LIFE.” We employed Schwartz’s [6] 9-level scale, anchored by “opposed to my values” (−1), “not important” (0), “important” (3), “very important” (6), and “of supreme importance” (7). The same scale was used to measure the nine values in the LOV [10] . Lastly, Materialism values were constructed from Richins and Dawson’s [11] 18-item scale with each item measured on a seven point scale anchored by “Strongly Disagree” (1) and “Strongly Agree” (7).
4. Data
Of the 141 observations in the final dataset, 53 percent of respondents are female; 53 percent have less than a college degree; 25 percent have a college degree; and 21 percent attended graduate school. Age ranges from 19 to 89 years with a mean of 44. Twenty-five respondents (17.7 percent) are between 19 and 29; 32 (22.7 percent) fall between 30 and 39; 35 subjects (24.8 percent) are between 40 and 49; 28 (19.9 percent) are in their 50s; and 21 subjects (14.9 percent) are 60 or older. The median household income range is $50,000 to $74,999.
To determine which items from Robert’s instrument (described in Section 3.2.1) capture ECCB rather than other societally-oriented behaviors, we used SPSS to perform a principle components factor analysis using oblimin rotation on the 20 items jointly measuring both ECCB and non-environmental, societally-oriented behavior. The scree plot suggested two factors. We dropped one item which loaded on both factors at the 0.40 level and one which did not load at the 0.35 level on either factor. After deleting these two items, we repeated the factor analysis, resulting in two clean factors capturing 65.57 percent of the variance. Table 1 contains the 11 items measuring ECCB (Variance Extracted = 51.0 percent). Cronbach’s alpha (α) measures the degree to which multiple survey items
Table 1. ECCB scale items and factor loadings.
jointly measure a single, latent construct. Alphas of 0.7 or greater are considered acceptable ( [18] , p. 264-65). The 11 items measuring ECCB meet this threshold (α = 0.93); consequently, the variable ECCB reports the mean of each subject’s responses to these 11 items.
The measures of Schwartz’s Self-Transcendence Values (universalism: α = 0.90 and benevolence: α = 0.88) and Richins and Dawson’s Materialism (success: α = 0.77, acquisition centrality: α = 0.76, and happiness: α = 0.81) also meet accepted standards for Cronbach's alpha. The only measure containing three items in the LOV motivational domain is enjoyment, and its alpha is also acceptable (0.78). Three LOV motivational domains contain two correlated items: self-direction (Spearman correlation coefficient (r) = 0.49, p = 0.00), achievement (r = 0.55, p = 0.00), and maturity (r = 0.51, p = 0.00). For statistical analysis, values variables contain scale means. We also include gender (a dummy variable with gender = 1 indicating female; gender = 0 indicating male) and age in years in subsequent analysis.
5. Regression Analysis
To determine which, if any, of the three values constructs is linked to ECCB, we first estimated a base model by regressing ECCB on a constant, social desirability, gender, and age. These estimates are reported in columns two and three of Table 2. Then, we estimated three models which augment the base model with variables representing the Self-Transcendence Values (columns four and five of Table 2), the LOV (columns six and seven), and Materialism (columns eight and nine).
Table 2 includes an incremental F-statistic (denoted ΔF) and its associated p-value. This statistic tests the null hypothesis that the added variables as a group do not have explanatory power versus the alternative hypothesis that they do. We also report an F-statistic (denoted F) and p-value testing the statistical significance of the model as a whole (excluding the intercept). To measure model goodness-of-fit, we report the coefficient of determination (R2), adjusted R2, and the Akaike Information Criterion (AIC) [19] . To test our research hypotheses, we report the estimated coefficient, standard error, t-value and two-tailed p-value for the explanatory variables.
5.1. Base Model
Table 2 columns two and three contain the estimates from regressing ECCB on a constant, social desirability, gender, and age. The goodness of fit measures are R2 = 0.068, adjusted R2 = 0.048, and AIC = 0.583. The model is significant at the 0.05 level (F = 3.34, p = 0.021). Aside from the intercept, only gender is significant among the explanatory variables (β = 0.684, t = 3.049, p = 0.003).
5.2. Self-Transcendence Model
Incorporating universalism and benevolence increases explanatory power, as
Table 2. OLS regression results.a,b,c
aDependent Variable is ECCB. bNumber of Observations is 141. cRegressions were performed with NLOGIT 4.0.
1Here we adjust the two-tailed p-values reported in Table 2 to one-tailed p-values to account for directionality.
In addition to the intercept, three independent variables are significant at the 0.01 level: universalism (β = 0.504, t = 4.104, p = 0.000), supporting H1; gender (β = 0.690, t = 3.200, p = 0.002); and benevolence (β = −0.385, t = −2.646, p = 0.009), supporting H2. Note that universalism and gender positively influence ECCB, while benevolence negatively affects it (consistent with an ingroup versus outgroup interpretation).
5.3. LOV Model
The LOV values are grouped into five motivational domains: maturity, security, self-direction, achievement, and enjoyment. Based on the associated incremental F-test, we fail to reject the null that the estimated coefficients of the added variables―maturity, security, self-direction, achievement, and enjoyment―jointly equal zero (∆F = 0.40, p = 0.848). In addition, the overall model fails to achieve significance (F = 1.48, p = 0.171) or improve fit (adjusted R2 = 0.027 and AIC = 0.639). Of the explanatory variables, only gender is significant (β = 0.547, t = 2.117, p = 0.036); consequently, we fail to find support for H3 through H7.
5.4. Materialism Model
Lastly, we add the three dimensions of Materialism―success, acquisition centrality, and happiness―to the base model. We fail to reject the null that the coefficients of the added variables jointly equal zero (∆F = 1.78, p = 0.154). Note, however, the model as a whole is significant (F = 2.59; p = 0.021), largely due to the individual coefficient of acquisition centrality exerting a negative effect on ECCB (β = −0.310, t = −2.137, p = 0.034), supporting H9, while gender positively affects it (β = 0.732, t = 3.176, p = 0.002). We fail to find support for H8 or H10.
5.5. J-Tests
A non-nested specification test, such as Davidson and MacKinnon’s J-test, represents an alternative strategy to select among competing sets of independent variables [20] . We conducted a series of trials with each values construct serving in turn as base, then added predicted values for the other constructs. The test statistic for model inclusion is the t-statistic of the added, fitted values.
For the first trial, we re-estimated the Self-Transcendence model including predicted values based on LOV and Materialism. At the 0.05 level, Materialism should be included in the model (t = 2.845, p = 0.005), but LOV should not (t = 1.238, p = 0.218). For the second trial, we re-estimated the LOV model including fitted values for Self-Transcendence and Materialism. The results suggest adding both Self-Transcendence (t-value = 4.181, p = 0.000) and Materialism (t = 2.728, p = 0.007) to the model. For the last trial, we re-estimated the Materialism model including fitted values for Self-Transcendence and LOV and obtained a similar result: a suggested final model specification incorporating Self-Transcendence (t = 4.021. p = 0.000), but not LOV (t = 0.919, p = 0.360). Should only a single construct be required, J-tests eliminate LOV at levels of significance less than or equal to 0.360. Levels of significance less than or equal to 0.005 are required to exclude Materialism in favor of Self-Transcendence.
5.6. Composite Model
At the more typical 0.05 level of significance, the J-tests suggest including elements of Self-Transcendence and Materialism as explanatory variables. As shown in columns four and five of Table 2, both universalism and benevolence are related to ECCB; however, only one Materialism construct, acquisition centrality, proved significant. Consequently, we estimated a composite model, presented in columns 10 and 11 of Table 2, regressing ECCB on a constant, social desirability, gender, age, universalism, benevolence, and acquisition centrality. The added variables (∆F = 8.33, p = 0.000) and the overall model (F = 6.10, p = 0.000) are significant at the 0.01 level; the individual coefficients of gender, universalism, benevolence, and acquisition centrality are also significant at the 0.01 level, yielding further support for H1, H2, and H9.
6. Conclusion
1Here we adjust the two-tailed p-values reported in Table 2 to one-tailed p-values to account for directionality.
As the composite model indicates, adding universalism, benevolence, and acquisition centrality to the base model improves fit and yields a significant model, demonstrating the potential of a values-based approach to modeling consumer behavior. The composite model is the best fit of the five models presented in Table 2 (R2 = 0.215, adjusted R2 = 0.180, and AIC = 0.455), and reveals significant estimated coefficients that conform to expectations. We find a positive relationship between ECCB and universalism (β = 0.526, t = 4.374, p = 0.000, supporting H1) and negative relationships between ECCB and both benevolence (β = −0.454, t = −3.142, p = 0.001, supporting H2) and acquisition centrality (β = −0.307, t = −2.710, p = 0.004, supporting H9)1. A significant model and a reasonable fit from an exploratory study such as this suggests further research may yield an increasingly effective addition to the economist’s toolkit for building robust models of consumer behavior.