Descriptive statistics of, firstly, the PEB items and, secondly, of the variables representing possible determinants of environmentally significant behavior are reported. Then, the findings of a PCA conducted to distinguish between different types of PEBs are presented. Finally, a repeated measures ANOVA representing an integrative environmental behavior model is applied to explain distinct types of private and public PEBs.
4.2. Predictors of the Environmental Behavior Model
The average agreement ratings for the items addressing environmental attitudes are listed in Table 2
along with the results of the reliability analysis. All six item-to-scale correlations were larger than r = 0.45, which reflects an acceptable level of consistency. The scale’s overall Cronbach’s alpha value was acceptable at α = 0.8. All items were retained in the scale, as omitting any items would have reduced the overall Cronbach’s alpha value (see last column in Table 2
presents the average ratings for the single items representing psychological constructs. In addition, the correlation between prescriptive and descriptive norms was calculated, which proved to be significant with r = 0.45 (p
< 0.001). Nonetheless, based on the above-elucidated theoretical grounds, this correlation seemed small enough to treat the two types of norms as two separate predictor variables in the encompassing behavior model.
The correlation between the two perceived behavioral control items “I can easily be environmentally friendly if I want to” and “It is very easy for someone like me to be environmentally friendly” was likewise calculated and proved to be significant with r = 0.52 (p < 0.001). Since both items were presumed to measure the same construct, an average value of both variables was considered appropriate to serve as the PBC measure in the encompassing environmental behavior model.
The participants’ average ratings in the four justification items are shown in Table 4
along with the results of a corresponding Cronbach’s alpha analysis. The overall α of 0.67 was acceptable, and the same was true for the corrected item scale correlations, which consistently exceeded 0.4. All items were kept in the scale, as omitting any item would have reduced the overall Cronbach’s alpha value. Among the four justifications, the statement “I behave very environmentally friendly in most areas of life, so it is also okay if some of my behaviors pollute the environment” obtained the highest score with M = 2.4, which roughly corresponds to a neutral rating between rather disagree (≈2) and rather agree (≈3).
4.3. PCA Results
A PCA was conducted on the 23 PEB items to investigate their factorial structure and identify factors that are able to account for common variance of certain (types of) behaviors. The scree plot was interpreted in order to determine the number of components to be extracted (Figure 2
). The upward bent in the plot from components 3 to 2 suggested the extraction of two principle components with initial eigenvalues of ʎ1
= 4.86 and ʎ2
= 1.98. The graphical scree-test was preferred to the Kaiser Gutman criterion because the latter tends to lead to the extraction of rather too many factors if the number of variables is quite large [118
The resulting initial factor-loading matrix was rotated based on the Varimax criterion to arrive at a clearer factorial structure. The resulting rotated components possessed eigenvalues of ʎ1
= 4.02 and ʎ2
= 2.82. Table 5
presents the corresponding rotated component loading matrix with the loadings of the different PEBs on the two extracted and rotated factors.
According to Guadagnoli and Velicer [119
], variables with component loadings of 0.30 or 0.40 are usually regarded as salient for the corresponding component, whereas loadings with smaller absolute values are ignored. A threshold of a = 0.35, which is the mid-point of these two values, was accordingly defined in this study as the criterion for a behavior item to form part of a certain component and thus be assigned to a corresponding behavioral scale. However, if an item loaded higher than 0.35 on both extracted components, it was assigned to the scale of the component on which it loaded highest, as this component was then assumed to best reflect the respective behavior item. Two items showed small loadings below 0.35 on both components and were therefore omitted from the subsequent analyses.
Accordingly, the five public sphere behaviors all loaded considerably high on component 1 (Table 5
, behavior items 19–23, loadings between a = 0.56 and a = 0.76). In addition, there were some private sphere behaviors that loaded considerably on this component, namely items 2, 3, 5, 15, and 17, for which loadings ranged between 0.40 and 0.56. Although these two sets of behaviors loaded on the same factorial component, they represent different types of environmental behaviors; as defined by Stern [20
], public sphere behaviors influence the environment indirectly via social and political influence, whereas private sphere PEBs have a straightforward, direct positive impact on the environment. Furthermore, there were some private sphere behaviors that loaded considerably on component 2, namely items 1, 4, 6, 9 to 14, 16, and 18, for which loadings ranged between 0.35 and 0.56. Accordingly, it was decided to use these three sets of items as the basis for the construction of three different PEB scales that were presumed to represent different types of PEBs: (1) public sphere behaviors with socio-political primary goals that are instrumental indirectly for ecological aims as secondary goals [20
]; (2) lighthouse private sphere behaviors; and (3) less socially salient private sphere behaviors. Thus, in addition to a separation of public sphere PEBs from private PEBs, two different types of private PEBs were derived from the PCA loading pattern (Table 5
). The term lighthouse PEBs seemed appropriate for the private sphere PEBs loading on component 1 for two reasons. Firstly, public sphere behaviors also loaded highly on component 1, which suggested some similarity of these private sphere behaviors to public sphere behaviors, thereby implying a high public visibility (social salience) of these behaviors. Secondly, when analyzing the content of these behaviors, it was recognized that they tend to involve the communication of an ecologically oriented message promoting sustainability-related values to others. Together, these two aspects—high visibility and communication function—raised the association to a lighthouse.
Some examples of lighthouse private sphere and less socially salient behaviors shall briefly illustrate this reasoning:
To refuse to buy products from companies with a bad reputation for environmental protection (behavior item 2, lighthouse private PEB) seems to include a strong sustainability-oriented message to others. The same is true for behavior item 5, “I reduce the number of miles I fly by taking the train (or other more environmentally friendly means of transport) or not travelling at all.” Similarly, if someone cleans up the environment, this may be well visible to others and sends a public message that keeping the environment clean is important (behavior item 17, lighthouse private PEB). On the contrary, littering waste often occurs when social control is lacking, and people tend to try to hide this misconduct from others. Hence, although previous research has shown that social norms influence littering behavior [41
], the corresponding social influence may be weakened if it is socially non-transparent and unobserved and thus unknown to others whether a person litters (behavior item 18, non-lighthouse private PEB). Such a non-transparency seems even more emphasized for at-home behaviors such as running the dishwasher when it is nearly empty or only about half full (behavior item 6) or taking very long hot showers (behavior item 10), which do not seem to be socially salient and are obscured from others and therefore represent non-lighthouse private PEBs.
A Cronbach’s alpha reliability analysis was performed for the three resulting scales. When considering the five public sphere behaviors as a scale, the resulting Cronbach’s alpha value was 0.73. The Cronbach’s alpha of the items representing the 5 lighthouse PEBs and 11 non-lighthouse private PEBs were 0.63 and 0.67, respectively. Deleting any items from any of these three scales would have reduced their respective Cronbach’s alpha value. Therefore, no items were omitted, and the average ratings for the 5 public sphere, 5 lighthouse, and 11 less socially salient private behavior items where considered as measures of how frequently each participant engaged in each of the three types of PEB.
4.4. Repeated Measures ANOVA Representing the Behavior Model
To investigate the roles of different predictors derived from previous research on PEBs in influencing the three types of PEB distinguished on basis of the PCA, a repeated measures ANOVA was conducted with PEB type as the within-subject factor (three levels encompassing public sphere, lighthouse private sphere, non-lighthouse private sphere PEBs). This method was deemed to be most appropriate because it calculates separate prediction models (parameter estimates) and significance tests for each PEB type and additionally enables testing for significant differences in the roles that certain predictors play in influencing the three different types of behaviors. Such differences are captured as interaction effects between the between-subject factor predictor variables and the within-subject factor PEB type.
The between-subject predictor variables of the model included personal motivational determinants such as (i) general environmental attitudes, (ii) green self-identity, (iii) positive affect (feeling good), (iv) awareness of consequences, and (v) justifications, as well as contextual social forces represented by (vi) prescriptive and (vii) descriptive social norms, and personal capabilities, namely, (viii) environmental knowledge, (ix) education level, (x) perceived behavioral control, (xi) income, and finally, the demographic control variables (xii) gender and (xiii) age. All predictors were entered as covariates into the general linear ANOVA model. The dependent measure and the psychological predictor variables were represented by the scales described in Section 4.2
. Gender was represented by a two-level variable (coded 1 = male, 2 = female) and the same was true for education level (0 = no academic degree, 1 = with academic degree). The two cases with missing values for the latter variable were randomly assigned to a category. Income was represented by a three-level variable (0 = income below the median category, 1 = income within the median category, 2 = income above the median category). For the 164 non-respondents, the median category was used as the statistically most neutral income estimate. Because income effects are not at the foreground of the present study, this procedure was deemed the best means to avoid losing the information these participants provided in relation to all of the other predictor variables.
Collinearity measures for the 13 predictors revealed no corresponding problems; however, the Mauchley test of sphericity revealed a significant violation of the homogeneity of covariance between the three repeated measures of the within-subject factor (χ2 = 74.97, df = 2, p < 0.001). A corresponding epsilon correction (Huynh-Feldt ε = 0.94) of the degrees of freedom of the F-distribution used to calculate the significances for the within-subject factors and corresponding interaction effects was conducted to take this into account.
The analyses revealed a significant main effect of the within-subject factor PEB type (p < 0.001). Public sphere behaviors (M = 2.5) were significantly less frequently displayed than lighthouse sphere behaviors (M = 4.1); however, less socially salient private sphere behaviors (M = 4.8) were most frequently evinced.
The tests of between-subject factors (Table 6
) show which variables of the behavioral model significantly contribute to the explanation of PEBs in general (i.e., when averaging the three PEB types for each participant). Accordingly, general environmental attitudes and (self-evaluated) environmental knowledge were the strongest predictors (for both: partial eta squared η2
= 0.087, p
< 0.001) of PEBs, followed by green identity, justifications, awareness of consequences, gender, and prescriptive social norms (for all five predictors, p
≤ 0.001 and η2
≥ 0.009), age (p
< 0.01), and perceived behavioral control (p
< 0.05, η2
= 0.005), respectively.
Differences in the importance of predictors for explaining the three different types of PEB were tested within the repeated measures ANOVA through interaction effects between the within-subject factors (type of PEB) and between-subject factors (predictor variables of the behavioral model), as shown in Table 7
. The analysis revealed significant interactions between PEB type and predictors for seven predictor variables. The strongest effect size was observed for general environmental attitudes X PEB type (η2
= 0.029, p
< 0.001), which were significantly more important for the public sphere (b = 0.52) and lighthouse private PEBs (b = 0.40) than for non-lighthouse private PEBs (b = 0.07; p
< 0.001 in both comparisons). The simple contrast between the influence of general environmental attitudes for public sphere and private lighthouse PEBs was likewise significant (p
< 0.05), which confirmed Hypothesis 1 that general environmental attitudes are particularly important in influencing public sphere behaviors. In addition, environmental attitudes proved more important for private lighthouse than less socially salient private PEBs.
A weaker interaction effect in the opposite direction was observed for green identity (η2 = 0.004, p < 0.05), which was in fact most strongly related to non-lighthouse private behaviors, followed by lighthouse private PEBs, and least strongly related to public sphere behaviors. Contrasts for the interaction effect showed significant differences in the b-weights of green identity for both lighthouse (p < 0.05) and non-lighthouse private PEBs (p < 0.01) compared with public sphere PEBs.
We also found a significant interaction effect (η2
= 0.024, p
< 0.001) between PEB type and environmental knowledge, the latter of which proved more important for the prediction of public sphere behaviors (b = 0.34) compared with lighthouse (b = 0.19) and non-lighthouse private PEBs (b = 0.08). A weaker (η2
= 0.005, p
< 0.01) but somewhat parallel interaction effect was found for education level, which was positively and significantly related to public sphere behaviors (p
< 0.05) but unrelated to lighthouse private behaviors and even significantly (p
< 0.05) negatively related to non-lighthouse private PEBs (see Table 7
A highly significant interaction effect was revealed for prescriptive social norms (η2 = 0.006, p < 0.001), which were found to be a significant predictor for public sphere PEBs (b = 0.145, η2 = 0.01, p < 0.001) but not for the two types of private behaviors. This finding clearly confirms Hypothesis 3 regarding a strong influence of social norms in the public sphere but implies the rejection of Hypotheses 5, which assumed a similar difference between private lighthouse and non-lighthouse PEBs with respect to the influence of social norms. However, descriptively, the smallest b-weight of prescriptive norms was indeed observed for less socially salient non-lighthouse PEB (b = 0.005).
The Hypotheses 2 and 4 had to rejected as there was no significant influence of descriptive social norms on any of the three types of PEBs.
A significant interaction effect was also observed for age (η2 = 0.008, p < 0.001) which proved to be significantly and strongly positively related to non-lighthouse private PEBs (η2 = 0.042, p < 0.001) but unrelated to public sphere and lighthouse private behaviors. This latter finding was only partially consistent with bivariate correlations between age and the three types of PEBs that were additionally calculated. These bivariate correlations did not prove significant for the public sphere PEBs (r = 0.03, p = 0.272), showed a very small but significant positive correlation between age and lighthouse private PEBs (r = 0.08, p < 0.01) and revealed a more substantial and highly significant positive correlation between age and non-lighthouse private PEBS (r = 0.27, p < 0.001). The seeming inconsistency between a small but significant bivariate correlation between age and lighthouse private PEBs and a non-significant b-weight of age in the regression model for the lighthouse PEBs is not unusual, however, because manifold predictor variables contribute to the explanation of variance in the regression model and therefore can make the variable age redundant there.
Finally, the repeated measurement ANOVA revealed a significant interaction effect for gender (η2
= 0.004, p
< 0.05), whereby females displayed significantly more private sphere PEBs than males (both lighthouse and non-lighthouse), whereas no gender effect was observed for public sphere behaviors. Additional t-tests conducted to compare males and females in relation to the three types of PEBs were consistent with this latter finding, as depicted in Table 8