Effects of an Environmental Education Course on Consensus Estimates for Proenvironmental Intentions

Environmental Activism and Descriptive Norms

Environmental educators have mostly focused on everyday, self-interest behaviors aimed at reducing personal and household impacts on the environment in contradistinction to prosocial behaviors necessitating the investment of considerable resources, cooperation, and joint action (McFarlane & Hunt, 2006). Environmental activism is defined as engagement in prosocial activities focusing on preserving or improving the quality of the environment and increasing public awareness of environmental issues (Seguin, Pelletier, & Hunsley, 1998). Environmental activism has received minimal attention by environmental educators, despite the fact that it can contribute substantially in financing environmental conservation and providing voluntary work for protected area management (Hovardas & Poirazidis, 2006).

Social norms can influence behavior and behavior intention substantially (Aronson, Wilson, & Akert, 2010). Injunctive norms and descriptive norms are two typical categories of social norms (Lapinski & Rimal, 2005). Although injunctive norms have a normative character and refer to actions that ought to be done, descriptive norms are people’s perceptions of what people are actually inclined to do (Rivis & Sheeran, 2003). Descriptive norms can be operationalized as estimates concerning the percentage of people within a social group of reference who would be willing to engage in given actions (Bauman & Geher, 2002). In this regard, descriptive norms reflect consensus estimates; that is, they include information about the number of people who would be willing to perform an action, and thereby, they allocate willingness to enact certain behaviors or rejection of these behaviors to either majorities or minorities.

Descriptive norms can have a considerable effect on proenvironmental behavioral intention (Bamberg & Möser, 2007; Kallgren, Reno, & Cialdini, 2000; Thøgersen, 2008). This is especially pronounced for environmental activism (Kolstø, 2006), where the contribution of a single individual only raises the probability of successfully providing a behavioral outcome by a small amount (Diekmann & Preisendörfer, 1998; Lubell, 2002). Although previous research investigated the influence of a number of variables on proenvironmental behavior intention, including sociodemographic variables, environmental attitudes, self-reported behavior, and environmental information sources (Clark & Finley, 2007; Machairas & Hovardas, 2005; Togridou, Hovardas, & Pantis, 2006), consensus estimates have not yet received proper attention. A fact too often ignored in empirical research is that people hold descriptive norms with some uncertainty. It is very likely for an individual to decline a behavior because he/she has underestimated the actual intention of similar others or, conversely, to intend to engage in a behavior because he/she has overestimated the intentions of others.

Collaborative learning environments can provide the social context necessary to fine-tune expectations one has for others and decrease errors in consensus estimation. This learning arrangement is very common for environmental education projects (Korfiatis & Paraskevopoulos, 2003). Previous research has shown that feedback about performance of relevant others may be decisive for behavioral choices (Staats, Harland, & Wilke, 2004) because it increases the saliency of descriptive norms (Abrahamse, Steg, Vlek, & Rothengatter, 2007; Ohtomo & Hirose, 2007). Furthermore, group discussion can influence behavioral intentions. Social interaction during education interventions can support participants in inferring what is popular and stimulate their intentions accordingly (Werner, Sansone, & Brown, 2008).

The False Consensus Effect

Previous research demonstrated that people make systematic errors in identifying social norms. A special case of the inaccurate assessment of social norms is the false consensus effect—the tendency to overestimate the degree to which others agree with one’s own beliefs or behavioral intentions (Ross, Greene, & House, 1977). False consensus has been reported as a relatively robust effect that occurs across domains (Kilianski, 2008; J. Krueger & Clement, 1994). If a false consensus effect occurs, consensus estimates given by participants who endorsed a given behavior should be significantly higher than consensus estimates given by those who would not be willing to engage in that same behavior (De La Haye, 2000, Marks & Miller, 1987).

Concerning environmental behavior, false consensus was identified among participants in a commons dilemma context, namely, a context where individual actors have to share a common resource (Gifford & Hine, 1997). Usually, individual and collective rationalities in commons dilemma situations collide. In particular, individual actors do not deliberate jointly over the use of the common resource, and each actor tends to consume more than his/her share, which leads to a consumption rate that overrides the recovery rate of the resource. Gradually, the availability of resource decreases and the resource is eventually exhausted. Gifford and Hine (1997) reported that when a student sample was confronted with a natural resources management issue, less than a fifth declared that they intended to increase their harvests out of the natural resources’ pool (e.g., increase the rate of the consumption of the common resource and surpass their share). However, these students greatly overestimated the total number of peers who would maximize their harvests, which reflects the false consensus effect. The false consensus effect has been also observed in recycling behavior (Kimura & Shinoki, 2007) and in the case of water conservation (Monin & Norton, 2003).

In line with Fishbein and Ajzen’s (1975) theory of reasoned action, the false consensus effect has been reported to influence behavior (Botvin, Botvin, Baker, Dusenbury, & Goldberg, 1992) and behavioral intentions (Bauman & Geher, 2002). Specific interventions could be successful in reducing the bias of the false consensus effect and promoting beneficial social norms (Bauman & Geher, 2002). The theory of reasoned action also states that behavioral intentions are a function of attitudes toward the behavior. If an intervention succeeds in providing novel information on descriptive norms, consensus estimates before the course will be partly based on environmental attitudes, although this correlation should be weakened after the intervention (Bauman & Geher, 2002).

Hypotheses of the Current Study

To our knowledge, no study up to date has attempted to reduce errors in consensus estimates in the context of proenvironmental behavior. Within the frame of the current study, we scheduled an environmental education intervention to increase the accuracy of consensus estimates and correct misperceptions about the prevalence of shared support for activist proenvironmental behaviors.

Hypothesis 1: A false consensus effect will emerge in the case of behavioral intention for activist environmental actions. In this case, consensus estimates given by those who would be willing to engage in a particular behavior will be significantly greater than the estimates given by those who would not endorse the same behavior.

Method

Measures

Students were asked whether they were willing to engage in two behaviors (yes/no), namely, (a) donate 50 Euros to adopt a tree in a Cypriot protected area (“donate”) and (b) offer voluntary work for 1 week to support environmental management in a protected area in Cyprus (“volunteer”). Behavioral intention items did only appear in the questionnaires before and after the course and were not part of the reference material or students’ assignments. Students were also requested to estimate the percentage of peers who would be willing to engage in these behaviors (on a scale from 0% to 100%).

Willingness to engage in the given actions reflects actual intention (statement but not engagement in the prescribed behaviors), that is, the number of respondents who stated that they would be willing to perform the given actions. Actual intention allowed us to determine the majority and minority pertaining to each item, namely, whether most participants were willing to perform or declined to perform the behaviors. By subtracting the actual consensus (based on the actual percentage of people who agreed with each item) from each participant’s estimate, we determined the extent to which participants overestimated or underestimated consensus among peers (Bauman & Geher, 2002; De La Haye, 2000; Hovardas & Poirazidis, 2006; Togridou et al., 2006). Positive scores indicate students who overestimated actual consensus, whereas negative scores suggest that students underestimated actual consensus.

Furthermore, we computed the accuracy in consensus estimates expressed as deviation of estimates from actual intention as a percentage of actual intention (percentage error). For example, let us suppose that a student participant who agreed with a behavioral intention statement estimated that 50% of peers would also agree (Table 1, first row). If the students who agreed with this same statement amounted to 60% of the sample, then this participant presented an error of −10, where the negative sign reflects underestimation of actual intention. Expressed as percentage of actual intention, this error equals to −16.67%.

OPEN IN VIEWER

Table 1.

An Example of Calculating Estimates and Errors

Own intention	Actual intention (%)	Consensus estimate (%)	Error	Percentage error (%)
Yes	60	50	−10	−16.67
Yes	60	70	10	16.67
No	60	30	−30	−50

Students completed the NEP scale, which presented an acceptable internal consistency both before and after the course (Cronbach’s α = .74 and .75, respectively). Responses were summed across items to calculate individual scores.

Students stated to what extent they carried out eight environmentally related activities during the last year. Each rating was made on a 4-point Likert-type scale with verbally defined alternatives (1 = never, 2 = seldom, 3 = sometimes, and 4 = often). The scale was used in the past in studies with Greek-speaking samples and proved reliable and efficient in assessing respondents’ proenvironmental behavior profile (Hovardas & Poirazidis, 2006; Korfiatis, Hovardas, & Pantis, 2004). The items of the scale referred to private sphere environmentalism (i.e., green consumerism in the form of purchasing organically grown foods), private sector household behaviors (i.e., sorting household waste for recycling), behaviors in the public sphere (i.e., attending a public meeting on environmental issues, providing financial support to a nongovernmental environmental organization, and participation in an environmental education project), and activist behaviors (i.e., attending a rally or protest about an environmental issue, participation in a reforestation program, and providing voluntary work for a nongovernmental environmental organization). The scale had a Cronbach’s alpha reliability coefficient of .78. Aggregate values were calculated for each respondent by adding responses across items.

Participants stated their environmental information sources for the last year among TV and radio, newspapers and magazines, friends and relatives, the university, and nongovernmental organizations. Again, we computed scores for each participant. Overall, participants’ scores for environmental information sources and self-reported behavior were quite similar to those derived by other Greek-speaking samples (Hovardas & Poirazidis, 2006; Korfiatis et al., 2004; Togridou et al., 2006).

Data Analyses

As data were nonnormally distributed, we conducted nonparametric tests to determine time (before vs. after the course) and group (actual majority vs. actual minority) effects on consensus estimates. We used the McNemar test to investigate whether actual intention will increase in time significantly. We conducted Mann–Whitney tests to examine differences between consensus estimates of contributors and noncontributors, and Wilcoxon tests to explore differences of consensus estimates before and after the course. We computed Spearman’s rho coefficients to examine the correlations of NEP scores and consensus estimates. We also conducted logistic regression analyses to identify significant predictors for behavioral intention items both before and after the course among students’ gender, age, year of study, NEP scores, self-reported behavior scores, environmental information scores, and absolute values of error in consensus estimates. For both behavioral intention items, we identified participants who did not change their intention, those who declined the behavior before the course and endorsed it afterwards (positive change) as well as those who presented the opposite pattern (negative change). We also calculated change in error estimates after the course. A multinomial logistic regression was conducted to identify significant relationships between change in behavioral intention items and change of error in consensus estimates after the course. Our hypotheses as well as measures and analyses for each hypothesis are presented in Table 2.

OPEN IN VIEWER

Table 2.

Hypotheses, Measures, and Analyses

Hypotheses	Measures	Analyses
1. Consensus estimates for contributors will be significantly greater than those of noncontributors (false consensus effect)	Consensus estimates	Mann–Whitney test
2. Mean errors in consensus estimates will decrease after the intervention	Consensus estimates	Wilcoxon’s test
3. Environmental attitudes will be significantly correlated with consensus estimates before the course and not after	NEP scores, consensus estimates	Spearman’s r; Williams’s test
4. Absolute value of errors in consensus estimates will be the most significant predictor of participants’ intentions	Behavior intention (dependent variable); gender, age, year of study, NEP scores, self-reported behavior scores, environmental information scores, absolute value of errors in consensus estimates	Logistic regression
5. For participants who will decline a behavior before the course and endorse it after (i.e., positive change), accuracy in consensus estimates will increase after the course	Change in behavior intention (dependent variable); change in error estimates	Kruskal–Wallis test, multinomial logistic regression

Note: NEP = New Environmental Paradigm. With the exception of Hypothesis 5, which refers to change from pre- to postintervention time frame, all other hypotheses include measures calculated both before and after the course.

Results

Consensus Estimates (Hypotheses 1 and 2)

The majority of the sample agreed with behavioral intention items before and after the course (Table 3). Although actual intention increased after the course, the McNemar test revealed that time effects were not significant. Majority and minority estimates were lower than actual intention. However, respondents who refused to engage in the given actions (i.e., the minority of the student sample) underestimated the number of peers who would be willing to engage to the given actions (i.e., the majority of the sample) to a greater extent than the majority itself. Indeed, all estimates of the minority remained lower than 50%, which indicates that they misperceived their minority status. As differences between those who endorsed behavioral intention items and those who declined them were significant (see last column of Table 3 for a detailed account of z scores derived by Mann–Whitney tests), the false consensus effect was displayed both before and after the course. This finding validates our first hypothesis (i.e., consensus estimates for contributors were significantly greater than those of noncontributors).

OPEN IN VIEWER

Table 3.

Actual and Estimated Behavioral Intention Before and After the Course

		Estimated intention
Items	Actual intention	Majority willing to perform the behavior	Minority not willing to perform the behavior	z scores
“Donate”
Before the course	60.29	51.43	32.24	−2.86**
After the course	73.53	52.18	35.00	−2.27*
“Volunteer”
Before the course	72.06	51.05	28.64	−3.04**
After the course	82.35	53.88	30.00	−3.05**

Note: Actual and estimated behavioral intentions are given as percentages of the sample; z scores refer to Mann–Whitney tests.

*

p < .05. **p < .01.

Table 4 shows mean error in estimated behavioral intention calculated as average percentage deviation of estimates from actual intention. Negative signs indicate underestimation of actual intention. Accuracy of estimates increased significantly after the educational intervention for both “donate” (Wilcoxon’s z = −4.71; p < .001) and “volunteer” (Wilcoxon’s z = −4.15; p < .001). This fact is reflected by the smaller absolute values of mean errors after the intervention. Indeed, this result was revealed for contributors and noncontributors (first and second column of Table 4, respectively). These findings confirm our second hypothesis (i.e., mean errors in consensus estimates decreased after the intervention).

OPEN IN VIEWER

Table 4.

Mean Error in Estimated Behavioral Intention Before and After the Course

Items	Majority willing to perform the behavior	Minority not willing to perform the behavior
“Donate”
Before the course	−15.70	−30.36
After the course	−5.40	−14.24
“Volunteer”
Before the course	−23.45	−43.11
After the course	−14.82	−31.29

Note: Mean errors correspond to average percentage deviation of estimates from the actual behavioral intention; negative signs reflect underestimation of actual intention.

Environmental Attitudes (NEP Scores) and Consensus Estimates (Hypothesis 3)

Items in the revised NEP scale showed increased median values already before the course. Of the 15 items, 10 items in the scale presented a median value of 4 in a 5-point Likert-type scale format (responses to the 7 even-numbered items of the scale were reversed so that agreement indicates pro-NEP responses across all items). In all, 4 items had a median value of 3, whereas 1 item had a median value of 5. Expressed as aggregate NEP scores, respondents revealed a median value of 55 (maximum 75; minimum 15). After the course, median values were maintained across items and in the total score of the scale. A Wilcoxon test revealed no significant trends. These results imply that respondents indicated a relatively high degree of environmental dispositions already before the intervention, which did not alter after the course.

Although the educational intervention did not have any impact on NEP scores, there was a significant differentiation in the correlation between NEP scores and consensus estimates. Before the course, NEP scores were significantly correlated with consensus estimates for both items (Spearman’s ρ = 0.27; p < .01 for “donate” and Spearman’s ρ = 0.24; p < 0.05 for “volunteer”). Indeed, the Williams’s test for differences between correlation coefficients showed that correlations were different from one another (t = 2.99; p < 0.01 for “donate” and t = 2.17; p < 0.05 for “volunteer”. On the contrary, no significant coefficient was computed after the course. Indeed, the Williams’s test for differences between correlation coefficients showed that correlations were different from one another (t = 2.99; p < 0.01 for “donate” and t = 2.17; p < 0.05 for “volunteer”). These findings validate our third research hypothesis (i.e., environmental attitudes were significantly correlated with consensus estimates before the course and not after) and indicate that estimates before the course were partly based on environmental attitudes, although this correlation was not observed after the course.

Absolute Value of Errors in Consensus Estimates As Predictors of Behavioral Intention (Hypothesis 4)

Binary logistic regressions were performed with behavioral intention as the outcome variable and gender, age, year of study, revised NEP scale score, self-reported behavior score, environmental information sources scale score, and absolute value of error in consensus estimates as predictors (Table 5). No significant correlations were found among independent variables, which excluded any potential effect of multicollinearity (Field, 2000; Wall, Devine-Wright, & Mill, 2007). A forward stepwise procedure was followed, where the predictor with the highest p value was added to the model at each subsequent step until all remaining predictors had nonsignificant coefficients. For all models, absolute value of error in consensus estimates was the only predictor that had a significant effect (see second and third column of Table 5 for variables in the equation and regression coefficients, respectively). This result confirms our fourth hypothesis (i.e., absolute value of errors in consensus estimates were the most significant predictor of participants’ intentions). Regression coefficients were negative, which implies that respondents were less likely to be willing to engage in given behaviors as errors increased. As we identified the false consensus effect for activist environmental actions, consensus estimates given by contributors were considerably higher than consensus estimates given by those noncontributors. Given that both contributors and noncontributors underestimated actual consensus, this resulted in minority members performing higher errors in consensus estimates. Therefore, higher errors were accompanied by lower levels of behavior intention.

OPEN IN VIEWER

Table 5.

Binary Logistic Regression Models of Behavioral Intention Items on Gender, Age, Year of Study, Revised NEP Scale Score, Self-Reported Behavior Score, Environmental Information Sources Scale Score, and Absolute Value of Error in Consensus Estimates

Regression model	Variables in the equation	Regression coefficient	Constant of the model	Change in −2 log likelihood	Hosmer and Lemeshov χ2	Nagelkerke’s R2	Percentage correct (%)
“Donate,” before the course	Absolute value of error	−0.07*	2.14**	6.38*	10.67(ns)	0.16	81.5
“Donate,” after the course	Absolute value of error	−0.06*	2.64**	6.68*	11.19(ns)	0.16	82.8
“Volunteer,” before the course	Absolute value of error	−0.06**	3.78***	10.81**	9.56(ns)	0.24	81.4
“Volunteer,” after the course	Absolute value of error	−0.11**	4.45***	13.35***	6.09(ns)	0.31	85.5

Note: Forward stepwise entry of predictors was used; ns = non significant.

*

p < .05. **p < .01. ***p < .001.

Fit indices across all logistic regression models were acceptable, namely, as changes in −2 log likelihood were significant for all equations, there was in each case a marked improvement over the constant-only model (Table 5, fifth column). The Hosmer–Lemeshow goodness-of-fit statistic was not significant across all equations, which indicates a good fit (Table 5, sixth column). Hence, the models adequately fitted the data. Nagelkerke’s R² values showed that binary logistic regression models for “volunteer” explained more variance compared with models for “donate” (Table 5, seventh column). All models correctly predicted behavioral intention for more than 80% of participants (Table 5, last column).

Change in Behavior Intention and Change of Error in Consensus Estimates From Pre- to Postintervention Time Frame (Hypothesis 5)

Table 6 presents change in behavior intention from pre- to postintervention time frame, namely, “positive change” corresponds to participants who declined the behavior before the course and endorsed it afterwards, whereas negative change corresponds to the opposite pattern. “No change” describes individuals who maintained their initial intention. Although about two thirds of the sample did not change their intention for neither item (Table 6, no change), one fifth changed simultaneously in favor of both items (Table 6, positive change). A chi-square test revealed that change in behavioral intention after the course was interrelated between the items, χ² = 24.96, p < .001; Cramer’s V = 0.45, p < .001; that is, those who changed in favor of one behavior intention item tended to also change in favor of the other.

OPEN IN VIEWER

Table 6.

Crosstabulation of Change in Behavioral Intention for the Item “Donate” by Change in Behavioral Intention for the Item “Volunteer”

	“Volunteer”
	Negative change	No change	Positive change	Total
“Donate”
Positive change	2	4	10	16 (23.53%)
No change	3	39	3	45 (66.18%)
Negative change	2	4	1	7 (10.29%)
Total	7 (10.29%)	47 (69.12%)	14 (20.59%)	68 (100.00%)

Note: χ² = 24.96, p < .001; Cramer’s V = 0.45, p < .001. Numbers present counts, while percentages of the sample are given in parentheses; positive change corresponds to participants who declined the behavior before the course and endorsed it afterwards, whereas negative change corresponds to the opposite pattern.

Table 7 presents change of error in consensus estimates across the three types of change in behavioral intention (i.e., negative change, no change, and positive change). Decrease in mean error was more pronounced in the case of positive change for both items, but this trend was significant only for “volunteer” (Kruskal–Wallis χ² = 7.17, p < .05). Specifically, participants who declined the behavior before the course and endorsed it afterward showed a marked improvement in their estimates compared with participants who revealed no change (Mann–Whitney z = −2.01, p < .05) and those who presented negative change (Mann–Whitney z = −2.61, p < .01). The above-mentioned findings partly validate our fifth hypothesis (i.e., for participants who declined a behavior before the course and endorsed it after accuracy in consensus estimates increased).

OPEN IN VIEWER

Table 7.

Change in Behavioral Intention and Change of Error in Consensus Estimates

	Negative change	No change	Positive change	Kruskal–Wallis χ2
“Donate”	−7.04	−7.79	−19.88	4.97(ns)
“Volunteer”	−3.29	−7.36	−26.19	7.17*

Note: Positive change corresponds to participants who declined the behavior before the course and endorsed it afterwards, whereas negative change corresponds to the opposite pattern. Negative signs reflect decrease in mean error after the course; ns = non significant.

*

p < .05.

To provide further support for our fifth research hypothesis, we performed multinomial logistic regressions, where the outcome variable can have more than two categories. Change in behavioral intention after the course was the dependent variable (i.e., negative change, no change, or positive change), whereas change of error in consensus estimates served as the predictor variable (Table 8). The predictor variable for both “no change” and “positive change” categories had significant coefficients with a negative sign (Table 8, third column). This implies that the likelihood of these response categories decreased with respect to the reference category, namely, participants who presented negative change, when errors in consensus estimates increased after the course. These results indicate that positive change in behavior intention was related to increased accuracy in consensus estimation, which supported our fifth hypothesis (i.e., for participants who declined a behavior before the course and endorsed it after, accuracy in consensus estimates increased). The odds of positive change for “donate” decreased by 10% (0.90 − 1 = −0.10) for each unit increase in error (Table 8, fifth column). Analogous inferences can be made for “volunteer”.

OPEN IN VIEWER

Table 8.

Mulitnomial Logistic Regression Models of Change in Behavioral Intention on Change of Error in Consensus Estimates

Regression model	Variables in the equation	Regression coefficient	Constant of the model	Odds (Exp[B])	Change in −2 log likelihood	Nagelkerke’s R2	Percentage correct (%)
“Donate”					6.05 *	0.13	70.9
	No change	−0.09*	3.09***	0.91
	Positive change	−0.10*	1.78*	0.90
“Volunteer”					8.48*	0.20	81.8
	No change	−0.11*	4.03**	0.90
	Positive change	−0.12*	2.14*	0.88

Note: The reference category for both models is the group of respondents who declined behaviors they had endorsed before the course, namely those who showed negative change. The pearson and deviance statistics were not significant for either model, which revealed that data were consistent with models’ assumptions.

*

p < .05, **p < .01, ***p < .001.

Fit indices across all multinomial logistic regression models were acceptable, namely, as the chi-square statistics of the likelihood ratio tests of the proposed models against the null models was significant, the proposed models outperformed the null models (Table 8, sixth column). The model for “donate” explained less variance compared with the model for “volunteer”, as indicated by the differences in the Nagelkerke’s R² statistic (Table 8, seventh column). Overall, 70.9% of the cases are classified correctly by the proposed model for “donate” and 81.8% for “volunteer” (Table 8, last column).

Discussion

Proenvironmental intentions and proenvironmental attitudes as assessed by the revised NEP scale were pronounced before the course and remained unimpacted by the educational intervention. These findings indicate that participants were quite proenvironmentally motivated already before the intervention. Furthermore, the fact that gender, age, and year of study did not influence behavior intentions indicated that this proenvironmental inclination was homogeneously distributed across the student sample. The absence of demographic effects could be attributed to the fact that participants were recruited out of a student population. However, such a confined influence of demographics on proenvironmental intentions is in line with recent research, for instance as far as gender is concerned (Clark, Kotchen, & Moore, 2003; Togridou et al., 2006; Urban & Zvĕřinová, 2009; Wall et al., 2007).

Our results reiterated findings of previous studies (Hovardas & Poirazidis, 2006; Togridou et al., 2006), according to which individuals not willing to perform proenvironmental behaviors were much lesser compared with those who were willing to perform these behaviors (i.e., the former comprised a minority whereas the latter a majority). Furthermore, minority members showed significantly higher errors in consensus estimates by underestimating the actual majority and overestimating the actual size of their group (Hovardas & Poirazidis, 2006; Monin & Norton, 2003; Suls, Wan, & Sanders, 2006; Togridou et al., 2006). This severe underestimation of a majority willing to act in a proenvironmental manner was accompanied in our research by the false consensus effect, where majority estimates were consistently higher than minority estimates. The present study revealed that the false consensus effect might appear in the case of proenvironmental behavioral intention as well (Ando, Ohnuma, & Chang, 2007).

De La Haye (2000) argued that in the case of desirable actions, such as proenvironmental behavior, the range of possible errors is not the same for people who endorse a given desirable behavior, on one hand, and people who decline that same behavior, on the other. Errors in the direction of overestimation are restricted to a small range, whereas errors in the range of possible underestimation are much larger. This automatically leads to a mean estimate that is lower than the true value. Such a measurement artefact has been identified as regression to the mean (De La Haye, 2000) and results in underestimation of actual intention for both contributors and noncontributors. When participants are in the majority (i.e., contributors), measurement artefact and social projection pull in opposite directions, and errors are relatively small. For minority members, however, the measurement artefact and social projection pull in the same direction and errors are much larger. Interestingly, this leads the minority of noncontributors to falsely perceive themselves as belonging to the majority of the group of reference (Gifford & Hine, 1997; J. Krueger & Clement, 1997; Monin & Norton, 2003).

Past research has advanced both motivational and cognitive explanations for the false consensus effect (Gershoff, Mukherjee, & Mukhopadhyay, 2007). A motivational account for false consensus can be the desire to see one’s own attitude or behavior as being in the majority, which can be reinforced by an internal locus of attention, that is, by using oneself as an anchor when making judgments regarding other people (J. I. Krueger & Funder, 2004; Marks & Miller, 1987). Cognitive explanations can be based on the availability heuristic (Alicke & Largo, 1995), namely, individuals typically find it easier to bring to mind individuals or reasons that support their own views as compared with alternative views. A possible explanation for this could be selective exposure, that is, the tendency to associate with similar others (Bosveld, Koomen, & Van der Pligt, 1996). The educational intervention we implemented in the frame of the present study obviously increased the probability of encounters with nonsimilar others. Taking part in “home” groups and “expert” groups as well as watching peers discussing various issues and arguing for and against various alternatives might have reduced both the impact of internal locus of attention as well as selective exposure. In this regard, the jigsaw approach might have addressed the availability heuristic by increasing the salience of alternative opinions and manipulating the relative ease with which participants recalled other individuals who did and did not share their beliefs. Our study complements previous research on the jigsaw approach showing that it does not only result in knowledge gains (Doymus, Karacop, & Simsek, 2010; Souvignier & Kronenberger, 2007) but might also catalyze the diffusion of beneficial descriptive norms.

In line with Bauman and Geher (2002), errors in consensus estimates were significant predictors of proenvironmental intentions. It follows that participants’ erroneous perceptions, namely, norm misperceptions, influence their decision to engage in proenvironmental behavior. Despite the fact that the false consensus effect was displayed both before and after the course, accuracy in estimating consensus increased significantly after the intervention. The results of our study lend strong support to the hypothesis that accuracy in estimating consensus can be increased by an intervention designed to address internal locus of attention and the availability heuristic (Bauman & Geher, 2002). Indeed, our research showed that errors in consensus estimation predicted behavior intention, and, more specifically, endorsement of proenvironmental actions after the course (i.e., positive change in behavior intention) was accompanied by better estimation of behavior intention.

However, there is an alternative explanation to what might be the cause for the decreased errors in consensus estimates that were revealed after the course. It could be that increased consensus estimates of participants after they had taken part in the educational program are misinterpreted as increased accuracy, namely, participation in an environmental education program, which advocates proenvironmental behaviors, is expected to increase participants’ anticipations of how many people are likely to engage in such behaviors. As participants’ initial estimates were lower than actual intention, increased estimates and increased accuracy are not readily discernable. Accuracy could have been plainly manifested in a context where initial estimates were higher than actual intention. Therefore, future research should incorporate a wider array of behavior intention items to eliminate the possibility that elevation of estimates is mistaken as increased accuracy in the estimation of behavior intention. We acknowledge this possibility of an alternative explanation of decreased errors in consensus estimates after the course as a limitation of our study. To this we should add two more limitations. First, our study used a confined sample of students in a university setting and a specific collaborative learning arrangement, namely, the jigsaw approach. Therefore, our results cannot be readily generalized to other contexts regarding environmental education interventions. Second, the pre–post design we implemented cannot justify causal inferences, so the description of our results of statistical analyses should not be misunderstood as implying any type of causal relationships between measures.

The present study showed that the theoretical and methodological background of environmental education interventions can be enriched by incorporating consensus estimates for proenvironmental intentions in assessment procedures. By operationalizing descriptive norms within assessment frameworks, scholars in the field of environmental education can translate educational objectives into concrete, assessable predictions. For instance, instructors can use estimates of proenvironmental behavior intention to formulate educational goals and assess environmental education projects. Especially for participants with high prior levels of proenvironmental intention, environmental educators should schedule interventions by taking into account not only actual intention but also consensus estimates.