How much versus who: which social norms information is more effective?

ABSTRACT

We conduct an experiment to investigate how different types of information about social norms affect individuals’ stated contributions to a specific pro-environment program, a student ‘green fee’, in the context of a referendum. Compared to students that receive no information about peer contributions, on average, students that receive information about the dollar value range of contributions at peer institutions contribute less while students that learn about the high percentage of students voting ‘yes’ on green fee programs at peer institutions contribute more. The results are economically significant as the absolute values of both effects represent approximately 25% of average contributions. These results suggest that information about participation rates can be more effective than information about dollar amounts in encouraging contributions to environmental initiatives. Of interest to stated preference researchers, we find that results do not change when controlling for self-selection into survey completion.

KEYWORDS:

• Social norms
• stated preference
• pro-environment behaviour
• referendum

JEL CLASSIFICATION:

• Q50
• C90
• H41
• D03

Notes

¹ Throughout the manuscript, we use the term ‘extensive’ as a measure of how many people participate in the behavior and the term ‘intensive’ as a measure of how frequently or intensely people participate in a behavior.

² The Association for the Advancement of Sustainability in Higher Education (AASHE 2012) provides extensive information about higher education green fees in the United States (AASHE, Dedicated Student Fees for Renewable Energy and Energy Efficiency).

³ See Whitehead and Cherry (2007) for an overview of many studies documenting hypothetical bias and the approaches that researchers have taken to mitigate the bias.

⁴ At the design stage, we expected both Treatment 1 and Treatment 2 to increase mean student contributions, and hence, Treatment 3 to perhaps increase contributions even further. Based on the focus groups, we expected most students in the control group would be willing to contribute $0–$10 a semester. Evidently, we underestimated the number of students who would be much more generous absent peer information.

⁵ Students were randomly assigned to one of the four groups.

⁶ Almost 100% of MLAU students are full-time students, corresponding to 8–9 credit hours per academic year. Thus, they would likely interpret this range as about $8–$80 per year ($4–$40 per semester).

⁷ For OLS regressions, contributions at the endpoints of 0 and 100 are coded at those levels. Other contribution amounts are coded as the mid-point of the payment card intervals.

⁸ We utilize Stata’s intreg command for the interval regression.

⁹ From the full model, we drop fullloadlaundry, lightingwaste, heatacwaste, drivingtrips, recyclepaper and foodwaste because these are all variables that measure environmental behaviors and attitudes and all have p-values higher than 0.2 in the full OLS regression. We also drop age because this information is already largely captured in yearinschool. Finally, we drop #roommates because of its high p-value and lack of theoretical relationship with the level of one’s contributions to a green fee. A partial F-test and Wald test confirm that the dropped variables are insignificant predictors of green fee contributions. Results for the full model are available upon request.

¹⁰ We also examine a specification that interacts the treatments with observable characteristics to investigate whether the magnitudes of Treatments 1 and 2 depend on these characteristics. However, none of the interaction terms are significant.

¹¹ Heckman’s (1979) two-step procedure first estimates a probit model for sample inclusion and then includes the inverse of the Mill’s ratio in the regression equation. Stata’s maximum likelihood version of the Heckman model supports the Huber/White/sandwich estimator of variance and is hence our model of choice. The two-step results are similar and available upon request.

¹² This information is given to us in a way that does not identify the individual.

¹³ First-stage results from the selection equation are omitted from the table for readability but available upon request.