THOU SHALLT NOT SELL NATURE
2014
A STUDY ON HOW TABOO TRADE-OFFS AFFECT OUR PRO-ENVIRONMENTAL BEHAVIOUR.
BRITT TAMAR STIKVOORT
STOCKHOLM RESILIENCE CENTRE & BEIJER INSTITUTE OF ECOLOGICAL ECONOMICS |
Supervisors: Therese Lindahl, Tim Daw Examiner: Garry Peterson
Date: 2nd June 2014
CONTENTS
ABSTRACT _______________________________________________________________________________ 3 1 INTRODUCTION ______________________________________________________________________ 4 2 THEORETICAL FRAMEWORK: CONCEPTUAL MODEL ________________________________ 9 3 METHODS __________________________________________________________________________ 13 3.1 WHY EXPERIMENTS _________________________________________________________________ 13 3.2 PREPORATORY WORK _______________________________________________________________ 14 3.3 PARTICIPANTS, MATERIAL AND SET-UP __________________________________________________ 14 3.4 MEASUREMENTS ___________________________________________________________________ 16 Independent variables ___________________________________________________________________ 16 Dependent variables: donation behaviour ___________________________________________________ 19 3.5 DATA ANALYSIS ___________________________________________________________________ 20 Research-question 1 ____________________________________________________________________ 20 Research-question 2 ____________________________________________________________________ 20 4 RESULTS ____________________________________________________________________________ 22 4.1 DEMOGRAPHICS ____________________________________________________________________ 22 4.2 PRELIMINARY ANALYSIS _____________________________________________________________ 22 Conceptual model revisited ______________________________________________________________ 22 Trade-off treatment _____________________________________________________________________ 24 4.3 ANALYSES ________________________________________________________________________ 24 Decision-to-donate _____________________________________________________________________ 25 Donation size __________________________________________________________________________ 29 Additional analysis: Model averaging ______________________________________________________ 30 5 DISCUSSION & CONCLUSION _____________________________________________________________ 32 5.1 PARK VALUATION AND SOCIAL CONSCIOUSNESS _________________________________________________ 32 5.2 THAT WHICH AFFECTS THINKING DOES NOT AFFECT BEHAVIOUR ______________________________________ 33 5.3 MULTIMODEL-INFERENCE _______________________________________________________________ 34 5.4 CONTRIBUTION TO SCIENCE ______________________________________________________________ 35 5.5 APPLICABILITY TO SOCIETY _______________________________________________________________ 36 APPENDICES ______________________________________________________________________________ 38 APPENDIX A:DEFINITIONS AND FURTHER CLARIFICATIONS OF CONCEPTS ______________________________________ 38 APPENDIX B:ADDITIONAL TABLES, FIGURES AND ANALYSES _______________________________________________ 39 Table B1: Literature overview _____________________________________________________________ 40 Table B2: Linear Regression Table for Donation Sizes __________________________________________ 43 Analysis B3: Closer look at Mindsets _______________________________________________________ 44 Analysis B4: Limitations to study and disucssion of validity _____________________________________ 45 APPENDIX C:EXPERIMENT MATERIAL ______________________________________________________________ 46 Appendix C1: Mindset priming task ________________________________________________________ 46 Appendix C2: Hypothetical Case description and CV study ______________________________________ 48 Appendix C3: Group discussion instructions __________________________________________________ 50 Appendix C4: General survey part I ________________________________________________________ 51 Appendix C5: Trade-off treatment and Survey part II __________________________________________ 56 REFERENCES ___________________________________________________________________________ 58
ABSTRACT
Background: Humans are part of social-ecological systems, and preferably these systems are resilient, as this increases security of societal benefits derived from them. However, ecosystem- resilience is often threatened by loss/degradation of natural areas. Ideally, nature is only developed after careful cost/benefit analyses, but non-marketable ecosystem-services are often left unaccounted in land-development plans, resulting in loss of these systems and services.
One solution is incorporating ecosystem-services into cost/benefit analyses by putting a price- tag on these services. However, people do not accept the ensuing trade-offs, which pit sacred values (nature)against secular values (money). Such (taboo) trade-offs are morally offensive, yet they are necessary if we want to preserve ecosystems from ongoing degradation.
Moral cleansing – attempts to reaffirm one’s own moral position - is a reaction towards taboo trade-offs (i.e. in the shape of donations to charities) However, little is known about people’s behavioural response to assaults on sacred values related to the environment.
Aim: This study focuses on how trade-offs between environmental ‘sacred’ values and monetary values affect expressions of moral cleansing, namely pro-environmental behaviour in the shape of donations to an environmental charity. It investigates whether taboo trade-offs have effects on people’s environmental donations, and consequently the relative importance of trade-offs in such behaviour compared to other behaviour-influencing factors. Laboratory experiments (N=139) were conducted followed by regression analyses, and Multimodel- Inference techniques for data-analysis.
Conclusion: Participants’ decision-to-donate to an environmental charity is affected by social
consciousness and taboo trade-offs. Thus taboos are a factor influencing donation behaviour.
Discussion: Results suggest that people with a non-anthropocentric worldview believe that they ought to donate more, but in reality, other factors influence the real decision-to-donate. In this study it is exposure to a taboo trade-off and social consciousness that affects the real decision-to-donate. This supports prior evidence for moral cleansing effects and expands it to environmental fields. It also shows the added use of the explorative MMI-approach in social science-topics. Societal applicability is found in improvement of CBAs, and potential usage as behaviour-change technique. However, such usage deserves more attention on practicalities, feasibility and ethics.
1 INTRODUCTION
The environment and natural ecosystems are a vital ingredient to human wellbeing and economic development, and this understanding is gaining increased scientific support (Loreau et al., 2006; MA, 2005). Ecosystems are often intertwined or integrated with social (man-made) structures and systems, and such systems are typically self-organising (Portugali, 2000, cf.
Andersson & Colding, 2014) and contain social networks, economic markets and other institutions as well as ecological systems (Andersson & Colding, 2014). There is a growing recognition among scientists that humans are not only drivers, but ‘parts’ of these social- ecological systems (Marzluff et al., 2008). The idea of linking social with ecological systems was introduced to the wider public in the Millennium Ecosystem Assessment (2005) and had a big impact in both policy and scientific spheres (Carpenter et al., 2009). Resilience1 - the capacity of a system to respond to perturbations or shocks from within or outside the system, without changing its basic state and processes (Walker & Salt, 2006) – is an important ingredient of such social-ecological systems because it keeps systems away from thresholds1 - the ‘tipping’ point where a system crosses from one to another state - which may result in the system getting ‘stuck’ in a specific, possibly undesired, state. Resilient systems are far removed from such thresholds; they have a ‘buffer zone’, a greater operating space (Rockström et al., 2009). In terms of human wellbeing and economic activity, resilient social-ecological systems are more desirable since the benefits society derives from these systems is more secure.
However, the resilience of many systems is threatened by loss of natural areas (e.g. by city expansion, migration, population growth (Breuste, Haase, & Elmqvist, 2013)2 or agricultural expansion (Foley et al., 2011). These developments are increasingly under scientific scrutiny (MA, 2005), particularly because there are many social and societal benefits to be lost with nature’s disappearance (e.g. natural resources, but also recreation areas, medicinal resources etc. (TEEB, 2010). Such benefits derived from ecosystems are referred to as ‘ecosystem services’ (ESS) (Breuste et al., 2013) and they often occur in bundles1 - packages of interdependent services that are invariably tied together, as a result of complex relationships
1 This term is discussed in more depth in Appendix A
2 15 % of the total world population lived in urban areas in 1900, in 2000 this was increased to nearly 50 % (MA, 2005 (Chapter 22)). Urban areas are expected to absorb all of the coming decennia’s increase of world population to 9.5 billion people in 2050 (UN, 2012).
between certain ESS and the environment that provides them (Raudsepp-Hearne, Peterson, &
Bennett, 2010). Among these bundles are services that are tangible (e.g. wood from forests and oil) or less tangible (e.g. education, inspiration, recreation). Due to the interconnectedness of many services, losing one may also mean losing others, which makes trading-off individual services against one another virtually impossible.
Ideally, nature is only developed after careful consideration of the benefits and costs for doing so, i.e. balancing services lost and gained by land-development versus preservation. Policy- makers (spatial-planners) ideally base their development-plans on such careful cost-benefit analyses (CBAs)1 (Ropke, 2005, cf. Parks & Gowdy, 2013). However, non-marketable and invisible services are often left unaccounted (Bräuer, 2003) in land development plans, with the loss of ecosystems as a consequence. Underlying this, is the failure of the neoclassical economic theory which uses perfectly rational actors (homo economicus) to explain the behaviour of real human beings when it comes to non-marketable, uncertain and complex services such as ESS (Gintis, 2000).
One solution is incorporating ecosystem services into land development plans by ‘putting a price tag’ on these services. In essence, this involves making a trade-off between monetary units and nature3. However, there is a big problem with pricing the environment; many people do not accept such a trade-off to be made for reasons of principle. The basic problem with the trade-off is that it weighs sacred values (nature)against secular values (money). Sacred values are absolute and inviolable (P. E. Tetlock, 2003) and pitting such values against secular ones is considered unthinkable and outrageous; in other words, ‘taboo’ (Fiske and Tetlock (1997) give a more in-depth discussion about the emergence of this phenomenon).
Taboo trade-offs are morally offensive to us, independent of whether it is our own decision (Tetlock, Kristel, Elson, Green, & Lerner, 2000) or whether we are judging others who are considering the trade-off (Hanselmann & Tanner, 2008). The normal reaction to taboo trade- offs in avoidance or dismissal; such trade-offs ‘simply cannot be made’ (Oppenheimer &
Tetlock, 2008).
3 One complicating matter is the difficulty of stating exactly how ‘valuable’ the environment is (Banerjee, Crossman, & de Groot, 2013). To incorporate the monetary value of environmental services, scientists have developed economic valuation methods for natural resources (Bräuer, 2003). However, finding the right price tag is complicated due to the susceptibility of the outcomes to many different factors (e.g. Nyborg, 2000).
So, partly, ESS are omitted because many such services are not easily monetized (e.g.
recreation, health, education, or relaxation benefits). Additionally, such monetization and trading against other monetized goods makes people feel bad, and induces them to avoid it altogether. It is easier to ‘forget’ to incorporate nature in a CBA than to try and monetize it.
However, despite the apparent unease such trade-offs induce, economists insist that they are often unavoidable (due to scarcity of resources) if we want to give nature a fair chance in land development planning (Tetlock, 2003).
If people are unable to avoid a taboo trade-off, they roughly respond in two ways: exhibiting moral outrage and performing moral cleansing behaviour. Moral outrage often arises when the taboo trade-off is made by others. Bystanders reassure their own moral feeling by responding with outrage against the decision-maker. However, when people are exposed to the trade-off themselves, or someone close to them experiences it, people also tend to respond with acts of moral cleansing. Moral cleansing is behaviour with which an individual tries to reaffirm his own moral position within his social community by acting in extremely moral ways. This behaviour is more prevalent among people who feel contaminated by a taboo trade-off, which can occur even by merely thinking about the trade-off for only a short time (mere contemplation effect) (Tetlock et al., 2000). It can result in any kind of ‘cleansing’ behaviour, from willingness-to-volunteer for a good cause (Tetlock et al., 2000) and donating money to a charity (Sachdeva, Iliev, & Medin, 2009) to even the physical act of washing hands (Zhong &
Liljenquist, 2006).
So far, studies of moral cleansing have focused on human- and society oriented values, but little is known about people’s behavioural response to assaults on sacred values related to the environment. Tetlock and Oppenheimer (2008) even go as far as stating that the entire taboo trade-off topic concerning the environment is not well investigated. Such insights would be very interesting for policy makers (as well as for environmentally oriented NGOs). Either inducing or avoiding exposing people to taboo trade-offs can be a potential tool for eliciting behaviour change. There are many question-marks around this, however, such as for instance which kind of behaviour it would influence, to what extent, and with which force. Ethical considerations are also required before actually using such an approach. But basically, even before implementing anything on an ‘evidence-based’ on the moral cleansing effect, empirical research is a first requirement.
This study therefore focuses on how trade-offs between environmental ‘sacred’ values and monetary consideration affect people’s moral cleansing. I focus on moral cleansing since this is the prevalent response among people who are actually experiencing a taboo trade-off or are very closely related to the decision-process (Oppenheimer & Tetlock, 2008). Moreover, moral cleansing is an effect on one’s own behaviour, whereas moral outrage is a behaviour oriented towards others. The specific moral cleansing behaviour this thesis will focus on is pro- environmental behaviour as compensation for thinking about a ‘bad’ trade-off between the environment and money. Pro-environmental behaviour is on its own account a very broad field of ‘behaviours’ that has to be more concrete to measure it. As a proxy for general pro- environmental behaviour this thesis considers people’s willingness-to-donate money to an environmental cause4.
This thesis focuses on two research-questions. Firstly, I am interested in the explanatory power of trade-offs on pro-environmental donation behaviour, on top of other donation-influencing factors.
RQ-1 will be tested with data from a controlled ‘laboratory’ experiment. As experiments happen in a pre-defined and constant setting, causality between variables can be tested, by keeping all variables constant (or randomized) except for the variable(s) of interest. In this case the variables of interest are trade-off type (either taboo or not taboo) and donation behaviour.
The other donation-influencing factors will be mentioned in the next chapter.
One problem with this approach, however, is that there is uncertainty about which factors contribute to pro-environmental behaviour (the next chapter will go deeper into these factors).
Model selection1 – decision on which variables to exclude or include in analyses – is not straightforward here. It is therefore advisable to take an information theoretic approach (Burnham & Anderson, 2002), and more specifically to take a so-called Multimodel-Inference (MMI) approach, rather than relying on the testing of a single model against an inflexible cut-
4 In the Methods chapter (3) I will elaborate on the way this is measured.
RQ-1: Does exposure to a taboo trade-off have considerable effect on people’s pro- environmental donation behaviour?
off point (usually a significance-level of p=0.05). MMI is a quite novel approach (although it is increasingly used in (behavioural) ecological disciplines (Symonds & Moussalli, 2011)). It looks all possible models given a certain set of variables, rather than testing one single model.
Because of uncertainty in model selection in this study, (which is the case in many studies in social sciences) I apply and test the MMI approach, both to find answers to the second research- question, and to assess the potential contribution of the MMI approach to a social-science topic.
The second research-question I will address with this approach is:
RQ-2: What is the relative importance of the influence of trade-offs on pro-environmental (donation) behaviour, in comparison to other factors?
2 THEORETICAL FRAMEWORK: CONCEPTUAL MODEL
In this section I outline the conceptual model that guided the experiment design and ongoing analyses. Many factors are believed to have an influence on environmental behaviour independent to the effects of moral compensation for taboo trade-offs5. In a study on assessing effects of trade-offs, it would be remiss to ignore effects of other factors, since these might add variation to the results that could drown out the effects of trade-offs. Measuring other factors allows taking these effects into consideration, and allows comparison of relative importance of trade-offs to other well-investigated factors. Therefore, based on evidence from previous research (see also Appendix B1) I develop a conceptual model in this chapter, containing the below-discussed factors.
Mindsets - here referred to as a person’s personal set of preferences, or perspectives, which influence decisions and behaviour. It is the set of thoughts and dispositions that is salient at a given moment in time. For instance, a person can have a social mindset, meaning he or she puts much emphasis on things that are important from a social perspective. People’s mindsets are diverse, but in this thesis the dichotomy between social and individual mindset are central. An individually minded person adopts “a personal perspective when confronted with personal, self-regarding wants and interests” whereas socially-minded people adopt a social perspective (Mill et al., 2007). Such social or individual mindsets have been proven to influence people’s valuation of nature (e.g. (Howley, Hynes, & O’Donoghue, 2010))6.
Mindsets can be affected by contextual factors as trivial as the task description during experiments (Ami, Chanel, Aprahamian, Joule, & Luchini, 2008)7. These contextual influences are explained by the mental effects of accessibility – a mental state where a whole set of preferences and mental concepts is more accessible than others, as a result of the activation of
5 For instance, a recent review of previous studies found 18 different variables to affect pro-environmental behaviour (Gifford & Nilsson, 2014).
6 The influence of social versus individual mindsets (also called ‘citizen’ versus ‘consumer’ mindsets) on valuation has been a point of critique in economic valuation methods (Nyborg, 2000).
7 For instance, both Ovaskainen and Kniivilä (2005) and Mills and colleagues (2007) used question framing – the way the questions are formulated – to distinguish between social and individual mindsets. Howley and colleagues (2010) asked participants to ascribe value to attributes of a nature area, according to either individual or society’s preferences.
one or a few mental concepts (Förster & Liberman, 2007), i.e. if one concept becomes salient (receives cognitive attention) then this makes related concepts more accessible (salient).
Salience of sets of concepts can be influenced by external factors, and activating these on purpose is called ‘priming’ (Förster & Liberman, 2007). In other words, priming one concept makes salient other related concepts8.
Priming techniques can thus make (social or individual) mindsets more or less salient, by simple means like for instance textual differences. As it was shown to affect environmental valuation among participants, it seems plausible that it could also have an effect on pro- environmental behaviour. However, to my knowledge this effect has not yet been investigated.
Altruism is a construct that can be defined as an individual’s desire to increase other people’s welfare (Piliavin, 2001). As such, people who are more altruistically inclined often think in ways that evoke more social community thinking, whereas people with less altruism are prone to think in more selfish terms (Hypothesis of Geller, cf. Kollmuss & Agyeman, 2002). That this can influence pro-environmental behaviour is discussed in a review by Steg and Vlek (2009). Altruism is therefore one of the factors that will be taken up in this conceptual model, along with the above mentioned mindsets.
Similarly, people who belong to a social setting or group have to conform to the social norms9 of that group in order to belong (Hackman, 1992). Prior research has shown that many kinds of behaviour, including pro-environmental behaviour, is affected by social norms (for instance Allcott, 2011; Chen, Lupi, He, & Liu, 2009; Goldstein, Cialdini, & Griskevicius, 2008).
However, not everyone is as sensitive to adhering to the norm (Allen, 1966). (Pro- environmental) behaviour has in the past often been looked at using the theory of planned behaviour (Kollmuss & Agyeman, 2002). This model was first posited by Ajzen and Fishbein and aimed to investigate which factors influence people’s behaviour (Ajzen, 1985). One of
8 For instance, the way a scenario was written up in WOII terms (e.g. ‘blitzkrieg’) or in Vietnam War terms (e.g.
‘quick strike invasion’) affected how US citizens decided to vote on direct foreign intervention (Gilovich, 1989, cf. Förster & Liberman, 2007). Also interesting is the effect of primes on behaviour: walking speed of
participants primed with the concept ‘elderly’ was slower than those primed with neutral words (Bargh et al., 1996, cf. Förster & Liberman, 2007). Such effects are found across many kinds of behaviour; e.g. in resource- dilemma tasks, participants who were primed for cooperation behaved more cooperatively than their non-primed counterparts (Bargh et al., 2001, cf. Förster & Liberman, 2007).
9 Social norms in this thesis are held to be ‘injunctive’ – a set of social rules that society feels people should uphold (Burger, 2001) and they are ‘subjective’ – it is not the norm itself, but a person’s belief about what others hold as a norm that matters (Rivis & Sheeran, 2003)
those factors is Social-Norm-Conformity10 - the extent to which people want to live up to the social norms that they think others hold. This Social-Norm-Conformity can be an influence to pro-environmental behaviour and is thus considered in this conceptual model.
Additionally, environmental or ecological beliefs (referred to from here on as worldviews) are believed to affect pro-environmental behaviour (Bamberg & Möser, 2007). Such beliefs can be measured in many ways, but for this thesis I have applied a frequently used and proven valid method called the New Ecological Paradigm Scale (Dunlap, Van Liere, Mertig, & Jones, 2000), which consists of 15 Likert-scale statements that together form a person’s ecological worldview score. Both altruism and NEP were found to influence pro-environmental behaviour independently from one another (Clark, Kotchen, & Moore, 2003) which indicates that ecological worldview is a factor to reckon with, and therefore is incorporated into the conceptual model.
Perceived behavioural control is a construct that measures how much control people perceive to have in the desired (in this case pro-environmental) behaviour (Ajzen, 1985). Bandura developed a similar concept named self-efficacy’ which is one’s belief of capability to do a certain action or behaviour (Bandura, 1977). Gifford and Nilsson (2014) report it as one of the factors influencing environmental behaviour, and consequently it finds a place in this conceptual model.
Gifford and Nilsson also mention demographic variables such as for instance age, gender and social class. The latter can be linked to the fact that people with a higher income11 have less of a budget constraint than those with less money to spend. Therefore the conceptual model is supplemented with these three demographic indicators: age, gender and income.
10 The other two are attitudes and perceived behavioural control
11 Income is not necessarily the same as social class but very often social class is measured by asking respondents for their income (e.g. Lachman & Weaver, 1998)
On the basis of the above presented factors found in prior literature, and the potential effects of taboo trade-offs, I propose the conceptual model of factors as proposed
in Figure 1.
Figure 1: Conceptual model of factors influencing environmental donation behaviour. The threefold divide into demographics, external and internal factors is only for clarity, but is not part of the analysis in this thesis.
3 METHODS
3.1 WHY EXPERIMENTS
The thesis took a behavioural experiment approach, which is the preferred method for measurement of participants’ actual, rather than self-reported behaviour. I focused on donation behaviour as an expression of pro-environmental behaviour because this was easy to measure objectively (people either donate, or not, no need for researcher interpretation) and unobtrusively. Another benefit to (laboratory) experiments)12 is that it allows investigation of causality between variables by keeping all but one variable constant in a controlled environment (Levitt & List, 2007). Although this controlled environment does decrease the strength of external validity13 (it makes the experiment more artificial), internal validity14 is strengthened, because there is more control over measurements and variables (Coolican, 2009).
I revisit the issues with validity, as well as other limitations to the study, in Appendix B4.
Many of the studies mentioned in the previous chapter are basing results on self-reported behaviours, which entails asking participants about what they would do. However, what people report they would do, and what they actually do, can differ. It is often the case in self-report surveys involving payment or donation behaviour that participants suffer from hypothetical bias (Loomis, 2011), which is when people report a different willingness-to-pay/donate in a hypothetical case, than what they actually do in a real situation15. Instead of relying exclusively on self-reports, this thesis is interested in the actual donation behaviour of people, and to get to that, it is necessary to go beyond survey designs. To say something about both the hypothetical and real donation behaviour, I measured both self-reported as well as real behaviour. For measuring real behaviour, an experimental set-up was necessary, since surveys generally do not allow the observation of real behaviour of participants.
12 With a laboratory setting is meant that the experiment was not performed ‘in the field’ or in participants’
natural surroundings, but in a setting that was kept constant unchanged throughout the experiment period, to keep all variables the same/constant, except for the one we changed deliberately.
13 External validity: how well results of a study can be generalized to other situations (Schram, 2005)
14 Internal validity: confidence with which causal conclusions can be drawn from a study (Schram, 2005)
15 The sources for such biases are as of yet still ill understood, and although the tendency of participants is to overstate their willingness in hypothetical cases, this is not always the case (Loomis, 2011).
3.2 PREPORATORY WORK
Before designing the actual experiment-material, a pilot study was conducted, aimed at checking if the treatment for social and individual mindsets, adopted from Howley and colleagues (2010) worked. They used nature-attributes relating to the countryside and found that participants’ values differed significantly between the two treatments. To test this, respondents of an online survey engaged in the task later used in the experiment (see Appendix C for main experiment-material) – they gave grades (one to ten, one being lowest) to twelve attributes of an urban park. One group was asked to give these values for themselves, the other group was asked to answer with society and their community in mind. This pilot-survey yielded 63 completed replies. Because not all results were normally distributed, a Wilcoxon rank-sum non-parametric was used to see if ascribed park values differed between the treatment groups16. It was decided that the task would be applicable to the real experiment.
3.3 PARTICIPANTS, MATERIAL AND SET-UP
The sample of participants consisted of people responding to poster-advertisements that were distributed over campus and public transport-hubs. Participants were asked to mail to the experiment leader and consequently they were scheduled into one of the groups. They were enticed with a participation fee of 200 SEK, which could in the end be partially donated to a charity. Participants did not know of this possibility, however, until the very end of the experiment.
Participants filled out a consent-form and then participated (randomly assigned) in a group either with social or individual mindset-primes: valuing an urban park by grading a list of attributes of that park according to importance (equal to that described in the pilot). They filled out a survey
16 There was marginally significant difference between the groups’ mean age, those in the citizen-treatment were older. This may rather show an income effect rather than age effect. Since income was not measured, this cannot be said for certain. Income was added as a demographic variable in the final experiment material.
Taboo trade-off
Option 1: Preservation of natural forest --> preventing
extinction of animals or plants
Option 2: Spending money on buying material items, such as brand clothing or designer
furniture
Tragic trade-off
Option 1: Preservation of natural forest --> preventing extinction of animals & plants
Option 2: Prevention of fishing methods that are harmful to marine mammals
Figure 2: Schematic representation of trade-off options per treatment. Left are the options for taboo treatment groups, right the options for tragic treatment groups.
measuring demographic variables such as age and gender, as well as personal characteristics such as Social-Norm-Conformity, altruism and ecological orientation (these variables are elaborated upon further down). Moreover, participants answered a question about willingness- to-pay for the continued existence of an urban park, which is a standard question used in economic valuation studies17. Next, all groups were randomly re-assigned to either a taboo or a tragic trade-off, and asked to give an individual answer to the choice provided by the experiment leader (i.e. choose between option 1 or 2, see Figure 2). To measure participants’
willingness-to-donate they were asked for their hypothetical donation to an environmental charity (reminding them of the 200 SEK fee they were given). Finally, participants’ real donation was observed after the experiment was over. Participants were given an envelope containing their participation fee, and consequently asked to go behind a green partitioning screen, take from the envelope the money they wished for themselves, and drop the envelope with whatever left for donation to Greenpeace18 in a money-box. This way the default option for participants was to donate, and the choice to take money for themselves cost (a little) extra effort. According to Thaler and Sunstein, defaults may nudge people into following the default option (2008). Moreover, it is often easier for people to forego money than to ‘give away’ what one already owns (Kahneman & Tversky, 1979). After participants had left their envelopes andl eft the room, the envelopes were recovered, and donation size noted down. Figure 3 gives a schematic overview of the entire experiment set-up.
Figure 3: Schematic overview of study set-up
17 This question and the average score on park attributes, are indicators for the success or failure of the prime.
18 I expected many international students to join the experiments, so having an internationally well-known NGO was pivotal. Hence I chose Greenpeace. Appendix B4 discusses this choice more elaborately.
The material used in the experiment consisted of several papers of reading material, as well as pen-and-paper survey questions. The experiment was conducted room that allowed participants to sit in a circle, approximately one-and-a-half meter apart, with a table in between to write upon. The experiment leader sat on the opposite side of the room, unless explaining something, keeping observer-influence at a minimum. Instruction and explanation language was English.
All written material can be found in Appendix C.
3.4 MEASUREMENTS
INDEPENDENT VARIABLES19
Mindset prime: Park Valuation
The first treatment was about priming participants with either social or individual mindsets.
Howley and colleagues (2010) used a list of attributes to put participants in either a consumer or citizen mindset, by asking participants to value these attributes either from an individual or a social/societal perspective (on a scale from one to ten). One half of participants was specifically asked to score the attributes (characteristics) in order of importance for themselves individually, whereas others were asked to score them on importance to their community and society. This thesis used a similar method as Howley and colleagues, although it adopted a more urban-related attribute list (see Appendix C) to fit more to the reality of expected participants. This park valuation task was both the prime to get participants into social or individual mindsets, and a measure for whether these groups differed in their park valuation because of this priming (ceterus paribus due to randomization). There ought to be a difference in park valuation averages between the citizen and consumer mindsets, if the prime succeeded.
Moreover, the average individual valuation in essence represents the importance a person ascribes to social attributes of (urban) nature. Thus, those scoring a high average on this task seem to assign more importance to the social characteristics of nature.
19 Presented in order of appearance in the experiment material (Appendix C)
Willingness-to-Pay
Participants did a classical economic valuation study, a stated preference ‘contingent valuation method’ (Carson & Hanemann, 2005). They were asked for their willingness-to-pay (WTP) for the preservation of a park, in terms of a percentage on top of their monthly housing rent, that they would be willing to spend on such preservation (asking for percentages of monthly rents reduces effects of income). WTP questions come in many different formats (Carson &
Hanemann, 2005) but for this study I used open-ended questions. Participants were asked to read a hypothetical scenario and give their willingness-to-pay extra monthly rent for preserving the (hypothetical) park close-by. The question was asked in two steps, first asking if participants were willing to pay, thereafter how much20. This measure was used to assess whether the priming of mindsets had been successful, as a successful prime ought to lead to higher willingness-to-pay.
Demographics
The survey asked standard demographic questions: age, gender, whether participants were studying or not, how long they were studying, educational background, nationality and how much they earn (5-point scale).
Social-Norm-Conformity
To measure people’s need to conform to what they think others see as the ‘norm’ Ajzen designed questions probing into Social-Norm-Conformity (author referred to this factor as
‘subjective norms’) (Ajzen, 1985). This study used a measurement averaging the score on three items Ajzen suggests (on 5-point scales ranging from strongly-disagree to strongly-agree, e.g.
opinion of important people on environment & importance of their opinions). High scores imply higher levels of need to comply with perceived social norms (based on self-report).
Altruism
This thesis adopted four items related to donation behaviour from the 20-item Self-Report Altruism Scale (Rushton, Chrisjohn, & Fekken, 1981). The four items (5-point scales ranging from strongly-disagree to strongly-agree, e.g. give money to charities, or likelihood of volunteering) were summarized into one score on Altruism. High scores on this scale imply higher levels of altruism (based on self-report).
20 This was done to distinguish protest responses from zero responses; participants stating ‘no’ (protest responses) were asked for a further explanation of their choice.
Environmental worldview
To see how environmentally oriented worldviews of participants were, the experiment survey incorporated a copy of the New Ecological Paradigm (NEP) Scale developed by Dunlap and colleagues (2000) to measure people’s environmentalism. Its averaged score was used in the data analysis as an indicator of ecological worldviews. High scores imply more inclination towards non-anthropocentric ecological worldviews.
Trade-off treatment
The values for the trade-offs were inspired by Baron and Spranca (1997). From their list of 14 sacred values the highest scoring environmental value (their 2nd experiment) was taken:
“Prevent the destruction of natural forests by human activity, resulting in the extinction of plant and animal species forever”. For the tragic trade-off, an equally sacred value from the same article that was valued equally high was used: “Prevent fishing methods that lead to the painful death of dolphins and other marine animals.” For the taboo trade-off, the environmental value was put against a secular value, which was defined as: “Spending the money on material objects, such as brand clothing, design furniture or going out more often”. See also Figure 2.
Participants had to choose to which of the options they wanted to give their freely spendable money21. They were explicitly asked to keep their choice private, because expressing moral outrage by talking about it could affect the expression of moral cleansing (Tetlock et al., 2000).
Sacred Value measurement
To check whether participants judged the options in the trade-off as sacred or not, the experiment included an adoption of the Sacred Value Measure from Hanselmann and Tanner (2008). In previous studies this measurement has shown good internal consistency (Tanner et al., 2007, cf. Hanselmann & Tanner, 2008). Thus, I used this measurement to judge whether the trade-off treatment was successful in inducing either taboo or tragic decision-situations. For each of the two options that were to be traded-off, participants were asked five questions about how sacred these were to them (5-point scal es ranging from strongly disagree to strongly agree, e.g. on quantifiability of option in terms of money or righteousness of adding it to a personal cost/benefit consideration). Higher scores indicated the option was more ‘sacred’.
21 The exact question was: “Suppose you were confronted with a situation that demands you to make a decision of spending some of your freely spendable money (which you have left after paying for living expenses, rent, bills etc.) to either of the following two options (tick one box only).”
Positive & Negative Affect Schedule
To see if the trade-off treatment had any effect on the emotions experienced by participants, I included a measurement on emotion. In the pilot the Negative Emotion Scale from Hanselmann and Tanner (2008) was used but highly criticized for being ‘out of sync’ with what respondents were feeling in the pilot phase. Therefore the experiment itself utilized instead Watson et al.’s (1988) PANAS (Positive & Negative Affect Schedule) which was improved by Thompson (2007) for international applications.
Self-efficacy
A third tenet of the behaviour-explaining Theory of Planned Behaviour was that of perceived behavioural control, or self-efficacy (Ajzen, 1985). Since donation itself is not a behaviour that is in any way ‘difficult’ given the situation, I instead decided to measure the perceived difficulty participants had with making the decision in the trade-off (specific question: “Was this a difficult decision, Y/N”). Although the behaviour measured was donation, it could still be the case that the feeling of self-efficacy (or lack thereof) related to the trade-off, had a spill-over effect on other behaviour as well (such as donations).
DEPENDENT VARIABLES: DONATION BEHAVIOUR
This study distinguishes two steps in donation behaviour. A first step for participants was to decide if they wanted to donate at all, i.e. decision-to-donate. Participants were asked if they were willing to donate part of their participant fee to an environmental charity (non-specified, but Greenpeace was given as an example). After deciding if they wanted to donate22, participants decided how much to donate, i.e. donation amount/size (“How much of your final participant fee (200 SEK) would you be willing to donate?”). Thus far the hypothetical donation behaviour was measured. The actual donation behaviour measurement has been discussed above, but briefly put: participants received an envelope containing their participation-fee, and were asked to individually decide if and what to donate, and then drop the envelope in a donation box, taking out whatever money they wanted to keep for themselves.
.
22 NB those responding ‘no’ to the donate question were asked if they would donate to another charity or none at all, to discover if participants were against donating or against the idea of an environmental charity only.
3.5 DATA ANALYSIS
RESEARCH-QUESTION 1
The first RQ is answered by using logistic and linear regressions, depending on whether the dependent variable is dichotomous (logistic regression) or continuous (linear regression). The regressions are performed with R (R-Studio Version 0.98.501, 2009-13). Regressions were run with models containing all independent variables (discussed above and introduced in chapter 2: conceptual model) as well as models containing all-variables-except-trade-offs.
As a next step, the AICc criterion for comparison was used to compare the models with and without the trade-offs-variable. The AICc is a derived measure for assessing the relative quality of a model, based on the Akaike information criterion (AIC) (Akaike, 1974). An AICc is the preferred alternative according to Burnham and Anderson (2002) for small sample sizes and large model-sets. AIC(c)s do not tell anything about quality of a model in absolute senses, but it can be used in a relative comparison between models (i.e. between the model with and without trade-offs). In addition to the suggested differences obtained from AICcs, I also ran ANOVA tests that compared the full model with the model-without-trade-offs, which would be the formal way to test whether trade-offs are a contribution to the model. Based on this, RQ- 1 could be answered.
RESEARCH-QUESTION 2
Research-question 2 investigated the relative importance of trade-offs as a factor in comparison to other factors on donation behaviour. Due to the large amount of potential factors of influence, it is best to take an MMI approach (Burnham & Anderson, 2002). MMI is particularly useful in exploratory research when the ‘right’ model is uncertain, and when there are many potential variables (Burnham & Anderson, 2002). Instead of judging hypotheses with arbitrary cut-off points that are used in conventional hypothesis-tests, MMI looks at the comparison of all possible models, their fit to ‘reality’ (data found in study) and the relative importance of variables within this list of models (Burnham & Anderson, 2002; Symonds &
Moussalli, 2011). Although it is less strong in giving ‘evidence’ for the effects of specific
variables, it can suggest relative importance of a set of variables and models23, and is therefore excellently suited for answering RQ2.
I ran a MuMIn package in R (R-Studio Version 0.98.501, MuMIn, 2009-13) that analysed regressions of all possible configurations of a predefined set of variables (those mentioned in the conceptual model) on a defined dependent variable (hypothetical or real decision to donate).
Degrees of freedom, coefficients, AICcs, relative likelihood and Akaike weights (indicating the normalised likelihood of a model being the best) were calculated, which allowed for comparison between models. From this analysis, inference could be made about which variables were strongest predictors across all models, as well as the effect size of these variables, which answered RQ2.
23 If more than one model is plausible, then model averaging – when an average model of all models is formulated to explain the dependent variable - can be insightful. In such an approach one looks at the overall importance of separate variables across all possible models and concludes which variables are more or less important ‘overall’.
4 RESULTS
This chapter is separated into three sections, the first of which briefly presents the participant pool in terms of demographics. A second section revisits the conceptual model of chapter 2, on basis of the quality of the data. It argues for the omission of two variables, and illustrates how the trade-off treatment seems to have been successful. The third section then reports on the effects of trade-offs on donation behaviour, by using regression analyses and the MMI approach to answer the research-questions. Variables are in italics in this chapter, to distinguish them from the conceptual factors they represent.
4.1 DEMOGRAPHICS
Table 1 gives an overview of demographic variables of the sample. In total 139 people participated, during the autumn of 2013. Although not specifically targeted, participants were predominantly students (126, ranging from bachelor to PhD levels). There were more female participants (59%) and although the age-range spread from 19 to 47, 80 % of participants was 28 or younger.
The participant-pool thus was predominantly female, Western and young.
4.2 PRELIMINARY ANALYSIS
CONCEPTUAL MODEL REVISITED
Before running regressions with the variables from the conceptual model, I first explored the data for accuracy and precision24. Two problematic variables require discussion on this behalf:
self-efficacy and mindsets.
24 Accuracy here is the closeness of a measure to the actual concept it is supposed to measure, whereas precision gives an indication of how replicable the measurement is (i.e. if repeated in similar conditions, will it give similar results or widely different ones) (Taylor, 1997).
Table 1: Descriptive Statistics of participant sample (N=139)
Mean or % SD
Age 25.63 4.796
Gender (male) 41 %
Student 91 %
Monthly income
(low-high=1-5) 1.63 .735 Nationality
Swedish 12 %
Greek 11 %
German 9 %
Other nationalities 69 %
I excluded the variable self-efficacy for two reasons. Firstly, the measurement was not accurate.
The measurement (question about perceived difficulty of the decision) was interpreted in two ways; it was not clear for participants whether the question was about the trade-off, or about the decision to donate or not. This conclusion was made from the comments participants were allowed to write in addition to answering the question. The data was promising in terms of effects on the dependent variable, however it was not clear what the data actually captured, since the question itself was not clear. In other words, its accuracy, ability to capture the actual thing I intended to capture, was uncertain due to the ill-constructed question.
Secondly, on a more theoretical basis it could be claimed that asking for difficulty is related to - but not a direct measurement of - self-efficacy. Self-efficacy (perceived behavioural control) affects difficult behaviour (Ajzen, 1985), so asking participants if they perceived a task to be difficult is a first step in finding out effects of self-efficacy. However, a next step ought to have been: enquiring after participants’ perceived behavioural control, i.e. their belief in ability to overcome said difficulty. This would have led to a theoretically sound measurement.
A second variable that was excluded from the conceptual model for the regression-analyses was mindsets. I measured ‘temporary’ mindsets, induced by a prime, which made certain mindsets more salient in the minds of participants (successfully (Howley et al., 2010)).
Comparing the data from the groups who received individually oriented with those receiving socially oriented instructions revealed no difference between the groups25. The prime possibly failed to induce (strong enough) mindsets, and therefore the variable of which prime participants received (binary) is excluded from the regression analyses. It would be unfair to involve a variable that measured nothing, but which suggests that it represents the non-effect of mindsets.
25 NB: no significant difference in park valuation between social versus individually primed participants (t=.985, p=.327, 2-tailed). Neither was there significant difference between these on behalf of the measure of willingness- to-pay in the economic valuation part of the experiment (binary: yes/no) (Mann-Whitney U=2382.5, p=.748). See Appendix B3 for a more detailed account of this analysis.
TRADE-OFF TREATMENT
As a manipulation check, the extent of participants’ association of sacredness with trade-off options was investigated.
The Sacred Value Measurement was used to assess participants’ endorsement of sacred values.
Participants were asked to choose between I) donating to a cause of preserving a natural forest and II) (in taboo groups) buying more material items or (in tragic groups) donating to a charity preventing harmful fishing methods (see Figure 2 in previous chapter). Participants filled out the SVM questions for each of the options.
The value ascribed to the first was equal in
both treatment groups (Figure 4, (a)). This was expected since the option (preserving natural forest) was the same for both groups. It underlines that participants were truly randomized in their assignment to treatments. The value ascribed to prevention of harmful fishing methods was significantly higher than the ascribed value to buying more material goods (Mann-Whitney Rank Sum Test Z=-3.520, p=.000). This suggests that the treatment was successful in exposing participants to either two sacred values of equal value (tragic treatment) or to one sacred and one secular value, where the sacred value is valued far above the secular one (taboo treatment).
4.3 ANALYSES
The choice of donating money was considered a two-step process; first deciding to donate or not, secondly deciding how much to donate. I mainly focus on the donation-decision, but I will also briefly mention donation-size (although no answers to the research-questions can be given
68.78
57.06 69.22
80.77
40 50 60 70 80 90
Value-score for "forest conservation" (a)
Value-score for "prevent harmful fishing" or
"material goods" (b)
Taboo trade-off groups Tragic trade-off groups
Figure 4:Mean Rank of Mann-Whitney Rank Sum Test (horizontal axis), comparing groups in taboo treatments with tragic treatments. For (a) there was no significant difference between groups, for (b) there was a significant difference.
due to too low sample-sizes), and I conclude this section with an additional analysis. For clarity, I repeat the research-questions here:
DECISION-TO-DONATE
Research-question 1
Table 2 presents the regressions on decisions-to-donate (donation-willingness) for the hypothetical and real situation respectively, with and without the variable trade-offs. In the full model very few variables were able to significantly contribute to hypothetical donation- willingness, except for the variable ecological worldview (z=2.035, p=0.042). Its contribution is positive, indicating that those with a more non-anthropocentric ecological worldview are hypothetically more willing to donate. Age, gender, and all the other variables did not contribute to the hypothetical decision. When we compare the full model with the model without the trade-offs, nothing significantly changes in the results. The only significant predictor is still ecological worldview (z=2.061, p=0.039). The AICcs suggest that the second model (without trade-offs) is even a little bit better, (although the difference is less than 2 AIC, which does not give enough evidence to assume that the general model is less than the second model (Burnham & Anderson, 2002)). In any case it seems safe to say that trade-offs do not have a contributing effect on the model in the hypothetical case.
In the real donation situation, the decision-to-donate26 is not affected at all by ecological worldviews, but by the variable park valuation (z=2.577, p=0.010). Moreover, in the real situation the trade-off variable affected the decision-to-donate (z=-2.654, p=0.008). This result indicates that participants in a taboo trade-off were more likely to donate (taboo was coded with zero, hence the inverted relationship).
26 NB. This ’decision’ is measured differently than the hypothetical one. In the hypothetical case I asked participants if they were willing to donate. In the real situation, participants are ascribed to a non-donation or donation decision on behalf of their actual behaviour (i.e. donation or not).
RQ-1: Does exposure to a taboo trade-off have considerable effect on people’s pro- environmental donation behaviour?
RQ-2: What is the relative importance of the influence of trade-offs on pro-environmental (donation) behaviour, in comparison to other factors?
Comparing the full model with the model without trade-offs results in a decrease of the AICcs with more than 5, which is according to Burnham and Anderson (2002) enough reason to doubt the higher scoring model in favour of the lower-scoring one (i.e. higher numbers on an AICc scale are worse). In other words, there is reason to suspect that the model containing the extra variable trade-offs is better, despite that it is larger and thus less parsimonious. Note that AICcs take into account this aspect of parsimony by ‘punishing’ models that are bigger. Despite the increase in size, however, the model with trade-offs is still better according to AICcs.
With ANOVA tests (Chi square) I compared the model with and without trade-offs (respectively in the hypothetical and real situation). As expected, the model with and without trade-offs in the hypothetical case do not significantly differ from one another (Chi2=-0.547, p= 0.459), whereas there is a significant difference in the real decision situation (Chi2=-7.483, p=0.006). Thus, it seems that in a real donation situation, people’s behaviour of deciding to donate can be partially explained by their prior experiencing of a trade-off.
Table 2: Binary Logistic Regression of decision-to-donate in (left) hypothetical & (right) real situation
Decision to donate
Hypothetical Donation Real Donation
With trade-offs Without trade-offs With trade-offs Without trade-offs Determinant z value Pr (>z) z value Pr (>z) z value Pr (>z) z value Pr (>z) Intercept -2.046 0.041 -2.053 0.040 -2.15 0.032 -2.236 0.025
Age -1.508 0.132 -1.436 0.151 -0.278 0.781 0.014 0.989
Gender 0.101 0.920 0.16 0.873 -1.283 0.200 -1.002 0.316
Income 1.206 0.228 1.123 0.261 1.089 0.276 0.71 0.478
Trade-offs -0.738 0.461 -2.654 0.008
Altruism 1.358 0.175 1.355 0.175 0.288 0.774 0.252 0.801
Social norm
conf. 0.343 0.732 0.246 0.806 1.449 0.147 1.164 0.244
Ecol.
worldview 2.035 0.042 2.061 0.039 0.314 0.754 0.509 0.611
Park
valuation 0.305 0.760 0.231 0.818 2.577 0.010 2.317 0.021
AICc 181.63 179.87 180.40 185.57
Intermediate summary
The regressions show that, although trade-offs do not affect people’s own self-reported donation behaviour, it does contribute to their actual donation behaviour. The direction is that those experiencing taboo trade-offs are more likely to decide to donate money to an environmental charity. RQ-1 can therefore be answered with a confirmative ‘yes, trade-off types do affect pro-environmental donation behaviour’.
Research-question 2
Research-question 2 focused on relative importance of trade-offs in comparison to other well- investigated factors on pro-environmental behaviour. To answer this question, a Multimodel- Inference (MMI) approach was used. All possible combinations of variables in this model lead to 2^8(256) models, which were (automatically performed by MuMIn) put into regressions.
In the hypothetical decision situation, the best model (out of a set of 256) included only ecological worldview and altruism as variables and had a weight of 4.3%, i.e. this is the probability that his model is the best one for the data. It took 33 models to get to the 50% weight point (meaning that the remaining 223 were able accounting for the other 50%). Model 2 to 11 (sorted according to delta AICc score) all had a delta AICc score <2, so according to Burnham and colleagues (2002, 2011) there is enough statistical support to consider these models as alternatives to the top one (NB. These authors consider all values below 7 to have ‘less’ support but still viable (Burnham et al., 2011), 154 of the 256 models fall within this range).
In the real decision situation, the best model included park valuation and trade-offs as variables, and this model had a weight of 6.2%. Seventeen models together accounted for 50%
of the weight. All these models contained the variable trade-offs, as well as park valuation.
Other variables were more sporadic in their occurrence. The first nine models all had a delta AICc lower than two, meaning they are considerable contenders for being the ‘best’ model (Burnham et al., 2011).
At a variable level, I investigated the relative importance of each variable as the sum of the weights of the models that contained that variable. Table 3 shows the results for (left) the hypothetical and (right) the real decision situation (see also Figure 5). It shows that the most important variable in the hypothetical case was ecological worldview which starred in models that accounted for 80.2% of the AICcs weight. A poor ‘runner up’ is altruism, with only 58.2%
of the AICcs weight explained by models containing this variable. In the real situation, both park valuation and trade-offs were in models that explained AICc weight in 91.8% and 88.5%
respectively.
0.80 0.58 0.46 0.42 0.30 0.28 0.28 0.27
0.28 0.28 0.25 0.38 0.50 0.88
0.37 0.92
Ecological worldview Altruism Age Monthly income Social norm conformity Trade-offs Gender Park valuation
1.00 0.50 0.00 0.50 1.00
Real decision Hypothetical decision
Table 3: Akaike weights of variables in hypothetical and real decision-to-donate
Hypothetical Real
Rank Variable Weight Rank Variable Weight
1 Worldview 80.2% 1 Park valuation 91.8%
2 Altruism 58.2% 2 Trade-off 88.5%
3 Age 46.4% 3 Social-Norm-Conformity 49.8%
4 Monthly income 42.2% 4 Monthly income 37.6%
5 Social-Norm-Conformity 29.6% 5 Gender (male=0) 37.0%
6 Trade-off 28.1% 6 Altruism 28.3%
7 Gender (male=0) 27.9% 7 Worldview 27.8%
8 Park Valuation 27.0% 8 Age 25.3%
Figure 5: Akaike weights of models summed per variable. For each variable, all the weights of the models including this variable are added and presented here. Orange/dark for real decision-to-donate, yellow/light for hypothetical decision-to-
donate.
