1 Introduction

Shifting individual behaviour is one important tool for addressing environmental issues like climate change and environmental degradation. Indeed, extensive work has investigated the drivers and levers of pro-environmental behaviours (for recent examples, see Bonan et al. 2021; Carlsson et al. 2021; Zemo and Termansen 2022). Nudges and monetary incentives are two of the most common interventions to affect pro-environmental behaviour (Carlsson et al. 2021; Maki et al. 2016; Schubert 2017; Sloot and Scheibehenne 2022). Nudges are defined as a change in the decision environment to alter individual decisions, without changing the choices available or significantly changing the economic incentives (Thaler and Sunstein 2009). Although monetary incentives may be the standard solution for an economist, and indeed incentives can affect pro-environmental behaviour (Diederich and Goeschl 2017; Maki et al. 2016; Viscusi et al. 2011), there have been concerns about motivational crowding out through the use of incentives in certain contexts (Gneezy et al. 2011; Ling and Xu 2021; Rode et al. 2015). On the other hand, nudges, such as the use of defaults or injunctive norm messages, can be an effective way to shift behaviour in some contexts, but less effective in other contexts (Carlsson et al. 2021; DellaVigna and Linos 2022; Gravert and Olsson Collentine 2021; Szaszi et al. 2022).

Naturally, policies can be combined, and it is of interest to find out if and how there are synergies between them (Al-Ubaydli et al. 2017; Brent et al. 2015; Gravert and Olsson Collentine 2021; List et al. 2017). In many cases, nudges are a relatively low-cost intervention, so if a nudge reduces motivational crowding out of a monetary incentive, then it might be efficient to combine the two. Indeed, authors have pointed to the importance of using nudges to make Pigouvian pricing policies effective (Dorner 2023; Gravert and Shreedhar 2022). Recently, Chan (2024) developed a theoretical model that shows the interaction between Pigouvian policies and behavioural factors, including nudges, social norms and spillovers. In this paper, we empirically evaluate the individual and combined effects of an information nudge and a monetary incentive on nature restoration volunteering in a field experiment setting.

We are not the first to empirically study the effects of combining nudges and monetary incentives. Furthermore, nudges can take many forms. We will therefore briefly present a set of previous studies. Experimental studies on the interaction between nudges and incentives show varied results. In a study on food choice and using a nudge in terms of recall of environmentally friendly behaviour, Panzone et al. (2021) find no synergies between the nudge and the monetary incentive. Similarly, using a goal setting nudge with feedback, Fanghella et al. (2021) finds no synergy between this nudge and a monetary incentive on energy saving behaviour. On the other hand, a negative synergy effect was found in a study on energy consumption where the nudge was a peer comparison (Sudarshan 2017), and in a study on water use where the nudge was a peer comparison (Hernández et al. 2024). In a meta-analysis of 44 experimental studies Alt et al. (2024) find moderate evidence for a negative synergy effect of policy mixes, but that this is primarily driven by mixes of two interventions from the same domain. Combining behavioural and traditional interventions, there is no evidence for a negative synergy effect. Moreover, the type of nudge does not seem to affect the extent and sign of a synergy effect.

It is likely that the specificity of the matching between the nudge to the context and behaviour is important, and it is one reason that more work needs to be done in this area (Drews et al. 2020; van Valkengoed et al. 2022). For example, in Khanna et al.‘s (Khanna et al. 2021) meta-analysis, the degree of complementarity depended on which nudges were paired with the financial incentive. In a small number of cases, there were no complementary effects. In a study on demand for LED lighting, Rodemeier and Löschel (2025) find that the spillover between information nudges and Pigovian subsidies critically depends on the informational content of the nudge. Informing about energy savings increases LED demand, but if the energy savings is expressed in monetary terms, it reduces LED demand. At the same time, both types of information nudges reduce price elasticities, making subsidies less effective.

The importance of understanding synergies between nudges and incentives matters not just for short-term policy success, but also for the long-term effects of interventions (Drews et al. 2020). For example, if changes in underlying pro-environmental motivation drive synergies, this could have broader implications for environmental policy support (Gravert and Shreedhar 2022). Most studies considering the potential synergies speculate that this is driven by motivational crowding in (or out). There could be “incentive crowding” effects where the nudge impacts the efficacy of the incentive; for example, by highlighting the pecuniary and non-pecuniary benefits of the incentive, which strengthens the incentive treatment effect (Drews et al. 2020). A monetary incentive could also crowd out (or in) intrinsic motivation (for one of several reasons – see Frey and Jegen 2001; Rode et al. 2015) and that could impact the efficacy of the nudge (Hilton et al. 2014; Martin and Rivers 2018; Schall et al. 2016). Few studies have empirically assessed the mechanism driving synergies (or the lack thereof). A recent paper by Fanghella et al. (2021) on energy conservation examines treatment-values interactions to rule out motivational crowding effects in their context. Alt et al. (2024) found that nudges can complement incentives by mitigating the crowding-out effects of financial incentives in an online experiment. Their work is in the context of donating to charity, and they can observe two donation events to track short-term crowding-out explicitly.

In this paper, we report on the findings from a field experiment where we test for synergies in an information nudge and a monetary incentive for first-time volunteering for a nature restoration group in Aotearoa New Zealand. Information nudges inform individuals of the benefits of a choice, providing new information and/or making existing benefits salient (Carlsson et al. 2021). In this study, we target the information nudge towards the intrinsic motivation of the relevant audience. We use a two-by-two design to evaluate the effects of an information nudge, monetary incentive, and combined treatment on volunteering behaviour. We also illustrate with a simple model how to think about the mechanism underlying potential synergies, building on the model of intrinsic and extrinsic motivation by Bowles and Polania-Reyes (2012). We also empirically explore the mechanisms using attitudinal data. This is an area that is important to understand from a policy perspective (Drews et al. 2020) and one that very few studies have been able to interrogate (notable exceptions being Alt et al. (2024) and Fanghella et al. (2021). Finally, we add to the limited experimental literature on interventions that can increase nature conservation behaviours. As Nielsen et al. (2021) asserts, nature conservation is an important and under-researched area in the behavioural science literature. Moreover, most studies on PEBs in the environmental economics literature focus on energy or water consumption (Brent et al. 2017).

There is a considerable shortage of behavioural science research that focuses on behaviours that directly impact nature and biodiversity (Brent et al. 2017; Grilli and Curtis 2021; Nielsen et al. 2021). This is concerning, given the enormous value populations place on nature, the fundamental role nature plays in society and because nature is declining rapidly. Volunteering for nature restoration groups is an impactful behaviour (in terms of environmental outcomes) that few people are engaged in, even though it creates significant benefits for society and the volunteers themselves (Ganzevoort and van den Born 2020; Meier and Stutzer 2008; Nielsen et al. 2021; Ryan et al. 2001). Moreover, we focus on the urban population because few studies focus on behaviours for biodiversity conservation and even fewer study them in an urban context (Brent et al. 2017; Nielsen et al. 2021; Truelove et al. 2014). We also select our target behaviour through an explicit selection approach, which prioritises the behaviours most impactful for end outcomes (see McLeod et al. 2024), which is uncommon but highly recommended in the literature (Al-Ubaydli et al. 2017; Nielsen et al. 2021).

We find that the information nudge and voucher alone are not effective at increasing volunteering. However, when we combine the two treatments, the total treatment effect is significantly greater than the sum of the individual effects of the nudge and incentive. This shows that there are significant positive synergies between an information nudge and incentives in this context. Based on our theoretical understanding of how synergies could arise and the design of the information nudge, we find that the positive synergy is likely driven by motivational crowding effects from the voucher, which are offset through the use of the nudge. This has important implications for those thinking of designing voucher programmes to encourage the uptake of pro-environmental behaviours.

The rest of the paper is organized as follows. In Sect. 2 we outline the experimental design. In Sect. 3 we outline a simple model illustrating how nudges and monetary incentives affect behaviour, and in Sect. 4 we present our three main hypotheses. In Sect. 5 we present the data, and in Sect. 6 we present our results. Finally, in Sect. 7 we discuss our results.

2 Experimental Design

Our field experiment aimed to test the effectiveness of an information nudge, a voucher and a combined information nudge and voucher on increasing volunteering for nature restoration. The experiment was approved by the Waikato Management School (WMS) Human Research Ethics Committee, application number: WMS 22/134. We pre-registered the study before collecting the data on AsPredicted. Our experimental design is summarised in Fig. 1.Footnote 1

Fig. 1
figure 1

Experimental design. Individuals completely unwilling to volunteer were screened out at the start of the survey. “Volunteering” in this design refers to nature restoration volunteering. As such, “first-time” volunteers may have volunteered before in other domains

Full size image

We constructed a sample of potential first-time volunteers by first recruiting into an online survey, administered through the Qualtrics platform. We initially asked about their willingness to volunteer for a nature restoration group, to identify those who were willing but had not volunteered in the last three years, to create a pool of potential first-time volunteers. Within this pool and the survey, we randomly assigned the participants to one of the treatment groups, as per Fig. 1. A copy of the full survey can be found in the Appendix. Participants were recruited via social media advertising and mailbox drops, targeted to all adult residents in or near Hamilton, New Zealand. The ads stated that we were interested in hearing from them about community engagement with restoration groups, and a prize draw consisting of five NZ $100 Prezzy cards (the draw was conducted after the study was completed).Footnote 2

After answering all demographics and environmental attitudes questions, respondents were asked whether they would like to pre-commit to a volunteering event sometime over the next month and specify days they may be available. This is a stated preference variable that we call “pre-commitment” because individuals are committing to attend but have not yet committed to a specific date or time; see Fig. 2. Based on treatment group status, individuals received variations of the question about pre-committing to an event; we describe the treatments in detail in the next section.

Fig. 2
figure 2

Outcome volunteering variables. Pre-commitment is a general commitment to attend a volunteering event in the next four weeks. Commitment is a commitment to attend a specific event, and attendance is actual attendance at an event

Full size image

After survey completion (between 20/01/2023 and 14/02/2023),Footnote 3 we reached out to all pre-committed individuals asking them to confirm whether they would attend one of two volunteering events: one on a weekday morning, and one on a weekend morning.Footnote 4 We call this variable “commitment” because we asked individuals to confirm their attendance at a specific event and inform us how many family members would be attending with them. We collected this information through another short survey (the commitment survey – a copy can be found in the Appendix). We sent a reminder to all pre-committed individuals about the commitment survey and reminded relevant groups about the voucher. Finally, we observed whether individuals attended a volunteering event, denoting this variable as “attendance.” We evaluated whether the treatments have any effects on all three measures of willingness to volunteer.

2.1 Experiment Treatments

Based on treatment group status, individuals received variations of the question about pre-committing to an event at the end of the survey. The literature emphasises the importance of crafting information and messages for the target population, given their specific motivations and barriers within their context (Carlsson et al. 2021; Michie et al. 2011). Thus, we discussed volunteer motivations and benefits with our study partner nature restoration group (see Section 2.3). We also drew on a social marketing survey conducted by ourselves and colleagues with a representative sample of 1,901 urban residents in Aotearoa New Zealand (Dorner et al. 2024). The survey allowed us to identify drivers of and barriers to volunteering for nature, using the Behaviour Change Wheel (BCW) framework, which is underpinned by the Capability, Opportunity, and Motivation model of Behaviour (COM-B) (Michie et al. 2011).

Through this prior research, we classified 10% of respondents as supportive; motivated to volunteer for nature due to its various private and social/environmental benefits, and not significantly affected by any of the barriers. Hence, many of these respondents were already volunteering for nature and thus were not included in the present study, while the others may only need basic information to be prompted to volunteer. The receptive but unsure group (29%) were motivated based on the benefits of volunteering, but not a strongly as the supportive, and their main barriers to participation were a lack of knowledge about when and where to volunteer and not knowing others who volunteer. They did, however, believe that people like them volunteered for nature. The final two segments had low motivation and high barriers across the COM-B questions (Dorner et al. 2024), so we do not consider these groups for treatment design. For the present study, we are informed by these profiles, but their proportions in the sample will likely not map onto ours due to differences in sample population (the whole of the country versus just Kirikiriroa | Hamilton city) and recruitment methods.

All respondents received the same baseline information with details about the length and type of event. This baseline information formed the conditions for the control group. We expected to observe some volunteering in the control treatment because our prior research showed that some non-volunteers are motivated but lacking basic information about when and where to volunteer. To separate this effect from the other treatments, we presented this information factually about the volunteering opportunity and without further framing. The other treatments can thus be interpreted as relative to this basic information control treatment.

Information nudges inform or make salient the benefits of a choice (Carlsson et al. 2021). Thus, for the information nudge treatment, we provided respondents with information that targeted relevant intrinsic motivations for volunteering in this context and population. The full nudge statement is shown in Fig. 3. We were careful not to include other types of nudges in the text (for example, social comparisons). However, we note that there is overlap between some of the intrinsic motivations (e.g. warm glow for providing environmental and community benefits) and moral suasion (hinting at the moral course of action) (Carlsson et al. 2021). The nudge also includes more than one piece of information about how it can increase individual utility (warm glow, personal wellbeing, socialising with similar people) to ensure we provide information that is widely applicable to the target population. It is beyond the scope of this study to separate the effect of specific types of information for the relevant population and to test the impact of combining multiple types of information into one message. Hence, we refer to this broadly as an information nudge treatment.

Fig. 3
figure 3

Question about pre-commitment for those in the combined treatment group

Full size image

For the voucher group, individuals received a one-off NZ $50 supermarket voucher for attending. This is equivalent to 1.3 hours of work based on the national average hourly earnings during the March 2023 quarter (Stats 2023). The voucher was not meant to fully compensate individuals for their time, but it still represents an attractive financial incentive. We carefully framed the voucher to reduce the risk of crowding out intrinsic motivation by specifically referencing that the voucher is a 1-off payment to support people trying something new (Gneezy et al. 2011). However, we deliberately did not overlap this framing with the information in the nudge to avoid confounding effects. Hence, the framing of the incentive is minimal, but provides a reasonable and clear motivation from the provider to reduce the chance of motivation crowding (Bowles and Polania-Reyes 2012) without the more complete framing provided by the nudge.

In Fig. 3, we show what the question looked like for someone randomly assigned to the combined treatment group. We have highlighted the specific components of the question that constitute the nudge and voucher offer.

2.2 Volunteering Events

In conjunction with our partner the Fairfield Project, we organised four volunteering events for people to attend as part of the experimental design. The events were held over one week from the 18th to the 25th of February 2023. There were two events for the voucher-group and two other events for the non-voucher groups to choose from.

For the non-voucher groups, volunteering events were held on Saturday the 18th and Monday the 20th of February. For the voucher groups, volunteering events were held on Wednesday the 22nd and Saturday the 25th of February. The Monday event for the non-voucher group was supposed to be on Wednesday the 15th of February, but adverse weather conditions the day before the event meant we rescheduled it to Monday the 20th of February.

The events all started at 10:00 AM and concluded at 12:00 PM, where a light lunch was provided for attendees and vouchers were handed out (at the relevant events). Fortunately, all four events had similar weather conditions – fine, with a mix of sun and cloud overcast. The temperature ranged from 17 °C to 24 °C throughout the events. Volunteers could choose between several volunteering activities on the day (for example, trapping, potting, and planting) and had two opportunities to select activities during the event. The purpose of this was to cater to a broad range of interests and skills with the hope of increasing enjoyment for those involved.

The final description of the event with full details and information was given via email after individuals committed to attending the event. We sent short reminders about the events to those who had committed in the days leading up to the events.

We monitored attendance at the original four events and the follow-up events using sign-in sheets from the Fairfield Project. As is standard at all Fairfield Project volunteering events, attendees needed to listen to a health and safety briefing and then sign in to the site. Our research team managed these sign-in sheets at the four volunteering events, informing participants that the sheet would be used only for health and safety purposes and to track whether they attended an event as part of our research study.

2.3 Field Partnership

The Fairfield Project is an urban biodiversity and gully restoration group in Kirikiriroa | Hamilton, Aotearoa | New Zealand.Footnote 5 They have a particular focus on environmental and sustainable education for people of all ages and backgrounds. As such, they carry out educational workshops and volunteering events for schools, businesses, and the wider community.

The Fairfield project does so alongside their primary activity, the restoration, and maintenance of the ecologically significant Kukutaaruhe Gully, for which they rely on the assistance of local volunteers. They serve a diverse community in Fairfield (a suburb of Hamilton city) which includes managing several large community gardens and providing community members with opportunities to cultivate their crops. Fairfield has a central location in the small city of Hamilton, and as such it is less than a 15-minute drive from any location within the city.

The Fairfield Project has a consistent base of volunteers, but always needs more volunteers. Like other community nature restoration groups, we informally spoke to the Fairfield Project find that volunteers tend to be older and that it is very difficult to attract and retain new volunteers. They also expressed concern that many residents were unaware of their work and the opportunities to get involved as a volunteer - a sentiment shared by other community groups and shown in recent research by the Ministry for the Environment (MFE) (2021) and Dorner et al. (2024).

3 A Model on Motivational Crowding and Treatment Interactions

We now discuss under what circumstances a combination of two interventions would influence behaviour greater than the sum of the individual interventions in isolation. We illustrate this by adapting Bowles & Polania-Reyes’s (2012) model of state-dependent preferences with intrinsic and extrinsic motivations. This is a useful starting point because our two interventions, a financial incentive and an information nudge, target extrinsic and intrinsic motivations, respectively. In the appendix, we report the full model; here we focus on the main implications.

In the model, the individual decides how much effort/volunteering to exercise. The benefit of volunteering depends on the public environmental benefits from their volunteering, the monetary incentive offered, \(s\), and the intrinsic rewards from volunteering\( v\). The intrinsic reward from volunteering depends on a baseline intrinsic reward, \({\lambda _0}\), and the potential crowding effect of the monetary incentive (\(s{\lambda _m}\)). If the incentive and intrinsic rewards are independent, \({\lambda _m} = 0\) and the intrinsic reward is equal to the baseline intrinsic reward \({\lambda _0}\). If the incentive has a crowding out (in) effect, \({\lambda _m}\) will be negative (positive) and reduce (increase) the marginal utility of volunteering and contributing to the public good.

The level of the subsidy thus affects the marginal benefit to the individual, MB, of volunteering in two ways:

$$\frac{{dMB}}{{ds}} = \underbrace 1_{Directeffect} + \underbrace {{\lambda _0}{\lambda _m}}_{Crowdingeffect}$$
(1)

The first is the direct effect of the subsidy on behaviour, the second is the crowding effect, which depends on the sign of (\({\lambda _0}{\lambda _m}\)). Equation (1) suggests that crowding out or in is larger for those with higher baseline levels of intrinsic motivation (captured by \({\lambda _0}\)), though this also depends on whether and to what extent \({\lambda _0}\) and \({\lambda _m}\) are correlated. It is, however, intuitive that crowding out cannot occur to the same extent if an individual has little motivation to begin with, and therefore, crowding out is more likely for those with high baseline motivation rather than low (Dorner and Lancsar 2023). Crowding out could occur for many reasons (Bowles and Polania-Reyes 2012). For example, it could be that the incentive makes intrinsic rewards less salient (Chao 2017), takes away perceived autonomy, leads to moral disengagement (Bowles and Polania-Reyes 2012; Rode et al. 2015) or undermines the recipient’s moral self-identity (Bénabou and Tirole 2006). Incentives could also crowd in motivation. For example, incentives could reinforce existing environmental attitudes, signal the social desirability of a certain behaviour or enhance the warm glow utility effects (Rode et al. 2015).

We next introduce the nudge. We allow the nudge to affect utility in two ways: i). The nudge could impact the intrinsic rewards, and ii) the nudge could impact the crowding in (or out) effects of the monetary incentives. The first point is the direct effect of the nudge, and is consistent with mechanisms associated with information nudges in the literature (Carlsson et al. 2021; Schubert 2017). The nudge could highlight specific identities and intrinsic motivation, or it could offer new information about the intrinsic rewards. This aligns with our nudge intervention, which provides information targeted towards individuals’ intrinsic motivations for volunteering, as described in the previous section. The second point is the indirect effect of the nudge and captures a synergy that may exist between incentives and nudges (Drews et al. 2020; Gravert and Shreedhar 2022). For example, a nudge targeting intrinsic motivation through moral suasion (Ito et al. 2018) or social comparison (Allcott 2011) might not only have a direct effect but also reduce the crowding out of the monetary incentive. In the case of the present study, the information nudge targets several highly relevant sources of intrinsic motivation (warm glow/moral, wellbeing, socialising) that could all be crowded by a financial incentive. Conversely, incentives may also impact the efficacy of nudges, for example, by undermining the intrinsic motivation channel that a nudge is targeting. Our experiment does not allow us to identify the direction in which the potential effect goes, and our model should be seen as an illustration of the interactions between the interventions.

The intrinsic reward from volunteering now depends on a baseline intrinsic reward, \({\lambda _0}\), the potential crowding effect of the monetary incentive (\(s{\lambda _m}\left( N \right)\)), and the effect of the nudge on intrinsic reward, \({\lambda _n}\). This direct effect could be positive, negative, or zero, but is most likely \( \ge 0\) (based on the results in the literature – see reviews by (Carlsson et al. 2021; DellaVigna and Linos 2022; Schubert 2017; Szaszi et al. 2022). To incorporate the potential synergy between nudges and incentives, we assume the marginal effect of the monetary incentive on intrinsic rewards is a function of the level of the nudge, through the term \({\lambda _m}\left( N \right)\).

Again, if we start at \(s = 0\), the effect of the monetary incentive on the marginal benefit can be expressed as:

$$\frac{{dMB}}{{ds}} = \underbrace 1_{Directeffect} + \underbrace {{\lambda _0}{\lambda _m}\left( N \right)}_{Crowdingeffect}$$
(2)

The indirect crowding-out/in effect of the incentive now also depends on the level of the nudge, and the sign of \({\lambda {^{\prime}}_m}\left( N \right)\). Equation (2) has implications for how synergies between a monetary incentive and a nudge might be observed. First, consider when \({\lambda {^{\prime}}_m}\left( N \right) = 0\); that is, there is no synergy/interaction effect. We then show in the appendix that if the cost function of volunteering is convex, the observed effect on effort will be less than the sum of separate effects of the subsidy and nudge; this could be incorrectly categorised as a negative synergy. On the other hand, if we observe the effect on volunteering effort of a monetary incentive and nudge combined to be equal to the effect of them being added separately, it is possible (and indeed likely) that there is a positive synergy at play, meaning \({\lambda {^{\prime}}_m}\left( N \right) > 0\). Finally, if implementing both a monetary incentive and nudge leads to effort to increase by more than if they are implemented separately, then this is a clear, unambiguous indication of a positive synergy.

Based on this model we formulate our two main hypotheses in the next section.

4 Hypotheses and Empirical Methods

Following from our theoretical model, we test the following hypotheses:

H1

All three treatments will increase the likelihood of volunteering (relative to the control).

This assumes that \({\lambda _n} > 0\) (the nudge impacts the net benefits function and shifts behaviour), \({{dMB} \over {ds}} > 0\) (the incentive’s overall marginal impact is positive), and that the combination of the nudge and voucher has a positive impact on the utility function. Previous literature suggests that the monetary incentive may have a crowding out effect on intrinsic motivation. However, within the context of this experiment, we chose a monetary incentive that a priori we expect to more than compensate for any crowding out effects (Gneezy and Rustichini 2000).

H2

There will be an interaction (synergy) between the nudge and incentive such that the combined treatment effect is at least as large as the sum of the individual treatment effects.

Thus, we hypothesize that there is a potential positive synergy effect; that is, \({\lambda {^{\prime}}_m}\left( N \right) > 0\). Based on our model, if we observe the sum of the individual treatment effects being the same as their individual effects, this may still show a positive synergy due to the convex cost of effort function. If we see the sum is greater than the individual effects, then there is unambiguously a positive synergy.

Finally, we consider heterogeneous effects of the treatments, by level of intrinsic motivation.Footnote 6 A priori, we expect motivational crowding to be strongest for participants who are more highly intrinsically motivated. This expectation is intuitive given our theoretical model, in which the crowding effect of money on marginal utility (Eq. 5) is \({\lambda _0}{\lambda _m}\), or intrinsic motivation multiplied by the marginal crowding effect of money. In other words, the more highly motivated have more motivation to crowd out. While other outcomes are possible, this intuition is also backed by recent empirical evidence (Dorner and Lancsar 2023). It also follows that any observed money incentive-nudge synergy effects may be driven, at least in part, by the nudge preventing this motivational crowding in the highly motivated, given \({\lambda _m}\) is a function of the level of the nudge (Eq. 2).

To test our hypotheses, we conduct a series of tests with predominantly non-parametric chi-squared hypothesis tests, in line with our pre-registration, to recover the causal effect of our randomly assigned treatments on our outcomes of interest. These tests rely on the assumption that treatment assignment is exogenous, which we verify with balance checks in the data section.

Finally, we estimate linear probability models on the decision to volunteer, in order to investigate heterogeneous treatment effects and therefore test the underlying mechanism in our theoretical model.Footnote 7 We interact treatment dummies with environmental self-identity (EID) as a measure of intrinsic motivation for the environment.Footnote 8 Hence, the basic model we estimate is:

$$\begin{aligned} {Y_i} = & {\beta _0} + {\beta _1}Nudg{e_i} + {\beta _2}Vouche{r_i} + {\beta _3}Combine{d_i} + {\beta _4}HEI{D_i} \\ & + {\beta _5}{\left( {Nudge*HEID} \right)_i} + {\beta _6}{\left( {Voucher*HEID} \right)_i} + {\beta _7}{\left( {Combined*HEID} \right)_i} + {\varepsilon _i} \\ \end{aligned} $$
(9)

where \({Y_i}\) is a indicator variable of decision to volunteer (pre-commitment, commitment or attendance), the \(\beta \) terms are the coefficients on each term in the regression, \(Nudg{e_i}\), \(Vouche{r_i}\) and \(Combine{d_i}\) are indicator variables for if individual \(i\) is a member of that treatment group, \(HEI{D_i}\) is a dummy variable indicating if individual \(i\) has high EID and \({\varepsilon _i}\) is the idiosyncratic error term. We define a high EID score as above 5 (out of a maximum 7). All models are run with Huber-White robust standard errors. Our primary interests are the signs of \({\beta _6}\) and \({\beta _7}\). If \({\beta _6}\) is negative, then the monetary incentive has a crowding effect for those with a high EID. If \({\beta _7}\) is positive, the nudge reduces the motivational crowding out of the incentive for those with a high EID.

5 Data

The data we use is from several sources, including our online survey (this is the main source of data), the commitment survey, and attendance sheets from the volunteering events (see section 2). Survey one received high engagement, and we ended up with a total usable sample of N = 757 (this includes those who are already volunteering). This was after excluding individuals who were strongly opposed to volunteering, under 18 years of age, were not the first household member to complete our survey,Footnote 9 and did not live near Hamilton. We also dropped responses that were less than 75% complete. Of the 757 respondents, 130 were already volunteering for nature restoration groups, and 627 were classified as “first-time” volunteers (did not volunteer for a nature restoration group over the last three years). This sample of first-time volunteers (N = 627) is our sample of interest. First-time volunteers are new to volunteering for a nature restoration group, though they may have volunteered elsewhere before.

5.1 Demographics

We collected data on a range of demographics. Table 1 reports the demographic summary statistics for our overall sample. This includes those already volunteering and not part of our treatment intervention as a descriptive comparison, given that the interventions aim to move first-time volunteers into the volunteering category. Our sample is not meant to be representative of the New Zealand population. Rather, our sample is aimed at being representative of those living near an urban centre with at least a minor interest in volunteering for a restoration group.

Table 1 Demographics summary statistics
Full size table

Our sample is highly female-dominated and tends to be well-educated, with 58% of respondents having attained at least a bachelor’s level education (this is higher than the overall population - Ministry of Social Development 2016). In terms of ethnicity, approximately 24.5% of New Zealand residents identify as either Māori or Pacific, aligning closely with our data (23% Māori or Pacific). Most respondents never or infrequently volunteer elsewhere. Those already volunteering are more likely to be male, less likely to be Māori or Pacific, more likely to live outside of Hamilton City, and less likely to have a dependent child. Those already volunteering have higher EID scores and our EID index variable has a Cronbach’s alpha of 0.90, indicating strong internal consistency (Cortina 1993; Cronbach 1951).

Our environmental identity (EID) index measures beliefs about how environmentally friendly one is. Environmental identity has been widely studied in psychology and has strong associations with pro-environmental behaviour (Sparks et al. 2021; Whitmarsh and O’Neill 2010). We deploy the widely used environmental self-identity scale (EID) from van der Werff et al. (2013). This is a three-item scale, constructed from Likert scale questions (1–7) that is replicated exactly for our surveys. We use this index to distinguish between individuals with high and low environmental identity (those above or below a score of five).Footnote 10

For those who pre-committed to attend a volunteering event over the following month, we also gathered information on their general availability to attend events. Immediately following pre-commitment, we asked individuals to select any dates over the upcoming four weeks (from an on-screen calendar) where they were likely to be available to attend a volunteering event between the hours of 10:00 AM and 12:00 PM. This gave us a count variable for the number of days each pre-committed individual was available (taking values between 0 and 28), providing us with a measure of availability.Footnote 11

5.2 Balance Test

In Table 2, we report demographic summary statistics for each treatment group and observe a good balance overall. Moreover, we formally assess whether randomisation was successful using a multinomial logit model to predict treatment status. We include a full set of demographic control variables, and the results are reported in the Appendix. None of the 48 estimated coefficients is significant at the 5% level. This indicates that our demographics have no true predictive power over treatment assignment. We confirm this by estimating a second multinomial logit model with intercepts only (no regressors or explanatory variables). We find that the Akaike Information Criteria (AIC) is lower (indicating a better model fit) for the model with no variables (1,742.7) than the model with our full set of controls (1,779.8). This is also shown in the Appendix. Finally, an LR test comparing the complete and empty models reveals that the covariates jointly are non-significant in predicting treatment status (p-value of 0.136).

Table 2 Demographic summary statistics by treatment group
Full size table

We can conclude that our treatments were successfully randomly assigned and proceed to our chi-squared hypothesis testing and regression results.

6 Results

6.1 Overall Treatment Effects on Volunteering

In Figure 4 we present a summary of the volunteering rates by treatment groups.

Fig. 4
figure 4

Summary graph of volunteering rates by treatment group. Error bars show 95% confidence intervals. Pre-commitment is in the top left, commitment is in the top right, and attendance is in the bottom left. The p-values are from chi-squared tests for differences between the groups

Full size image

There is a non-negligible group of people willing to volunteer for the first time even in the control condition. This suggests that the provision of information and being asked directly to volunteer has a positive impact on volunteering rates, and we know this because our sample only includes those who are not already engaged in nature volunteering. Hence, our treatment comparisons to the control should be interpreted as relative to the provision of basic information about a volunteering opportunity.

Volunteering rates are highest at the pre-commitment stage and decrease significantly at the commitment stage, with a further slight decline at the attendance stage. At the pre-commitment stage, there are no significant differences between treatment groups. For the two other stages, the volunteering rate is highest for the combined treatment, followed by the voucher treatment.

6.2 Treatment Effects

Our first hypothesis (H1) concerns the effect of the various treatments on volunteering rates, relative to the control. In Table 3, we report the pairwise average treatment effects and p-values from our set of pre-registered one-sided chi-squared tests. There are no statistically significant differences between the nudge treatment and the control group for any outcome. There are no statistically significant treatment effects in the voucher treatment either, although the treatment effect sizes are larger compared with the nudge treatment. On the other hand, there are statistically significant differences in commitment and attendance rates between the combined treatment and the control.

Table 3 Average treatment effects (ATE) in percentage points
Full size table

The first hypothesis is thus only partially supported. The nudge and voucher alone are not effective at increasing volunteering. However, the combined treatments significantly increase the probability of committing to and attending a volunteering event. We also find that none of the treatments affect pre-commitment to volunteer.

Our second hypothesis (H2) concerns the comparison between the combined treatment effect and the individual nudge and voucher treatment effects. Firstly, in Table 4, we compare the willingness to volunteer in the combined treatment group with those in the nudge and voucher groups separately. We find that the combined treatment is more effective than both the nudge and voucher alone at promoting commitment to and attendance at volunteering events. For example, the average treatment effect (ATE) of the voucher on commitment is an increase by 4.8% points, and when the voucher is combined with the nudge, the ATE is 12.2% points.

Table 4 Results for combined vs individual treatment effects on volunteering behaviour
Full size table

To test if the combined treatment effect is larger than the sum of the two individual treatment effects, we calculate an upper bound on the average treatment effect under no synergy (simply the sum of the two individual average treatment effects) and compare this to the treatment effect we observe for the combined group. Simply adding the nudge and voucher treatment effects gives an upper bound on no synergy because of the convex nature of the cost of effort function (we would expect diminishing returns to intervention, so the linear addition will likely overestimate the true effect on effort under no synergy). We show these results in Table 5. For example, for the effects of the treatments on commitment, our upper bound treatment effect estimate under no synergy is 6.9%, but we see a treatment effect almost double in size (12.2%). These differences are statistically significant under basic one-sided and two-sided tests. This shows that there are likely positive synergies between the nudge and the incentive because their combined effect is much greater than the sum of the individual effects.

Table 5 Comparison of the actual ATE and the expected ATE under no synergy for the combined treatment group
Full size table

Thus, our results support our second hypothesis that the combined treatment is significantly more effective than either the voucher treatment or the nudge treatment alone. Moreover, we show that the combined treatment effect is greater than the sum of the individual treatment effects from the nudge and the incentive. This shows there are positive synergies between nudges and incentives in this context.

6.3 Exploring the Mechanism Behind the Interaction Effect Between Nudge and Voucher

In this section of the results, we explore the mechanisms behind the positive synergy between the nudge and the voucher. As discussed in Section 4, we predict that crowding in- and out will be stronger among those with a high EID. The regression results are presented in Table 6; we include a set of controls in each model, and the full set of results are presented in the appendix. As a reminder, if the voucher crowds out intrinsic motivation, we expect the coefficient on the interaction between high EID and the voucher to be negative because we expect crowding out to be stronger if EID is high. Likewise, if the nudge reduces crowding out, we expect the interaction between high EID and the combined treatment to be less negative.Footnote 12

Table 6 Linear probability model results for treatment-EID interactions
Full size table

The first section of Fig.4 shows the results for pre-commitment. The high EID-treatment group interactions are all non-significant, meaning we detect no differences in marginal treatment effects by level of EID. The second column shows the results for commitment. On average, the voucher increases commitment probability by 15.7% points and the combined treatment increases it by 13.7% points for individuals with low EID. However, in the voucher treatment, the interaction term with EID is negative and statistically significant, and the total marginal effect for those with a high EID is zero. This suggests that there is motivational crowding out from the incentive among those with a high EID. On the other hand, for the combined treatment, the interaction term is not statistically significant and is close to zero (−0.021), which means that the combined treatment is just as effective for those with high EID and those with low EID. Contrasting this with the voucher treatment, this suggests that the nudge reduces the motivational crowding out of the monetary incentive.

The third column shows the results for actual attendance at a volunteering event. Here, all the high EID-treatment group interactions are non-significant. Thus, the crowding out for high EID individuals that we observed at the commitment stage is no longer present at the attendance stage.

Overall, this additional analysis does not undermine our finding that the combination of the nudge and voucher reduces motivational crowding. However, it provides a mixed picture for our conjecture that motivational crowding will be strongest among the initially highly motivated. This conjecture is supported in the commitment model, whereby the voucher crowds out motivation among those with high initial intrinsic motivation, while the combined treatment removes this crowding effect. However, in the attendance model, both low and high EID individuals are subject to crowding out. Our theory does not provide an ability to distinguish between commitment and attendance, noting that attendance is a revealed preference, subject to individuals randomly being unable to attend the actual events. This also assumes that the drop off between commitment and attendance is not endogenous to EID; given all the other results for attendance are similar to commitment, this assumption is robust. Therefore, we leave it to future research to understand this difference between the results in the commitment and attendance models. Our theoretical model does not, however, rule out the possibility of motivational crowding occurring for both low and high types. Our revealed preference data in actual attendance shows that this indeed appears to be the case for the voucher, contrary to our a priori prediction.

7 Conclusions

In this paper, we add to the growing literature on the effects and synergies of combining nudges and incentives (Dorner 2023; Drews et al. 2020; Gravert and Shreedhar 2022). This is an under-researched area of the literature where there are mixed results and important policy implications (Drews et al. 2020). We also add to the very limited experimental literature on interventions to encourage PEBs that relate specifically to nature and biodiversity (in our case, nature restoration volunteering). This is important now more than ever, as we are rapidly diminishing the natural environment and the ecosystem services that global communities critically depend upon (Costanza et al. 2017). Across the behaviour change literature, there has been a lack of focus on the behaviours that matter most for the end outcomes of interest (in our case, environmental outcomes - Al-Ubaydli et al. 2017; Grilli and Curtis 2021; Nielsen et al. 2021). We identified volunteering as a highly impactful behaviour using an explicit selection method.

Offering a one-off financial incentive increases nature restoration volunteering behaviour, but the effect is not statistically significant. Clearly, with a larger sample size, we would have been able to say something more definitively, but there are thus indications that financial incentives may crowd out intrinsic motivation (particularly for behaviours with high intrinsic motivation components, like volunteering). Next, we find that nudging participants alone (providing information about several relevant intrinsic benefits) does not affect first-time volunteering behaviour compared with providing basic information about the volunteering opportunity. This adds to the growing number of review studies that find that nudge effects can be very small (and often zero) in many contexts (Szaszi et al. 2022). However, there are considerable positive synergies between the information nudge and voucher incentive, with the voucher effectiveness being significantly enhanced when coupled with a nudge. For policymakers, this suggests that the efficacy of incentive-based interventions to encourage the uptake of nature restoration volunteering (and potentially other behaviours) could be enhanced by coupling the intervention with a low-cost nudge, targeting relevant intrinsic motivation. The prediction that the positive synergies are caused by the nudge reducing any motivational crowding out arising from the incentive is only limitedly supported.

These results add to recent literature that examines the presence of synergies between nudges and financial incentives (Drews et al. 2020; Fanghella et al. 2021; Sloot and Scheibehenne 2022). Most studies focus on energy consumption as the behaviour of choice, and we are the first to study this synergy in the context of nature restoration volunteering. This is pertinent because synergies can be positive, negative, and null, depending on the context, and policymakers need more empirical evidence to evaluate possible synergies in different contexts. Some studies show negative synergies (nudges distract participants from incentives and vice versa), so it is important to assess synergies empirically before coupling nudges and incentives in large-scale interventions (Drews et al. 2020; Fanghella et al. 2021).

In addition, we only deployed one variation of the financial incentive (in terms of value and the framing of the incentive). We designed our incentive to limit crowding out of intrinsic motivation by emphasising the one-off nature of the incentive and that it was to help people try volunteering for the first time. Our results are found in the context of this specific incentive design, and we cannot say what role the framing, value of the incentive, or context (volunteering for nature restoration) had on our results. The literature on the crowding effects of financial incentives is mixed, and future research could consider deploying different values of incentives and using different framings to evaluate the crowding-in or out potential in the context of nature restoration volunteering.

The focus of this study was the effect of interventions on revealed first-time volunteering behaviour. However, as we measured pre-commitment, commitment and attendance, we found an intriguing difference between pre-commitment and commitment/attendance. Given the lack of treatment effects at the pre-commitment stage, our results show that a financial incentive, coupled with an information nudge targeted at intrinsic motivation, may have reduced the intention-behaviour gap relative to the other treatments and control.Footnote 13 We were, however, unable to investigate this gap further with our experiment as we did not optimally design the experiment to understand intentions versus behaviour and drivers of any gaps. Exploring intention-behaviour gaps in relation to volunteering and other pro-environmental behaviours represents an interesting avenue for future research. Including different stages of commitment, as we do here, may provide additional insights into how individuals make decisions and the steps researchers and policy-makers can take to reduce gaps between intentions and behaviour.