When we promise, we voluntarily commit ourselves to doing the promised action in the future (Austin, 1975; Rawls, 1999). Some theorists have argued that, in the absence of sanctions or reputational costs, promises are not binding or just cheap talk (Hobbes, 1651/1960; Farrell, 1987). Empirical studies have shown that this suspicion about the fallibility of promises is widely shared. For example, people distrust politicians to keep their election promises, even though most of these promises are usually at least partially fulfilled (Artés, 2011; Naurin, 2014; Thomson et al., 2017). Moreover, people consistently underestimated others' promise keeping by up to 40% in tasks where, in fact, the vast majority of participants kept their word (Woike and Kanngiesser, 2019). Research has also shown that promises and commitments increase, for example, cooperation (Ostrom et al., 1992; Sally, 1995; Bicchieri, 2002; Ellingsen and Johannesson, 2004; Charness and Dufwenberg, 2006; Balliet, 2009; Koessler, 2022) and truth-telling (Bhanot, 2017; Jacquemet et al., 2021; Kanngiesser et al., 2021). Promises thus appear to be powerful devices for boosting beneficial behaviors and fostering behavioral change.

To date, promises have been studied predominantly in one of four variants: (1) as spontaneous communication in social dilemma or trust experiments in the lab (Ostrom et al., 1992; Ellingsen and Johannesson, 2004; Charness and Dufwenberg, 2006), (2) as hypothetical decisions in vignette studies (Mischkowski et al., 2019), (3) as non-incentivized decisions in the lab (e.g., to take part in a survey; Conrads and Reggiani, 2017), and (4) as forced promises (e.g., to respond honestly; Kataria and Winter, 2013; Heyman et al., 2015). However, all variants have some shortcomings. First, participants in social dilemma experiments could primarily use promises to signal their intentions to cooperate, making it difficult to disentangle the effect of promises per se from the coordination effect of signaling prosocial intentions (van den Assem et al., 2012; Ismayilov and Potters, 2016; Woike and Kanngiesser, 2019). Second, hypothetical or non-incentivized decisions may not be indicative of people's behavior with monetary stakes. Third, promises are, per definition, voluntary commitments (Rawls, 1999), and participants who are forced to promise would have every right to renege on their promise because they did not give their word voluntarily. Moreover, from an ethical point of view, forced promises disrespect individual autonomy and decision making (Hertwig and Grüne-Yanoff, 2017), which raises questions about their suitability as intervention tools.

One approach that may simultaneously overcome each of these shortcomings investigates the effectiveness of voluntary promises under incentivized conditions. A recent study with crowd sourced participants (Amazon's Mechanical Turk) found that most participants kept their voluntary promises even when it was costly for them (Woike and Kanngiesser, 2019). Yet, promise keeping involved relatively little effort and the online setting may not have fully captured the multitude of factors that influence decisions in everyday settings. Incentivized field experiments with voluntary promises can provide such insights.

There is some evidence from field studies that voluntary promises can be effective under real-world conditions. For example, analyses of participants' behavior and communication in a high stake game show (Golden Balls) revealed that voluntary commitments increased cooperation as compared to elicited commitments (Belot et al., 2010).¹ In a hotel setting, voluntary commitments to behave pro-environmentally (e.g., to re-use towels) were particularly effective if guests made a specific pledge (rather than a more general pledge) and signaled their intentions via a pin (Baca-Motes et al., 2013). Overt signals of voluntary commitment also increased office workers' donations to charity in a field study of more than 250 work places (Kessler, 2017). However, some field studies have found strong selection effects: people who voluntarily promised to pay their taxes on time (and entered a lottery if they complied) had already been more likely to pay their taxes on time in the past as compared to people who did not promise (Koessler et al., 2019).

We contribute to this literature by reporting two preregistered field experiments with students in the vicinity of a large university cafeteria (N = 406). We used an incentivized paradigm that gave participants a choice between committing or not committing to a course of action (Woike and Kanngiesser, 2019; Kanngiesser et al., 2021). In Experiment 1, participants had a choice between (a) receiving € 1 and doing nothing else and (b) receiving € 4 under the condition that they promised to pay back € 2 in a sealed envelope at the cafeteria's exit (promise condition). They then indicated their chosen action by ticking a box (e.g., “I choose € 4 and I promise that I will pay back € 2.”). Our paradigm minimized potential selection effects as even (and, in fact, especially) participants who had no intention of keeping their word had an incentive to choose the (higher pay-off) promise option. Participants in the control condition of Experiment 1 were offered a choice between receiving € 1 and receiving € 4, but the higher payment involved no promise and simply gave a choice to pay back some, all or none of the money. All monetary incentives in the study were windfall money. In both conditions, the higher payment amount was sufficient to cover the costs of a full student meal at the cafeteria.² We ensured that payback could not be linked to individual participants to minimize any reputational costs for breaking the promise or not returning any money. We predicted higher payback rates in the promise condition compared to the control condition. Specifically, we expected that more people would pay back at least € 2 in the promise than in the control condition and that average payback rates would be higher in the promise than in the control condition.

In Experiment 2, promise keeping required substantially more effort. Moreover, we tested whether removing the word “promise” when asking participants to agree to a course of action would influence their behavior (Woike and Kanngiesser, 2019). It has been argued that promissory obligations arise even without explicitly using the word “promise” as long as the general conditions for promising are fulfilled (Searle, 1969; Scanlon, 1990). Participants in the promise condition had a choice between (a) receiving € 0.10 and doing nothing else and (b) receiving € 3, a € 0.60 stamp, and a postcard under the condition that they promised to add the € 0.60 stamp to the postcard and mail the (anonymous) postcard to a university postbox within a week of taking part in the experiment. A second group of participants also had a choice between (a) receiving € 0.10 and (b) receiving € 3, a € 0.60 stamp, and a postcard under the condition that they agreed to add the € 0.60 stamp to the postcard and mail it to a university postbox within a week of taking part in the experiment (ask condition). As in Experiment 1, participants in both groups indicated their chosen action by ticking a box. All monetary incentives were windfall money. If using the word “promise” has an effect beyond agreeing to what one was asked to do, we would expect that more participants send back the postcard in the promise than in the ask condition.

Promises can be powerful devices for fostering prosocial behavior and cooperation (Ostrom et al., 1992; Bicchieri and Lev-On, 2007), but their effectiveness has been predominantly studied in lab settings. We found in two preregistered field experiments that the majority of participants (German university students) kept their word even when it was costly and effortful for them.

Our study extends previous work with US crowd-sourced workers and Indian adolescents (Woike and Kanngiesser, 2019; Kanngiesser et al., 2021) and shows that incentivized, voluntary promises are effective commitment devices. Our findings connect with recent research investigating whether participants returned a supposedly lost wallet, which found that most German participants were honest and returned the wallet (Cohn et al., 2019). In contrast to Cohn et al. (2019), our paradigm is deception-free, does not rely on covert confederates, and asks participants for their informed consent. This ensures that participants are not unknowingly burdened with participating in a study they did not consent to. Moreover, the voluntary nature of the promise ensures that participants' autonomy was respected (Hertwig and Ortmann, 2008).

Our findings from Experiment 2 revealed similar return rates in the promise and the ask condition, indicating that the wording of the commitment had little impact in our study. Analyses of communication in a high stakes TV show have found that explicit statements of commitment led to higher cooperation rates than implicit statements (Turmunkh et al., 2019). Moreover, Woike and Kanngiesser (2019) found that MTurk participants more likely to keep a costly commitment to return money when they explicitly promised as compared to being asked to do it; though payback rates in the ask condition were substantially higher than in a no-commitment control condition. Depending on context, pilot testing of wording may be needed to ensure that commitments are formulated in a way that ensures high compliance rates. Moreover, overt signals of commitment such as pins may further increase compliance (Baca-Motes et al., 2013; Kessler, 2017).

Our field study recruited German university students as participants and we chose incentives that were appropriate and proportionate for this context and population. Applications in other contexts or with other populations may require some adjustment of incentives. For example, more than 90% of US crowd-sourced workers kept their promises for (bonus) incentives as small as $0.10 and raising those incentives ten-fold from $0.20 (and a promise to pay-back $0.10) to $2.00 (and a promise to pay back $1.00) only had a small effect (Woike and Kanngiesser, 2019).

Previous studies have found cross-cultural variation in people's reasoning about promissory obligations (Miller and Bersoff, 1992; Song et al., 2009). They have also found that promises are expressed differently in different languages (Egner, 2006). Systematic cross-cultural studies on promise-keeping under incentivized conditions are still lacking and would be an interesting avenue for future research.

The study focused on investigating the effectiveness of a voluntary promise intervention under field conditions, but was not designed to test why people keep their promises. Different proposals have been made for why people keep their commitments including a preference for keeping one's word (Ellingsen and Johannesson, 2004; Vanberg, 2008), an aversion to disappointing others' expectations (Charness and Dufwenberg, 2006), or a combination of both accounts (Mischkowski et al., 2019). As decisions in our study could not be linked to individual participants (and this was stressed in the instructions), we can rule out reputational costs or fear of sanctions as potential motives. However, we are unable to say whether participants kept their word because they preferred to do so or because they thought the experimenters expected them to keep it (or both).

In summary, the majority of participants in our study kept their word at a personal cost while engaging in everyday activities even when there were no foreseeable negative consequences or reputational costs for breaking it. Our paradigm offers an easy-to-implement tool for eliciting voluntary commitments that could be leveraged, for example, to increase pro-environmental behavior (Baca-Motes et al., 2013), timely tax payments (Koessler et al., 2019), or school attendance (Benhassine et al., 2015).

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: https://osf.io/hkna8/.

The studies involving human participants were reviewed and approved by Ethics Committee of the Faculty of Education and Psychology, Freie Universität Berlin. The patients/participants provided their written informed consent to participate in this study.

PK, DS, and JW designed the study. DS collected the data. PK analyzed the data and wrote the first draft of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.