Inter-group conflict over limited resources is thought to be one of the significant drivers of human evolution, shaping human-specific psychology (Bowles, 2008; Choi & Bowles, 2007; Henrich & Muthukrishna, 2021) as well as cultural beliefs and practices that may help outcompete less-cooperative parties (Eckel, Fatas, Godoy, & Wilson, 2016; Handley & Mathew, 2020; Richerson et al., 2016; Zefferman & Mathew, 2015). Among the key factors predicting success in conflict is the level of intra-group cooperation, where efforts are directed either to increase group resources or to disadvantage competing groups through premeditated acts of aggression (Bowles & Gintis, 2011; De Dreu, Gross, Fariña, & Ma, 2020), which in extreme cases includes even self-sacrificial acts (e.g., suicide terrorism).

Evidence from geographical areas recently perturbed by intergroup conflict lends initial support to this hypothesis: in experimental economic games, participants who experienced violent oppression from other groups play more cooperatively with their ingroup members (Bauer, Cassar, Chytilová, & Henrich, 2014; Gilligan, Pasquale, & Samii, 2014; Voors et al., 2012; albeit not everywhere: Werner & Lambsdorff, 2020), and directly experiencing conflict-related violence predicted later engagement in a community's collective action (Bellows & Miguel, 2009). While these studies were conducted post-conflict and cannot speak to the dynamics of conflict-related cooperation, laboratory studies that manipulated the presence of between-group competition in economic games showed that participants in conflict situations contribute more to a common pool of their group (Majolo & Maréchal, 2017), punish ingroup non-contributors (Sääksvuori, Mappes, & Puurtinen, 2011), and that cooperative groups have a higher probability of success (Francois, Fujiwara, & Van Ypersele, 2018).

However, coordinating people to align their interests in order to defeat other parties through increased cooperation is no small feat, given the allure of free-riding that may significantly endanger the whole endeavor. Since intergroup competition is often a numbers game, securing commitment to the joint action among party members is of utmost importance. As illustrated by raiding parties in small-scale societies, people are sensitive to the imbalance of power and are willing to partake in a raid only when having sufficient advantage (Wrangham & Glowacki, 2012), especially when shirking may quickly shift the balance of powers in favor of the opposing party (see Mathew & Boyd, 2014 for consequences of deserting a raiding troop). Scaling the commitment problem from raiding parties to oppressed groups, having the means for the committed members to assort and self-organize similarly predicts the probability of initiation of insurgencies and ethnic conflicts (Ellingsen, 2000; Fearon & Laitin, 2003; Jakobsen & De Soysa, 2009).

While people may verbally commit to helping during intergroup conflict (Glowacki et al., 2016), a verbal commitment is often unreliable and gives individuals with Machiavellian strategies a chance to exploit others (Bereczkei, Szabo, & Czibor, 2015; Számadó, 2010). This problem is amplified in one-shot cooperative dilemmas like conflicts where one side can lose viable resources or risk significant personal harm. A potential solution to communication dishonesty is attaching a cost to communicating commitment such that uncommitted individuals would not be willing to pay the cost if they are not planning to cooperatively partake in the conflict (Sosis, Kress, & Boster, 2007). As formalized by costly signaling theory, when both the signaler and receiver may benefit from reliable communication of a signaler's hidden quality (communicating conflict-related cooperative intentions in this case), the high-quality signaler will endure a communication cost to demonstrate signal reliability, and this cost will be disadvantageous for individuals low on the signaled trait, given the purported benefits (Grafen, 1990). In other words, the hidden quality affects the cost/benefit ratio that the potential signallers face (note that the cost may also be zero for high-quality signallers as long as it is positive for low-quality signallers; Számadó, Samu, & Takács, 2022).

The cost/benefit trade-off of signaling was suggested to guarantee the reliability of the human communication of cooperative intent (Bliege Bird & Smith, 2005) and received initial support from mathematical models showing that such signals may evolve under various constraints (Gintis, Smith, & Bowles, 2001; Roberts, 2020; Salahshour, 2021) and have stable equilibria (Barclay, Bliege Bird, Roberts, & Számadó, 2021; Lotem, Fishman, & Stone, 2003; McNamara & Houston, 2002). A laboratory study showed that generosity is associated with cooperative intentions (Fehrler & Przepiorka, 2013), and evidence from communities in Oceania, South America, and South Asia further documented that this generosity is repaid with cooperative opportunities and support from others (Bliege Bird & Power, 2015; Lyle & Smith, 2014; Power & Ready, 2018). Several authors further argued that generosity is not limited to the expectation of reciprocity and may be used as a costly signal of cooperative intention since people increase their generosity when observed (Bereczkei, Birkas, & Kerekes, 2010; van Vugt & Hardy, 2010) and when having a chance to be chosen by cooperative others, a phenomenon labeled as competitive altruism (Barclay & Willer, 2007; Sylwester and Roberts, 2010, Sylwester and Roberts, 2013).

Nevertheless, the boundary between generosity being a costly signal or an investment with the expectation of reciprocal repayment is rather thin and permeable. Although some cultural mechanisms, such as indiscriminate generosity, may prevent expectations of reciprocity (Bliege Bird & Power, 2015), the reputation for being generous is often in the eyes of the people with whom an individual frequently interacts (Power & Ready, 2018) and who, therefore, are most likely to reciprocate. Moreover, generosity is a broad and vague quality, and its utility may be dubious in specific collective risky joint actions such as raids or intergroup conflicts. Although people may be generous, it does not guarantee that they are committed to a specific collective action in a particular domain (e.g., a raid). To overcome the vagueness problem, commitment signals to a joint action are often embedded within cultural conventions that prescribe the form and expected costs of the signaled message such that it could be easily decoded by receivers (Barker, Power, Heap, Puurtinen, & Sosis, 2019; Lang & Kundt, 2023; Soler, Batiste, & Cronk, 2014).

For example, the Tsembaga of New Guinea had a complex ritual system used to signal war allegiances with various visual markers, performative dances, and pig sacrifices (Rappaport, 2000). These rituals indicated a willingness to participate in ensuing warfare with a costly signal (pig sacrifices, energy spent), effectively allowing ritual organizers to assess the troop's potential strength (i.e., how many people showed up/danced/sacrificed pigs). Using cross-cultural ethnographic databases, Sosis et al. (2007) investigated the association between costly male rituals (such as teeth-pulling, scarification, piercing, tattooing, and learning secret knowledge) and external warfare in 60 small-scale societies, finding that the number and intensity of required costly rituals were positively associated with the frequency of intergroup conflict. Albeit correlational, this result hints at a causal process where conflict pressures groups to enhance their cooperative efforts, which are bolstered by assorting cooperators through costly signaling. Of course, costly signals would often be embedded within a complex system of cultural/religious beliefs, myths, traditions, and identities that may further fuel inter-group conflict (Akbaba & Taydas, 2011; Brubaker, 2015; Neuberg et al., 2014). Yet, it is the potentially causal role of costly signals in mobilizing a competing troop that we experimentally examine here.

To add validity to our experimental setup, we note that the costly signals associated with warfare differ from the usually studied signals of cooperative intent (e.g., generous giving) in three inter-related ways, requiring a novel approach to this question: 1) conflict-related signals often have low or non-existent personal value for signal recipients (unproductive signal costs), 2) cooperative dilemmas may be non-iterative due to the potentially dire consequence of inter-group conflict, limiting future repayment of the signal, and 3) if cooperative dilemmas are one-shot, other factors than cooperative reputation are needed to explain signal stability.

Regarding the first aspect, Sosis et al. (2007) showed that conflict-related costly signals often include pain, body modifications, and similar signals that do not benefit recipients (in contrast to a signaler's generous giving). This may be due to the pluripotency of such signals (signaling both a commitment to joint action and specific qualities related to conflict – e.g., bravery, pain tolerance, anxiety management etc.; Barker et al., 2019; Lyle, Smith, & Sullivan, 2009). Nonetheless, using unproductive costs rather than generosity has further advantages, such as limiting reciprocity expectations, thereby effectively increasing signal trustworthiness (i.e., recipients perceive that selfish motives of expected future rewards do not drive the signal; Bliege Bird, Ready, & Power, 2018; Raihani & Power, 2021). Furthermore, in cooperative dilemmas with long return rates (e.g., offspring quality when reaching adulthood), or dilemmas where a potential cooperator may become a competitor (e.g., warfare), signals with unproductive costs limit the temptation of signal recipients to exploit the signaler (Bergstrom, Kerr, & Lachmann, 2008; Bolle, 2001). Signaling cooperative intent with generosity may be risky if such signals strengthen a possible opponent, creating a second-order signaling problem (signalers need to trust recipients of their signals). Thus, rather than looking at generosity as a costly signal, it is crucial to investigate signals with unproductive costs that may better capture conflict-related signaling.

The case of inter-group conflict also differs from previous studies of generous giving because conflict carries essential risks of injury and even fatality when betrayed by others, making the cooperative dilemmas potentially one-shot rather than repeated. However, previous models using generosity as a costly signal often relied on the repetitive nature of interactions that secures' signaler benefits (Roberts, 2020), and the same is true of a recent experimental study with unproductive costs (Lang, Chvaja, Purzycki, Václavík, & Staněk, 2022). That is, due to the pre-determined number of interactions, people paying the signal cost and then defecting would be worse off than just plainly defecting. Yet, this is not necessarily the case if defectors could, for instance, desert to the other side during conflict. This caveat calls for a better understanding of how could costly signals work in one-shot interactions.

Importantly, the canonical costly signaling model (Grafen, 1990) postulates that costly signals should be effective in one-shot scenarios (e.g. when handicaps signal genetic quality). The assumption of differential costs has been built into the previous costly signaling models of cooperative intent (Gintis et al., 2001; McNamara & Houston, 2002), but it is unclear why generosity should be differentially costly for people with different intent. While genetic quality may directly affect signal intensity in animal models, this link is more flexible in human intention signaling (Fehrler & Przepiorka, 2013; Sosis, 2003). One possible mechanism facilitating the willingness to pay the signal cost is the perception of the costs and benefits of signaling, which may be biased by the signaler's intention (Sosis, 2003). For instance, blood donors perceive the health risks of blood donations as lower than non-donors, affecting the decision to send the signal (blood donation) or not (Lyle et al., 2009). Yet, this assumption remains largely untested in cooperative signaling (with the exception of Lang et al., 2022), hence this manuscript aims to shed light on this potential mechanism.

In summary, while previous studies suggest that costly signals may facilitate the reliability of communicating cooperative intent in iterated interactions, it is not clear whether and how costly signals may help assort cooperators during inter-group conflict. To fill this gap, we conducted four pre-registered studies with a general US population. We used an experimental framework where participants were first scored on their cooperative strategies, randomly divided into high and low cost conditions (cost manipulation), and then asked to choose a group in which they will play a PGG. They could choose between a group requiring a commitment signal (burning resources; we manipulated the amount of resources needed for the signal) and a group without such a signal. After this assortment, participants played one-shot PGG and competed with other groups.

In Study 1, we tested the effects of costly signals in a one-shot PGG with no conflict to get a benchmark result for later studies. In Study 2, we added between-group conflict. We achieved this by awarding the more cooperative group ¼ of the earnings from the less cooperative group. This modification allowed us to investigate how the possibility of sending costly commitment signals changes the within- and between-group dynamics during conflict. We also gave participants an opportunity to sacrifice part of their endowment to disadvantage a competing group and tested whether costly signals play a role in the decision to sacrifice for the group. In Study 3, we replicated Study 2 but disassociated the signal cost from resources used during the conflict to avoid disadvantaging signaling groups. Finally, in Study 4, we tested a rival proposition that forced costly signals create commitment in participants rather than signals it. We randomly assigned participants into the signaling and non-signaling groups, testing whether forced signaling would push people to stronger parochial cooperation (including self-sacrifice) despite their preferred strategies.