1. Introduction
Economic and evolutionary theories suggest that cooperation and prosociality are learned behaviors, often shaped by family interactions and normative factors. Yet, how these behaviors extend beyond the family unit remains unclear. We ask whether prosocial tendencies associated with sibling relationships generalize to non-kin contexts, and whether behavior systematically differs across what we term sibling status—having at least one sibling growing up or not, where sibling children (SC) did, and only children (OC) did not. We also test whether minimal kinship cues about a partner’s sibling background serve as coordination signals or identity heuristics in strategic settings.Footnote 1
We examine these questions using incentivized experimental games that model cooperation, public goods provision, and altruism. We find that, while subjects uniformly acted prosocially, SC consistently exhibited elevated prosocial behavior (e.g., 50 percentage points more cooperation in the stag hunt, up to 14 and 6 additional tokens in transfers to the public good and others in the dictator game, respectively, out of a 100 token endowment) compared to OC and controlling for other factors and without information about one’s counterpart.
We also test whether revealing a counterpart’s sibling background (i.e., a sibling status cue) affects behavior by randomly providing credible information about a counterpart. We do not claim that sibling status is a commonly salient social identity. Rather, we treat it as a minimal, experimentally controllable cue to show that family-structure background can serve as an identity heuristic in otherwise anonymous interactions. Our experimental intervention suggests that people react differently to sibling status cues. We find that when both partners are OC, this cue produces a surprising negative effect: OC–OC pairs cooperate and contribute even less than when sibling background is not revealed. Understanding why requires examining how sibling background relates to the norms that govern behavior. We proceed to reveal differentials in perceived social norms across groups and document systematic differences in these perceptions, which provide a plausible channel for the observed behavioral divergence. This design allows us to contribute to the literature in 3 ways. First, we provide experimental evidence that sibling background is associated with systematic differences in prosocial behavior. Second, by varying a sibling background cue, we isolate how a simple, observable family-structure signal shapes behavior in anonymous interactions. Third, by jointly measuring norms and actions, we show that SC and OC differ in their perceptions of norms, linking sibling background to distinct norm profiles.
1.1. Background literature
Why do humans often go out of their way to help or cooperate with others, even at personal cost or when facing overt risk? Becker (Reference Becker1974) developed foundational theories that emphasize how rational utility maximization accounts for altruism within families and links intrafamilial interactions to broader norms. In parallel with economic science and Becker, the kin-selection theory (Hamilton, Reference Hamilton1964; Smith, Reference Smith1964) offers a compelling explanation: altruistic behavior is more likely to be directed toward relatives, as helping family members promotes genetic survival. While these models explain intrafamilial cooperation, prosocial behavior often occurs between unrelated individuals—raising questions about how early socialization relates to extrafamilial interactions. Our findings are consistent with the idea that prosocial behaviors learned within the family environment influence decision-making in broader, non-kin economic environments.
When people exhibit prosociality, they are generally unrelated and often randomly paired; thus, these behaviors cannot be accounted for by the kin selection theory alone. Axelrod and Hamilton (Reference Axelrod and Hamilton1981) suggest that repeated kin interactions foster cooperative norms that may apply to broader, extrafamilial contexts as an evolutionary byproduct. Research shows that sibling dynamics, such as comforting, sharing, conflict resolution, reinforce prosocial tendencies (e.g., Hughes et al., Reference Hughes, McHarg and White2018; Sulloway, Reference Sulloway2001), and that early peer interactions shape children’s aversion to free-riding (Fehr et al., Reference Fehr, Bernhard and Rockenbach2008). These effects extend to economic preferences and educational attainment (e.g., Dai and Heckman, Reference Dai and Heckman2013; Detlefsen et al., Reference Detlefsen, Friedl, de Miranda, Schmidt and Sutter2024). Moreover, the development of social preferences is substantially impacted by the presence of older siblings (Andreoni and Miller, Reference Andreoni and Miller2002; Fehr and Fischbacher, Reference Fehr and Fischbacher2003).
Contrary to common stereotypes that OC are less cooperative or socially maladjusted (e.g., Falbo and Polit, Reference Falbo and Polit1986; Polit and Falbo, Reference Polit and Falbo1987), recent studies find no robust personality differences (Dufner et al., Reference Dufner, Back, Oehme and Schmukle2020). Detlefsen et al. (Reference Detlefsen, Friedl, de Miranda, Schmidt and Sutter2024) demonstrate that social preferences (e.g., trust and trustworthiness) may be influenced by birth order. Courtiol et al. (Reference Courtiol, Raymond and Faurie2009) find firstborns reciprocate less than later-borns in investment games.Footnote 2 We conjecture that such stereotypes about OC may germinate in differences in how they perceive norms compared to those with siblings (a cognitive development story similar to that in Evans et al., Reference Evans, Athenstaedt and Krueger2013). While OC may be more prone to making strategic decisions that appear selfish, they are more likely to act with a strategic, contextually focused mindset than to be solely self-interested. Many studies describe the effects of growing up with siblings and the behavior between siblings and kin (e.g., birth order), but we identified none that examine how having siblings or not relates to strategic or prosocial behavior toward non-familial counterparts or random strangers. After an exhaustive search, we were unable to identify a secondary dataset with suitable measures of incentivized prosociality and sibling status, motivating our experimental design.
At the same time, informational cues play a critical role in achieving socially efficient outcomes through coordination and social action (Schelling, Reference Schelling1980; Skyrms, Reference Skyrms2001); thus, a sibling status cue may serve as a focal heuristic, leading to an apparent intergroup effect. However, making outgroups easier to recognize can make interactions more unsociable; Astorne (Reference Astorne2023) proposes that clearer identity signals can sometimes reduce cooperation by making group boundaries more salient. Experimental support for these identity-driven models is limited, but Wild et al. (Reference Wild, Flear and Thompson2023) tested the basis in an evolutionary setting using ultimatum games, which inspired the informational intervention experiment in this study. Thus, we might expect prosociality to be robust when tied to a shared group cue or related shared norms; conversely, we may expect a negative cue effect leading to reduced prosociality 1) in cross-group dyads or 2) in shared groups with inside knowledge (i.e., ‘I know how our type behaves
$\ldots $
’).
Social norms play a key role in shaping economic behavior. Henrich et al. (Reference Henrich, Boyd, Bowles, Camerer, Fehr, Gintis, McElreath, Alvard, Barr and Ensminger2005) and Ostrom (Reference Ostrom1990) document wide variation in cooperation across cultures, driven by local norms. Cialdini (Reference Cialdini1984), Jonas et al. (Reference Jonas, Martens, Niesta Kayser, Fritsche, Sullivan and Greenberg2008), and Bicchieri (Reference Bicchieri2005, Reference Bicchieri2016) highlight how perceived social norms shape social behavior, making norms a key channel for understanding prosocial actions. Krupka and Weber (Reference Krupka and Weber2013) use incentivized coordination games to measure injunctive social norms and norm perception (i.e., appropriateness), demonstrating a strong link between norms and observed actions in economic games. These distinctions align with Tomasello and Vaish (Reference Tomasello and Vaish2013), who argue that moral norms are internalized through childhood cooperative routines. Norm adherence arises from shared intentionality, making childhood collaboration and parental influence critical channels through which norms relate to actions. When norm learning is less rooted in repeated sibling-like interaction, it is plausibly more situational (resulting in more strategic behavior). This tapestry of normative relationships motivates our inquiry. Yet, surprisingly few studies examine the nexus of norms, prosociality, and sibling background.
This inquiry is closely related to a growing body of research examining the One-Child Policy (OCP) in China and its impact on childhood development and individual personality differences. Jiao et al. (Reference Jiao, Ji, Jing and Ching1986) show that OC (aged 4–10) were more egocentric and less cooperative than children with siblings using 180 matched pairs. Cai et al. (Reference Cai, Kwan and Sedikides2012) used large internet samples of Chinese adults and found higher levels of narcissism among OC. In contrast, others have presented a more positive view of OC status. Liu et al. (Reference Liu, Lin and Chen2010) use surveys of Chinese middle school students and found that OC are more psychologically and behaviorally flexible than adolescent SC. Similarly, using data from the National Children’s Study of China, Li et al. (Reference Li, Li, Chen and Yang2024) find that OC exhibit more prosocial personalities, especially altruism. Still, larger families dilute parental attention, and parental tradeoffs may also impact prosocial development. While most studies focus on the Chinese context and child samples, Stronge et al. (Reference Stronge, Shaver, Bulbulia and Sibley2019) examined a sample of over 20,000 New Zealanders, finding that those without siblings were slightly less conscientious and honest, but more open to experience, than SC peers. What remains understudied is how such a policy may alter social preference formation and norm adherence.
There is a sparse experimental literature exploring behavioral differences between OC and SC that may result from the OCP. Cameron et al. (Reference Cameron, Erkal, Gangadharan and Meng2013, the closest study to ours) combine a quasi-experimental design with behavioral data from incentivized economic games to find that individuals in a Beijing-only cohort affected by the OCP are less trusting, less competitive, and more risk-averse compared to those in a cohort born before the policy. Li and Qiu (Reference Li and Qiu2021) find no differences in a dictator game setting (see also Chen et al., Reference Chen, Zhu and Chen2013), but provide some evidence that SC may be less demanding in the ultimatum game and defect less in the sequential prisoner’s dilemma. These papers are important in that they suggest that sibling absence may influence strategic behavior and social preferences.
Our study extends the literature beyond Cameron et al. (Reference Cameron, Erkal, Gangadharan and Meng2013) in several ways. Although those authors argue that their findings are generalizable to other urban parts of China, the results may not extend to broader contexts. We use a U.S.-heavy but globally diverse sample of adults, making our results potentially applicable across a wider range of socioeconomic groups (we argue that adulthood is interesting because it is in adulthood that individuals’ actions carry greater weight). Cameron et al. treat the OCP as a natural experiment and examine behavioral outcomes, such as trust and risk preferences. Our study goes further by introducing a randomized informational intervention that reveals subjects’ sibling status as a cue, thereby allowing us to capture differential effects in dyads. Perhaps most importantly, we establish more consistent and significant differences in prosociality (across several dimensions) and perceived norms than in the current literature, and we explore how those differences relate norms to actions: OC appeared more responsive to behavioral expectations, while SC did not.
2. Data and methods
Our design and first three hypotheses were pre-registered at Aspredicted.co #130273 and #192219, and approved by WIU IRB # 24-01.
2.1. Design
We conducted an informational 1
$\times $
3 between-subjects intervention experiment, in which the sibling status of a subject’s counterpart in each dyadic economic interaction (a sibling status cue) was revealed, isolating differences in prosocial behavior between sibling groups. We focused on decision-making in classic bargaining games that model settings in which humans must coordinate to achieve a socially efficient outcome. Decisions depend on their perception of risk, trust, precedents, norms, and how the other players may act (Binmore, Reference Binmore1994).
Each subject participated in three interactions common in experimental economics to capture prosocial behavior:
-
• Stag hunt (hereafter SH, to capture cooperation in the face of a risky outcome, Figure 1a). The prosocial action is the cooperative, Pareto-dominant choice (what is typically the ‘Stag’ strategy, which we label ‘Blue’ in the experimental interface for neutrality), compared to the risk-dominant (Yellow for ‘Hare’). SH involves 2 pure-strategy Nash equilibria: one (Blue, Blue) in which each player coordinates on a risky yet mutually (or socially) beneficial outcome, and one (Yellow, Yellow) in which each player opts for a diminished but guaranteed payoff. Online and remote work groups, as well as other team settings, often resemble the classic coordination problem. We consider collaboration and cooperation as the same semantically.
Figure 1 Normal form for two of the strategic games: payoffs are written
$(u_A, u_B)$
, where A is the row player and B is the column player, and show the number of tokens received. Panel (a) is stag hunt (SH) with blue corresponding to the cooperative stag strategy and yellow to the safe hare strategy. Panel (b) is public goods game (PGG) with VCM/social interest rate of 1.5. PGG actions are simplified here to illustrate the strategic form, thus resembling an n-dimensional prisoner’s dilemma; in the experiment, players were permitted to transfer any amount between the endpoints of 0 and 100 (
$s_i \in [0, 100]$
). Public pool increases in
$s_i$
, as do the benefits to j for free riding. -
• Public goods game/VCM with
$\alpha =1.5$
(hereafter PGG, Figure 1b). PGG is commonly used to capture contributions out of a personal endowment at a rate of return that balances mutual benefits (public wealth) with individual incentives to free ride. The model resembles real-world scenarios such as working on group projects. -
• The Dictator game (hereafter DG, Figure 2) is a standard, non-strategic measure of altruism. One player unilaterally allocates from a private endowment to another who has no endowment or say in the matter. DG resembles decision-making involving inequitable allocations of wealth, such as layoffs, workplace bonuses, and philanthropy.
Figure 2 Extensive form for the dictator game (DG): Payoffs are vertical vectors with top outcome for A as the first mover and bottom outcome for B (second or non-mover). Player A can allocate an amount of tokens
$s_A \in e_A=[0,100]$
to B; player B has no action. The rational, self-interested player allocates 0; players with social preferences (a different specification of
$u_i$
) are more likely to commit to non-zero actions.
In each game, payoffs are a function of mutual actions. A benefit of using these games as models of prosocial behavior is that the instructions are easy for subjects to understand. The games were presented in random order to avoid order effects (e.g., learning or fatigue) and used neutral language. We were mindful of the potential for game-dependent heterogeneity, particularly in DG, which is quite sensitive to context (List, Reference List2007). Players (N = 859) interacted asynchronously online via Prolific.co.Footnote 3
Each game involved stakes of approximately 100 tokens (U.S. $1 = 100 tokens). Subjects earned a $1.00 participation fee and subsequently a bonus of up to $2.40 depending on 1 of the 3 games’ outcome (randomly determined) and other bonuses, paid to their Prolific account. The design employs complete randomization, embedded question timers, and comprehension checks to adhere to assumptions necessary to establish causality of the intervention in an experimental environment (e.g., the Stable Unit Treatment Value Assumption [SUTVA], independence, compliance, etc.) and internal validity.
The experimental design involves three conditions. In the control condition, subjects participated in the games without knowing their counterparts’ characteristics. In the first treatment, t1, subjects are informed that their dyadic counterpart grew up with at least 1 sibling (SC sibling status cue). In t2, subjects are informed that they are playing with an only child (OC sibling status cue). Note that in t1/t2, both members of the dyad receive a sibling status cue about the other before choosing action. In the data collection instrument, subjects are told: ‘and everyone will be given information about their counterpart
$\ldots $
’). This design provides a one-sided cue about the counterpart, with implied but not explicit reciprocity. We subsequently matched counterparts randomly into dyads that conform to the information they were provided (to avoid deception); instructions stated that they would receive a different counterpart in each game. As a result, t1/t2 are best interpreted as testing a belief/cue effect (i.e., how people react to information about the other), rather than as a full-strength shared identity/common-knowledge manipulation.
The instrument included questions that elicited normative measures in accordance with the typology proposed by Bicchieri (Reference Bicchieri2016). We collected players’ empirical expectations (hereafter EE, i.e., expectations about what players actually did) in the mindset of Bicchieri and Xiao (Reference Bicchieri and Xiao2009). After each game, participants were asked to predict how others in their session had behaved in that same game—the modal action decision makers chose. These reports were incentivized: participants earned a small bonus when their prediction matched the actual mode, analogous to Bicchieri and Xiao’s use of incentivized guesses about how many dictators chose the ‘fair’ option in their study. Given the ‘wisdom of the crowd’ consensus effect, this approach provides a robust and simple proxy for subjects’ actual beliefs, arguably simpler than other mechanisms (see Schmidt et al., Reference Schmidt, Heinicke and König-Kersting2022). We used the method in Capraro and Rand (Reference Capraro and Rand2018) to elicit normative beliefs (hereafter NB, i.e., what people say is moral/right) for each game; subjects selected options in response to the question, Which action in this interaction, if any, do you think was the morally right or socially correct choice? Footnote 4 After collecting EE and NB, we compared them between the groups and experimental conditions (Section 3.3).
Considering the relationship between risk preferences and strategic choice documented in Büyükboyacı (Reference Büyükboyacı2014), we had players choose a gamble from 5 lottery options (following Eckel and Grossman, Reference Eckel and Grossman2008). Similar scenarios to our games may be considered socially risky (Wilson and Eckel, Reference Wilson and Eckel2011). The average completion time was 9 minutes. There was no deception throughout.
2.2. Sample
Power tests
$(\delta>0.5,power=0.8$
, and
$\alpha =0.05)$
, indicated a target of 540–600 subjects. We collected a sample of 891 from Prolific.co.Footnote
5
Prolific offers integrated payment and demographic profiling, minimizing survey length. We opted to oversample in the control group to support subgroup analysis and to balance by gender. Additionally, we desired a large corpus for conducting text analysis (natural language processing involves considerations distinct from statistical power). Oversampling the control improves the precision of baseline estimates and enables more reliable comparisons across heterogeneous treatment arms, as we were benchmarking against unconditioned priors. Power tests suggested that this imbalance would not lead to a statistical problem for the conditional average treatment effects (CATEs). Mindful of the potential for cultural differences, we collected a demographically balanced, international sample (using screening filters) with a strong emphasis on U.S. users.
Critical to our analysis, we could identify the sibling backgrounds of our subjects in two ways. First, Prolific provided an indicator whether they had any siblings. ‘Having siblings’ was not narrowly defined (i.e., biological siblings only, step-siblings, and half-siblings). Subsequently, we asked our subjects 1) to confirm whether they had any siblings in the context of the environment they grew up in, 2) questions about the gender distribution of their siblings, and 3) how close they considered themselves to those siblings (scale of 1–10). We observed negligible disagreement between these measures in the sample. This check helps ensure accurate classification of sibling background. Combined with asynchronous play, complete randomization, and the absence of communication between participants, this supports SUTVA. SUTVA matters because our treatment is a family background information cue, so we require participants’ outcomes to depend only on their own assigned information condition (and not on spillovers, communication, or cross-participant interference) for the estimated treatment effects to be interpretable as causal.
Table 1 provides a balanced summary of our full sample of 859 observations across treatment conditions, although we observed that females were slightly over-represented in t2. We find no differences in risk preferences or other demographic characteristics between the groups, suggesting that our randomization was sufficient to support the empirical assumptions.
Table 1 Summary statistics, by treatment
Note: Race and U.S. residence are reported as sample proportions; standard deviations are omitted. Approval captures the average number of approved (completed) prolific studies across subjects. We omit nationality indicators for conciseness.
2.3. Hypotheses
Our experimental design and sample were selected to test the following hypotheses:
-
•
$H_1{:}\ {}$
Sibling status and prosociality: SC are more prosocial (e.g., cooperative and altruistic) on average than OC in the selected economic games versus
$N_1{:}\ {}$
No systematic differences. Differences may be game-dependent.
$H_1$
aligns with Becker (Reference Becker1974), where family structures shape altruistic preferences and resource-sharing norms; Axelrod and Hamilton (Reference Axelrod and Hamilton1981), in that sibling relationships may serve as a foundation for cooperative norms that extend to extrafamilial social and economic contexts; and Cameron et al. (Reference Cameron, Erkal, Gangadharan and Meng2013), who show that Chinese ‘OCP’ cohorts were less trusting than the earlier generations. We test
$H_1$
with statistical differences in prosociality measures within the control group. -
•
$H_2{:}\ {}$
Impact of sibling status cues: Revealing a counterpart’s sibling status influences prosocial behavior (i.e., there exists a CATE) versus
$N_2{:}\ {}$
No effect.
$H_2$
is supported by Hamilton (Reference Hamilton1964), which posits that cooperation is partly driven by recognition of shared traits and builds on Choi and Bowles (Reference Choi and Bowles2007) and Li et al. (Reference Li, de Oliveira and Eckel2017), where the perception of status may impact cooperation. We test
$H_2$
using CATEs in t1 and t2 by sibling status.Footnote
6
-
•
$H_3{:}\ {}$
Normative differences across sibling status groups: SC and OC groups display different norm profiles (i.e., patterns of EE and NB across games), versus
$N_3{:}\ {}$
No relationship.Given any behavioral differences, we explore whether the overall norm profile (e.g., individually reported modal and mean levels of EE and NB by game) differs between OC and SC.
-
• We explore a fourth, exploratory relationship involving norms and game actions, in the mindset of Charness and Fehr, Reference Charness and Fehr2015, who propose using such measures to identify behavioral regularities. We expect a noticeable relationship between norms and behavior in economics games; magnitudes may differ by sibling group and game, resulting in gaps. The sibling status cue may reveal group differences. Intuition from game theory suggests a strong connection between EE and actions, whereas the norms literature provides strong evidence of the relationship between NB and actions (e.g., Bicchieri, Reference Bicchieri2005; Krupka and Weber, Reference Krupka and Weber2013)—particularly in DG, which is sensitive to many social factors (List, Reference List2007). This exploration entails our belief that there may be notable differences between norms and actions for informed subjects in each treatment.
3. Results
3.1. Sibling status group analysis
$Result_1{:}\ {}$
Subjects with siblings exhibit elevated levels of prosociality on average compared to those without across all 3 games.
All statistical analyses in the following sections were conducted within-game, and effect sizes are interpreted relative to each game’s scale (i.e., not normalized). As such, direct comparisons of effect sizes across games (e.g., between PGG and DG) should be made with caution. Figure 3 displays the distribution of actions across all 3 games by the sibling group. Cooperation in SH was 13.6 percentage points (Fisher’s exact test,
$p=0.002$
) lower in OC on average (3a). In PGG, SC transferred approximately 9 tokens more than OC on average (t-test,
$p=0.002$
), and we observe fewer Nash actions (i.e., zero transfers) in SC, with spikes at 50 and 100 tokens. In DG, mean allocations are 4 tokens higher for SC than OC, on average (t-test,
$p=0.012$
). We observe more 50–50 allocations and instances of pure altruism (i.e., allocations > 50) in SC, and more zero actions in OC. Two-sided K–S tests confirm the statistical difference between the distributions by sibling group (
$p<0.01$
for each game). In summary, while prosociality is prevalent in both groups, those who grew up with siblings exhibit elevated prosociality across all measures.
Figure 3 Comparison of prosocial actions by sibling status, uninformed, N = 568.
Table 2 provides a richer analysis, detailing point estimates from regression analysis of the relationship between having siblings and behavior across the 3 games, controlling for other factors. Our principal measures of interest are (1) playing prosocial action ‘Blue’ (binary) in the SH, (2) transfers in PGG, and (3) allocations to the counterpart in DG; the critical predictor of a sibling-based differential is a binary indicator for the subject being an SC (with OC the omitted). We include several common covariates, including risk tolerance (using the incentivized Eckel and Grossman, Reference Eckel and Grossman2008, method), demographic factors, and racial background/ethnic covariates to control for unobserved background heterogeneity (e.g., social capital, trust priors, and socioeconomic status [SES]-correlated experiences). In this study, race isn’t salient; however, it’s reasonable to treat race controls as capturing correlated cultural background factors, particularly since we cannot include nationality indicators (due to degree-of-freedom constraints). Several papers demonstrate behavioral differences across subjects by racial background (e.g., Castillo and Petrie, Reference Castillo and Petrie2010; Fong and Luttmer, Reference Fong and Luttmer2009). Following the narrative that higher SES of parents is associated with more patient, risk-tolerant, and often, more prosocial children (as described in Dohmen et al., Reference Dohmen, Falk, Huffman and Sunde2012), we control for growing up in a poor household (binary).
Table 2 Estimates of prosocial behavior (control group only)
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
. Column 1 reports average marginal effects on the likelihood of a subject playing Blue in SH (avg. marginal effects are obtained from the Firth penalized logit estimator, projected into a standard logit model with coefficients held fixed). Columns 2 and 3 detail estimates from a Tobit model for transfers in PGG and Allocations in DG, respectively, and bounded at the [0,100] endowment (7 top censored obs.).
Column 1 presents marginal effects from a Firth’s logit, indicating that, on average, being an SC is associated with a 0.501 higher predicted probability of playing Blue (about 50.1 percentage points), holding other covariates at their observed values (average marginal effect).Footnote
7
Columns 2 and 3 detail PGG transfer and DG allocation estimates derived from Tobit models, respectively. SC transfer nearly 14 more tokens in PGG (
$p = 0.005$
) and 5 more in DG (
$p = 0.092$
) on average, compared to OC, controlling for other factors. In all 3 games, significant risk tolerance coefficients conform to theoretical expectations.
3.2. Experimental results
$Result_2{:}\ {}$
Revealing the sibling status cue of one’s counterpart influences behavior, leading to a negative cue effect in OC only.
Our conceptual approach offers insight into the diverse behaviors exhibited by sibling groups. It allows us to explore an interactive sibling status cue-based effect that we identify in this section. Figure 4 illustrates the mean actions in each game, by treatment and sibling group. Figure 4a shows proportional differences across the treatments in SH. In the control, OC played Blue 54% of the time compared to 68% in the SC—a difference of 13.6% (Fisher’s,
$p=0.002$
). Thus, OC tended to cooperate less often, on average, in the absence of information. The differential between sibling groups playing Blue is 23% in t1 (
$p=0.006$
) and 47.8% in t2 (
$p<0.001$
), where OC played Blue only 22% of the time. Figure 4b displays the box plots for transfers in PGG. Mean transfers in the control condition were approximately 59 tokens for SC and 51 for OC (t-test,
$\Delta = 8, p<0.001$
). In t1, mean transfers were 20 tokens more by SC (
$p<0.001$
). In t2, SC transferred 24 tokens more than OC (
$\Delta = 24, p<0.001$
). Thus, OC–OC matches (in t2) resulted in less cooperation in SH and far lower transfers in PGG. Figure 4c depicts boxplots for the DG across subgroups and conditions. In the control, SC allocated, on average, 4 tokens more than OC (t-test,
$p=0.141$
). In t1, SC allocated 10 tokens more than OC (
$p<0.001$
), and in t2, the SC allocated 11 tokens more (
$p<0.001$
). In sum, we note that convergence to the socially efficient outcomes is robust as cooperation and contributions remain relatively high under all conditions (in contrast to Nash behavior). With information, OC mean allocations diverged and exhibited thicker left tails, indicating smaller transfers than those by SC.
Figure 4 Comparison of prosocial actions by sibling status and different treatments.
Regression analysis can isolate the cue effects of our informational intervention, controlling for other factors that might predict cooperation or altruism. We remind the reader that t1 reveals to the decision maker that the counterpart grew up with siblings, and t2 reveals that they were OC. The goal is to investigate whether being informed about a dyadic counterpart’s sibling background has a significant influence on behavior (through cues). Again, direct comparisons of effect sizes across games should be made with caution. We advise the reader that while the cue treatment effects are causal, baseline SC–OC differences reflect correlation, not causation (see Section 4.2).Footnote 8
Table 3 details the CATE estimates on prosociality for the 3 economic games, comparing across sibling status models. The primary predictor for cue effect is a binary indicator for treatment condition (control omitted); the prosociality measures are as before. Columns 1 and 2 display marginal effects from a standard logit listing the probability of a player choosing the Pareto-dominant Blue strategy. We observed in Section 3 that OC were less likely to play Blue than SC in the control condition. We find no sibling cue effect in t1, controlling for other factors. However, when they learned they were matched with an OC (in t2), the probability that they played Blue was 34 percentage points lower than the control (
$p<0.001$
). Columns 3 and 4 detail estimates from a Tobit regression over transfers in PGG. While there is again no treatment effect on SC players, OC sent 24.82 fewer tokens in t2 (
$p<0.001$
); we observed no such cue effect in t1. Columns 5 and 6 capture the CATEs for each sibling group in the DG. We observe that OC allocated 7.62 fewer tokens in t2 with a diminished confidence level (
$p=0.071$
), but there is again no effect in t1. We observe that OC counterparts were less likely to cooperate in SH and contributed less to the public good (i.e., they were more prone to free-riding) on average in t2 when informed about the OC pairing. In this regard, the sibling status cue did not lead to an intuitive, positive identity effect. We note that the signs of the risk tolerance coefficients are consistent with game-theoretic predictions.
Table 3 Intervention treatment effects in 3 games, by sibling status
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
. Columns 1and 2 report marginal effects from logit regression on the likelihood of a subject playing Blue in SH. Columns 3–6 detail estimates from a Tobit model for transfers in PGG and allocations in DG, respectively, and bounded at the [0,100] endowment. Seven obs. dropped due to missing control values.
There are 2 main takeaways. First, OC exhibited, on average, diminished prosociality compared to SC across all 3 games. This is the case both when informed and uninformed. Second, the sibling status cue in each experimental condition does not appear to impact SC actions. However, we do see OC acting differently when receiving information about a counterpart with a similar sibling background.
3.3. Norms
Social norms is an overarching concept encompassing perceptions of the rules governing behavior. Game theory dictates that strategic situations should weigh heavily toward EE (Bicchieri, Reference Bicchieri2005; Binmore, Reference Binmore1994). We would generally expect a relationship between expectations and actions in SH (a coordination game) and PGG (a non-cooperative game involving incentives to free ride). However, personal normative beliefs are what people believe they should do, even if it doesn’t align with actual behavior (Bicchieri, Reference Bicchieri2016). Such norms are typically described as channels that lead to prosocial action; players with so-called social preferences might act in accordance with their reported NB. Familial upbringing is critical to the formation and adoption of norms in children.
3.3.1. Norm profile differences
In this section, we explore differences in norm profiles and the relationship between norms and action within sibling groups and across experimental conditions. Because norms were elicited post-action, they may reflect motivated reasoning (Bénabou and Tirole, Reference Bénabou and Tirole2005); we therefore omit them from regressions because they violate empirical assumptions. We use norms in an alternative manner, drawing inferences from differences in EE and NB (gaps) and between norm types and actions across sibling groups in the cue treatments.
$Result_{3}{:}\ {}$
Comparing the norm profiles across sibling groups, OC expectations differ consistently from those for SC across all 3 games (often leaning toward the Nash EQ). There are no notable differences in the NB.
Table 4 compares SC and OC norm profiles across 3 games, detailing modes and means for observed actions, EE, and NB. Across all interactions, norms differed systematically by sibling status. We focus on the first 3 columns, comparing actions and norm differentials across sibling groups. In SH (Panel 4a), SC cooperated substantially more than OC (i.e., the modal action for SC is efficient Blue outcome, whereas OC’s modal action is Yellow; mean percentage of subjects playing Blue is 20.92% higher for SC (Fisher’s
$p<0.001$
). Although both groups viewed cooperation in SH as normative on average, SC expected significantly higher mean cooperation levels than OC (EE
$\Delta =23.05\%$
,
$p<0.001$
). We note both groups deviated from EE to some degree in the pooled sample. Panel (4b) reports PGG actions and norm profiles. Across both sibling groups and informational conditions, the modal contribution is 50 (the equal split), as is the modal EE. Despite this, mean contributions differed by sibling status: SC contributed roughly 13 tokens more than OC (
$p < 0.001$
). However, OC expected others to contribute significantly less on average than SC did (
$\Delta $
= 12.01, Mann–Whitney
$p < 0.001$
). Again, both groups viewed higher transfers as socially normative on average, with NB for transfers hovering around 50 tokens. In DG (Panel 4c), again the 50/50 allocation is modal, but SC allocated more on average than OC (
$\Delta $
= 6.02 tokens, Mann–Whitney
$p < 0.001$
). Consistent with the other 2 interactions, both sibling groups have similar normative beliefs about socially appropriate behavior, on average. However, we again observe a significantly lower expected allocation by OC (
$\Delta $
= 7.96,
$p < 0.001$
).
Table 4 Summary of norm profiles by game and sibling status (pooled)
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
. The table represents a pooled sample of SC (N = 568) and OC (N = 291). For differences in means of actions and norms by group, we use Fisher’s exact tests on a binary variable for blue = 1 in SH, Mann–Whitney U-tests for PGG and DG. Action–EE and Action–NB columns report the gap between mean action and mean expectation/belief, tested with McNemar’s Chi for SH, Wilcoxon signed-rank tests on actions for PGG and DG; negative values indicate actions fell below stated norms.
$\Delta $
columns reflect the difference in means between SC and OC (SC–OC). We note that a proper diff in diff approach between action–norms gaps by group was statistically limited.
The Action–Norm gaps listed in Columns 4 and 5 in Table 4 reveal an asymmetry between groups. Action–EE gaps are positive and similar in magnitude for both SC and OC across all 3 games, indicating that both groups act slightly above what they expect others to do. However, Action–NB gaps diverge, particularly in SH, where OC actions fell well below their normative beliefs about playing Blue (Action–NB
$\Delta =-12.05\%$
, McNemar’s Chi
$p<0.001$
), and where actions deviate quite a bit from the mean NB (Action–NB
$\Delta $
=
$-24.57$
, Wilcoxian rank
$p < 0.001$
). We speculated that the NB would generally be similar to DG allocations, suggesting that players viewed the game less as a strategic exercise and more as an ethical dilemma, but there are gaps in both groups. In all games, the action–norm gap is larger in OC; however, we omit reporting the action–norm differences in gaps (diff in diffs) between groups because this second-order comparison lacks a direct substantive interpretation and an appropriate test. This pattern suggests that, while both groups anchor on empirical expectations (as expected in strategic settings), OC actions decouple more sharply from what they report as normatively appropriate, consistent with behavior driven by expectations rather than NB. We interpret this pattern as consistent with EE generally playing a larger role in shaping actions in OC, but note that this pooled sample includes subjects that received sibling status cues. In the following section, we explore norms in the context of the informational treatments.Footnote
9
3.3.2. Treatment effects on norm–action gaps
Having identified differences in norm profiles between sibling groups, we now use similar differentials in the informational treatments to explore the relationship between sibling cues and norms. Tables 5–7 detail actions and norm profiles across 3 games, across the 3 experimental conditions.
$Result_{4}(exploratory){:}\ {}$
Actions, empirical expectations, and normative beliefs gaps exist in both groups, but are more pronounced and more asymmetric in OC. For OC, actions more closely track EE but diverge significantly from NB, a pattern most prominent in t2 (with the OC sibling status cue) and most evident in SH and PGG.
Table 5 Summary of norm profiles by game, treatment, and sibling status (SH)
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
. For differences in means of actions and norms by group, we use Fisher’s exact tests on a binary variable for Blue = 1. Action–EE and Action–NB columns report the gap between mean action and mean expectation/belief, tested with McNemar’s Chi; negative values indicate actions fell below stated norms.
$\Delta $
columns reflect the difference in means between SC and OC (SC–OC). We note that a proper diff in diff approach between action–norms gaps by group was statistically limited.
Table 6 Summary of norm profiles by game, treatment, and sibling status (PGG)
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
. For differences in means of actions and norms by group, we use Mann–Whitney U-tests. Action–EE and Action–NB columns report the gap between mean action and mean expectation/belief, tested with Wilcoxon signed-rank tests; negative values indicate actions fell below stated norms.
$\Delta $
columns reflect the difference in means between SC and OC (SC–OC). We note that a proper diff in diff approach between action–norms gaps by group was statistically limited.
Table 7 Summary of norm profiles by game, treatment, and sibling status (DG)
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
. For differences in means of actions and norms by group, we use Mann–Whitney U-tests. Action–EE and Action–NB columns report the gap between mean action and mean expectation/belief, tested with Wilcoxon signed-rank tests; negative values indicate actions fell below stated norms.
$\Delta $
columns reflect the difference in means between SC and OC (SC–OC). We note that a proper diff in diff approach between action–norms gaps by group was statistically limited.
Starting with SH (Table 5), overall SC are more cooperative, playing Blue 13.64% (
$p<0.01$
) more than OC in control and 22.92% and 47.76% (
$p<0.01$
) more in both treatments, respectively. The gaps in the EE between the 2 groups widen across treatments, culminating in a 45.55 token difference in expectations in t2 (
$p<0.01$
), and 20% widening between t1 and t2 (Fisher’s
$p=0.032$
). Overall in SH, OC appear to align actions closely with EE, but not NB, more so when receiving the OC cue. Within groups, both SC and OC play Blue more often than they expect others to do (Action–EE gaps,
$\Delta =12.44\%$
and
$16.93\%$
, both
$p<0.01$
). In t1, OC actions aligned closely with their EE. With OC sibling status cue, SC exhibited no meaningful difference between actions and either EE or NB. Whereas OC cooperated far less than they believed was socially correct (Action–NB
$\Delta =-38.78\%$
,
$p<0.01$
).
Table 6 shows the norms profiles in PGG. Without intervention (Panel a), OC transferred, on average, 8 fewer tokens than SC (
$p<0.01$
). When matched to SC (t1, 6b), OC transferred 20 fewer tokens than SC (
$p<0.01$
). When matched to OC (t2, 6c), OC transferred fewer tokens than SC (
$\Delta =24$
tokens,
$p<0.01$
), a result similar to those in Table 3. Both groups transferred less than their expectations of others, and both contributed less than they believed they should, suggesting no systematic normative difference between groups in the control. There is a significant gap between action and expectation for SC (Action–EE
$\Delta =7.40$
tokens,
$p<0.01$
) in t1, but no such gap for OC, suggesting that OC actions appear more aligned with their EE compared to SC. In t2, the actions of both groups deviated from their expectations; however, OC appear much more aligned with EE, with the Action–EE gap being less pronounced and precise and suggesting sensitivity to expectations. For OC, we find that their actions differ greatly from what they believed they should do (Action–NB
$\Delta =35.67$
tokens,
$p<0.01$
), but not for SC.
Finally, Table 7 depicts the norm profiles in DG. Without the intervention, we observe SC allocating about 4 tokens more than OC on average (Panel a,
$p<0.01$
). With the sibling cue in t1, the allocation gap between groups is wider (
$\Delta =9.5$
tokens,
$p<0.01$
); both groups gave significantly more than their expectations but less than their NB. Consistent with the t2 results in Table 3, we observed OC gave nearly 11 tokens fewer than SC when matched with OC (
$p<0.01$
). Both groups gave amounts consistent with their expectations, but at least 12 fewer than their normative beliefs. Normative beliefs are very similar across treatments and similar (in magnitude) between groups in all treatments. However, OC had drastically lower expectations of other OC than SC did (
$\Delta =19$
tokens,
$p<0.01$
), but ultimately gave more to those OC (Action–EE
$\Delta =15.90$
tokens,
$p<0.01$
). The key pattern in the DG norms profiles we observe is that there is little deviation between actions and expectations in either group in the absence of a sibling status cue. However, there was more variance in EE sensitivity. This is in contrast to the other games, where OC appeared far more sensitive to EE than SC.
In summary, we have consistent evidence that empirical expectations may be more important than normative beliefs in determining actions in OC when the OC status cue is received. We also see that normative beliefs remained stable across treatments and sibling groups. Kolmogorov–Smirnov tests comparing distributions of expectations and actions reinforce the idea that OC acted more aligned with expectations, on average.
3.4. Text analysis supports norms results
The previous sections revealed that the empirical expectations differed consistently between sibling groups, whereas normative beliefs did not. In what follows, we use text analysis to examine how participants described their experiences of playing the games, relate them to normative themes, and discuss a few potential channels. We show, in subjects’ own words, that OC frame decisions more in terms of expectations and cooperation (or the lack thereof), while both groups share similar fairness and ethics language, thereby supporting the view that sibling background is related to how norms are applied.
Our data collection instrument included an open-response prompt asking subjects to compose ex post advice to a future player on the ‘best’ way to play each game (not framed as ‘norm questions’). We used recent advances in language AI topic modeling to examine how participants described their decisions, connecting normative language to behavior. We pooled all text responses across games into a single corpus and then split it by sibling group, resulting in 568 documents for SC and 291 for OC, with an average length of 36 words. Each corpus is a mixture of topics. Our preferred analytical method, topic modeling, extracts conceptual topics from unstructured text data to capture semantic relationships. We used the BERTopic package (Grootendorst, Reference Grootendorst2022), which leverages a pre-trained large language model (LLM) framework and employs density-based (DBSCAN) clustering to generate topics that reflect contextual themes from each advice document that share similar semantic content—what the advice tends to emphasize, rather than requiring us to specify topics ex ante.Footnote 10
From the full set, we retained 7 norm- or interaction-oriented topics post-hoc for analysis and assigned simplified topic labels based on representative words and example documents. For each topic, we coded whether a document was represented (i.e., binary prevalence) and compared the proportion of documents in which the topic appears across sibling groups using
$2 \times 2$
chi-square tests with Holm correction to control the familywise error rate across the 7 comparisons (reported in Table A9). Document-level prevalence captures how widely a concept appears across subjects (rather than being dominated by a few long responses), helping isolate which norm-related themes systematically differ by sibling background. For the sake of clarity, we provide a representative document example for each topic:
Fairness: ‘Just read what it says its simple math not to difficult figuring out give at least 30% and each gets some bonus it only seems fair’.
Norms/Ethics: ‘Try and decide on what would be a morally good action to take’.
Equality/Balance: ‘I decided to give you half of my tokens so that we can get equal benefits. If you are in the role of A, please do the same, let us maintain fairness’.
Sanctions/Social pressure: ‘If you want a guaranteed payoff, choose yellow. But prepare to be judged and to feel guilty’.
Rules/Law: ‘I believe a minimum of 25% should be the rule given in this scenario’.
Expectations: ‘Trust your instincts. I’m under the assumption both people know the rules; the other will choose yellow, so don’t get stuck in the bad outcome if you play blue’.
Cooperation: ‘Both will the max pay at the end, if we contribute all. I want to cooperate though I know some people may not’.; ‘Don’t play Blue, your counterpart is not going to cooperate with you’.
Our novel use of language AI reveals distinct key themes in how subjects describe prosocial actions across groups (Figure 5), supporting the norm profile differentials highlighted in Section 3.3 and providing insight into a potential channel through which our main results may manifest. OC documents contain language fitting into the Cooperation topic much more commonly than SC (
$\Delta =7.95\%$
pp, Holm-adjusted
$p < 0.001$
). OC also used Expectations language more often than SC (
$\Delta =7.64\%$
pp, Holm
$p = 0.012$
when describing how best to play these games. The difference favoring OC in the Rules/Law is smaller and borderline significant once correcting (4.81% vs. 1.94%;
$d=2.87$
pp, Holm
$p = 0.088$
). Sanctions, Equality, Norms/Ethics, and Fairness show no significant differences across how the sibling groups talk about playing the games (i.e., both groups use these themes commonly).
Figure 5 Group text prevalence differences in norms topics by group corpora:
$SC$
= 568 and
$OC$
= 291.
Taken together, these patterns show the same qualitative pattern as in the structured norm data. The stronger emphasis on Expectations and Cooperation topics in OC advice echoes the empirical differences in expectations documented earlier: OC more often frame ‘what to do’ in terms of what others are likely to do and whether mutual cooperation can be sustained. We find no differences in the frequency of Fairness, Equality, and Norms/Ethics topics used across sibling backgrounds. The text evidence thus supports the view that sibling status shapes how people use contextual information, offering a plausible channel for the observed treatment effects.
4. Discussion
Our results show that sibling background is associated with meaningful differences in prosocial behavior. However, prosociality is present in both groups. We interpret OC actions as being more sensitive to empirical expectations and sibling cues, consistent with strategic behavior. These patterns are robust to controls and consistent across domains, supporting a norms-based interpretation.
4.1. Channels
In this section, we discuss some plausible, non-mutually exclusive channels, ordered from most to least well-established in prior literature. A complementary line of work suggests that family background shapes prosocial behavior not only through sibling interactions but also via parental investments and SES. Higher-SES families tend to raise children who are more altruistic and less risk-seeking, with differences linked to systematic variation in parenting style and the quantity and quality of time spent with children (Deckers et al., Reference Deckers, Falk, Kosse and Schildberg-Hörisch2015); at the same time, high-SES families have fewer children on average. Using stylized Dictator games with child subjects, Chi et al. (Reference Chi, Malmberg and Flouri2024) conclude that parental interaction is the single most modifiable factor in improving a child’s life trajectory. Kosse et al. (Reference Kosse, Deckers, Pinger, Schildberg-Hörisch and Falk2020) demonstrate that children’s prosociality is associated with SES, the intensity of mother–child interaction, and mothers’ own prosocial attitudes. Li et al. (Reference Li, Li, Chen and Yang2024) argue that family size shapes children’s prosociality through 2 channels and a trade-off: increased sibling interaction, which promotes prosocial skills, and reduced parental attention per child, which may dampen them.Footnote 11 Parents also transmit economic preferences and social attitudes by investing time and effort in their children, where higher parental involvement is associated with children who display more ‘favorable’ economic attitudes and personality traits, including risk and trust preferences. Children are more similar to their parents in these domains when parents are more engaged in upbringing and socialization (Dohmen et al., Reference Dohmen, Falk, Huffman and Sunde2012; Zumbuehl et al., Reference Zumbuehl, Dohmen and Pfann2021). OC, lacking siblings, may rely more heavily on parental relationships and interactions with adults, which can shape their sensitivity to hierarchical social cues (see Conzo and Zotti, Reference Conzo and Zotti2020). Overall, these findings, in tandem with our results, highlight a parental and family–environment channel through which prosocial and preference-relevant traits are shaped. In our setting, sibling status should therefore be viewed as 1 component of the broader family environment: our results suggest that, even after controlling for proxies of economic background, variation in sibling experience is associated with stable differences in prosocial behavior, over and above the intergenerational transmission of preferences that operates through parents.
We observed that SC and OC expressed different expectations of prosocial behavior, while generally having the same normative beliefs. Two concurrent channels could explain this finding. One is Norm internalization—stable injunctive standards that people regularly implement in behavior—in which norms are deeply embedded and learned through repeated, reciprocal interactions (e.g., Bicchieri, Reference Bicchieri2005; Bicchieri et al., Reference Bicchieri, Muldoon and Sontuoso2023). Early sibling experience and repeated opportunities for negotiation, cooperation, and conflict resolution support the internalization of prosocial expectations (Fehr et al., Reference Fehr, Bernhard and Rockenbach2008; Hughes et al., Reference Hughes, McHarg and White2018). This internalization leads to strong alignment between stated norms and actual decisions and is consistent with Becker (Reference Becker1974). Norm susceptibility describes when behavior is more responsive to specific situations, expectations, contextual cues, or identity signals, and where norms mainly influence behavior when people lack strong prior preferences (see Anderson and Dunning, Reference Anderson and Dunning2014). OC who lack sibling interactions may develop prosocial behavior through adult-centered or observational learning, leading to a pattern of more context- and cue-dependent strategic behavior. As a result, their prosocial behavior may depend more on perceived features of the interaction or counterpart (see Choi and Bowles, Reference Choi and Bowles2007; Grusec and Goodnow, Reference Grusec and Goodnow1994; Holm, Reference Holm2000; Tomasello and Vaish, Reference Tomasello and Vaish2013). We observed that OC exhibited greater EE dependence, and we believe that their actions shifted more in response to informational cues (particularly when matched with another OC), suggesting reliance on situational inference over stable internalization.
Finally, we interpret the negative cue effect observed in the OC–OC dyads as a sort of coordinated pessimism. Our results provide some empirical support for theoretical predictions, such as in Astorne (Reference Astorne2023), that clearer identity signals do not necessarily improve cooperation. In contrast to a Chen and Li (Reference Chen and Li2009)-style shift in social preferences, we suspect that OC may understand the cultural landscape (and maybe their own tendencies), leading to a cue effect defined by shared doubts about each other’s cooperativeness or intent.Footnote 12 When OC learned they were matched with another OC, the cue changed what they inferred about the partner’s type: a statistical cue. This may result in OC–OC pairs moving toward a sort of “pessimistic equilibrium” (i.e., a relatively safe action in SH, lower contributions in PGG, given lower expectations). This conjecture is supported by the comparative analysis in Sections 3.3.1 and 3.3.2, which show that OC–OC dyads had lower EE about their counterparts compared to others.
4.2. Limitations
We recognize that our online Prolific sample approach may not capture the realities of other populations or directly mirror some real-world scenarios. While these factors can make it difficult to draw broad conclusions about how people behave in real-world situations or how our results scale without additional policy-based evidence (List, Reference List2024), the results are valid for this population and setting, and numerous papers demonstrate that these game-theoretical situations effectively model real-world interactions (Falk and Heckman, Reference Falk and Heckman2009). We do not claim that the overall difference between sibling groups (being in one group or the other) has a causal effect on prosociality (i.e., our initial prosociality results). Being an only child or having siblings is not randomly assigned. Parents with one child versus multiple children may differ systematically on unobserved dimensions, particularly in parental influence. Our controls for socioeconomic background and race help absorb some of this heterogeneity, but cannot fully address selection. Parenting practices may differ between single-child and multi-child households in ways that affect social development independent of sibling interaction itself. We therefore interpret the baseline SC–OC differences as documenting a robust association, rather than a causal sibling effect. The treatment effects from our randomized informational intervention, by contrast, support causal inference about how sibling status cues shape behavior. While we tested and ruled out differences in risk preferences as a factor, we acknowledge that other factors may exist (e.g., differences in cognitive reflection). Our norms measures cannot capture common dimensions, such as dynamic norms (e.g., how norms evolve, Axelrod and Hamilton, Reference Axelrod and Hamilton1981), network effects (e.g., how norms spread across groups Young, Reference Young2009), or targeted norm nudging within social networks (Bicchieri et al., Reference Bicchieri, Dimant and Sonderegger2023). However, our insights help motivate future research into how sibling-driven prosociality might work across diverse social and institutional contexts.
5. Conclusion
We provide some evidence that prosocial tendencies may be learned through sibling relationships and generalize to non-kin contexts. Consistent with Becker’s framework, we demonstrated settings in which individuals with siblings exhibited greater prosociality than OC. However, those who grew up in different familial environments tended to hold different expectations of prosocial behavior, while generally sharing the same normative beliefs. OC appear to respond differently to sibling status cues when encountering other OC. These results highlight the formative roles of early social learning and intrafamilial socialization in shaping behavior beyond the family, particularly relevant in the context of increased remote work and reduced interpersonal familiarity. Our results also provide insight into the relationship between norms and action between sibling groups.
As OC become more prevalent in low-fertility societies (Salari et al., Reference Salari, Heidarian, Abdolmaleki, Salim, Hashemian, Daneshkhah and Mohammadi2025), these patterns may have broader implications for collaborative networks, entrepreneurship, and economic development (Cameron et al., Reference Cameron, Erkal, Gangadharan and Meng2013). Widespread lack of sibling-oriented relationships and fewer extended kinship ties could mean weakened social support networks, unstable marriage and labor force patterns, and intergenerational care (following the OCP analysis in Zhang, Reference Zhang2017). Although sibling status is unlikely to be a salient intergroup characteristic, we provide evidence that it matters. In team-based environments where initial trust is crucial (e.g., hiring, remote collaboration, or short-term group work), sibling background may influence how individuals interact with unfamiliar colleagues or how people adapt to leadership roles. Future research should explore these areas.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/jdm.2026.10032.
Data availability statement
Replication data, instructions, and data collection instrument are publicly available in the Harvard Dataverse at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/8LIGQ5.
Acknowledgements
The authors benefited significantly from the support and input of John A. List, J. Braxton Gately, Anna Klis, Andrew Hussey, Gabriela Galindo-Rubio, and especially the careful consideration of the editor and two anonymous referees. The article was presented at the 2024 Midwest Economic Association meetings, and the authors appreciate the thoughtful comments that improved the manuscript.
Author contributions
Conceptualization: J.J.B., H.C., and M.T.; Data curation: J.J.B. and H.C.; Data visualization: J.J.B. and H.C.; Methodology: J.J.B., H.C., and M.T.; Writing original draft: J.J.B., H.C., and M.T. All authors approved the final submitted draft.
Funding statement
The authors declare that no specific funding has been received for this article.
Competing interests
The authors declare none.
Ethical standards
The research adheres to all ethical guidelines, including compliance with legal requirements in the USA.
A. Appendix
Table A1 Intervention treatment effects in 3 games, by sibling status (US)
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
.
Table A2 Intervention treatment effects in 3 games, by sibling status (Non-US)
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
.
Table A3 Intervention treatment effects in 3 games, by sibling status (US as a control)
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
.
Table A4 Intervention treatment effects in 3 games, by sibling status (female)
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
.
Table A5 Intervention treatment effects in 3 games, by sibling status (male)
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
.
Table A6 Estimates of prosocial behavior (control group only) additional control-gender
Note: Robust standard errors in parentheses; *
$p<0.1$
, **
$p<0.05$
, and ***
$p<0.01$
. Column 1 reports average marginal effects on the likelihood of a subject playing Blue in SH (marginal effects are obtained from the firth penalized logit estimator, projected into a standard logit model with coefficients held fixed). Columns 2 and 3 detail estimates from a Tobit model for transfers in PGG and allocations in DG, respectively (7 top censored obs.).
Table A7 Estimates of prosocial behavior (control group only) additional control-closeness
Note: Robust standard errors in parentheses; * p<0.1, ** p<0.05, and *** p<0.01. Column 1 reports average marginal effects on the likelihood of a subject playing Blue in SH (marginal effects are obtained from the firth penalized logit estimator, projected into a standard logit model with coefficients held fixed). Columns 2 and 3 detail estimates from a Tobit model for transfers in PGG and allocations in DG, respectively (7 top censored obs.).
Figure A1
Figure A2
Figure A3 Actions, normative beliefs, and empirical expectations in SH (all p-values are Fisher’s or Mann–Whitney tests unless otherwise specified).
Figure A4 Actions, empirical expectations, and normative beliefs in PGG (p-values from the Mann–Whitney tests).
Figure A5 Actions, empirical expectations, and normative beliefs in DG (p-values from the Mann–Whitney tests).
Table A8 Phi coefficient correlations, actions and norms in SH by sibling group.
Note:
$^{*}p<0.10$
,
$^{**}p<0.05$
, and
$^{***}p<0.01$
. n/a denotes an uncalculable correlation coefficient due to a large number of constant values.
Table A9 Document-prevalence differences in norm-language topics (SC vs. OC).
Note: SC% and OC% are the percentages of documents in each group containing at least one term from the category. Diff = SC%–OC% (percentage points). CI_lo/CI_hi are 95% confidence intervals for Diff_pp. p_holm uses Holm correction across 7 tests. Significance stars reflect the thresholding used in the provided output.
Open access
