Introduction
The ability to use an individual’s behavior in one context to make inferences about how that individual will behave in a different context (Wernimont and Campbell, 1968; Lammers et al., 2017), provides a basis for coordinating our actions with other people, and enables us to make decisions regarding whom to approach and whom to avoid. One means by which (Western) adults predict others’ future behavior is through the rapid attribution of traits based on limited behavioral evidence (Lupfer et al., 2000; Uleman et al., 2008; Shimizu, 2012). The tendency to infer morally relevant traits, in particular, is prevalent in adults, as an individual’s moral character may have consequences for one’s own welfare (Goodwin et al., 2014). At present the earliest developmental origins of this ability is relatively uncharted, given debates concerning the nature of young children’s moral understanding (see Killen and Smetana, 2015 for a review), and the emergence of trait reasoning in childhood (see Heyman, 2009 for a review). Here we investigate the developmental origins of the ability to use information from one moral sub-domain to make inferences about how an individual will behave in another sub-domain in infancy, focusing on fairness and help/harm given their prominence in theories of morality (Ting et al., 2019; Decety et al., 2021).
The developmental origins of children’s trait inferences
Research indicates that children’s ability to make trait inferences, broadly construed, varies considerably according to the experimental approach and task requirements. Initial work suggested that it is not until mid-childhood that children reason about others’ traits: children first describe others’ behaviors using trait terms at about age 8 (Livesley and Bromley, 1973; Peevers and Secord, 1973; Heyman and Gelman, 1999), and whereas 9- and 10-year-old children expect behavioral consistency across contexts, younger children do not (Rholes and Ruble, 1984; Kalish, 2002). Subsequent studies, however, demonstrated that children as young as 4 infer traits when multiple exemplars of an initial behavior are provided (Heller and Berndt, 1981; Cain et al., 1997; Boseovski and Lee, 2006), when children are told about two different actors that act in opposing directions (e.g., generous versus selfish; Heller and Berndt, 1981; Cain et al., 1997), or when they are asked to identify a trait from a given behavior, or use a trait to infer a subsequent behavior (Heyman and Gelman, 1999; Liu et al., 2007).
Moral trait inferences may emerge before the preschool years given the importance of this tendency in everyday social interactions (i.e., Rakoczy et al., 2016). Indeed, moral trait terms are some of the first that are utilized in children’s language output (Yuill, 1992; see Franchin et al., 2019): as early as 19–22 months, infants use terms such as ‘good,’ ‘bad,’ ‘naughty,’ and ‘nice’ (Bloom et al., 1975; Fenson et al., 1994), and by 30 months children start to apply words like ‘good’ and ‘bad’ to moral contexts (Snow, 1987). These findings raise the possibility that young children may make moral trait inferences, even if they struggle to form trait inferences for non-moral traits.
Moral sensitivities in infancy
Recent work suggests that infants possess moral sensitivities in the sub-domains of fairness and help/harm. Infants expect other agents to approach those that previously helped versus those that harmed (Kuhlmeier et al., 2003), and prefer agents that help over agents that hinder in third-party settings by 6 months of age (Hamlin et al., 2007; Hamlin and Wynn, 2011; Hamlin, 2015). By roughly 10 months of age (or earlier, when distributions featuring 2:0 versus 1:1 are used; see Buyukozer Dawkins et al., 2019) infants show a nascent sensitivity to fairness, expecting equal distributions (Meristo et al., 2016; Ziv and Sommerville, 2017), and by 13 months of age, evaluate others based on their distributive behavior, showing preferences for those that distribute resources equally over those that do so unequally (Burns and Sommerville, 2014), and link fair and unfair behavior to positive and negative stimuli, respectively (DesChamps et al., 2016; see also Geraci and Surian, 2023). Infants’ sensitivity to equal distributions can also be seen in cases where a distributor intended to perform an equal distribution but failed: 9-month-olds prefer a distributor that tried to divide resources equally but failed to a distributor who tried but failed to divide resources unequally (Geraci et al., 2022; but also see Strid and Meristo, 2020).
Of course, mature moral understanding entails a recognition that behavior from one moral sub-domain may have consequences for how an individual will act in another sub-domain; adults may expect a stranger who returns their dropped wallet to also hold the door open for them as they exit the bus. In a recent study, Gill et al. (under review)1 demonstrated that children aged 4- to 7 show similar tendencies: children reported greater surprise to fair (versus unfair behavior) after a protagonist hindered versus helped another individual. In addition, children reported greater surprise to the protagonist helping (versus hindering) another person after she previously distributed resources unfairly versus fairly. While children generalize from help/harm to fairness, and from fairness to help/harm, they did so uniquely from negative behavior: children’s surprise reports following fair behavior or helping behavior did not vary based on the test event. Thus, by age 4 children generalize across moral sub-domains after seeing moral transgressions.
This generalization tendency may extend to infants. Surian and colleagues (Surian et al., 2018) investigated infants’ ability to engage in moral generalization from help/harm to fairness (see also Ting and Baillargeon, 2021) by familiarizing 14-month-old infants to a protagonist helping another agent by pushing them up a hill or hindering the agent by pushing them down a hill. On test, infants in both conditions saw the previously helpful/hindering protagonist distribute two strawberries to two recipients equally (1:1) or unequally (0:2). After seeing the protagonist help another agent, infants looked longer to unfair than fair behavior. However, after seeing the protagonist hinder another agent, infants looked equally to fair and unfair behavior. These findings suggest that infants suspend expectations for fair behavior after learning that an agent hindered another agent’s goals.
In the current study, we investigated 14- to 27-month-old infants’ tendency to generalize across the sub-domains of fairness and help/harm. We first sought to conduct a conceptual replication of Surian et al. (2018): after seeing an actor help or hinder another individual, infants saw that actor perform equal or unequal resource distributions and their looking was recorded. Our study differed from Surian et al. (2018) in three ways. First, we used real-world actors rather than animations; a demonstration that infants show similar patterns documented by Surian et al. (2018) with real-world actors would provide increased confidence that infants apply such generalization tendencies in everyday life. Second, we used different helping and hindering events, modeled after Hamlin and Wynn (2011) and Hamlin (2015), to provide increased confidence that the results generalize across multiple instances of helping/hindering. Third, and perhaps most critically, our distribution events featured 5:1 versus 3:3 distributions, rather than 2:0 versus 1:1 distributions given concerns that 2:0 distributions may conflate social exclusion with unfairness because the actor who receives no resources is not included in the exchange (see DesChamps et al., 2016). As in Surian et al. (2018), we predicted that infants shown helping will look longer to unfair versus fair events, but that these expectations would be suspended following hindering leading infants to look equally at the fair versus unfair events.
In Experiment 2, infants watched these events in reverse: they saw fair or unfair distributions followed by helping or hindering behavior. Thus, we investigated whether infants possess a tendency to generalize across fairness and help/harm in a bi-directional manner. We also sought to compare the relative strength of these generalizations based on the original sub-domain. On the one hand, one might predict that infants will generalize more strongly from hindering versus unfairness given that hindering is often see as more egregious than unfairness (Yucel et al., 2022), and that infants appear to be sensitive to help/harm before fairness/unfairness in their evaluations (Sommerville, 2022). On the other hand, we might expect the opposing pattern given that some work suggests that while infants have baseline expectations for fair distributions, they do not have baseline expectations for helpful behavior (but see Hamlin et al., 2007; Lucca et al., 2018).
Discussion
We sought to investigate whether infants use information from one sub-domain to form expectations of how an individual will act in another sub-domain, whether they do so in a bi-directional manner, and whether the strength of their tendency to do so varies according to the source sub-domain. In Experiment 1, we found that after infants were familiarized to an agent who helped another agent in obtaining her goal, infants looked longer to that individual perform unfair versus fair resource distributions. In contrast, after seeing an agent hinder another agent, infants suspend expectations for equal resource distributions for that agent. Together, these results provide a conceptual replication of Surian et al. (2018) and extend these results to events that involve human actors (rather than animated agents) and novel exemplars of helping/hindering behavior. Additionally, whereas Surian et al. (2018) used a 2:0 versus 1:1 distribution, our experiments featured a 5:1 versus 3:3 distribution. Distributions that feature 2:0 outcomes may conflate social exclusion (since one recipient is not included in the exchange at all) with unfairness; in our experiment, by ensuring that each recipient was involved in the exchange but one recipient received more resources than another, our experiments deconflated unfairness from social exclusion, and demonstrate that infants use hindering behavior to make inferences to guide their expectations about fairness/unfairness per se.
In Experiment 2 infants were familiarized with a fair or unfair distribution and then saw helping or hindering on test. On test trials, infants’ attention to the helping and hindering varied based on their familiarization type. Infants looked equally at helping or hindering actions performed following an agent’s fair distribution. In contrast, infants looked longer at helping than hindering after seeing an agent perform an unfair distribution. These results suggest that whereas prior fair behavior has no impact on infants’ helping and hindering expectations, unfair behavior leads infants to believe that the individual will subsequently hinder, rather than help, another agent achieve her goal.
Across our two experiments, we found differential effects of the initial event on infants’ subsequent reactions to a second morally valenced behavior. Specifically, in Experiment 1, witnessing helping (a positive moral behavior) led infants to expect subsequent fairness (versus unfairness) whereas witnessing hindering (a negative moral behavior) led infants to have no expectations for fair or unfair behavior. In Experiment 2, witnessing fairness (a positive moral behavior) led to no expectations for helping versus hindering, whereas witnessing unfairness (a negative moral behavior) led infants to expect subsequent hindering behavior (vs. helping behavior). This pattern of findings is consistent with two distinct possibilities.
The first possibility is that infants’ fairness/unfairness and help/hinder expectations are differentially affected by the valence of prior moral behavior; from this perspective, witnessing helping led infants to form subsequent prosocial expectations whereas as witnessing unfairness led to subsequent anti-social expectations. Another way to put this is that infants may see hindering as more influential than helping for forming specific subsequent expectations in the fair/unfair sub-domain, and, conversely, that infants may see unfairness as more influential than fairness in forming subsequent expectations for helping/hindering. Other work has revealed differential effects of positive and negative information on competence versus morality judgments, showing that positive information more strongly influences competence judgments and negative information more strongly influences moral judgments (Wojciszke et al., 1993; Trafimow, 2001). It is possible that such differential effects of initial positive versus negative information also exist between sub-domains of morality, particularly since many prior studies manipulate morality primarily through harm/care scenarios (i.e., Wojciszke et al., 1993).
A second possible interpretation of our findings is that the results reflect initial differences in baseline expectations across the two sub-domains: much work has found that infants have baseline expectations for fair over unfair behavior (Geraci and Surian, 2011; Schmidt and Sommerville, 2011; Sloane et al., 2012; Ziv and Sommerville, 2017; Sommerville and Enright, 2018), whereas no particular expectations for whether a given individual will be helpful or hindering (Hamlin et al., 2007; Fawcett and Liszkowski, 2012; Hamlin, 2013; Margoni and Surian, 2018; Tan and Hamlin, 2022). Although infants did not look significantly longer at the unfair distribution (versus fair) during familiarization in Experiment 1, as might be expected by prior work (Geraci and Surian, 2011; Schmidt and Sommerville, 2011; Sloane et al., 2012; Burns and Sommerville, 2014), infants’ attention declined to the fair distribution across familiarization trials but not the unfair familiarization (whereas this was not true for either helping or hindering behavior during familiarization trials; in both cases, infants’ attention declined significantly). Thus, it is possible that in the current experiment infants may show a weak baseline expectation for fair distributions which may be due to either the difference in the task structure (i.e., no preview of familiarization trial, recipients’ faces not visible to infant) or the fact that testing occurred via Zoom. From this perspective, across both experiments witnessing initial prosocial behavior (i.e., helping, fairness) has no impact on baseline expectations for subsequent prosocial or anti-social behavior, whereas initial antisocial behavior (unfairness, hindering) shifts baseline expectations. Critically, this perspective is in keeping with work from adults showing stronger effects of negative moral information on moral trait inferences than positive information (Wojciszke et al., 1993); however, our current findings do not allow us to distinguish these alternatives.
Regardless, both interpretations are consistent with the broader claim that in the second year of life infants are capable of generalization across moral sub-domains. But are there qualitatively distinct interpretations possible that do not necessarily involve moral reasoning on the infants’ part at all? One possibility is that infants may have construed the protagonist’s behavior in the familiarization trials in terms of whether it facilitates another agents’ goals or disrupts it; in other words, infants may represent these events solely in terms of the protagonists’ role in an inter-personal interaction. Agents that help and enact fair distributions act as goal facilitators (by helping another agent get their desired toy or get an optimal number of resources), whereas those that hinder and enact unfair distributions (by hindering another agent’s access to a desired toy or minimizing the number of resources obtained) disrupt others’ goals. It is possible that infants respond on test based on the role that the protagonist adopts as either a goal facilitator or goal disrupter, and whether it is consistent with their prior role (i.e., looking longer when a goal facilitator becomes a goal disruptor and vice versa). However, it seems less likely than other interpretations given the asymmetry we observed in our data; thus, we favor the broader conclusion that our data support the interpretation that infants generalize across moral sub-domains while recognizing the exact way that they do so requires further study.
While our findings suggest a bidirectional tendency to generalize from moral norm violations from help/harm to fairness and vice versa, a descriptive characterization of the effect sizes across studies indicates that infants generalize more strongly from hindering behavior to unfairness (d = 1.067) than they do from unfair behavior to hindering (d = 0.815). Coupled with the fact that infants show a sensitivity to help/harm prior to when they show a sensitivity to fairness/unfairness (Sommerville, 2022), and the fact that individuals tend to see hindering as more egregious than unfairness (Yucel et al., 2022), these findings raise the possibility that the degree to which infants will generalize from a given moral behavior may vary according to the severity of that behavior. Future work can directly test this idea.
There are several possible limitations of our work that bear consideration. One possible limitation is that we only used single examples of helping/hindering actions, and fair/unfair distributions. Future work should confirm whether our findings generalize more broadly to other exemplars of these behaviors. Another limitation of our work could be that our sample did not formally test infants younger than 14-months, and as such we cannot speak to the developmental origins (or lack thereof) of infants’ generalization. However, preliminary findings in our lab testing 12- and 13- month-old infants provide no evidence that infants of this age generalize across moral sub-domains. Assuming this developmental transition is replicated, one possibility is that the transition to generalization reflects a domain-general change in infants’ generalization abilities. Alternately, this transition may be better explained by a domain-specific shift in behavioral generalization that is spurred by infants’ increasing language comprehension and exposure to common moral trait terms (i.e., nice, mean) provided by parents and caregivers. Young children show evidence of using moral trait terms between 19- and 22-months (Bloom et al., 1975; Fenson et al., 1994), and by 30-months, they even begin to apply these terms to morally relevant contexts (Snow, 1987). Given that infants’ language comprehension frequently exceeds their production, future studies can empirically test the role of exposure to common trait terms in the face of both adherence to and transgressions of moral norms to determine their role in infants’ generalization across moral sub-domains.
In closing, here we demonstrate that infants possess an ability to generalize across moral sub-domains of fairness and help/harm. Given findings that young children show similar patterns of generalization across moral sub-domains (Gill et al., under review, see footnote 1), these results point to striking commonalities between older children and infants’ tendency to rapidly generalize from moral transgressions. These results open the door to a more fulsome investigation into infants’ tendency to engage in behavioral generalizations, and raise important directions for future work, including the age of emergence of this tendency, the scope of generalization, and the underlying mechanisms supporting these generalizations.