Introduction

Empathy, broadly defined as the sharing and understanding of others’ emotional states and thoughts, is a core component of human socio-emotional functioning [1,2]. Despite the important role empathy plays in shaping social interactions and responses to others’ emotional needs, understanding of its early development, especially during the first year of life, remains limited. Additionally, a persistent bias in the existing literature towards industrialised Western samples means we know relatively little about the extent to which empathy development is shaped by socio-cultural factors.

Empathy is broadly considered a multi-dimensional process involving both affective and cognitive components [3]. These include the affective sharing or experiencing of others’ emotional states, through evolutionarily ancient processes such as emotional contagion and mimicry [2]. The cognitive components of empathy relate to understanding others’ states, via perspective-taking and self-other differentiation. There remains active debate as to when and how such processes emerge and the extent to which the cognitive and affective components are dissociable, or whether they represent an integrated system [4–6].

Studies with infants during the first year of life primarily focus on two main behavioural markers of empathy: facial expressions of concern and comforting behaviour. Although mapping external behaviours onto underlying mechanisms is challenging, it seems plausible that facial expressions of concern may reflect emotional contagion or a deeper understanding of others’ emotional needs. In contrast, comforting seems to be a clearer indicator of understanding emotional needs in others and has been commonly used as a marker of empathy in both children and adults. Facial expressions of concern following simulated distress by an adult have been observed in 8-month-olds from Israel and the US [7,8]. Although promising, a lack of control condition in these studies makes it difficult to rule out whether such expressions might reflect concern about lack of attention and normal responsivity in the adult or the infant’s own distress rather than concern for the adult. In a more recent study [9], the inclusion of a control condition of a neutral adult not paying attention to the child showed that even by 3 months, infants showed more signs of facial concern in response to a distressed adult than a neutral adult (albeit weakly and briefly at this age). Facial concern was found to increase with age but with steepness gradually decreasing over time [7–9]. Supporting the integrated view of empathy, the early trajectory of facial expressions of concern appear to predict other aspects of empathic development, including comforting behaviour [9] but also more broadly to social competence [10].

In contrast to facial expressions of concern, comforting behaviours (such as hugging the distressed individual) appear later in development. For instance, comforting has not been detected at 3 and 6 months in Israeli infants [9]. It then appears to become more likely over the second year of life (e.g., from 14% of infants at 12 months to 41% of infants at 18 months, [9]). However, another study [11] found no evidence of comforting behaviours in 18- and 24-month-old Canadian infants who witnessed a 10-second-long distress simulation by an experimenter, though this could have been due to the short duration of the simulation. Overall, the evidence suggests that across the second year of life, comforting behaviours are observable but can still be rare and sensitive to context, whereby children may need sufficient time to both process and respond to observed distress. It is also possible that such behaviours could potentially represent self-comforting rather than other-oriented responding [7] and control conditions can be important in ruling out comforting behaviour being driven by a desire to simply elicit a response from an unresponsive adult [9,11].

Another key issue hampering progress in our understanding of empathy development is a persistent sampling bias towards a limited number of cultural settings, particularly high-income, educated families in industrialised settings, particularly in Western Europe and North America. This sampling bias, which is persistent across psychology, including developmental research [12,13], ignores the rich and important socio-cultural variability that shapes the social nature of our species. Only by addressing it and including more culturally diverse samples can the foundations of complex capacities, such as empathy, be fully understood.

The importance of including more diverse samples becomes apparent when considering the handful of studies on empathy development that have already been conducted in more than one sociocultural setting. These have found evidence for cross-cultural variation in empathy, albeit mostly beyond infancy: For example, 6- to 7-year-old US-American children matched the emotional state of characters in affective stories more closely than did children from Greece [14]. 7- to 12-year-olds from Brazil scored higher on an empathy questionnaire than 6- to 13-year-olds from the US [15], suggestive of possible cross-cultural variation in the expression of empathy during middle and late childhood. When might such variation start to emerge? In a study on Malaysian, Indonesian, German, and Israeli 5-year-olds [16], all showed similar intensity of empathic concern when seeing an adult in distress over a broken toy. However, Malaysian and Indonesian children exhibited more self-distress and fewer empathic behaviours than German and Israeli children, suggesting that some cross-cultural differences in empathy may already be present in earlier childhood. A similar study has been conducted with infants, in which 19-month-olds from urban Germany and urban India witnessed an adult being upset over a broken toy [17]. Children from both contexts displayed self-distress and other-oriented empathic behaviours, such as comforting or engaging another adult’s help, without significant differences between the two sites, despite variation in maternal prosocial and socialisation goals. It should be noted, though, that this study [17] measured empathic concern and comforting behaviours in one combined score. It is therefore unclear whether early cross-cultural differences might exist for one marker of empathy but not the other. Taken together, these previous cross-cultural studies highlight the potential for important variation in the development and expression of empathy in early childhood, yet many questions remain. Importantly, only a single cross-cultural study [17] has examined empathy in infancy, so the age at which empathy development starts to show sensitivity to the cultural environment needs considerably more investigation. Most previously studied samples were (as far as that information is available) from urban, industrialised, middle-class settings, providing some homogeneity. To our knowledge, no such study has been conducted in a rural subsistence context, leaving a significant gap in our knowledge of variation in the development of empathy. Given that empathy is considered a key facet of socio-emotional development, an increased effort to understand it in more diverse contexts is needed if we want to fully understand its developmental origins and trajectory.

There are several reasons why cultural variation may be seen in early empathic behaviours. Variation in socialisation goals and practises of parents and other caregivers may affect the development of empathic behaviours in a number of ways. It has been proposed that observed differences in socialisation practices arise from differences in the cultural models of the broader community [18,19]. For instance, individuals from relational sociocultural contexts have been found to put more value on interpersonal relatedness, responsiveness, and obedience than individuals from autonomous communities, who tend to put more emphasis on individuality and independence (e.g., [17,20]). It is plausible that infants of caregivers with more relational socialisation goals attend more to the needs of others. It is also possible that variation in socialisation of emotional expression and regulation could affect the development of empathetic behaviours. Emotion regulation is considered important for expressing other-oriented empathic responses [21,22], and cross-cultural variation in the socialisation of emotion regulation has been found. For instance, comparing Germany and Cameroon, German mothers prioritised having emotionally expressive children, whereas Cameroonian Nso mothers placed more value on their children being calm and emotionally inexpressive [23].

Given the need for more diverse infancy research, the aim of the present study was to systematically investigate the early development of empathy in two distinct socio-cultural settings – participants sampled from Uganda and the UK. These samples varied in potentially relevant ways, as outlined below, however it was beyond the scope of this study to examine whether individual variation in these factors was predictive of infant empathetic behaviour: this should be a focus of future research. The Ugandan site consisted of multiple villages set on the edge of a large rainforest. Most people in the area rely on subsistence farming, as opportunities for and engagement in wage labour is low. Families often live in multi-generational households and distributed caregiving, including the presence of child caregivers, is common [24]. This site in Uganda poses an interesting context to study socio-emotional development, as ethnographic research in Uganda and other rural African contexts has documented the concealment of emotions and generally low levels of affect and emotional engagement in mother-infant relationships [25–27]. Additionally, Ugandan mothers from this research site have been shown to value relational over autonomous goals more than mothers from the UK site [24] – two cultural values which have been shown to be linked to differences in emotion socialisation [23]. Socio-cultural differences between our sites in Uganda and the UK with respect to an infant’s learning environment, caregiving arrangements, and maternal socialisation goals and strategies, allow for an investigation into the previously established trajectories for empathy development outside of and within traditionally studied populations.

Building on previous work (e.g., [7,9,11]), in this study, we conducted a comforting experiment when infants were 9 and 18 months old. In the experiment the infants observed an adult model (mother or adult experimenter) lightly injuring themselves. In the experimental condition, the model expressed distress after the pretend injury and did not respond to communication attempts by the child. Following Dunfield et al.’s protocols [11], in the control condition, the model did not express any signs of distress following the injury but remained emotionally neutral and unresponsive to the child. Although other previous studies have implemented other types of control conditions (e.g., [7,9]), we opted to include this control condition for two reasons: first, to test whether the child acted in response to the adult’s distress signals that express a need for comforting and not just the injury itself (as 18-month-olds have been shown to respond empathically after an injury even in the absence of distressed signalling [28]); and second, to rule out that comforting during the experimental condition was not just an attempt to elicit a social response from an unresponsive adult. To measure behavioural markers of empathy, we coded the frequency and duration of two behavioural components in the infants: facial expressions of concern and comforting behaviours. Many previous studies (e.g., [7,9]) have also included ‘hypothesis testing’, i.e., inquiry behaviours thought to represent attempts to understand the distress. A key behavioural marker of hypothesis testing is child gaze between the adult’s hurt body part and face or another adult. As our study included the distressed adult gaze alternating between their hurt body part and the child, this may have encouraged gaze alternation in the child, making interpretation of this behavioural marker very difficult. Therefore, we only present the data relevant to hypothesis testing in the supplementary materials, as we are not confident that they represent a valid measure of hypothesis testing within our paradigm.

We predicted that by 9 months of age, infants across both sites would display concerned facial expressions in response to an adult in distress but would not yet show comforting behaviour [7,9]. We expected that both concerned facial expressions and comforting would be present at 18 months, and that a greater proportion of children would show these behaviours at 18 months compared to 9 months [7,9,29]. Since findings on the predictiveness of early empathic behaviours from one timepoint to the next are mixed (e.g., [7,9]) we took an explorative approach to assessing if individual variation in concerned facial expression at 9 months would predict comforting at 18 months. Moreover, as comforting, but not concerned facial affect, has previously been found to be socially biased towards familiar individuals [29], we expected to see more comforting behaviour (but not concerned facial expressions) towards the mother than the experimenter in both socio-cultural contexts.

Regarding potential cultural variation, we took an exploratory approach, given limited previous research in rural Uganda and at this site in particular. We identified the following possible drivers for population-level differences in early empathetic behaviour: Sociocultural contexts emphasising social connectedness (i.e., relational models, as documented at this site in Uganda; [24]) may foster earlier empathic behaviours but may also hinder the development of empathic behaviours if repression of negative emotions is valued and thus fewer opportunities to respond to those in distress are provided [23]. Additionally, the higher value placed on emotional expressivity in autonomous contexts, such as the UK, may encourage more attention to and thus an increased tendency to respond empathically to others’ emotional states. These competing influences could plausibly make infants either more or less likely to show facial expressions of concern and comforting.

Methods

Statistical analyses

A total of 465 valid control and experimental trials across the two sites and timepoints were available for analysis. Whilst some infants had valid trials at 9 months with their mother and valid trials with both the mother and experimenter at 18 months, other infants did not have the full set of trials due to missing trials, or trials being excluded (see supplementary materials S1 in S1 File for details). All measures were extracted for each valid trial.

We first tested the effect of condition and if this was influenced by age or site to ensure our experimental condition had an effect on infant behaviour. Then, for the experimental condition only, we examined (i) the effects of age and site, (ii) the effect of familiarity with the distressed adult and site on the proportion of the trial where the infant displayed concerned facial expressions, and on the likelihood of the infant comforting; and (iii) whether showing a concerned facial expression at 9 months predicted comforting at 18 months (because comforting at 9 months was extremely rare, we did not expect to find a predictive relation between comforting at the two timepoints).

All analyses were conducted in R [34] using the package lme4 [35] for GLMMs. For each statistical model, we compared the full model, which included an interaction term, to a null model lacking the fixed effects and interaction terms of interest but retaining control fixed effects and random effects using a likelihood-ratio test [36]. If this was significant, we compared the full to a reduced model, where the interaction term was removed and constituent factors were entered as main effects, also using a likelihood-ratio test. The drop1 function was used to extract p-values for individual variables. See Table 1 below for model structures. See supplementary materials (S4 in S1 File) for model assumptions and stability assessment procedures. To test the robustness of our findings we also ran Bayesian equivalents of all generalised linear mixed models (see supplementary material S9 in S1 File for details).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Overview of model structures for full model GLMMs and GLMs.

https://doi.org/10.1371/journal.pone.0320371.t001

We also examined the spontaneity of the comforting response and the influence of explicit distress cues at 18 months. As comforting behaviour remained rare, even at 18 months, we maximised our sample by including data from both mother and experimenter trials into one analysis. Since there was no case where an infant at 18 months comforted both their mother and the experimenter, pseudo-replication was not an issue. We used a chi-squared test to compare the distribution of infants who comforted in Uganda vs. the UK before the mother or experimenter explicitly labelled their distress vs. afterwards.

Results

Table 2 shows descriptive statistics for concerned facial expression and comforting, by age, site, and identity of the distressed model, in the experimental and control conditions. See supplementary materials for a more detailed overview of specific comforting behaviours (S5 in S1 File), results from hypothesis testing (S6 in S1 File), and full model estimate tables (S7a–g in S1 File). The supplementary materials also include Bayesian equivalents of all generalised linear mixed models (S9 in S1 File), the results of which mirror the pattern of the frequentist models presented below.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Descriptive statistics for concerned facial expressions and comforting, across conditions, sites, and timepoints.

https://doi.org/10.1371/journal.pone.0320371.t002

Age and site effects in the experimental condition

Concerned facial expression.

The full-null model comparison did not reach significance (χ²(3) = 3.89, p = .274), indicating that Ugandan and UK infants spent similar proportions of the trials displaying concerned facial expression, with no change from 9 to 18 months. For full model estimates, standard errors, and confidence intervals, see in S7c Table in S1 File.

Comforting.

The full-null model comparison was significant (χ²(3) = 10.83, p = .013) but the reduced model comparison was not (χ²(1) = 0.01, p = .909). Infants’ likelihood to comfort increased significantly with age (estimate±SE = 1.21 ± 0.43, χ²(1) = 8.70, p = .003, OR = 3.34; see Fig 1) but there was no significant effect of site (estimate±SE = -0.59 ± 0.41, χ²(1) = 2.12, p = .145, OR = 0.55). For full model estimates, standard errors, and confidence intervals, see S7d Table in S1 File.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Comforting of mother by infant age across sites.

Comforting of mother by age (9, 18 months) across sites (Uganda, UK) (Fig 1). Collapsed across sites given there were no detected site differences. Horizontal lines represent model estimates and whiskers represent 95% confidence intervals around model estimates.

https://doi.org/10.1371/journal.pone.0320371.g001

Familiarity and site effects in the experimental condition

Concerned facial expression.

The full-null model comparison did not reach significance (χ²(3) = 1.43, p = .699), indicating that infants at both sites showed similar amounts of concerned facial expression towards their mothers and the experimenters. For full model estimates, standard errors, and confidence intervals, see S7e Table in S1 File.

Comforting.

The full-null model comparison was significant (χ²(3) = 13.11, p = .004) but the reduced model comparison was not (χ²(1) = 0.00, p = .981). Infants’ likelihood to comfort was significantly higher for mothers than for experimenters (estimate±SE = 1.50 ± 0.48, χ²(1) = 11.44, p < .001, OR = 4.47; see Fig 2) but there was no effect of site (estimate±SE = -0.55 ± 0.43, χ²(1) = 1.63, p = .202). For full model estimates, standard errors, and confidence intervals, see S7f Table in S1 File.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Infant comforting at 18 months by model familiarity across sites.

Comforting by model familiarity (mother, experimenter) across sites (Uganda, UK) at 18 months (Fig 2). Collapsed across sites given there were no detected site differences. Horizontal lines represent model estimates and whiskers represent 95% confidence intervals around model estimates.

https://doi.org/10.1371/journal.pone.0320371.g002

Effect of verbal cues

The distribution of infants who comforted before and after the explicit verbal cue of injury was given varied significantly with site (χ²(1) = 5.99, p = .014, OR = 7.33). Of the children who comforted either their mother or an experimenter at 18 months, most children from the UK site (77%) did so before the model explicitly labelled their distress, whereas the majority of children from the Ugandan site (69%) comforted after this verbal cue (see Fig 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Percentage of infants who comforted before the explicit verbal cue at 18 months by population.

Percentage of infants who comforted before the verbal cure, out of those infants who comforted at 18 months (N = 30), not the total sample.

https://doi.org/10.1371/journal.pone.0320371.g003

Discussion

Until recently, the early development of empathy had been understudied and as a result, theoretical frameworks may have underestimated its onset during the first year of life [37,38]. However, previous studies [7–9] have indicated that infants during the first year of life show two markers of empathy: expressions of concern and hypothesis testing, with some evidence also of comforting. Here, we expand on these findings with evidence across two culturally different settings to provide evidence of not only empathic concern but also comforting at 9 months. Although still rare in our cross-cultural sample, the detection of comforting behaviour at 9 months at both sites which is considered a reliable marker of empathy, is particularly notable, and supports the occurrence of comforting in the first year of life.

A key goal of our study was to provide a richer understanding of early empathy development that goes beyond a Western-centric model that has persisted in developmental science, but also in psychology more generally [39]. Despite differences in socio-cultural contexts and maternal socialisation attitudes between the two sites we sampled [27], we found similarities in infants’ empathic behaviours. This apparent continuity complements previous work [17] which also found no differences in empathic behaviour in 19-month-old infants from Germany and India. While such findings cannot demonstrate universal patterns in empathy development, they do show that these behaviours show similarities in sites that differ in child-rearing practises and caregiver attitudes.

In line with previous studies [7,9], the number of infants who engaged in comforting increased from 9 to 18 months. In our sample, approximately a third of infants at 18 months showed at least one comforting behaviour in response to a distressed model, with most infants expressing this through patting or hugging the distressed adult. This supports the notion that comforting becomes more common and robust in the second year of life.

In contrast to comforting, the production of facial expressions of concern was stable across both age points at both sites in our study. Previous studies have revealed a mixed picture of variation in facial expressions of concern in early development: Some [7] found no increase between the first and second year of life, whereas others [9] showed modest growth slopes during the first year of life which became weaker over the course of the second year. Yet others, [29] found facial concern increased from 14 to 36 months in a US sample, with the biggest change occurring from 14 to 20 months, indicating an increase over time which slows down with age. While concerned facial expressions in these previous studies were measured with a holistic rating of duration and intensity of facial concern, we decided to base our ratings on relative duration of concerned facial expression only. We decided against combining duration with intensity because facial expressions of emotions can vary with cultural norms around expressiveness [32], which may also affect the intensity with which they are expressed. Thus, whilst the relative duration of concerned facial expressions was stable from 9 to 18 months in our study, the intensity may have varied across timepoints or sites.

Since our procedure was based on prior work [11] which deviates slightly from the comforting paradigm used by other studies (e.g., [7,9,33]), our study took a slightly different approach to the adult’s distress simulation (including gaze exchanges with the infant). We focussed on the occurrence of at least one comforting event and as explained above, opted for duration without the inclusion of intensity when coding concerned facial affect. Despite these methodological differences, our results are in line with previous research (e.g., [7,9,29]) conducted in the US and Israel. This supports the apparent robustness of the presence of concerned facial expressions and comforting in early development, despite methodological adaptations and cross-cultural variability in caregiving practices.

Although our main analysis focused on the experimental condition, we also statistically compared the control and experimental conditions. We found that infants were significantly more likely to comfort the model in the experimental compared to the control condition, and that in the experimental condition they also displayed concerned facial affect for a greater proportion of the trial. The descriptive data (Table 1) show that very few infants engaged in these behaviours in the control condition. These findings indicate that it is highly unlikely that comforting behaviours and concerned facial affect in the experimental condition were a response to encountering an unresponsive/ distracted adult. Although it has been shown that 18-month-olds show concern and comforting to an adult who has injured themself but does not act as if in pain [28], we found little evidence that infants in our sample behaved in this way. This suggests that while infants may be able to infer emotional needs after seeing causes for distress [28], they are also sensitive to communicative cues about emotional needs, such as vocalisations of pain and visual attention and rubbing gestures on the injured body part and respond empathically to such cues.

By using a longitudinal approach, we could assess whether early empathy predicted empathic responding during the second year of life. In our study, concerned facial expressions at 9 months did not predict comforting at 18 months. These results contribute to a mixed picture of how empathy develops: On one hand, prior work has found that concerned facial expressions at 8 months predicted empathic comforting at 12 months but not at 14 months [7]. On the other hand, others have found that concerned facial expressions at 6 months (but not at 3 or 12 months) predicted comforting at 18 months [9], or that different developmental trajectories of early facial expressions of concern predicted social competence at 36 months [10]. While these studies used intensity ratings of concern and comforting, we used a duration-based approach for facial expressions of concern. Although this accounted for potential variation in cultural norms for expressing emotions such as concern, it may have contributed to the lack of a predictive relationship between the two. More research is needed to understand how different behavioural markers of empathetic behaviour relate to each other and how stable individual variation in early empathetic behaviours is over development, accounting for socio-cultural influences on the expression of both concern and comforting.

In line with previous studies (e.g., [29]), our findings on comforting at 18 months support the social bias of empathy, whereby empathy is evolutionarily biased towards kin and other individuals perceived as similar or familiar [2]. In our study, infants were more likely to comfort their mother than an experimenter, regardless of population, however facial expressions of concern were comparable regardless of the identity of the adult model, mirroring previous studies. It is possible that reduced comforting towards an unfamiliar adult may also be explained by other factors, such as being more reluctant to approach or interact with them. Future research examining underlying psycho-physiological mechanisms is needed to address this question further.

Although we found many similarities in the overall occurrence of empathy behaviours, our study revealed some interesting site level variation. Compared to Ugandan infants, the UK infants in our sample were more likely to comfort before a verbal cue to the injury was provided (“Look, I hurt myself!”). Although this result is based on a modest number of infants who comforted at 18 months (N = 30), it suggests variation in the occurrence of spontaneous comforting. Generally, empathy is conceptualised as being spontaneous [40], occurring because of the observer sharing a target’s emotional state, without the need for the emotional state to be verbally labelled. However, it may be that verbal cues to distress are used differently across socio-cultural settings and expectations of how others respond to those cues are culturally shaped. Early anthropological research with Ugandan agriculturalists described a tendency toward more regulated expressions of emotions and fewer affective exchanges between caregivers and infants [26,27]. This observation overlaps with findings from Holden et al. [24] that Ugandan mothers were more likely to prioritise relational attitudes (emphasising compliance and obedience) than UK mothers from these sites. These socialisation practices may contribute to the Ugandan infants responding more readily to explicit labels of emotions rather than engaging in spontaneous comforting. However, as we did not directly assess specific caregiving practices and their link with infant comforting behaviours, this possibility remains speculative. Future research should address this limitation by investigating local cultural norms around affective expression and socialisation practices, such as frequency of emotional exchanges or scaffolding of empathic behaviours, at each study site. Assessing these aspects alongside infants’ comforting behaviours could clarify how different cultural approaches to expressing and responding to emotions shape infants’ empathic development.

An additional future direction would be the inclusion of measures of parental empathy at the individual level, going beyond cultural generalisations: In samples from North America, Australia, China, and the Netherlands, parents’ empathic tendencies and abilities have been found to positively correlate with those of their offspring at various stages of development (e.g., [41–44]) and that parental empathy is itself affected by the socio-cultural environment of the parent. In this study, we did not assess individual mothers’ empathy, in part due to the challenges of assessing parental attitudes using self-report in a low-literacy environment. However, future research examining the influence of both group-level and individual-level factors on empathy development would aid a more in-depth understanding of socialisation influences on empathy.

Taken together, our findings provide support for the presence of empathic concern and comforting in infancy. However, using behavioural markers of such a complex process as empathy is not without challenges. Concerned facial expressions can be subtle and may be similar to other forms of negative affect. Although we addressed this by excluding co-occurrence with signs of distress, such as crying or whimpering, an important future direction would be to validate concerned facial expressions coding using systematic methods, like the Facial Action Coding System (FACs) [45]. Similarly, comforting behaviour may have alternative explanations, such as seeking reassuring contact in response to someone else’s distress. Although visible signs of infant distress, which may have led participants to seek out physical touch with their mother, were as rare in our study as in others (e.g., [7,9]), it would be beneficial for future research to supplement behavioural measures with physiological measures, like eye-tracking. Eye tracking has been successfully used in the realm of instrumental helping to capture understanding of others’ needs or intrinsic motivation to help (e.g., [46,47]), and a similar approach could be taken to comforting.

The extent of alloparental care represents a striking difference between our samples from Uganda and the UK: Most infants in our UK sample were primarily cared for by their mother (especially during the first year while on maternity leave) and their father. In our Ugandan sample, caregiving was distributed between a larger number of people, including other members of the household, neighbours, and siblings [24]. In this study, we focused on mothers to aid comparability with previous research [e.g., 7–9] however, future work should also consider the role of other caregivers, including fathers and where ecologically relevant, non-parental caregivers.

A broader constraint of our study was the use of an empathy paradigm originally designed for Western urbanised/industrialised settings, and the experimental situation was likely more novel to the Ugandan participants. We attempted to mitigate this novelty and ensure that the paradigm was culturally appropriate by piloting, co-producing, and conducting it with local Ugandan researchers. By the first timepoint of this study, all mother-infant dyads at both sites had already participated in two previous timepoints as part of the broader longitudinal project, thus reducing the novelty of the situation for all participants. If, despite our efforts, differences in the ecological validity of the task remained across our two study sites, then it is possible that the current study underestimated the likelihood of Ugandan infants to comfort or show concerned facial affect, but we can be confident that early empathic behaviours in our Ugandan sample were at least as common as in our UK sample.

In conclusion, by taking a rigorous longitudinal approach in two different cultural contexts, our study supports the view that multiple behavioural markers of empathy start to emerge in the first year of life. Infants from both sites showed similar patterns of empathy-related behaviour, being more likely to comfort their mother compared to an unfamiliar experimenter and more likely to engage in comforting with age. The only key socio-cultural difference that emerged was that the UK infants were more likely than Ugandan infants to offer spontaneous comfort before the model gave an explicit verbal cue. Overall, our results highlight the early emergence of behavioural markers of empathy at two sites with differences in socio-cultural settings.

Supporting information

Acknowledgments

We are very grateful to the families and communities in Uganda and the UK that have participated in this research. We thank the directors and team at Budongo Conservation Field Station for supporting our research in Uganda, as well as Prof Robert Newton for his support and advice. We thank Dr Kirsty Graham for their significant contribution to the conceptualisation, methodology, investigation and writing of this study. In Uganda, we are grateful to research assistants Emma Sopelsa-Hall, John McCutcheon, Charlie Ives and the late Hellen Biroch who assisted with data collection. Thank you to Businge Ambrose, Anguzu Julius, Masereka Brenda, Monday Mbotella and Mua Simon for translation of materials. In the UK, we are grateful to research assistants Rebecca Anderson, Molly Bowns, Lucy Dunn, Megan Earl, Keshy Emmanuel, Harold Green, Charlie Ives, Harshanaa Patel, Rhiannon Pearce, Maddy Peryer, Barbora Sodomkova, Emma Standley, Sonnie Tan, Maisie Thurman, Joseph Vogliqi, Daisy Whitwood, and Caitlin Woods, who assisted with data collection.