Why do we pick similar mates, or do we?

Abstract

Humans often mate with those resembling themselves, a phenomenon described as positive assortative mating (PAM). The causes of this attract broad interest, but there is little agreement on the topic. This may be because empirical studies and reviews sometimes focus on just a few explanations, often based on disciplinary conventions. This review presents an interdisciplinary conceptual framework on the causes of PAM in humans, drawing on human and non-human biology, the social sciences, and the humanities. Viewing causality holistically, we first discuss the proximate causes (i.e. the ‘how’) of PAM, considering three mechanisms: stratification, convergence and mate choice. We also outline methods to control for confounders when studying mate choice. We then discuss ultimate explanations (i.e. ‘the why’) for PAM, including adaptive and non-adaptive processes. We conclude by suggesting a focus on interdisciplinarity in future research.

1. Introduction

Humans often mate with those resembling themselves [1], a phenomenon described as positive assortative mating (PAM). Observed since the 1800s [2], PAM has become an important topic of interdisciplinary research [3–7] and is now reported for diverse physical, cognitive, behavioural and sociocultural traits [1,5,6,8,9]. The rarer phenomenon of negative assortative mating (NAM), which describes mating among dissimilar individuals [10], has also received attention [11,12].

Here, we address PAM's causes. As the term's meaning can be ambiguous [13], we first define PAM as the act of mating (i.e. a process or behaviour [14]) between similar individuals more often than expected by chance in a given population [13]. At the population level, this may manifest as a phenotypic or genetic association between mates (i.e. a pattern) [15–19]), measured for either continuous or discrete traits (see [14,20] on measurement). It is important to note that by defining PAM as a process, we make no assumptions about the existence of specific cognitive or behavioural biases that may precede mating (e.g. mate choice).

PAM can be explained at two causal levels [21,22] (figure 1). Proximate explanations concern how the phenomenon arises [22], describing the genetic, physiological, cognitive and behavioural bases of its expression, and the environmental factors influencing these (figure 2). For example, a population may be segregated into phenotypically (and genetically) differentiated clusters [4,7,9,23], creating mating pools of similar individuals. We call this ‘stratification’ [4,24–26], but elsewhere it is referred to as ‘social homogamy’ [27–30]), ‘propinquity’ [7,9] or ‘segregation’ [31]. PAM may also reflect mate choice based on preferences [1,32], either for similarity itself [1,5,33–35] or other traits/characteristics that produce PAM indirectly. For example, a widespread, directional preference for particular traits (irrespective of similarity) may create a competitive ‘mating market’ [36] in which individuals with high ‘mate value’ pair up and PAM occurs indirectly for traits defining value [14,37–44]. PAM can also reflect post-mating ‘convergence’, whereby shared environmental exposures or imitation cause mates to become more similar over time [45–47].

Figure 1.

An ultimate-proximate causal framework for PAM. Primary (i.e. mate choice, stratification, and convergence) and subordinate proximate mechanisms are shown in orange boxes, with causal pathways indicated by black arrows. Adaptive ultimate processes (i.e. those relating to the fitness effects of PAM itself) are shown in light blue boxes, while non-adaptive ultimate processes (i.e. those relating to fitness effects unrelated to PAM) are shown in dark blue boxes. Selective causal pathways linking fitness effects to proximate mechanisms are also indicated by black arrows. This figure is not comprehensive, and some processes and causal pathways are excluded because they fall outside the scope of this review. For example, cultural phenomena may influence many proximate mechanisms, but they are displayed here only in the context of stratification. Cultural behaviour may also be under selection, which is not considered here.

Figure 2.

How proximate mechanisms drive PAM. The left-hand boxes show geographically separated ancestral populations (blue and red) containing economically differentiated subpopulations (small and large circles). Blues (mainly low-economic status individuals) migrate, forming a stratified admixed population, where subpopulations are differentiated by both ancestry and economic status. Below, mating takes place, mainly within (sub)populations. PAM occurs by economic status and, in the admixed population, this covaries with genetic ancestry. Where subpopulations overlap, ancestrally and economically mismatched pairs form. Generations later, admixed individuals appear. In the middle boxes, individuals choose similar or high-value mates. In the former case, phenotypically similar individuals (squares or triangles) form pairs over time. In the latter case, high-value (1) individuals pick each other, leaving low-value (2) individuals unmated. Later, low-value individuals mate in the absence of other options. In the right-hand boxes, initial mating results in both matched and mismatched pairs. Over time, the mismatched pairs converge, increasing levels of PAM.

By contrast, ultimate explanations account for why PAM occurs in terms of the evolutionary forces shaping trait values/distributions, especially selection [22]. For example, if similarity is beneficial, it may increase reproductive success [48–51] and result in positive selection on traits proximately mediating PAM [16], such as mate choice [5,35,52] or dispersal [51]. We describe this as an ‘adaptive’ explanation. Alternatively, if traits mediating PAM come under positive selection for reasons unrelated to similarity, this may result in a level of PAM that is not necessarily beneficial in its own right. We describe this as a ‘non-adaptive’ explanation. In both cases, ultimate explanations require complementary proximate mechanisms describing how PAM arises in actual populations (i.e. the why and the how) [22].

Because PAM may reflect numerous causes, all explanations at both causal levels should be considered. However, many studies focus on a handful of phenomena, often at one causal level. This results in incomplete perspectives and, arguably, false causal inferences [26,53]. For example, a review [29] dismissed stratification as largely irrelevant but excluded biological studies suggesting the contrary (e.g. [25,53–55]).

Here, we present an interdisciplinary conceptual framework on the causes of PAM. We first consider proximate mechanisms and how they interact, including ways to control for confounders when studying mate choice. We then discuss ultimate hypotheses for PAM, covering adaptive and non-adaptive explanations.

2. Proximate explanations

(a) . Stratification: the effect of non-random encounters

Stratification describes a situation in which genetically or phenotypically differentiated members of a given population are heterogeneously distributed across groups or clines [9,56,57] (figures 1 and 2). This increases the likelihood of similar individuals interacting (i.e. non-random encounters [14,43]) and therefore mating (i.e. PAM), even in the absence of preferences [58]. In non-human species, this may drive reproductive isolation and eventual speciation [59,60]. Although we do not discuss it here, stratification may also be a consequence (as well as a cause) of PAM if assortment reinforces or generates population subdivision [4,61,62]. Importantly, the effects of stratification on PAM are scale-dependent, meaning that PAM may occur even if mating is random or disassortative at the local scale (e.g. within groups) [23]. Therefore, when stratified populations are examined at a coarse scale, inferences of mate choice should be made cautiously. Nonetheless, at the scale of the whole population (i.e. the given population of interest comprising heterogeneous groups), PAM will be greater in a stratified population than a well-mixed one, all other things being equal.

Human stratification involves spatial clustering across discrete units (e.g. neighbourhoods, cities and countries) and continuous clines (e.g. a north–south gradient) [57,63–65], sometimes with a temporal dimension (i.e. differentiated groups may occupy the same area at different times) [66–69]. At local scales (e.g. individuals sharing a neighbourhood), stratification by discrete (e.g. language groups across which there is no diffusion) or continuous (e.g. a ‘left–right’ political spectrum) cultural forces is also common [9,24,25,65,70], as is segregation based on behavioural traits (e.g. friendship groups) [26,71–74]. Despite the continuous nature of much human stratification [75], we only consider discrete ‘subpopulations’ here (figure 2) for simplicity.

Patterns of human stratification vary depending on the size and demographic complexity of study populations [76], which range from relatively isolated small-scale societies [77–79] or larger national populations [80] to vast continental melting pots [18,65]. In many populations, stratification often covaries with genetic ancestry [18,25,53,65] (figure 2). Unlike its categorical proxies (e.g. ethnicity), genetic ancestry is a continuous trait describing the proportion of an individual's/population's genome shared with a recent common ancestor in ancient or contemporary reference populations, which range from continental (e.g. African, European) to finer-scale regional clusters [81,82]. These reference populations have identifiable genetic profiles due to differentiated processes of genetic drift and/or natural selection coinciding with limited gene flow [83]. These genetic differences often result in extensive phenotypic variation, which is frequently accompanied by sociocultural divergence [56,84]. Consequently, a covariance between stratification and genetic ancestry (e.g. a heterogeneous distribution of ancestry groups across neighbourhoods) may lead to complex, unpredictable patterns of PAM. Differences between continental groups range from the conspicuous (e.g. skin pigmentation, facial morphology and hair colour [84–86]) to the obscure (e.g. susceptibility to baldness [65] and health outcomes [87]), while at finer scales (e.g. within or between countries), subtle variation abounds [18,88]. For example, there are height and facial morphology differences between the British and Irish [89] and the British and Dutch [90], respectively.

Ancestry-related stratification is complicated by admixture (figure 2), which is when (typically limited) mating occurs across permeable barriers dividing subpopulations with distinct genetic ancestry profiles [91]. This produces admixed individuals with different proportions of ancestry from several source populations, who may have proportionately eclectic phenotypic and sociocultural profiles [89,92,93]. These individuals may form distinct population clusters of their own [18,65,94] and, over generations, populations can become suffused with cryptic layers of stratification via this process. Admixture is usually intertwined with distant historical phenomena, such as empires [95], trade [96] and invasions [97]. In the UK, for example, Viking and Anglo-Saxon invasions produced concentrations of Scandinavian and Germanic ancestry that are still based loosely around ancient geopolitical boundaries [97–99]. The consequences of admixture for PAM, however, remain poorly understood.

Although many questions remain about how ancestry-related stratification affects PAM [76], controlling for genetic ancestry (box 1) can offer insights into its contribution. Ancestry corrections reduce mate correlations for morphological [119], sociocultural [100] and physiological [120] traits in several populations, although sometimes only minimally [3,101]. Genome-wide similarity in mates [121] and friends [74] also disappears after ancestry corrections [26,53] (but see [122]). Because genetic ancestry is fractional, controls based on categorical proxies (e.g. race, ethnicity) [6,7,9,117,123] may be less effective than a continuous, objective measurement from genetic data (see [82] for discussion). This is especially true when populations are intricate ancestral mosaics [4,65,97,98] (box 1 below).

Box 1. Methods to identify similarity preferences.

Many studies aim to identify the extent to which similarity preference drives PAM. This requires methods to control for the effects of stratification, competition (i.e. mate value preference) and convergence.

Stratification

Controlling for the effects of stratification usually involves modelling predefined spatial, sociocultural or phenotypic variables (i.e. stratifiers) as covariates [7,9,100,101]. However, principal components (PCs) or other algorithmically identified clusters representing latent genetic (or phenotypic) structure can be used to identify hidden stratification [54,89], such as that linked to cryptic variation in genetic ancestry [24,97,98]. Both approaches pose challenges. For example, predefined controls rely on researcher judgement in choosing variables and defining their scale/granularity. Spatial variation, for instance, exists at many scales [19,23] (e.g. regional recruitment sites [102], principalities [103], neighbourhoods [104]), but only one scale is usually modelled. PCs and other latent variables avoid these issues, but they can be hard to interpret [82]. When based on genetic data, these variables also do not capture stratification that lacks a genetic footprint [3,101]. For example, a meritocratic education system may cause clustering by intelligence and associated alleles [100], but unless this is strongly correlated with other genetic variation, it will not be detected in latent genetic structure.

An alternative approach to dealing with stratification involves dizygotic and monozygotic twin pairs (T1 and T2) and their mates (M1 and M2). Twins are assumed to share a common environment, which, without mate choice, accounts entirely for PAM. Trait correlations between T1 and M1 should match those between T1 and M2, as both twins pick mates reflecting the average of the common environment [55,105–107]. Deviations from this (i.e. correlations between T1 and M1 that exceed correlations between T1 and M2) suggest choice [55,105]. However, this has a little-discussed problem, namely that gene–environment interactions may violate the common environments assumption. Dizygotic and, to a lesser degree, monozygotic twins can differ [108] and, by adulthood, may occupy phenotypically differentiated niches (i.e. stratification). If twins mate within local subpopulations, correlations between T1 and M1 could exceed those between T1 and M2 even without choice, which would be unrecognized by the model [55]. Widespread use of twin approaches may also be impractical given the rarity of monozygotic twins, as demonstrated by the small number in the UK Biobank cohort of approximately 500 000 people [109].

Competition

There are fewer methods to control for the effects of competition. A proxy of mate value can be modelled as a covariate, although measuring this is complicated [110]. Self-assessments of mate value [111–113] are subjective, so composite variables derived from objective value-related traits (e.g. body mass index (BMI) and health) may be preferable [113,114]. If many variables influence mate value, as will be the case in most populations, creating composites may be streamlined via principal components analysis [115]. The effects of competition may also be explored by testing PAM in populations (or samples) with different sex ratios [116]. If PAM is lower when sex ratios are skewed, value-based choice may be tentatively inferred. However, except for small, highly controlled studies, there is a risk that variation in PAM will reflect other differences between samples.

Convergence

Convergence can be controlled for by modelling relationship duration when testing PAM [45,117] (although it may still have occurred during pre-mating friendships [117]). If using mixed-effects models with within-subject repeated-measures data, mate-pair identity can be fitted as a random effect to account for couple-specific sources of convergence (e.g. a shared environmental exposure) [20,118].

There may also be stratification within relatively ancestrally homogeneous populations due to the creation of physical (e.g. universities, leisure spaces) or abstract (e.g. language, politics, race) sociocultural niches [124–127]. Despite theoretical work on human niche construction [126,128–130], phenotypic and genetic variation across niches is poorly understood in biology. Most human niches are probably differentiated by traits that act as organizing principles (e.g. cognitive abilities in universities), but they can vary more cryptically. For example, occupational groups differ in anthropometry [131], personality traits vary by religious affiliation [132] and psychiatric conditions fluctuate with political alignment [133]. While urban niche construction produces the most obvious (and perhaps complex) forms of stratification [127], comparable patterns exist at larger scales. In the UK, for example, historical migration has produced geographical clustering of several traits and associated alleles (although see [134] on the challenges of interpreting genetic associations) [24]. (See Glossary).

(b) . Convergence

Individuals in stable mated pairs may converge behaviourally through imitation (including mimicry), lifestyle synchronization and shared environmental effects [20,47,48,118] (figures 1 and 2), potentially leading to additional convergence in labile physical traits linked to behaviour, such as health or body mass index (BMI) [135,136]. Convergence may involve one mate shifting towards the other (i.e. asymmetrical adjustment) or mutual movement towards a midpoint [48]. Convergence also occurs in other species, where it is viewed as a form of adaptive phenotypic plasticity [137] that increases behavioural compatibility in mismatched pairs [48]. Evidence in humans—including in studies with large (ca 20 000) sample sizes [105,138]—suggests that convergence on lifestyle and behavioural traits [6,46] as well as BMI [135,136] has only a modest effect on PAM, even in relationships lasting decades [29,45,105,117,138]. Nevertheless, convergence in humans is still relatively understudied, so further research is warranted.

(c) . Mate choice

(i) . Similarity preference

Individuals may express preferences for similar traits or trait values, leading to active choice of similar mates (figures 1 and 2) [13,35,103]). For example, in hypothetical mating scenarios [5] and existing mated pairs [139], people prefer those of similar height [5], facial appearance [140,141], personality [33], sociocultural background [142,143], educational attainment [144] and other traits [1,5]. Preferred similarity in one trait may also cause PAM indirectly for correlated traits [1] via causal pathways that are predictable (e.g. a height–weight association) or cryptic (e.g. an association between genetic ancestry and lactose tolerance [61]).

How similarity preferences affect actual mating is often unclear [145]. Practical constraints (e.g. limited mate availability, competition or sociocultural barriers) can prevent preferences being realized [34,39,146], or the existence of mutually exclusive preferences [34,146] may mean that some preferences are neglected. Furthermore, preferences may be labile, changing between the point of measurement and actual mating [20,147]. More fundamentally, there is uncertainty about whether preferences and choice are mediated by the same cognitive processes [148]. There are several models for how preferences are integrated into choice [34,146,149], but these are yet to arrive at a consensus. The difficulty of interpreting preference data applies to preferences of all kinds.

(ii) . Mate-value preference

Preferences for absolute values of particular traits, irrespective of similarity, may also generate PAM (figures 1 and 2). Given some consensus on what constitutes ‘mate value’ [1,139], members of one sex are expected to compete for mating access to high-value members of the opposite sex [41,150–153]. If high-value individuals outcompete others and are chosen by opposite-sex counterparts [34,52,146], then, assuming free choice [154], they will mate. Given monogamy, these individuals will exit the mating pool, and the process will repeat itself down the value scale [14,43]. This will lead to a cascade of value-matched mating [39] (figure 2), causing PAM for traits covarying with mate value. Competition over mates (henceforth just called ‘competition’) is sometimes conceptualized in terms of ‘mating markets’ [36,39,111].

Identifying traits contributing to mate value is challenging. Individuals may express preferences for display traits covarying positively with fitness, such that they signal condition or quality [32,152,155,156]. Traits such as bilateral facial symmetry [77,157,158], facial averageness [78,158], sexually dimorphic facial morphology [159], skin condition [160–162], age [14,34] and prosociality [1] arguably fulfil this role, although this is debated [163]. Despite global trends in preferences [32,155], mate value may be population specific [158]. Local cultural (e.g. social learning) and ecological factors (e.g. food availability and pathogen prevalence [147,164,165]) may influence preferences for traits ranging from behaviour [33] to skin pigmentation [161,162,166]. In Thailand, for example, affluent urban populations prefer low BMI, while poorer mountainous groups prefer the opposite, possibly because high BMI buffers against nutritional and climatic stress [167].

The effects of competition are unpredictable for other reasons. First, models of competitive mating suggest that PAM may vary along with the value spectrum, being greatest among low-value individuals (i.e. a heteroscedastic relationship between mates’ value). This is based partly on the assumption that mating pools become more homogeneous as they dwindle to concentrated groups of the least attractive individuals (see [14,43] for descriptions of these models, their assumptions, and their predictions). Second, mating markets may be inefficient. For example, unequal sex ratios [116,156,168] can create supply/demand imbalances that empower individuals to mate above their rank [116,169], or coercive mating may override choice [170]. Third, competition only generates PAM for preferences that are shared by both sexes. Sexually discordant preferences [155,156,171–173], on the other hand, may produce NAM, with each sex preferring opposing trait values [174]. Fourth, qualitatively distinct value-related traits (e.g. intelligence and kindness [146]) may be exchangeable, producing cross-trait assortment [1] and seemingly random, non-assortative mating patterns [38,115]. In a US study, for example, poorly educated people with high social status were found to frequently marry those who are well educated but of comparatively low status [175].

3. Ultimate explanations

Ultimate hypotheses (figure 1) explain why traits shaping PAM might have emerged due to evolutionary processes. In behavioural ecology, a level of PAM may be viewed as the outcome of an optimal evolutionary strategy. In this context, ‘optimal strategy’ describes the phenotype (or set of phenotypes) that confer the highest available fitness payoff [176,177], including both ‘direct’ (fitness gained from producing offspring) and ‘indirect’ (fitness gained from the reproductive success of relatives) components [22]. Traits mediating PAM may evolve because similarity itself is beneficial, although this need not necessarily be the case [176]. For example, if a trait producing PAM is also advantageous for another function, selection may favour larger values of that trait (and therefore PAM indirectly) regardless of the fitness consequences of mating with similar individuals, which may even be negative [178,179]. Hypotheses positing that we see PAM because mating with similar individuals is, on average, beneficial are sometimes described as ‘adaptive’, while those positing that PAM emerges without being beneficial may be referred to as ‘non-adaptive’ [180]. We follow this convention, but it is important to note that both types of hypotheses concern evolutionary explanations for PAM.

(a) . Adaptive hypotheses: the benefits of positive assortative mating

(i) . Direct inclusive fitness benefits

Mate compatibility. PAM may lead to improved reproductive success if similar mates experience compatibility due to positive interactions between male and female genotypes/phenotypes [181–183]. Compatibility is generally recognized as arising from two sources. First, it may reflect phenotypic interactions between mating individuals. For example, behavioural phenotypes can affect ‘social compatibility’ [50,184], mitigating intra-pair aggression [185] or aiding behaviours such as foraging, parenting [50,182,186] and conjugation [187,188]. As pair-bonding monogamists engaging in biparental care [189], humans are candidates for social compatibility [13,50], and there is some evidence that social similarity improves cooperation [71] and relationship satisfaction in mated pairs [46,117,190,191] (although see [192]; also see [29,143,191,193] on possible compatibility arising from dissimilarity). However, little is known about social compatibility and reproductive outcomes in humans [183].

Second, ‘genetic compatibility’ [181] may arise from the allele combinations parents pass on to their offspring, with more compatible parental genotypes leading to better reproductive outcomes (e.g. embryo survival [194] or postnatal viability [16]). In humans, genetic compatibility (reviewed broadly in [181,195]) is rarely discussed in the context of PAM [11,12,181]. However, some studies suggest that mating with genetically similar people maintains coadapted (or compatible) sets of alleles (i.e. avoidance of outbreeding depression). For example, human genetic admixture may interfere with mitochondrial DNA function (via interactions with nuclear DNA), so that positive ancestry assortment improves fitness [196,197]. In addition, parental height difference predicts the incidence of emergency caesarean sections under some circumstances, such as when male height greatly exceeds female height. This is presumed to be because the larger fetuses produced by taller men pose birth challenges to shorter women [198,199].

However, caution is required when discussing human mate compatibility in an evolutionary context, not only because evidence on the topic is sparse, but also because compatibility may reflect ephemeral (in an evolutionary sense) sociocultural phenomena (e.g. religious background) [6,200,201]. For example, the relationship between moderate inbreeding (e.g. between 3^rd/4^th cousins) and offspring number/viability was hailed as providing the ‘biological basis of third cousin crush’ [202]. However, this association could reflect material/social support for unions between relatives in some cultures [108,111], which may not have existed across evolutionary time scales.

(ii) . Indirect inclusive fitness benefits

Improved mating success of relatives. Relatives share genes through identity-by-descent, so individuals gain indirect inclusive fitness benefits from the reproductive success of kin [203–205]. Consequently, it has been argued that where PAM increases the likelihood of mating with relatives (i.e. inbreeding), such as when it occurs on strongly heritable traits (e.g. face shape), it improves relatives’ mating success and will be favoured by kin selection [206–208]. In support of this, simulations suggest that strategies of weak kin-based mating outcompete others without increasing direct fitness, measured in this case as survival probability [209].

(iii) . Optimal outbreeding

While PAM may confer benefits through mate compatibility and the improved fitness of kin, it may also produce strong negative effects due to inbreeding depression. These effects, which arise from excess homozygosity, include a greater likelihood of deleterious recessive mutants being expressed in offspring [210,211]), and lower resistance to pathogens or environmental perturbations [212–214]. These opposing fitness consequences may favour an intermediate level of PAM that achieves the highest average fitness, sometimes described as the point of ‘optimal outbreeding’ [206]. For example, in several human populations [80,215], reproductive success appears to peak at moderately high levels of relatedness (e.g. 3^rd or 4^th cousins [80]), although evidence is mixed [216].

(b) . Non-adaptive hypotheses: benefits unrelated to positive assortative mating

Because the proximate forces controlling PAM are unlikely to be shaped by the fitness consequences of PAM alone, selection from other sources may lead to non-adaptive levels of PAM [217]. For example, in the case of stratification, selection on geographical dispersal propensity can arise from resource distribution, local sex ratios, local kin competition, physical constraints on movement and a myriad of other factors [51,58,217–220]. If the fitness consequences of these factors outweigh those of PAM, as will commonly be the case, an optimal strategy may include levels of dispersal (and therefore stratification) that are not adaptive with respect to PAM. Similarly, individuals expressing preferences for display traits (subsequently mate-value signals) indicating that mates are ‘good partners’ [221] or possess heritable ‘good genes’ [222,223] may experience greater fitness (reviewed in [206,223]). This may favour mate choice behaviours focussed on maximizing mate value, rather than similarity. Patterns of PAM emerging in resulting mating markets [1] (see §2c(ii)) may not be adaptive in their own right, but the fitness consequences of this may be outweighed by those of value-based mating. Given that the proximate forces mediating PAM may be under strong selection for reasons unrelated to mating, it is possible that the majority of PAM observed in contemporary populations is non-adaptive.

4. The fitness effects of mating behaviour may result in selection on traits beyond mate choice

It is often assumed that the fitness effects of mating behaviour will primarily result in selection on mate choice, but this is not necessarily true [48]. For example, stratification affects the likelihood of mating with similar individuals (see §2a), so traits regulating this phenomenon may evolve in response to the costs and benefits of PAM described in §3 [218]. In small, stratified populations, there may be an increased risk of inbreeding depression, so geographical dispersal away from birth locations may evolve as a mechanism of optimal outbreeding [217,219]. Conversely, mating within local communities or ethnolinguistic groups may improve behavioural compatibility, facilitating cooperation in defence, subsistence and biparental care [51,215,218,219,224], which would favour the evolution of localized (or limited) dispersal [224]. Post-mating behavioural mechanisms may also evolve in response to the benefits of PAM. For example, it is hypothesized that behavioural convergence is a mechanism for increasing compatibility in mismatched pairs, and stronger convergence is associated with improved reproductive outcomes in non-human species (see §2b for discussion) [48,187]. With some exceptions [46], this possibility has received little attention in human studies.

Where the fitness effects of mating behaviour do result in selection on mate choice, individuals must be able to identify signals that reliably advertise suitability (e.g. high mate value, compatibility). This may be feasible in many cases, such as when compatibility depends on concordance in stable, observable traits (e.g. height). However, identifying suitability in other cases may be more challenging [225]. For example, mating with distant relatives in order to optimize outbreeding requires a mechanism to discriminate subtle degrees of genetic relatedness. However, human kin detection may rely primarily on heuristics such a coresidence or shared perinatal association, which only signal close kinship [226]. A candidate mechanism for more distant kin detection in optimal outbreeding is positive sexual imprinting (see [208,227] for discussion), whereby offspring develop a mating template from relatives (or non-relatives [228]) in the developmental environment. Humans may exhibit a positive imprinting-like behaviour (ILB) [105,229–232], with some studies suggesting that people's mates resemble their opposite-sex parents, especially with respect to facial traits [103,228,233]. However, both the existence and function of ILB in humans are contested [226,234]. Therefore, while the fitness benefits of mating behaviour may often result in selection on mate choice, this should not be assumed as a baseline.

5. Conclusion

Studies on PAM's proximate causes should recognize the full range of causal mechanisms involved in driving the behaviour and be aware of their complex manifestations. For example, the covariance between stratification and intricate patterns of genetic ancestry is little discussed in many studies, as is the possibility that this may be associated with cryptic phenotypic variation. A greater challenge lies in moving from a general to population-specific understanding of PAM's proximate causes. Stratification, for example, may vary substantially due to local sociocultural, ecological and geographical circumstances (see [24,25] for examples), and mate choice itself may differ between subpopulations. Understanding PAM's proximate causes relies on methods to identify (or control for) the effects of different mechanisms. Many of these methods are well known, but their recognition and adoption in some fields are limited (see box 1 for discussion). Studies on ultimate processes require an equally broad perspective. They should recognize that an observed level of PAM may not be adaptive in its own right, but that this does not preclude evolutionary explanations. They should also consider that the fitness effects of PAM may cause selection on traits beyond mate choice, including dispersal propensity and convergence. In general, empirical, theoretical and methodological insights from across the literature should be used when studying PAM. For example, findings from non-human studies [50,181,182,185] on topics such as convergence [48,118] and compatibility [50,181,182] may be valuable to researchers of human mating behaviour. If the causes of PAM remain elusive, broadening horizons will at least decrease the likelihood of false causal inferences and highlight potentially promising directions for future research.

Data accessibility

This article has no additional data.

Authors' contributions

T.M.M.V. conceived the review and wrote the initial draft of manuscript, with subsequent edits and comments from A.M.-S., E.O.F. and V.S.

Competing interests

We declare we have no competing interests.

Funding

T.M.M.V., E.O.F. and V.S. are funded by the UK National Environmental Research Council. A.M.-S. is funded by a Leverhulme Trust Research Project grant no. RPG-2018-208.

Glossary

Stratification

where individuals in a population do not interact randomly and are heterogeneously distributed across groups or clines

Admixture

mating between genetically differentiated groups

Imitation

copying the behaviour of others

Mimicry

the automatic imitation of gestures, postures, mannerisms and other motor movements

Kin selection

when traits are favoured due to beneficial effects on the fitness of relatives

Direct fitness

the component of fitness gained by producing offspring

Indirect fitness

the component of fitness gained from reproduction by relatives

Inclusive fitness

the sum of direct and indirect fitness

Inbreeding depression

reduced fitness of offspring resulting from reproduction between relatives

Genes identical by descent

inherited copies of the same ancestral gene

Relatedness

a measure of genetic similarity between two individuals, relative to the average, often the probability of sharing genes identical by descent

Phenotypic plasticity

a genotype's ability to express multiple phenotypes in different conditions

Labile trait

one that is adjusted across an individual's lifetime

Niche construction

when organisms modify their ecological niche, including cultural changes in humans.