A worldwide view of matriliny: using cross-cultural analyses to shed light on human kinship systems

Abstract

Although matriliny and matrilocality are relatively rare in contemporary human populations, these female-based descent and residence systems are present in different cultural contexts and across the globe. Previous research has generated numerous hypotheses about which cultural traits are associated with the stability or loss of matrilineal descent. In addition, several studies have examined matrilineal descent with phylogenetic analyses; however, the use of language phylogenies has restricted these analyses to comparisons within a single language family, often confined to a single continent. Cross-cultural comparisons are particularly informative when they account for the relationships between widely distributed populations, as opposed to treating each population as an independent sample or focusing on a single region. Here, we study the evolution of descent systems on a worldwide scale. First, we test for significant associations between matriliny and numerous cultural traits that have been theoretically associated with its stability or loss, such as subsistence strategy, animal domestication, mating system, residence pattern, wealth transfer and property succession. In addition, by combining genetic and linguistic information to build a global supertree that includes 16 matrilineal populations, we also perform phylogenetically controlled analyses to assess the patterns of correlated evolution between descent and other traits: for example, does a change in subsistence strategy generally predict a shift in the rules of descent, or do these transitions happen independently? These analyses enable a worldwide perspective on the pattern and process of the evolution of matriliny and matrilocality.

This article is part of the theme issue ‘The evolution of female-biased kinship in humans and other mammals’.

1. Introduction

In matrilineal systems, descent is traced along female lines [1]. Matrilineal kinship organization occurs relatively infrequently in human populations [2,3], and much research by anthropologists and ethnographers has discussed the rarity and apparent instability of matriliny [2–14]. Some of this work has focused on whether matrilineal systems are inherently unstable from the male perspective, since men in these cultures do not belong to the same kin group as their biological children; instead, with matrilineal kinship organization, a man is a member of his own mother's kin group while his children are members of their mother's (his wife's) kin group. Therefore, in matrilineal systems, men are expected to invest resources in their sister's children instead of their own, which appears to violate the evolutionary predictions of inclusive fitness that individuals will prefer to invest resources in their closest kin [6,9]. This has been dubbed the ‘matrilineal puzzle'. With inclusive fitness and kinship theory in mind, the evolutionary puzzle of matriliny could be resolved by paternity uncertainty: a man might be uncertain that his wife's children are his own, but, given that he and his sister share a mother, he is certainly related to his sister's children [15], providing an explanation for inheritance from a mother's brother to his sister's son. However, the level of paternity uncertainty required for this explanation of matriliny appears to be much higher than is observed in real human populations [16], although more recent theoretical approaches have shown that mother's brother–sister's son inheritance can also be stable in the context of high paternity certainty if certain restrictive assumptions are relaxed [8,17].

Although matriliny is relatively rare across human populations, matrilineal populations are distributed around the world (figure 1). Previous studies of matriliny have often focused on one geographical region or a single language family, and relatively few analyses have accounted for phylogenetic relationships between studied populations [14,19]. In a key phylogenetically controlled study, Holden and Mace assembled data on rules of descent (binarized as ‘matriliny' versus ‘patriliny') and on the presence or absence of cattle across 68 Bantu- and Bantoid-speaking populations [14]. Using a phylogeny of the Bantu language family to control for ancestral relationships between populations, this study was able to demonstrate a directional evolutionary association in which the spread of cattle domestication led previously matrilineal populations to transition to patriliny, supporting the adage that ‘the cow is the enemy of matriliny' [12].

Figure 1.

Map of major descent types in worldwide populations, adapted from D-PLACE [18]. Out of 1291 populations in the Ethnographic Atlas, there are 160 matrilineal populations, 590 patrilineal populations, 362 bilateral populations, 52 duolateral populations, 50 populations with mixed descent, 48 ambilineal populations and 12 populations with quasi-lineages. 17 populations are missing data and are not included.

Anthropological research on matriliny and matrilocality in different parts of the world has generally predicted that female-based descent and residence patterns are becoming increasingly rare [6,9,12,20]. Accompanying this general prediction, however, has been a wide variety of proposed traits that have been predicted to influence matrilineal descent, either reinforcing it or destabilizing it [9]. In addition to cattle as mentioned above [14], large domesticated animals, in general, have been predicted to destabilize matriliny [12]. Other hypotheses hinge on subsistence strategy; for example, horticulture has been predicted to be positively associated with matriliny, whereas plough agriculture has been linked to the loss of matriliny [12]. Still other authors predict that the transition from matriliny to patriliny is associated with factors related to inheritable resources, such as increased wealth, property and inequality [2,7], but that matriliny can be fostered by daughter-biased investment of resources [12,13]. Related to the inclusive fitness interpretation of the matrilineal puzzle, some authors predict that matriliny is associated with cultural traits that would reduce paternity certainty; for example, external warfare, which requires men to be absent for long periods, is thought to promote matrilocal residence [21,22], and high divorce rates are hypothesized to promote or stabilize matriliny [10,11]. In addition, populations with matrilocal or other female-based post-marital residence systems are often also matrilineal; in these cases, it has been hypothesized that matrilocality precedes and promotes the transition to matriliny [2–4]. We summarize these hypotheses in table 1.

Table 1.

Existing hypotheses linking matriliny to other cultural factors. The ‘SCCS p-value' columns give the significance of the association between each trait and matriliny/matrilocality in the Standard Cross-Cultural Sample (χ² analysis). The ‘BayesTraits p-value’ columns give the significance of the BayesTraits analysis of correlated evolution between each trait and matriliny/matrilocality, using the supertree. Bold red text indicates values that were significant after Holm-Bonferroni correction. * indicates that only one economic specialization (metalworking) was statistically significant.

			matriliny		matrilocality
hypothesis	ref.	related D-PLACE category tested here	SCCS χ2 p-value	BayesTraits p-value	SCCS χ2 p-value	BayesTraits p-value
domestic factors: marriage, family, residence
‘matrilineal descent groups have probably developed independently… on the basis of prior matrilocality'	[2–4]	matrilocal residence	6.1 × 10−14	6.7 × 10−10	n.a.	n.a.
the rule of exogamous marriage is difficult to reconcile with matriliny (matrilineal puzzle)	[6]	exogamy	0.05991	0.35762	0.00772	0.03189
matriliny is incompatible with a focus on nuclear family	[7]	nuclear family	0.80878	1	0.30207	0.03631
polygyny/polyandry associated with matrilineal inheritance	[8]	polygamy	0.99909	0.61088	0.15830	0.30649
communal breeding promotes matrilineal system	[23]	(no equivalent variable)
high frequencies of divorce are associated with matriliny	[10,11]	(no equivalent variable)
low paternity certainty promotes matriliny	[2,12]	(no equivalent variable)
economic factors: wealth transfer, inequality, economic specializations
matriliny can arise from daughter-biased investment by parents and/or grandparents	[13]	matrilineal inheritance of movable property matrilineal inheritable of real property	6.9 × 10−7 6.9 × 10−8	3.0 × 10−15 4.4 × 10−12	1.2 × 10−6 7.7 × 10−7	2.3 × 10−10 7.3 × 10−9
matriliny is vulnerable with increasing wealth	[2]	(no equivalent variable)
‘economic changes brought about by contact with Western industrial nations’ disrupt matrilineal systems	[5]	(no equivalent variable)
economic differentiation (presumably within a population) is incompatible with matriliny	[7]	economic specializations class caste	none sig. 0.05130 0.00755	none sig. 0.74781 0.02065	0.00022* 0.02660 0.03970	none sig. 0.00125 0.00096
matriliny is incompatible with inequality: ‘power, property and prestige’	[2]	class caste slavery	0.0513 0.00755 0.53279	0.74781 0.02065 0.00012	0.02660 0.03970 0.14228	0.00125 0.00096 0.00377
warfare promotes matrilocal residence	[21,22]	(no equivalent variable)
subsistence factors: food sources, agriculture, domestication
matrilineal societies are less likely to have ploughs or large domesticated animals, more likely to be horticultural	[12]	plough agriculture large domesticated animals pastoralism intensive agriculture extensive agriculture animal husbandry	0.02110 0.10952 1 0.09594 0.22234 0.46334	0.00191 0.02905 0.81965 0.02038 0.00043 0.53215	0.00279 0.00206 0.19944 0.00103 0.06572 0.02041	9.2 × 10−5 2.1 × 10−6 0.18108 0.00011 3.4 × 10−6 0.00030
cattle domestication destabilizes matriliny (the cow is the enemy of matriliny)	[14,24]	cattle domestication milking leather working	0.06909 0.00311 0.02418	0.08807 0.00311 0.02418	0.00389 0.00024 0.32645	0.00016 1.9 × 10−7 0.01898
increased reef density promotes matriliny (the fish is the friend of matriliny)	[19]	the presence of fishing boat building	0.13718 0.75392	0.08403 0.51263	0.01612 1	0.00045 0.02705

Previous evolutionary studies of descent and post-marital residence have primarily been conducted within language families [14,25–28], which limits their ability to draw conclusions about the evolution of descent and residence patterns on a larger scale, both because different language-family trees cannot be intuitively merged and because language groups demonstrate their own dynamics of change [29]. To date, there has not been a global evolutionary analysis to detect which cultural traits appear to maintain, generate or destabilize matriliny. Here, we evaluate hypothesized relationships between cultural traits and female-based descent and residence patterns on a global scale. We first assess these associations across 1291 populations in the Ethnographic Atlas; we then account for the non-independence of these populations (Galton's problem [30,31]) in two ways. First, we repeat our analyses in the Standard Cross-Cultural Sample [32], which is a subset of populations whose cultures are relatively independent from one another. Second, we construct a worldwide phylogeny that is a composite of numerous genetic and linguistic phylogenies merged with an existing supertree of human populations [33] and use this phylogeny to perform analyses that account for the ancestral relationships between populations [31]. With these phylogenetically controlled analyses, we aim to assess the likelihood that traits in the Ethnographic Atlas show correlated evolution with matrilineal descent and matrilocal residence patterns, enabling us to both test and generate hypotheses about the evolutionary dynamics of kinship system organization.

2. Methods

(a). Constructing the D-PLACE dataset

We gathered ethnographic data from D-PLACE [18], an online resource that stores digitized data from the Ethnographic Atlas [34] (data downloaded from github.com/D-PLACE/dplace-data), and compiled it into a matrix of 94 cultural categories from 1291 populations. For 76 of these categories, we were able to reclassify the cultural data into binary traits for analysis. For example, from ‘Settlement patterns [EA030],' we could encode sedentary (coded as 1) or non-sedentary (coded as 0). In some cases, we could not have coded a single binary trait for a given cultural category without excluding a significant amount of information from the analysis. In these cases, we created multiple binary traits from one category in the Ethnographic Atlas. For example, ‘Descent: major type [EA043]' was binarized twice, once according to the presence or absence of matrilineal descent and then according to the presence or absence of patrilineal descent (table 2). For 17 of the cultural categories in D-PLACE, there was not an intuitive way to binarize the data, so those categories were excluded from the analysis. Additionally, for the ‘Age or occupational specialization: gathering [‘EA061]' and ‘Age or occupational specialization: agriculture [‘EA065]' only one of the populations exhibited age or occupational specialization, so those categories were also excluded from the analysis. Data were not available for all cultural traits for each population; missing values for a population were coded as ‘NA'. We classified D-PLACE data from 76 categories into 126 binarized traits; for a full list of these binarized traits, see electronic supplementary material, Dataset S1.

Table 2.

Binary descent and post-marital residence traits. The binary categories of descent as classified for our analyses.

D-PLACE Category from Ethnographic Atlas	our categorization	D-PLACE type
descent: major type [EA043]	matrilineal	matrilineal
	non-matrilineal	patrilineal, bilateral, duolateral, quasi-lineages, ambilineal, mixed
	patrilineal	patrilineal
	non-patrilineal	matrilineal, bilateral, duolateral, quasi-lineages, ambilineal, mixed
marital residence with kin: prevailing pattern [EA012]	matrilocal	matrilocal, avunculocal, uxorilocal, ambilocal with marked preponderance of uxorilocal practice
	non-matrilocal	patrilocal, virilocal, ambilocal with marked preponderance of virilocal practice, ambilocal, optionally uxorilocal or avunculocal, optionally patrilocal (or virilocal) or avunculocal, neolocal, non-establishment of common household
	patrilocal	patrilocal, virilocal, ambilocal with marked preponderance of virilocal practice
	non-patrilocal	matrilocal, avunculocal, uxorilocal, ambilocal with marked preponderance of uxorilocal practice, ambilocal, optionally uxorilocal or avunculocal, optionally patrilocal (or virilocal) or avunculocal, neolocal, non-establishment of common household

(b). Tests for association between descent/residence systems and other cultural traits

We next determined whether cultural traits significantly co-occurred with certain descent or residence systems more than would be expected by chance, based on the frequency of each trait across 1291 populations in the Ethnographic Atlas [34]. We tested for significant associations between each of the 126 binary traits with each of our focal traits: ‘matrilineal descent,' ‘patrilineal descent,' ‘matrilocal residence' or ‘patrilocal residence'. For each of these comparisons, we made a contingency table (as in table 3). If all expected counts for the 2 × 2 contingency table were five or more populations, we performed a χ² test; otherwise, we performed a Fisher's exact test since the χ² test is unreliable with small expected numbers in any one category.

Table 3.

Contingency table of female-based descent and female-based post-marital residence patterns. Here, the matrilocal category also includes other ethnographic classifications indicating female-based residence patterns, such as avunculocal and uxorilocal residence.

	matrilineal	non-matrilineal	total
matrilocal	110	88	198
non-matrilocal	47	1008	1055
total	157	1096	1253

We then conducted a similar set of tests, but only included the subset of populations that are in the Standard Cross-Cultural Sample; this sample of 186 populations was chosen in an effort to survey populations that are as independent from one another as possible [32]. For each set of tests, we performed a Holm–Bonferroni correction for multiple hypothesis testing.

(c). Compiling a genetic dataset

We downloaded mtDNA sequences from the National Center for Biotechnology Information (NCBI) database using the following search terms for each population in D-PLACE: (1) ‘(Population) mitochondrial genome' and (2) ‘(Population) AND human[organism] AND mitochondrion[title] AND genome[title]', where ‘(Population)' was the name or alternative name of a human population. Additional sequences were obtained from Kivisild et al. [35], the Human Mitochondrial Genome Database [36] and the 1000 Genomes Project [37]. The genetic data generally included metadata with the population name and geographical sampling location for each sequence; when this was absent (such as when the population name in the NCBI search term occurred in the title of the study but not in the sequence metadata) we checked the citation given in NCBI and its electronic supplementary material to assign a population name and sampling location to the sequence.

(d). Matching the genetic data to D-PLACE population names

For each mitochondrial genome, we first determined whether the population name given in NCBI was identical to a population in D-PLACE; if so, we considered the mtDNA and the ethnographic data to correspond to the same population. To validate these cases, we checked whether the D-PLACE population location and the mtDNA sampling location were from the same geographical region. If populations could not be associated in this way, we examined the languages spoken by each population. D-PLACE provides a standardized language code (ISO 639-3) for most populations in the downloadable data tables, and we had previously matched many of the sequenced populations to their spoken language using the Ruhlen database of phoneme inventories from Creanza et al. [38]. If a population in D-PLACE had an identical ISO code to a sequenced population, we matched them and checked the validity of the match with the geographical locations of each population. For populations that were still unmatched, we checked alternative population, language and dialect names given in Ethnologue [39] for the sequenced populations to attempt to match an alternative name to a the D-PLACE population. For populations that were matched by population name and location, we also verified the match with language information whenever possible. In sum, we collected 4210 mitochondrial genome sequences from a total of 193 populations, 165 of which were successfully matched to populations listed in D-PLACE (electronic supplementary material, Dataset S2).

(e). Making the supertree

In order to test our hypothesis using a global phylogeny of D-PLACE populations, we built a supertree: a single tree assembled from a set of smaller phylogenies. This method enables us to combine trees that were constructed using different data types. Our supertree was constructed by combining eight continent-level mtDNA phylogenies, 12 linguistic phylogenies from D-PLACE and a previously constructed supertree from [33]. This previously published supertree, which has 186 populations and four non-human outgroups, was generated by merging phylogenies constructed from several types of genetic and linguistic data. We included in subsequent analyses those populations from D-PLACE that we could associate with genetic data, enabling us to estimate their genetic distance from one another, and we incorporated both genetic and linguistic phylogenies from these populations into the published supertree.

For each population included in this study, we acquired full mitochondrial DNA sequences from at least one individual. Most matrilineal populations that have been genotyped had only mtDNA available, but mitochondrial DNA alone is not well suited to building a large-scale phylogeny of populations; mitochondrial phylogenies are generally constructed at the haplogroup level [35,40]. Accordingly, we assigned each of the genotyped populations to a region based on the geographical location of that population and the United Nations Geoscheme to make smaller, continent-level mtDNA phylogenies within the following regions: North-Central America (13 populations), South America (12), Central-South Asia (17), Europe (31), North-East-Southeast Asia (50), West Asia (12), Africa (47) and Oceania (11). To construct these genetic trees, we aligned the genetic data using MAFFT [41] and constructed a pairwise genetic distance matrix with the Kimura 2-parameter distance metric. The distance matrix of the individual sequences was then averaged within populations, resulting in a population-level pairwise distance matrix that quantified the average genetic distance between the populations being studied. We then constructed continent-level neighbour-joining trees from these distance matrices to merge with the published supertree, as described below and in the electronic supplementary material.

We included data from 12 linguistic phylogenies available for download on D-PLACE, which had been pruned to only societies present in the D-PLACE database [42–52]. Each linguistic phylogeny used in the construction of the supertree included at least one population from our genetic dataset.

We added our additional genetic and linguistic phylogenies to the published supertree using TNT (Tree analysis using New Technology) [53] following the methods in [33], resulting in a phylogeny with 290 human populations and four non-human outgroups. Detailed methods are included in the electronic supplementary material. Then, we trimmed this large phylogeny (which included populations from the previously published supertree not included in this study) to include only the 165 populations for which we had ethnographic and genetic data. Branch lengths were necessary for our evolutionary analyses, but the supertree methods do not produce a tree with branch lengths, so we used mtDNA genetic distance data to add branch lengths to the supertree (electronic supplementary material). The result was a global phylogeny of the 165 populations that included branch lengths. Finally, we rescaled the phylogeny to have a mean branch length of 0.1, based on the recommendation in the documentation for the BayesTraits package used in subsequent analyses, which states that this scaling prevents the predicted evolutionary rates from ‘becoming small [or] hard to estimate' [54].

(f). Testing for correlated evolution

We next tested for correlated evolution between binary traits coded from the Ethnographic Atlas [34]. If two cultural traits are evolving independently from one another, then the transitions between the presence and absence of one trait should not depend on the other trait. In contrast, if the evolution of the two traits is correlated, then transitions in one trait are significantly dependent on the state of the other trait. For example, Holden and Mace found that Bantu populations with matriliny were significantly more likely to transition to patriliny when cattle were present compared to when they were absent [14]. To test this type of prediction on a worldwide scale, we used the function ‘Discrete’ in the program ‘BayesTraits (V2)’ via the R package ‘BayesTraitsWrapper V1’ (btw v. 1.0). We tested each of the traits ‘matrilineal descent,' ‘patrilineal descent,' ‘matrilocal residence' or ‘patrilocal residence' in combination with each of 124 other cultural traits using each of the four trees that contained at least 30 populations (global supertree, and the Bantu, Austronesian and Indo-European language trees) (4 × 124 comparisons for each of four trees). For each combination of factors, we performed 20 runs of Discrete using the maximum-likelihood method to generate independent and dependent models of correlated evolution. We performed a likelihood ratio test using lrtest (package: btw v1.0) between each pair of models to obtain a p-value indicating whether or not the dependent-evolution model fit the data substantially better than the independent-evolution model. The reported p-values are the medians of the p-values from these 20 runs. We obtained the final values for transition rates between states by calculating the means and 95% confidence intervals of the computed rates from the 20 runs of the dependent model for each trait combination. We performed a Holm–Bonferroni correction for multiple hypothesis testing to determine the threshold for significance for each test. Two traits (male-biased house construction and female-biased house construction) did not converge in BayesTraits and were omitted from these results.

3. Results

(a). Tests of cultural trait association

It has been hypothesized that the evolution of matrilineal descent correlates with female-based post-marital residence; specifically, that matrilocal residence facilitates the transition to matrilineal descent in a population [2–4]. In line with this hypothesis, we found that matrilineal descent often co-occurred with matrilocal residence in D-PLACE populations, with 70% of matrilineal populations also having matrilocal residence, even though both characteristics are relatively rare across human populations (table 3). We tested this association with a χ² test and found that it would be extremely unlikely for this degree of co-occurrence between matriliny and matrilocality to occur by chance (p-value 2.8 × 10⁻⁸⁷).

We repeated this process for each of the 126 binary cultural traits that we coded from D-PLACE to determine whether they were significantly associated with matrilineal descent, patrilineal descent, matrilocal residence or patrilocal residence (electronic supplementary material, table S1 and see table S2 for full results). We conducted these χ² tests on both the 1291 populations in the Ethnographic Atlas and, to correct for Galton's problem, on the Standard Cross-Cultural Sample of 186 populations (table 1; electronic supplementary material, tables S1 and S2). Broadly speaking, we found that matrilineal descent was positively associated primarily with other female-based aspects of the kinship system (matrilineal inheritance of real and movable property, matrilocal residence and female-biased hereditary political succession) and negatively associated with their male-based counterparts (table 1; electronic supplementary material, table S1). In addition, when we considered the full Ethnographic Atlas dataset (1291 populations), matrilineal descent was positively associated with extensive agriculture and female-biased participation in agriculture, and negatively associated with cows, large domestic animals and milking. However, these associations were not significant for the 186 populations in the Standard Cross-Cultural Sample (table 1; electronic supplementary material, table S1).

We conducted a similar analysis to identify cultural traits that were significantly associated with matrilocal residence. The same cultural traits that were associated with matriliny in the Standard Cross-Cultural Sample were also associated with matrilocal residence: matrilocal residence was significantly positively associated with several female-biased patterns (e.g. inheritance of real/movable property and hereditary political succession) and negatively associated with their male-biased counterparts (table 1; electronic supplementary material, table S1). In addition, matrilocal residence was negatively associated with several subsistence-related traits, such as intensive agriculture, plough cultivation, milking and specialized metal working.

(b). Tests for correlated evolution

In addition, we used phylogenetically controlled analyses to shed new light on existing hypotheses about the cultural factors that have been proposed to either foster or destabilize female-based descent and residence patterns (table 1). For these analyses, we used four phylogenies—the global supertree and Bantu, Austronesian and Indo-European language trees (figure 2)—and we tested whether each of the binarized cultural traits from D-PLACE showed correlated evolution with matrilineal descent, patrilineal descent, matrilocal residence or patrilocal residence (electronic supplementary material, tables S1, S3, S4 and figures S1–S7). These tests allowed us to assess whether these kinship and residence traits were evolving independently from other cultural traits or whether the transition rates of one were significantly dependent on the state of the other trait (see electronic supplementary material, table S4 for full results). For example, under the prediction that matrilocality facilitates the emergence of matriliny, we might expect to see an elevated rate of transition from no matrilineal descent to matrilineal descent in the presence of matrilocal residence.

Figure 2.

The evolution of matrilineal and non-matrilineal lineages across multiple phylogenies. The tips of each tree show whether populations were coded as matrilineal (green) or non-matrilineal (white). At each node, the probability that the common ancestor was matrilineal is represented by the fraction of the bar that is green; a white bar indicates a small probability that the common ancestor at that node was matrilineal. The Indo-European tree does not include any matrilineal populations and is not shown. For a discussion of the estimated ancestral states, see the electronic supplementary material.

Broadly, we found support for the hypothesis that matriliny and matrilocal residence show correlated evolution, corroborating the results of the association test (tables 1 and 3; electronic supplementary material, table S1). However, our global phylogenetic analysis suggests a different directionality from previously hypothesized. Murdock suggested that populations that were already matrilocal would be more likely to develop matrilineal descent systems [2,3]. Contrary to this prediction, our results indicated that matrilocal populations have a very low transition rate from non-matrilineal to matrilineal systems (figure 3a). Instead, it appears more likely that populations transition first to matriliny and then subsequently to matrilocality.

Figure 3.

The evolution of matrilineal descent is correlated with matrilocal residence, matrilineal inheritance and hereditary political succession, and extensive agriculture across a global supertree. We tested for correlated evolution between pairs of binary cultural traits, assessing whether the two traits were evolving independently or whether the transition rates of one trait depended significantly on the state of the other trait. (a) Matrilineal descent was highly unstable without matrilocal residence. (b) Matrilineal descent was significantly correlated with matrilineal inheritance of real property, such that the presence of this trait was associated with an increased rate of transition towards matrilineal descent. (c) Matrilineal populations without matrilineal hereditary political succession were unstable, and tended to lose matrilineal descent. (d) Matrilineal descent was lost more quickly in the absence of extensive agriculture. The predicted rate of transition between each state is indicated by the numbers adjacent to each arrow, and the arrow thickness is scaled to the rate. Arrows with a rate of zero are indicated in black.

Daughter-biased investment has also been hypothesized to foster matriliny [13]. While we did not have a cultural variable to directly test the daughter-biased investment hypothesis, we examined maternal inheritance of both real and movable property as a proxy. We found that both of these inheritance patterns form a stable state in combination with matriliny: matrilineal descent and matrilineal inheritance were stable together in the sense that there were low rates of evolutionary transitions away from this combination, but one trait without the other was not stable (figure 3b; electronic supplementary material, table S3). In addition, corroborating the results of the association tests above, we found that both matrilineal descent and matrilocal residence showed correlated evolution with other sex-biased aspects of the kinship system (matrilineal/patrilineal inheritance of real and movable property, matrilocal/patrilocal residence and female/male-biased hereditary political succession; figure 3c and table 1; electronic supplementary material, tables S1, S3 and S4).

Other cultural aspects of family structure have been putatively associated with female-based descent and residence: polygamy is hypothesized to associate with matrilocality and matrilineal inheritance, exogamy is thought to destabilize matriliny and nuclear family structure is hypothesized to be incompatible with matriliny [6–8]. We did not find significant evidence for correlated evolution between matriliny and any of these factors in the global analysis, although we found support for correlated evolution between matriliny and the presence of both matrilineal and patrilineal exogamous groups (electronic supplementary material, tables S1, S3 and S4). In particular, matrilineal descent is stable in combination with matrilineal exogamous groups but not stable in combination with patrilineal exogamous groups.

Various forms of economic differentiation have also been hypothesized to be incompatible with matriliny [2,5,7]. Although in the χ² analysis we found that matrilocal residence and metalworking co-occurred less often than would be expected by chance in the Standard Cross-Cultural Sample, we did not find evidence for correlated evolution between matriliny and any sex-biased economic specializations, including metalworking (table 1; electronic supplementary material, table S1). Matriliny is also hypothesized to be incompatible with economic inequality—'power, property and prestige' [2]. We tested several traits related to economic inequality and observed evidence for correlated evolution between matriliny and the existence of slavery in a population, with slavery increasing the rates of transition between modes of descent. However, we did not find correlated evolution between matriliny and caste or class systems across global populations. Interestingly, class systems showed significant correlated evolution with matriliny and matrilocal residence only for the Bantu populations. In these populations, class systems were evolutionarily unstable in combination with both matriliny and matrilocal residence, with populations quickly transitioning away from class systems or away from female-based descent or residence.

Various aspects of subsistence strategy have been frequently hypothesized to associate with matriliny; in particular, the intensification of agricultural practices and the domestication of cows and other large animals have been predicted to destabilize matriliny and promote the transition to patriliny [12,14,24]. In support of this hypothesis, we found significant evidence of correlated evolution between matriliny and extensive agriculture—the practice of farming over a large land area with relatively small labour investment and low yields (figure 3d). This type of agriculture system appeared to stabilize matriliny, as matriliny was lost more rapidly when extensive agriculture was absent. However, across numerous tests, we found more statistical support for this suite of hypotheses when considering matrilocal residence instead of matrilineal descent. For example, matrilocal residence was more stable in the presence of female-biased participation in agricultural practices (electronic supplementary material, tables S3 and S4), and matrilocal residence was never gained in the presence of intensive agriculture (figure 4). In addition, the combination of matrilocal residence with plough agriculture was not stable, and populations with this combination appeared to quickly lose either ploughs or matrilocality. Moreover, populations almost never transitioned to matrilocality once they acquired plough agriculture (figure 4). We found a similar pattern for other agricultural traits: matrilocal residence was not stable in combination with large domestic animals, animal husbandry or milking (figure 4 and table 1; electronic supplementary material, tables S3 and S4). However, matrilocal residence was also gained and lost rapidly without these associated traits, so the absence of agricultural and domestication practices did not fully stabilize post-marital matrilocal residence (figure 4).

Figure 4.

The evolution of matrilocal residence is correlated with the presence of large domestic animals, intensive agriculture, plough cultivation and milking on a global scale. As above, we tested for correlated evolution between pairs of binary cultural traits to determine if the rate of transition of one trait was significantly dependent on the state of the other trait. The presence of large domestic animals (a), intensive agriculture (b), plough cultivation (c) and milking (d) all reduced the rate of transition between types of residence system. Once gained, these traits seemed to preclude the evolution of female-based systems. Arrows are labelled with predicted transition rates (also indicated by thickness).

The spread of cattle specifically has been hypothesized to destabilize matriliny in Bantu populations [14,24]. We did not observe correlated evolution between cows and matriliny on any scale. We did, however, observe significant correlated evolution between cows and matrilocal residence, but only in the Bantu lineage; a matrilocal society was more likely to adopt cows than a non-matrilocal one, but then matrilocal residence was quickly lost in the presence of cows (electronic supplementary material, tables S1 and S4).

4. Discussion

The long-term dynamics of matrilineal descent and matrilocal residence, including the factors contributing to their stability and loss, have been a focus of research since the late nineteenth century but only rarely studied in the context of the evolutionary relationships between populations [2–15,22]. Here, we take a worldwide view of the evolution of matrilineal descent and matrilocal residence patterns across globally distributed populations, and we study the interactions between cultural traits while accounting for the cultural non-independence of related populations. In particular, we test multiple hypotheses that have been proposed in the literature that pinpoint cultural traits that might either stabilize or destabilize patterns of matrilineal descent and matrilocal residence (table 1).

Taken together, our results indicate that only a subset of these hypotheses are supported by worldwide analyses to detect cultural trait associations or correlated evolution. For matriliny in particular, our significant findings pointed to associations between matrilineal descent and other patterns of cultural inheritance through the female line, such as female-biased hereditary political succession, matrilocal residence and matrilineal inheritance of real and movable property (table 1; electronic supplementary material, table S1). Our results lend support to the hypothesis that daughter-biased investment (using a proxy of wealth transmission through the maternal line) fosters matriliny [13], but we also note that matriliny might in turn also foster maternal inheritance of wealth (figure 3). In addition, these quantitative results suggest that female-biased cultural patterns of descent, inheritance and residence might be parts of a suite of traits or a broader cultural norm of female-based transmission rather than independently evolving cultural traits. Our results also shed new light on Murdock's hypothesis that matriliny evolves on the basis of prior matrilocality [2–4]: in contrast to Murdock's prediction, our evolutionary analysis indicates that it might not be uncommon for matriliny to develop first, prompting a transition to matrilocality (figure 3).

There is an intuition from previous literature that matriliny is a potentially unstable form of kinship system organization; interestingly, our results appear to support this hypothesis in the sense that most of our observed associations were negative (with the exceptions of the female-based inheritance patterns mentioned above), with matriliny and matrilocality occurring less often with other cultural traits than would be expected by chance (electronic supplementary material, table S1, blue cells). In particular, a number of hypotheses have linked the loss of matriliny to subsistence and economic factors, particularly shifts from horticulture and relative economic equality to intensifying agriculture, animal domestication and increasing economic inequality [2,12]. In line with these hypotheses, we found a stabilizing effect in which extensive agriculture (as opposed to intensive agriculture) appeared to reduce the rate of loss of matriliny (figure 3). However, we found no significant association or correlated evolution between matriliny and of the subsistence or economic factors hypothesized to destabilize it [2,12,14,24]. Instead, we found significant associations between these factors and matrilocality: for example, the presence of intensive agriculture, animal husbandry, milking or plough agriculture in a population all appeared to be unstable in combination with matrilocal residence. Many of these same trends were similar for matriliny but were not statistically significant; this might indicate that matrilocality has a stronger pattern of being destabilized by intensifying agriculture or animal domestication than matriliny does, or it could simply suggest that we have more statistical power to detect this phenomenon in our larger sample of matrilocal populations.

Another related hypothesis predicts that the ‘cow is the enemy of matriliny' [14,24]. In testing this hypothesis, we found a negative association between cows and matriliny, and between cows and matrilocality, across the full dataset of 1291 populations. However, these associations were no longer significant when we accounted for the non-independence of populations, either by using the subset of populations in the Standard Cross-Cultural Sample or by conducting phylogenetically controlled analyses (table 1, electronic supplementary material, table S1). Earlier studies tested the specific hypothesis that the domestication of cattle was correlated with a loss of matriliny in Bantu populations [14,24]. We repeated these analyses with a Bantu language phylogeny that included more populations (101 compared to 68 in [14]). In this Bantu subset of populations, we found significant evidence for correlated evolution between cows and matrilocality (p = 3.1 × 10⁻⁵), but the observed correlated evolution between cows and matriliny was not significant when we corrected for multiple hypotheses (p = 0.00028). The significant association that we observe between cows and post-marital residence patterns could be an example of a lineage-specific evolutionary trend [29] that is particularly salient in Bantu populations. Alternatively, since we observe significant associations between matrilocality and other cultural traits that are related to domestication but not limited to cattle, such as animal husbandry, large animal domestication and milking, on a worldwide scale, it is possible that framing the hypothesis as the ‘cow is the enemy of matriliny' limits our ability to detect similar processes that are occurring in societies that have domesticated other large animals besides cattle.

There are several caveats to this and any cross-cultural analysis. Any subset of populations is likely to be not fully representative of the diversity of worldwide human populations. Our two approaches for addressing Galton's problem (that related populations are not independent samples) make different sets of assumptions: using the Standard Cross-Cultural Sample makes the assumption that choosing populations from different regions produces a subset of populations that are functionally independent from one another, whereas conducting phylogenetically controlled analyses of cultural traits makes the assumption that those cultural traits are transmitted in a way that is well represented by the phylogeny (i.e. vertically transmitted). We constructed a global phylogeny from several regional language- and genetic-based phylogenies to maximize the number of matrilineal populations that could be studied in a phylogenetic context. However, the matrilineal populations on this global phylogeny are spread out across the tree, making it difficult to evaluate the evolutionary patterns that might occur among closely related populations that show evidence of recent gains and losses of matriliny, such as those clades with multiple matrilineal and non-matrilineal populations seen on the Bantu and Austronesian trees (figure 2). This underscores the need to genotype populations more broadly, with cultural diversity in mind.

In addition, the results of our evolutionary analyses were generally similar between the global and Bantu trees, with matrilineal descent and matrilocal residence showing significant evidence of correlated evolution with other sex-biased inheritance patterns and with traits related to intensive agriculture and animal domestication (electronic supplementary material, table S1). However, we found fewer significant results with the Austronesian tree, with matrilineal descent correlated with matrilineal inheritance of property, residence and political succession but no economic- or subsistence-related cultural traits. This result warrants further investigation in the context of hypotheses that are tailored to this region; for example, the ‘fish is the friend of matriliny' hypothesis hinges not just on the existence of fishing in a population but also on reef density [19], which might correspond to the relative effort investment required to catch fish. This nuanced view of fishing is not captured in the Ethnographic Atlas, demonstrating a limitation of this type of large-scale comparative analysis.

In sum, on a worldwide scale, we find support for several of the previous hypotheses linking the evolution of female-based descent to other cultural factors, and our global analysis adds to the discussion of these hypotheses by proposing the direction of these associations across many cultures and by hypothesizing whether they apply to descent, post-marital residence, or both. Taken together, our analyses suggest that complex and nuanced factors contribute to the evolution of matriliny and matrilocality, and both are best studied in the rich cultural context of human populations.

Data accessibility

We include all data and code analysed in this article as electronic supplementary material.

Authors' contributions

N.C. conceived of the experiment. A.S. collected data. A.S., K.T.S. and N.C. designed research, analysed data, made figures and wrote the manuscript.

Competing interests

We declare we have no competing interests.

Funding

Funding was provided by Vanderbilt University to A.S., K.T.S. and N.C., and by the Vanderbilt University Summer Research Program to A.S.