Social media summary
Karo cousin marriage decline due to change in demographic and alliance factors, not rejection of the cultural norm.
Introduction
Marriage between relatives has long occupied a central place in anthropological research because it links individual mate choice to broader questions of kinship, alliance formation, inheritance, and social organisation (Levi-Strauss, Reference Levi-Strauss1949; Needham, Reference Needham1986). Although many societies express preferences for particular categories of relatives as spouses, the frequency with which such marriages occur often differs markedly from the idealised patterns described in cultural norms. Understanding why marriage preferences persist, weaken, or disappear therefore offers valuable insights into the relationship between cultural prescriptions and observed behaviour. Current marriage patterns within one ethnic group in Indonesia provide an especially informative case, as community members have historically expressed a strong preference for marriage to a particular type of cousin, yet the practice itself has undergone a substantial decline.
The Karo people are a suku (ethnic group or tribe) from North Sumatra, Indonesia, with a long-standing stated preference for matrilateral cross-cousin (impal) marriage (Kipp, Reference Kipp1993; Kushnick,Reference Kushnick2006; Kushnick & Fessler, Reference Kushnick and Fessler2011; Singarimbun, Reference Singarimbun1975; Steedly, Reference Steedly1993). Traditionally, ‘proper’ impal marriages encompass both first cousin (MBD-FZS) and second cousin (MFBSD-FFBDS) unions (Fig. 1a). This preference is embedded within the broader ‘circulating connubium’ (van Wouden, Reference van Wouden1935) or asymmetric alliance systems described by alliance theorists (Leach, Reference Leach1954; Lévi-Strauss, Reference Levi-Strauss1949; Needham, Reference Needham1986), in which ‘wife giver’ (kalimbubu) and ‘wife taker’ (anakberu) relationships structure a patrilineal social order (Fig. 1b).
(a) First- and second-cousin forms of proper matrilateral cross-cousin (impal) marriage among the Karo; (b) the asymmetrical alliance structure fostered by impal marriage, in which X are wife-givers (kalimbubu) and Z are wife-takers (anakberu) of Y; and (c) the locations of the 62 Karo villages in which data were collected. In the kinship diagrams shown in Panels (a) and (b), triangles represent males, circles represent females, double lines indicate marriage, and lines indicate descent.

Figure 1 Long description
A) Kinship diagrams show first and second cousin marriages among the Karo. Triangles represent males, circles represent females, double lines indicate marriage and lines indicate descent. B) Diagram illustrates the asymmetrical alliance structure with wife-givers (X), ego (Y) and wife-takers (Z). C) Map of North Sumatra shows locations of 62 Karo villages, marked by white dots, near Gunung Sinabung and Kabanjahe. The map includes a compass rose and a scale bar indicating 20 km. An inset map shows Sumatra′s position relative to the Equator and Java.
Despite its cultural importance, the practice of impal marriage has declined markedly over the past century. Archival and village-census data show that while 11–13% of marriages in the 1930s–1960s were proper impal unions, this proportion fell to just 3–5% between the 1990s and 2014 (Kushnick et al., Reference Kushnick, Fessler and Zuska2016). Although marriages to genealogically defined impal have become less common, Karo continue to marry other categories of relatives provided clan exogamy is maintained, and spouses may be socially classified as impal through clan relationships even when they are not genealogically close impal. Since Karo customary law prohibits marrying an impal if one’s sibling has already done so (Singarimbun, Reference Singarimbun1975), the maximum historical prevalence was always constrained by demographic realities (Kushnick & Fessler, Reference Kushnick and Fessler2011). Contemporary Karo fertility rates are around 3–4.5 children per woman (Kushnick, Reference Kushnick2010). Such a moderate to low fertility rate would significantly limit opportunities to marry one’s impal. Although historical fertility rates are unknown, it is likely that, as has happened throughout Indonesia, fertility has declined in the modern era, reducing the number of impal available as potential marriage partners.
One of the prime goals of cultural evolutionary theory is to understand what factors and causal mechanisms drive social and cultural change (Boyd & Richerson, Reference Boyd and Richerson2024), and to explain why behaviour changes when the norms driving it remain unchanged (Boyd & Richerson, Reference Boyd and Richerson1985; Mesoudi, Reference Mesoudi2011). Within this framework, culture is treated as a system of socially transmitted information that evolves through differential learning, imitation, and payoff-biased adoption. Hypotheses about cultural change are typically tested by examining how variation in transmission pathways, ecological conditions, or social networks alters the frequency of beliefs and behaviours across generations (Henrich & McElreath, Reference Henrich and McElreath2003; Richerson & Boyd, Reference Richerson and Boyd2006). Some approaches have focused on people’s perceptions and representations, focusing on how ideas and values spread, mutate, and persist even when behaviour diverges from stated norms (e.g., González, Reference González2025). Other approaches emphasise observable behaviour, drawing on methods from human behavioural ecology and demography to study actual choices, actions, and fitness outcomes (e.g., Harris et al., Reference Harris, Boyd and Wood2021).
Adopting the first of these two approaches, in a study conducted during the same period and at the same locations as the present research, Kushnick et al. (Reference Kushnick, Fessler and Zuska2016) examined Karo perceptions of impal marriage through a factorial vignette experiment exploring the effects of demographic and socio-cultural factors on attitudes towards marrying an impal. Variables included age, gender, cosocialisation with one’s impal, availability of marriageable kin, practical alliance advantages, and church opposition. Findings revealed several unexpected patterns: cosocialisation increased both the perceived likelihood of marrying an impal and reduced disgust at the prospect; age influenced male and female respondents’ disgust judgements in opposite directions; and female respondents placed greater weight on alliance considerations than males. Although participant cosocialization served as an indirect proxy for the early-life experiences hypothesised to underlie the Westermarck Effect, it necessarily encompassed substantial variation in childhood experience, and therefore did not constitute a direct test of the hypothesis. Nevertheless, taken together, these results suggest that the decline of impal marriage cannot be fully accounted for by the Westermarck Effect – where sexual aversion develops through early negative imprinting (Fessler, Reference Fessler2007; Kushnick & Fessler, Reference Kushnick and Fessler2011). Likewise, the relatively low levels of third parties’ negative emotional reactions to impal marriage suggest that valuation of this prescription has not substantially eroded in recent times. Instead, the pattern appears to reflect a complex negotiation among multiple evolutionary conflicts: sexual conflict between men’s and women’s mating interests, parent–offspring conflict over the timing and choice of marriage partners, and sibling conflict arising when brothers and sisters have competing reproductive or alliance-related goals. Because cousin marriage involves the interests of the spouses’ parents, who are siblings, it represents a unique social domain where these conflicts intersect and may be addressed simultaneously (Chagnon et al., Reference Chagnon, Lynch, Shenk, Hames and Flinn2017; Kushnick, Reference Kushnick and Shackelford2023; Kushnick et al., Reference Kushnick, Fessler and Zuska2016).
To investigate these processes beyond perceptions alone, we melded ethnographic and epidemiological approaches to study cultural change, applying methods typically used in public health to the analysis of marriage practices. Epidemiology has long recognised parallels between disease processes and culture, whether in how culture shapes disease risk (e.g., food habits, beliefs about treatments) (Dressler, Reference Dressler2004) or in the analogy between cultural transmission and the spread of contagion, as in Sperber’s (Reference Sperber1985) ‘epidemiology of representations’. Our approach draws on a third parallel – using epidemiological research designs to study cultural practices themselves. Rather than directly examining the mechanisms through which marriage norms are transmitted across generations, we investigate the demographic and socio-cultural conditions associated with the persistence and decline of a culturally prescribed marriage practice. Our aim is therefore not to identify specific processes of social learning but to understand how changes in demographic opportunity, alliance incentives, religious influences, and other social factors are associated with changes in marriage behaviour. This complements previous work on Karo attitudes towards impal marriage by shifting attention from perceptions of the practice to the factors associated with its occurrence.
Specifically, we employed a case–control design, a method well suited to studying rare outcomes (Gamble, Reference Gamble2014; Thiese, Reference Thiese2014), to compare marriages involving impal (cases) with those that did not (controls). In case–control studies, researchers begin with individuals who have experienced a focal outcome (cases) and a comparable group who have not (controls), then work retrospectively to identify factors associated with the outcome. A classic example first identified cigarette smoking as a risk factor for lung cancer by comparing the smoking histories of cancer patients and healthy controls (Paneth et al., Reference Paneth, Susser, Susser and Morabia2004). This design allows efficient retrospective analysis of multiple risk factors without requiring large longitudinal samples or waiting for outcomes to emerge. Given the likely interdependence among variables, we paired the case–control approach with structural equation modelling (SEM) to disentangle direct and indirect effects within a unified framework (Amorim et al., Reference Amorim, Fiaccone, Santos, Santos, Moraes, Oliveira, Barbosa, Santos, Santos, Matos and Barreto2010; Beran & Violato, Reference Beran and Violato2010; Kline, Reference Kline2016). While SEM carries strong assumptions, such as potential unmeasured confounding, its exploratory power has been demonstrated in numerous studies integrating case–control designs with complex causal modelling (Gili et al., Reference Gili, Poletta, Campaña, Comas, Pawluk, Rittler and López-Camelo2013; Le et al., Reference Le, Zhang, Liu, Li, Ren and Ning2017; Lyu et al., Reference Lyu, Shi, Wang and Zhang2016; Oliveira et al., Reference Oliveira, Almeida, Silva, Assunção, Silva, Santos and Alencar2019; Porru et al., Reference Porru, Pavanello, Carta, Arici, Simeone, Izzotti and Mastrangelo2014). Applying this combined approach enabled us to quantify the social, demographic, and cultural factors shaping the decline of impal marriage in Karo society.
Methods
Data collection
From March to July 2014, we conducted fieldwork in 62 randomly selected Karo villages across 19 sub-regencies in the Karo and Deli Serdang regencies of North Sumatra (Fig. 1c). (For brevity, we refer to ‘we’ throughout this section; a credit statement at the end of the manuscript provides more details on the specific contributions of each co-author.) In each village, the village head identified individuals ever married to an opposite-sex impal (currently or previously). We interviewed the available spouse (husband, wife, or both) and then randomly selected volunteers from the same village who had never married an impal for comparison. In total, 278 individuals were interviewed: 152 who are or were married to their impal and 136 who were never married to their impal. For analysis, we created a subsample of 176 respondents by excluding those with multiple marriages, retaining only one spouse per couple, and matching the two groups by age class and gender across all sub-regencies as shown in Table 1. This comprised 88 cases (those ever married to impal) and 88 controls (those never married to impal) (Gamble, Reference Gamble2014; Thiese, Reference Thiese2014). Interviews, conducted privately and lasting 1–1.5 hours, collected information on marriage histories, impal relationships, and motivations for, and familial sentiment towards, marital choices. Data were recorded using ODK Collect (v1.3) on Android devices (Hartung et al., Reference Hartung, Lerer, Anokwa, Tseng, Brunette and Borriello2010).
Sample by gender and age class

Table 1 Long description
Counts are listed by Age Class in years old, with columns for Male impal, Male Non-impal, Female impal, Female Non-impal, and Total. For 30 or less, counts are 6 male impal, 6 male Non-impal, 9 female impal, 9 female Non-impal, total 30. For 31 to 40, counts are 13, 13, 14, 14, total 54. For 41 to 50, counts are 17, 17, 8, 8, total 50. For 51 to 60, counts are 3, 3, 8, 8, total 22. For 61 or more, counts are 4, 4, 6, 6, total 20. The Total row lists 43 male impal, 43 male Non-impal, 45 female impal, 45 female Non-impal, and 176 overall.
Probit SEM models
We are using SEM with a probit family link since the main endogenous variable – Is your marriage with a proper impal? (bion) – and others are binary (Kline, Reference Kline2016, p. 44). Probit, like logistic (aka. logit) regression, aims to predict the likelihood of a binary outcome and is expected to give similar results for substantive effects (Aldrich & Nelson, Reference Aldrich and Nelson1984; Hoffmann, Reference Hoffmann2016). Probit regression coefficients have a different interpretation from logit regression coefficients and cannot be transformed to odds ratios, so we analyse them by indicating significance (α = 0.05) and the sign of the effect and then produce marginal effects to aid interpretation. We used the lavaan package in R for SEM analyses, which drove the decision to use probit rather than logit models (Rosseel, Reference Rosseel2012).
There are two types of variables in an SEM model: endogenous and exogenous (Amorim et al., Reference Amorim, Fiaccone, Santos, Santos, Moraes, Oliveira, Barbosa, Santos, Santos, Matos and Barreto2010; Beran & Violato, Reference Beran and Violato2010; Kline, Reference Kline2016). An endogenous variable is any measured variable that can be treated as a response somewhere within the model. That is, any measured variable that has an arrow pointing into it. An exogenous variable is any variable that does not have an arrow pointing into it and is treated like an independent variable in a regression. SEM models often also contain unmeasured or latent variables, but we do not use them here. The advantage of SEM is that we can pool variance across the paths of the equations. Rather than treating each predictor as independent, the equations are structured in such a way that variance is considered along the structural paths, providing us with better estimates for the relationships of interest.
A limitation of our SEM models is that potential confounding variables were excluded. Given the constraints of the sample size, incorporating too many parameters would have risked overfitting and estimation instability (Kline, Reference Kline2016; Wolf et al., Reference Wolf, Harrington, Clark and Miller2013). While the model prioritises theoretically central variables, the omission of additional confounders means that some residual bias cannot be ruled out (Arif & MacNeil, Reference Arif and MacNeil2023; Greenland et al., Reference Greenland, Pearl and Robins1999). Spatial clustering similarly cannot be ruled out, but it is unlikely to account for the observed associations as cases and controls were drawn from the same villages. The small number of marriages sampled from each village precluded a formal analysis of spatial clustering at the village level.
To assess the models’ goodness of fit, we use two statistics: Comparative Fit Index (CFI) and Root Mean Square Error of Approximation (RMSEA; Schermelleh-Engel et al., Reference Schermelleh-Engel, Moosbrugger and Müller2003). The CFI is a revised form of the Normalised Fit Index, which asks whether the target model is an improvement over the independence model (i.e., an unspecified model). A good rule of thumb is a CFI value >0.9. The RMSEA is a descriptive measure of overall fit. Values closer to 0 represent a good fit, with values less than 0.5 indicative of models that have a close fit to the data (Xia & Yang, Reference Xia and Yang2019). The p-value presented with RMSEA tests the hypothesis that it is less than or equal to 0.05 (a cut-off sometimes used for good fit) and thus should be non-significant.
Our approach to building the combined/final SEM model (i.e., with all risk factors and risk factor groups) had elements of exploration. Our original model did not include a connection between Alliance and Cosocialization, treating these groups as independent. This model returned model fit statistics below what is considered satisfactory (CFI = 0.868; RMSEA = 0.083). We added additional links using modification indices. The largest improvement indicated by modification indices was a link from Friends (nnar2) to Marriage Type (nnr1b), with a modification index of 48.644. Rule-of-thumb estimates suggest adding links of >10 if they support theoretical positions. The added link made sense in terms of the overall causal reasoning used to build the original model (i.e., there is a strong link between unarranged marriages being between people who were friends prior to getting married), and it is the only difference between the original and the one used for final inference (see Supplementary Material). Adding this link improved model fit to above the rule-of-thumb thresholds (CFI = 0.962 [>0.9]; RMSEA = 0.045 [<0.05]).
Causal model
The foundations of an SEM model are the underlying causal assumptions (Greenland et al., Reference Greenland, Pearl and Robins1999; Kunicki et al., Reference Kunicki, Smith and Murray2023). The assumptions of our model are illustrated in the Directed Acyclic Graph (DAG), as shown in Fig. 2. It consists of several groupings of risk factors, each with two or more interconnected variables, and one or more causal links to whether the individual had married their impal (bion). The specific variables used, and their connections, are justified as follows:
• DEMOGRAPHY: Whether or not one marries their impal should be influenced by the total number of marriageable impal they have, which in turn should be influenced by the total number of opposite-sex impal they have regardless of age difference. This type of demographic argument has been central to explanations of cousin marriage rates (Kunstadter et al., Reference Kunstadter, Buhler, Stephan and Westoff1963) and is relevant for Karo communities given lower fertility rates today than in the past (Kushnick, Reference Kushnick2010).
• RELIGION: Religion can play a role in shaping whether one is able to marry a cousin. Although the religious authorities in Taneh Karo have been relatively lenient (Rae, Reference Rae1994), elsewhere in Indonesia, notably among the Manggarai of Flores, religious authorities have prohibited similar practices (Hicks, Reference Hicks1988). The religion that one practices may shape cousin marriage rates. The Protestant and Catholic churches, for instance, have different stances on, and requirements for entering, cousin marriages. When two potential impal spouses have different religious practices, it can also have an effect, such as causing familial strain and strife.
• ALLIANCE: The role of marriage in creating and maintaining alliances between descent groups has been central to explaining cousin marriage (Leach, Reference Leach1951; Levi-Strauss, Reference Levi-Strauss1949; Needham, Reference Needham1986). The necessary prerequisites for alliance-based marriage – arranged marriages, inter-familial variation in landholdings, and the potential for strategically keeping land in the family, for instance – still exist in Karo society, but there may have been changes in its importance which could have driven the decline of impal marriage (Needham, Reference Needham1978; Singarimbun, Reference Singarimbun1975). As land fragmentation, urban migration, and alternative avenues for social mobility have expanded, the instrumental value of marital alliances has weakened, potentially diminishing both the incentive and the opportunity for impal marriage.
• COSOCIALISATION: The variable cosocialisation reflects the potential influence of the Westermarck Effect – the hypothesis that individuals who grow up in close physical and emotional proximity during early childhood develop a form of sexual indifference or aversion towards one another. This could decrease the likelihood of later sexual or marital attraction between an impal pair (Fessler, Reference Fessler2007; Kushnick & Fessler, Reference Kushnick and Fessler2011). This will depend on the childhood living arrangements of the two individuals, and the sort of relationship that developed between them.
• MODERNISATION: Modernisation can reduce the rate of cousin marriage by reshaping educational and urban contexts (Shenk et al., Reference Shenk, Naz and Chaudhry2024). Greater access to schooling expands social networks beyond kin, elevates individual autonomy, and promotes norms that emphasise personal choice over family obligations. Urbanisation disperses kin networks, fragmenting traditional households and opening space for non-kin interactions, reducing the structural incentives for endogamous marriage (Greenfield, Reference Greenfield2016). Access to improved roads is a proxy measure for general mobility and access to cities. Together, these processes weaken both the supply of suitable cousins (Colleran, Reference Colleran2020) and the demand for kin-based unions, accelerating the decline of cousin marriage in many contexts.
Directed Acyclic Graph with coefficients from the final SEM model (***p < 0.001, *p < 0.05).

Figure 2 Long description
The directed acyclic graph shows various risk factors and their connections to the outcome of impal marriage (bion). The risk factors are categorized into demography, religion, alliance, cosocialization and modernization. Each factor is linked to impal marriage with specific coefficients. Demography factors include: - Same House (nnsh) with a coefficient of 1.04. - Same Village (nnsv) with a coefficient of 0.00. - Friends (nnrlb) with a coefficient of -0.07. Religion factors include: - Protestant/Other (brel2_protestantOther) with a coefficient of -0.30. - Religion Problem (nnlr) with a coefficient of -0.99*. Alliance factors include: - Spouse Land (nnld2) with a coefficient of 0.55*. - Strategic (nnaz) with a coefficient of 1.74*. - Family Land (nnla) with a coefficient of 0.71*. Cosocialization factors include: - Marriage Type (nnar2) with a coefficient of 0.56*. Modernization factors include: - percent of Roads Asphalt (kpas.s) with a coefficient of -0.02. - At Least High School (highschool) with a coefficient of 0.03. - Educational Institutions (ksch.s) with a coefficient of 0.13. Additional connections include: - Total Marriageable Impal (mimp.s) with a coefficient of 0.37*. - Total impal (bimp.s) with a coefficient of 0.60*. The graph uses different colors to represent each category of risk factors and outcomes.
Results
Descriptive statistics
Tables 2 and 3 summarise bivariate differences between impal and non-impal marriages across measured and categorical variables. Consistent with demographic expectations, individuals in impal marriages report higher mean numbers of both total opposite-sex impal and marriageable impal than those in non-impal marriages (Table 2), indicating greater exposure to eligible cross-cousin partners. In contrast, village-level modernisation indicators (school availability and road paving) show little difference between marriage types.
Descriptive statistics for measured variables

Table 2 Long description
Descriptive statistics report mean and SD for Impal, Non impal, and Total, plus Missing counts, across demography and modernisation variables. Total Marriageable Impal 1, mimp.s, has mean 2.35 and SD 1.67 for Impal, mean 1.42 and SD 1.50 for Non impal, and Total mean 1.89 and SD 1.65, with 0 missing. Educational Institutions, ksch with footnote marker 2, has mean 25.36 and SD 14.32 for Impal, mean 25.30 and SD 13.69 for Non impal, and Total mean 25.33 and SD 13.97, with 0 missing. Percent of Roads Asphalt, kpas with footnote marker 2, has mean 27.73 and SD 19.09 for Impal, mean 25.15 and SD 17.36 for Non impal, and Total mean 26.43 and SD 18.23, with 13 missing.
1 Defined as within 5 years of age of each other.
2 In kecamatan (sub-regency).
Descriptive statistics for categorical variables

Table 3 Long description
Counts are reported for categorical variables by group columns labeled Impal and Non, plus Total and Missing. Religion, Protestant or Other has 53 Impal and 61 Non, total 114, while Catholic has 35 Impal and 27 Non, total 62, with missing 0. Religion problem has 2 Impal and 11 Non yes, total 13, and 86 Impal and 77 Non no, total 163, with missing 0. Same village has 33 Impal and 26 Non yes, total 59, and 55 Impal and 62 Non no, total 117, with missing 0; same house has 3 Impal and 4 Non yes, total 7, and 85 Impal and 84 Non no, total 169, with missing 0. Friends while growing up has 54 Impal and 76 Non friend or close, total 130, and 34 Impal and 12 Non not or distant, total 46, with missing 0. Spouse land relative to average has 14 Impal and 16 Non below average, total 30; 48 Impal and 55 Non average, total 103; and 26 Impal and 17 Non above average, total 43, with missing 0. Family land purpose has 22 Impal and 6 Non yes, total 28, and 64 Impal and 79 Non no, total 143, with missing 5. Strategic purpose has 70 Impal and 39 Non yes, total 109, and 16 Impal and 49 Non no, total 65, with missing 2. At least high school has 52 Impal and 49 Non yes, total 101, and 36 Impal and 39 Non no, total 75, with missing 0.
Clear contrasts appear in alliance-related variables (Table 3). Impal marriages are substantially more likely to be arranged, strategically motivated, and framed in terms of land retention, whereas non-impal marriages are more often described as choice-based and non-strategic. Cosocialisation measures show a mixed pattern: impal spouses are more likely to report close or friendly relationships prior to marriage but are not more likely to have grown up in the same village, suggesting that kin-based interaction rather than shared residence underpins early familiarity.
Religious affiliation itself differs little between marriage types, but respondents in non-impal marriages are considerably more likely to report religion-related problems within their marriages (Table 3). Taken together, the descriptive patterns indicate that impal marriage remains closely tied to demographic opportunity and alliance-based decision-making, while modernisation proxies show minimal differentiation. These bivariate associations motivate the multivariate causal modelling presented below.
SEM model
We evaluated sub-models, one for each explanatory grouping (demography, religion, cosocialisation, alliance, and modernisation) before fitting the full model, which combined all the sub-models into one amalgamated model. In the full model, we achieved model fit values exceeding the rule-of-thumb thresholds, with CFI = 0.96 and RMSEA = 0.04 (p = 0.61). The model results, including probit coefficient estimates and p-values, are shown in Table 4 and Fig. 2. Several effects are significant, including three of seven (43%) direct effects and five of eight (63%) indirect effects. Note that Fig. 2 includes one pipe linking two risk-factor groups – from Marriage Type (nnar2), part of the Alliance group, to Friends (nnr1b), part of the Cosocialisation group. The rationale for including this pipe is discussed in the ‘Methods’ section; the original DAG, without the additional between-group pipes, is included in the Supplementary Material.(Fig. 2)
Coefficients and additional statistics for the full causal path model

Table 4 Long description
Coefficients are listed for multiple dependent variables and their predictors, with columns for Coeff., SE, z, p, and 95 percent CI lower and upper. For impal Marriage, bion, predictors include Protestant or Other, brel_2_prot, coefficient minus 0.30, SE 0.23, z minus 1.34, p 0.18, CI minus 0.75 to 0.14; Religion Problem, nnlr, minus 0.99, SE 0.48, z minus 2.06, p 0.04, CI minus 1.93 to minus 0.05; Total Marriageable impal, mimp.s, 0.37, SE 0.08, z 4.64, p 0.00, CI 0.22 to 0.53; Friends, nnr1b, minus 0.07, SE 0.09, z minus 0.83, p 0.41, CI minus 0.24 to 0.10; Ar Least H. S., highschool, 0.03, SE 0.12, z 0.22, p 0.83, CI minus 0.20 to 0.25; and percent of Roads Asphalt, kpas, minus 0.02, SE 0.11, z minus 0.21, p 0.84, CI minus 0.25 to 0.20. Total Marriageable impal, mimp.s, is predicted by Total Oposite Sex impal, bimp.s, with coefficient 0.60, SE 0.07, z 8.32, p 0.00, CI 0.46 to 0.74. Same Village, nnsv, is predicted by Same House, nnsh, with coefficient 1.04, SE 0.65, z 1.59, p 0.11, CI minus 0.24 to 2.32. Friends, nnr1b, is predicted by Same Village, nnsv, with coefficient 0.00, SE 0.14, z minus 0.01, p 0.99, CI minus 0.28 to 0.27, and by Marriage Type, nnar2, with coefficient minus 0.42, SE 0.07, z minus 6.00, p 0.00, CI minus 0.55 to minus 0.28. Strategic, nnaz, is predicted by Spouse Land, nnld2, coefficient 0.55, SE 0.23, z 2.45, p 0.01, CI 0.11 to 0.99, and by Family Land, nnla, coefficient 1.74, SE 0.40, z 4.38, p 0.00, CI 0.96 to 2.51. Marriage Type, nnar2, is predicted by Strategic, nnaz, with coefficient 0.71, SE 0.09, z 7.91, p 0.00, CI 0.54 to 0.89. At Least High School, highschool, is predicted by Educational Institutions, ksch, with coefficient 0.13, SE 0.17, z 0.78, p 0.44, CI minus 0.20 to 0.46.
The following direct effects on impal Marriage (bion) are significant: a marriage is more likely to be with an impal if an individual has a larger number of Total Marriageable impal (mimp); a marriage is more likely to be with an impal if the Marriage Type (nnar2) is arranged; and a marriage is less likely to be with an impal if there was a Religion Problem (nnlr) between the married couple.
Using the marginal effect of Total Marriageable impal (mimp.s) for the 25%, 50%, and 75% quantiles (approx. 1, 2, or 3 marriageable impal), the predicted probability of marrying impal is 0.31, 0.38, and 0.46. The other two direct effects are illustrated in Fig. 3 and described as follows: if Marriage Type (nnar2) is arranged, there is an approximately 47% chance that it is with an impal, but only a 29% chance if it is not arranged; and if there is a Religion Problem (nnlr) in the marriage, there is a 21% chance the marriage is to an impal, but if there is no Religion Problem then there is a 55% chance.
Marginal effects for Marriage Type (nnar2) and Religion Problem (nnlr).

The following first- and second-order indirect effects are significant: having a higher number of Total Opposite-Sex impal (bimp.s) means individuals are more likely to have a higher number of Total Marriageable impal (mimp.s); Marriage Type (nnar2) that arranged are more likely to occur between individuals who are Friends (nnr1b); marriages that were classified as Strategic (nnaz) are also likely to be of the Marriage Type (nnar2) that was arranged; and marriages that were made to retain Family Land (nnla) or gain through Spouse Land (nnld2), are also likely to be Strategic (nnaz).
Discussion
We used a case–control design combined with SEM to identify demographic, socio-cultural, and modernisation factors associated with the likelihood of impal (matrilateral cross-cousin) marriage in Karo communities in North Sumatra, Indonesia. The analysis revealed three key pathways: (1) a demographic chain linking the number of total impal to the number of marriageable impal, which in turn predicts the likelihood of impal marriage; (2) an alliance chain linking property-related motivations to strategic and arranged marriage decisions, which are strongly associated with impal marriage; and (3) a religion pathway, whereby the presence of religious problems between spouses is negatively associated with impal marriage. Indicators of modernisation, education, and infrastructural development showed no significant effects. Together, these findings suggest that the decline of impal marriage can be understood primarily as the product of demographic contraction and the partial erosion of alliance motives, rather than as a wholesale abandonment of the cultural norm itself.
Demographic constraints and the shrinking pool of marriageable kin
The strong positive association between the number of marriageable impal and the probability of marrying one’s impal underscores the demographic basis of the practice. In traditional Karo society, moderately sized sibling sets and fertility rates, often exceeding four children per woman (Kushnick, Reference Kushnick2010), created ample opportunities for cross-cousin unions. If fertility rates have declined, which we suspect but is undocumented, the pool of eligible impal would have contracted, reducing not only actual marriages but also the exposure necessary for such unions to remain culturally salient. This pattern parallels the demographic constraints mathematically modelled by Kunstadter et al. (Reference Kunstadter, Buhler, Stephan and Westoff1963), and the relatively lower fertility rates coupled with lower rates of cousin marriage in Bangladesh and India compared to Pakistan, where both fertility and cousin marriage rates are high (Shenk et al., Reference Shenk, Naz and Chaudhry2024). Demographic effects on cousin marriage extend to kinship terminology as well (Dole, Reference Dole1969). From an epidemiological perspective, this represents a decline in ‘exposure opportunities’, the equivalent of fewer potential ‘contacts’ through which the practice can be transmitted.
This finding complements earlier ethnographic observations that the Karo system of asymmetric alliance, while ideologically stable, has always been demographically fragile (Needham, Reference Needham1978; Singarimbun, Reference Singarimbun1975). Here, ‘ideologically stable’ refers to the persistence of the cultural prescription favouring impal marriage rather than to the frequency with which such marriages occur. Throughout the paper, we distinguish between the cultural ideal and its observed prevalence. The latter may decline because of demographic or social constraints even while the former continues to shape expectations and evaluations of appropriate marriage partners. This distinction is reflected in contemporary Karo practice, where individuals may continue to recognise and value impal relationships in a broader classificatory sense even as marriages to genealogically defined impal have become less common. The prohibition against marrying an impal when a sibling has already done so (Kushnick & Fessler, Reference Kushnick and Fessler2011; Singarimbun, Reference Singarimbun1975) further constrained prevalence. Our model quantifies this long-theorised demographic limitation: a reduction in total impal translates statistically into a reduced number of marriageable impal, which in turn lowers the probability of impal marriage. Demography, rather than changing values, thus appears to be a principal proximal driver of the decline.
Persistence and weakening of alliance motives
Despite the overall decline, alliance-related variables remain significant predictors of impal marriage. Marriages undertaken for strategic reasons or to retain or gain property were more likely to be arranged and, in turn, more likely to involve impal partners. This finding reinforces the continuing, though diminished, importance of alliance logic in Karo marriage systems. The impal tie has historically served as a mechanism to maintain the reciprocal flow of goods, labour, and ritual obligations between kalimbubu (wife-givers) and anakberu (wife-takers) (Leach, Reference Leach1951; Levi-Strauss, Reference Levi-Strauss1949; Needham, Reference Needham1986). Our data suggest that this structural logic persists in some contexts, particularly where property retention remains salient, but that its functional scope has narrowed.
Although most rural Karo communities engage in agriculture to some degree (Kushnick, Reference Kushnick2010), many households are increasingly integrated into non-agricultural economies, and landholdings are fragmented by inheritance, sale, and conversion. Under these conditions, the economic incentives for arranged or alliance-oriented marriages diminish (Schaffnit et al., Reference Schaffnit, Page, Lynch, Spake, Sear, Sosis, Shaver, Alam, Towner and Shenk2023). Yet the statistical persistence of alliance effects indicates that such motivations retain social legitimacy even as their material foundations weaken. This is consistent with cultural evolutionary theory’s prediction that norms may outlive the ecological or economic conditions that originally favoured them (Boyd & Richerson, Reference Boyd and Richerson1985; Mesoudi, Reference Mesoudi2011). The weakening of alliance motives therefore represents not the disappearance of the cultural schema but its adaptive relaxation in the face of new constraints.
Religion and moral recalibration
The negative association between religious problems and impal marriage highlights the complex role of Christianity in reshaping marriage ideals. While the Karo Protestant and Catholic Churches have historically tolerated cousin marriage to varying degrees (Rae, Reference Rae1994), Christianity may nonetheless have contributed to a moral and emotional ambivalence towards kin-based unions. Our data suggest that impal marriages are less likely when couples report religion-related problems. Such problems may arise through several pathways, including disapproval from clergy, kin, or the broader congregation; tensions associated with conversion between denominations; or differences in religious affiliations between spouses’ parents. This pattern differs from regions in Indonesia such as Flores, where religious authorities explicitly banned cousin marriage (Hicks, Reference Hicks1988), but instead suggests a gradual normative recalibration within Karo Christianity in which impal marriage is not forbidden but may nonetheless become socially or morally problematic.
Because our measure captured only respondents’ reports of religion-related problems within the marriage, we cannot determine whether these arose during courtship, family negotiations over marriage, pressure from religious authorities, conversion between denominations, or conflicts that emerged after marriage. Nor can we determine whether the reported problems reflected temporary obstacles that were ultimately resolved, or more persistent tensions within the relationship. The present findings therefore identify religion-related conflict as an important correlate of marriage type but do not permit finer discrimination among the social processes through which religion may influence marriage decisions.
Interestingly, the direct variable for religious denomination (Protestant/other versus Catholic) was non-significant, suggesting that the observed association is unlikely to reflect denominational affiliation alone. Rather, it is the experience of religion-related conflict, regardless of source, that distinguishes impal from non-impal marriages. Religion, then, functions less as a source of top-down prohibitions and more as a cause of social friction that may subtly discourage cousin marriage by generating discord or reputational costs. It is also possible that religious affiliation serves as a basis for social cooperation and alliance formation, making religious and kinship ties partially overlapping rather than independent domains. Evaluating this possibility would require more detailed data on congregational relationships and social networks than were available in the present study.
Limited role of modernisation
The absence of significant effects for modernisation indicators (i.e., educational attainment and road infrastructure) was initially unexpected, given modernisation theory’s emphasis on education and urbanisation as drivers of cultural change (Greenfield, Reference Greenfield2016). One possible interpretation is that modernisation in the Karo highlands is now sufficiently widespread that the remaining variation among villages is too small to generate detectable differences. Alternatively, cousin marriage may have declined earlier in the modernisation process, leaving a residual norm that persists independently of current infrastructural conditions.
Another possibility is that modernisation exerts its influence indirectly through the very mechanisms already captured in the model (e.g., declining fertility, reduced land dependence, and altered alliance patterns), rendering its direct statistical effect redundant. These results, thus, caution against assuming that modernisation acts as an independent causal force. Rather, modernisation may operate through the reconfiguration of demographic and alliance systems, supporting the argument that the erosion of impal marriage is best understood as a downstream consequence of broader socioecological shifts. In a sense, conceptualising modernisation as a direct effect on cousin marriage, rather than as a higher-level effect that exerts its influence on the other variable blocks, is a shortcoming of our model. The effects of modernisation are unlikely to be uniform across societies, as cultural institutions, kinship systems, and local historical circumstances may shape both the form and magnitude of its influence on marriage practices. The limited effects observed in Karo communities should therefore not be taken to imply that modernisation plays a similarly constrained role in other cultural contexts.
Cultural evolutionary and methodological implications
Although we frame our study within cultural evolutionary theory, the present analyses do not directly examine the mechanisms through which marriage norms are transmitted across generations. Rather, they identify the demographic and socio-cultural conditions associated with changes in marriage practices. By focusing on these conditions, our aim is to explain why a culturally prescribed practice may become less common over time, rather than to identify the specific processes of social learning through which culturally prescribed marriage practices are acquired, maintained, or modified. By framing cultural change in epidemiological terms, this study illuminates how demographic contraction and structural transformation can jointly reduce the ‘transmissibility’ of a cultural practice. The analogy to epidemiology is more than metaphorical: cultural practices, like pathogens, depend on exposure, susceptibility, and transmission pathways (Sperber, Reference Sperber1985). As the number of potential impal declines and alliance-based incentives wane, the ‘reproductive number’ of the practice falls below the threshold needed for its persistence. Yet, crucially, the normative preference for impal marriage remains widely acknowledged among the Karo (Kushnick et al., Reference Kushnick, Fessler and Zuska2016). The behaviour thus declines not because the norm is rejected, but because the ecological and social conditions for its enactment have eroded, a pattern akin to a population achieving herd immunity through environmental change rather than deliberate avoidance.
In addition to cultural evolutionary theory, other perspectives may also help explain the observed decline in impal marriage. From a demographic perspective, the shrinking pool of marriageable impal may itself account for much of the decline, independent of changes in cultural prescriptions (Kunstadter et al., Reference Kunstadter, Buhler, Stephan and Westoff1963). Likewise, developmental or life-course perspectives suggest that marriage preferences may be calibrated to prevailing social conditions during development (West-Eberhard, Reference West-Eberhard2003). Such processes could contribute to differences between cohorts in marriage preferences and behaviour, potentially producing a lag between changing practices and changing cultural expectations. Indeed, results from Kushnick et al. (Reference Kushnick, Fessler and Zuska2016) hint at such a possibility, as, although third-party observers do not display markedly negative reactions to impal marriage, such reactions are more evident among young men than among middle-aged and older participants.
From a human behavioural ecological perspective, changing socioeconomic conditions may have altered the costs and benefits of alliance-based marriage, leading individuals and families to adopt different marriage strategies. Historically, cousin marriage optimised inclusive fitness by preserving resources within kin groups and reinforcing cooperative ties (Chagnon et al., Reference Chagnon, Lynch, Shenk, Hames and Flinn2017; Dalzero et al., Reference Dalzero, Beheim, Kaplan, Stieglitz, Hooper, Ross, Gurven and Lukas2024, Reference Dalzero, Ross and Lukas2023; Guillot et al., Reference Guillot, Hazelton, Karafet, Lansing, Sudoyo and Cox2015). As market integration and migration diversify resource flows, the fitness payoffs of such marriages may diminish. Under a human behavioural ecological interpretation, individuals and families may increasingly adopt alternative strategies, and these are likely to vary by sex and other factors (Chagnon et al., Reference Chagnon, Lynch, Shenk, Hames and Flinn2017; Kushnick et al., Reference Kushnick, Fessler and Zuska2016).
The above interpretations are not mutually exclusive, and the present findings do not permit us to adjudicate among them. Rather, they identify demographic and socio-cultural factors that any satisfactory explanation of the decline of impal marriage will need to accommodate.
Methodologically, this study demonstrates the utility of combining a case–control design, traditionally used in epidemiology, with SEM to explore complex causal networks relevant to cultural evolution. The case–control design efficiently addresses rare cultural outcomes such as impal marriage, while SEM allows simultaneous estimation of direct and indirect pathways among interdependent variables. Although this approach does not directly identify the mechanisms through which cultural practices are transmitted, it provides a framework for evaluating competing hypotheses and quantifying the demographic and socio-cultural conditions associated with their persistence or decline. Limitations include the cross-sectional nature of the data, potential recall bias in self-reports, and the exclusion of some confounders due to sample-size constraints (Kline, Reference Kline2016; Wolf et al., Reference Wolf, Harrington, Clark and Miller2013). Future research could address these limitations through longitudinal designs, demographic simulations, or agent-based models that explicitly examine how demographic, social, and cultural processes interact over time.
Conclusion
In summary, the decline of impal marriage among the Karo appears to stem less from ideological rejection than from demographic and structural change. The shrinking pool of marriageable kin, the weakening of alliance-based incentives, and the subtle discouragement introduced by Christianisation together reduced the opportunity and desirability of impal unions. By integrating epidemiological reasoning with cultural evolutionary and human behavioural ecological perspectives, this study illustrates how social practices may decline not because they are culturally repudiated, but because their ecological and social ‘hosts’ have diminished. In this sense, the waning of impal marriage reflects the adaptive transformation of Karo kinship in response to modern demographic and economic realities. More broadly, our findings illustrate how epidemiological approaches can help identify the demographic and socio-cultural conditions associated with cultural change, providing a useful framework for investigating the persistence and decline of culturally prescribed practices.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/ehs.2026.10071.
Data Availability Statement
For more details about the project, see https://karoimpal.wordpress.com/. The data, a variables codebook, and Supplementary Material, which provides additional methodological details and R code, are available at: https://doi.org/10.5281/zenodo.22759971.
Acknowledgements
Thanks to the Karo people who shared intimate details of their lives with us, and the research assistants who helped in the field. A version of this paper was presented at the Cultural Evolution Society 2024 Annual Conference in Durham, UK. Fieldwork permits were provided by Indonesia’s Ministry of Research and Technology (RISTEK), with help from the American-Indonesian Exchange Foundation (AMINEF). Human subjects approval was granted by the Human Subjects Office at the University of Washington, Seattle (with approval to use those permissions from the Ethics Office at the Australian National University, Canberra). All participants provided informed consent before being enrolled in the study.
Author contributions
Writing was done by G.K. with help from S.P. Editing and manuscript approval were done by G.K., S.P., F.Z., K.K., and D.M.T.F. Research design was done by G.K. and D.M.T.F. Analytical design was done by G.K. and S.P. Analyses were carried out by S.P. with help from G.K. Fieldwork was done by G.K. with help from K.K. Research permits were arranged by G.K. and F.Z.
Financial support
Funding was provided by grants to G.K. from the Fulbright Scholars Progam, American Institute for Indonesian Studies (AIFIS), and ANU’s Research School of Humanities and Arts, and a grant to D.F. and G.K. from UCLA’s Center for Culture, Brain, and Development. S.P. was supported by the Evolution of Cultural Diversity Initiative at the ANU.
Competing interests
The authors declare no conflicts of interest.



