1 s2.0 S0305750X1930316X Main

World Development 125 (2020) 104668
Contents lists available at ScienceDirect
World Development
journal homepage: www.elsevier.com/locate/worlddev
Interethnic and interfaith marriages in sub-Saharan Africa

Juliette Crespin-Boucaud
Paris School of Economics (PSE), Office R6-01, 48 Boulevard Jourdan, 75014 Paris, France
a r t i c l e i n f o a b s t r a c t
Article history: This paper documents interethnic and interfaith marriage patterns to better understand which identity-
Accepted 28 August 2019 related cleavages matter in sub-Saharan Africa. Using Demographic and Health Surveys (DHS) spanning
Available online 19 September 2019 15 countries, I build a representative sample of women born between 1955 and 1989. Extrapolating to
the population of these countries, I find that 20.4% of marriages are interethnic and 9.7% are interfaith,
JEL classification: indicating that ethnic and religious differences are not always barriers. Accounting for diversity levels,
J15 both shares are actually similar. Regarding the pooled sample of these 15 countries, the share of intereth-
J12
nic marriages increased, and there is no country where interethnic marriages became less frequent. The
Z12
N37
share of interfaith marriages decreased in the pooled sample. Only in Cameroon did interfaith marriages
become more frequent. The share of Muslim-Christian marriages remained stable in the pooled sample.
Keywords: The increase in the share of interethnic marriages can only partly be explained by increases in urbaniza-
Ethnicity tion and education levels, suggesting that changes in preferences and in social norms may also be at play.
Marriage The decrease in the share of interfaith marriages is due to decreasing levels of religious diversity: tradi-
Religion tional religions were replaced by Islam and Christianity. These results show that some ethnic boundaries
Sub-Saharan Africa
became more porous whereas religious boundaries did not. However, religious boundaries shifted as a
result of changes in the religious landscape.
Ó 2019 Elsevier Ltd. All rights reserved.
1. Introduction we would expect interethnic marriages to be rare in societies where

ethnic cleavages are rigid: marrying within one’s group is a means for
Social identity is an individual characteristic that has long been an individual to ensure that her/his identity is passed down to her/his
demonstrated to be complex and multi-layered (Posner, 2005). children.3 Intermarriages have long been used to measure the
However, a large fraction of the literature on sub-Saharan Africa strength of cleavages within societies (Kalmijn, 1998), as they com-
has relied on a unidimensional view of identity, equating identity bine a measure of segregation (who meets whom and where) and a
with ethnicity. Even though surveys and censuses in most countries measure of who is thought to be an acceptable spouse. However,
now include categories for ‘‘mixed race” or ‘‘mixed ancestry”, very there is no quantitative evidence on interethnic marriages in the case
few large-scale surveys conducted in African countries include such of sub-Saharan Africa. This paper aims to fill this gap.
an option, perpetuating the idea that ethnic identity is the alle- I study interethnic and interfaith marriages in 15 countries in
giance to one group or tribe, and to one homeland. Moreover, while sub-Saharan Africa using data from the Demographic and Health
the impact of ethnic fractionalization has been investigated,1 the Surveys (DHS). This paper has four aims: providing descriptive
manner in which ethnic identities are formed and maintained has statistics on interethnic and interfaith marriages, discussing results
received little attention. As ethnicity is transmitted through descent,2 at the extensive margin (marrying outside one’s ethnic group) and
the intensive margin (how far outside one’s ethnic group?), assess-
E-mail address: [email protected] ing time trends, and analyzing which factors have contributed to
1
Since the seminal paper of Easterly and Levine (1997), several works have pointed the time trends. Contrasting interethnic with interfaith marriages
out the detrimental effects of ethnic diversity on growth and public good provision provides a broader picture of intermarriage patterns in African
(Alesina & La Ferrara, 2005; Churchill & Smyth, 2017; de la Cuesta & Wantchekon,
2016; Gershman & Rivera, 2018; Gören, 2014), with only a few surveys reaching
countries, but it must be emphasized that religious identity is more
different conclusions (Gisselquist, Leiderer, & Nino-Zarazua, 2016).
2 3
Chandra (2006) defines ethnic membership as ‘‘determined by attributes Bisin and Verdier (2000) develop a model of transmission of identity across
associated with, or believed to be associated with, descent”. Kanbur, Kumar generations and find that intermarriages result in minority-group parents having
Rajaram, and Varshney (2011) summarize debates over definitions of ethnic identity access to a weaker socialization technology. In sub-Saharan Africa, ethnic fraction-
and of related measures. In this paper, I define ethnic groups based on DHS alization does not necessarily result in majority-minority settings. I nonetheless
classification, and I do not discuss how intermarriages could have influenced or expect the theoretical result put forward to hold: high shares of intermarriage should
determined these classifications. be associated with weaker ethnic/religious affiliations for parents and their children.
https://doi.org/10.1016/j.worlddev.2019.104668
0305-750X/Ó 2019 Elsevier Ltd. All rights reserved.
2 J. Crespin-Boucaud / World Development 125 (2020) 104668
fluid than ethnic identity, as conversion allows individuals to The rest of the paper is organized as follows: Section 2 presents
change their religious affiliation. the data. Section 3 lists factors that could explain the prevalence of
First, I find that 20.4% of married women are in an interethnic intermarriages. Section 4 presents the empirical strategy used.
union, contrasting with 9.7% being in an interfaith union. Intereth- Section 5 reports results on the pooled sample and Section 6
nic unions are hence far from rare events in sub-Saharan countries, results at the country-level. Section 7 tests alternative stories and
and their share ranges from 10.4% in Burkina Faso to 46% in Zam- provides robustness checks on the findings. Section 8 concludes.
bia. Interfaith marriage shares range from 1.8% in Niger to 19.3% in
Cote d’Ivoire. Second, using a sample of women born between 1955 2. Data
and 1989, I find that interethnic marriages became more common
for later-born cohorts relative to earlier-born ones, while interfaith In this section, I present the data sources used and explain how
marriages became less common. There is no country where the the sample is built.
share of interethnic marriages decreased, and interfaith marriages
increased only in Cameroon. Third, building on recent research on
2.1. Data sources: DHS and Ethnologue
how to measure ethno-linguistic diversity (Desmet, Ortuño-Ortín,
& Wacziarg, 2016; Gershman & Rivera, 2018), I compute new lin-
I use Demographic and Health Surveys (DHS) that were imple-
guistic distance measures that allow me to take into account diver-
mented in sub-Saharan Africa (surveys used listed in Table 7,
sity within and across countries. In the case of interfaith marriages,
Appendix A). DHS questionnaires procure information on who is
I do not use a distance measure but instead study Muslim-
married4 to whom (within the household), and in specific countries
Christian marriages separately, as this type of union is arguably
and waves, these questionnaires also procure information on
the most distant kind of interfaith marriage in sub-Saharan Africa.
respondents’ ethnic and religious identity. The descriptive sample
I find that changes at the extensive margin do not translate into
includes the 25 countries with information on ethnic identity.5 The
changes at the intensive margin. Interethnic marriage shares
main sample is made up of 15 countries for which there are at least
increased, but there is no clear pattern regarding variation in lin-
two survey waves that gather information on the ethnic and reli-
guistic distance. Interfaith marriage shares decreased, but
gious identity of respondents: Benin, Burkina Faso, Cameroon, Cote
Muslim-Christian marriage shares remained stable. Fourth, I exam-
d’Ivoire, Gabon, Ghana, Guinea, Kenya, Malawi, Mali, Niger, Senegal,
ine whether time trends on intermarriage shares can be explained
Uganda, Togo, and Zambia. The main sample is made up of women
by increased education and urbanization levels. To do so, I compare
born between 1955 and 1989 and of their husbands.
time trends across specifications with controls and without con-
Additionally, I exploit the Ethnologue dictionary (Simons &
trols. The results for interethnic marriages point at the fact that,
Fennig, 2017) to get information on the classification of each ethnic
while education and urbanization play a role in the increase of
group’s traditional language. I use these classifications to compute
interethnic marriage shares, part of the increase could come from
the linguistic distance of all of the pairs of ethnic groups. For each
changes in norms and preferences about interethnic marriages.
pair, I identify the lowest common linguistic node that they share
Likewise, I find that urbanization and education are not the key dri-
and compute the number of nodes between each group and the
vers of the decrease in interfaith marriages: this decrease is mostly
common node. The mean of these two distances is the linguistic
due to the decreased levels of religious diversity over time. This
distance of this pair (detailed methodology in Appendix B).
study of intermarriages finds that some – though not all – ethnic
boundaries became more porous. Religious boundaries did not
become more porous, but the religious landscape changed as tradi- 2.2. Comparability over time: reweighting and recoding
tional religions were replaced by Islam and Christianity. Finally, I
confirm that my results are robust to varying definitions of inter- The main sample includes at least two data waves for each
marriages. I also test the hypothesis that spouses become more country, thus raising issues about comparability over time. I
similar as the length of their marriage increases. Ethnic ‘‘assimila- explain briefly the steps taken to ensure that I can identify time
tion” does not drive the results. However, there is evidence of con- trends using this sample (the online Appendix details the processes
version during marriage: my estimate is a lower bound on the used in this study).
decline of interfaith marriages. The main sample is made up of women born between 1955 and
This paper contributes to three strands of the literature. First, this 1989: for each year within this period, the sample includes women
paper extends the empirical literature on intermarriages (Fryer, from all of the 15 countries of the main sample. I reweight the sam-
2007; Furtado & Theodoropoulos, 2011; Kalmijn & Van Tubergen, ple to make it representative of the population of married women
2006; Monden & Smits, 2005; Qian & Lichter, 2007, 2011). Second, in each country. Reweighting and selecting the time period 1955–
it contributes to a growing literature that nuances or contests the 1989 ensure that the share of each country remains (roughly) con-
idea that ethnicity is always the key cleavage in sub-Saharan Africa. stant over time. Changes over time are hence not due to changes in
Contributions have suggested that which identity category is salient the respective weights of countries in the sample over time.
depends on the context (Eifert, Miguel, & Posner, 2010; Miles & I recode both ethnic and religious categories to build a classifi-
Rochefort, 1991). Looking at the micro-level literature, Berge et al. cation that fulfills two criteria. First, the classification does not vary
(2018) show that there is little evidence of co-ethnic bias in behavior within a country. Second, for all of the cohorts and survey waves,
games, contradicting results based on Implicit Association Tests all of the groups listed in this classification have a least one mem-
(IAT) (such as Habyarimana, Humphreys, Posner, & Weinstein ber of each gender. Grouping in fewer categories mechanically
(2007) in Uganda; Lowes, Nunn, Robinson, & Weigel (2015) in 4
Throughout the paper I use the terms ‘‘marriage,” ‘‘spouse,” ‘‘husband,” and
DRC). The complexity of relationships between ethnic groups (at
‘‘wife” to refer to married couples as well as to cohabiting couples.
the political level) was emphasized by Francois, Rainer, and Trebbi 5
These 25 countries are the 15 countries from the main sample, plus Central Africa
(2015) and Mozaffar, Scarritt, and Galaich (2003) regarding electoral Republic, Chad, the Republic of the Congo (Congo-Brazzaville), the Democratic
coalitions and power sharing. Simson (2018) shows that once educa- Republic of the Congo (DRC), Ethiopia, Liberia, Mozambique, Namibia, Nigeria, and
tion is controlled for, public sector jobs in Kenya and in Uganda are Sierra Leone. The countries included in this study are not a random sample of African
countries. Including a question on the respondent’s ethnic identity is not a decision
rather equitably distributed along ethnic lines. Third, this paper adds made independently from whether ethnicity matters in a country: the sample above
to the literature comparing the evolution and salience of ethnic and cannot be considered to be representative of countries that did not include such a
religious cleavages (McCauley, 2014). question in DHS.
J. Crespin-Boucaud / World Development 125 (2020) 104668 3
decreases the number of unions appearing as interethnic/inter- intermarry, such as one’s occupation. Using this variable thus results
faith, so a time-invariant classification is needed to measure in overestimating the relationship between urban residence and
changes over time. After recoding, ethnic classifications are specific intermarriage.
to each country and include less than 10 categories for most coun-
tries.6 The category ‘‘other (ethnicity)” groups together members of 3. Intermarriages and marriage markets: preferences, norms,
ethnic groups that were not listed in all waves, people who did not and diversity
identify with a specific group, and foreign nationals. Recoding reli-
gious classifications makes apparent a key change in the religious I provide in this section a framework for interpretation of the
landscape: the surge of Pentecostalism in Africa (Mayrargue, 2004; models and of the results presented in the paper.
Meyer, 2004). Changes in classification are likely to reflect the
agenda of church leaders, as ‘‘new Churches” have an interest in
3.1. Types of factors influencing intermarriage shares
being formally recognized in order to proselytize, which is not the
behavior of faiths less invested in proselytizing, such as traditional
In his seminal paper, Kalmijn (1998) distinguishes three factors
religions. Because new faith groups continue to be listed, harmoniz-
that could explain the prevalence of intermarriages: individual
ing nomenclatures across waves requires a high level of aggregation:
preferences, diversity levels within (local) marriage markets, and
Christian, Muslim, other. The category ‘‘other (faith)” includes fol-
the influence of norms and of third parties.
lowers of traditional religions, atheists, and members of new reli-
The individual preferences factor gathers all of the preferences
gious movements that cannot be linked to Christianity or Islam.
that individuals have concerning their matches on the marriage
Among the women who belong to the ‘‘other (faith)” group, at least
market. Two main characteristics of matches on the marriage mar-
41.8% identify with a traditional religion. It is a lower bound on their
ket are socio-economic resources and cultural resources: people
share, as many survey waves do not distinguish traditional religions
are likely to want to marry someone whose economic prospects
from other faiths that do not belong to Christianity or Islam.
are good and with whom they share values and preferences.
The diversity level factor encompasses all of the channels related
2.3. Variables: intermarriages and individual characteristics to how diverse marriage markets are. Some societies are highly
heterogeneous, and others are more homogeneous, for instance if
To study intermarriages, I build variables that measure there is a majority group. Moreover, spatial segregation affects
intermarriages as well as variables that are likely to influence the how diverse local marriage markets are. Low levels of diversity
likelihood of intermarriage. are associated with low levels of intermarriages (by sheer limita-
Ethnic and religious identity are self-declared in the DHS: tion due to the numbers of potential spouses from other groups).
I hence consider that the respondent’s answer is a measure of The third parties/norms factor includes the channel of group
her/his ‘‘true identity”. A marriage is interethnic (interfaith) if the identification, the one of group sanctions, and in the case of the set-
spouses’ answers correspond to different ethnic (faith) categories ting studied, the fact that members of one’s kin may be directly
in the recoded classification. Ethnic and religious identity cate- responsible for choosing one’s spouse. Field studies have shown
gories may be fluid and change, especially in the case of conversion transitions from kin-selected to self-selected marriages, for instance,
for marriage. I discuss how religious conversion and ethnic Bertrand-Dansereau and Clark (2016) (Malawi) and Clark, Kabiru,
‘‘assimilation” might affect my results in Section 7. and Mathur (2010) (Kenya): elders and parents are less involved
I consider two main variables that can lead to intermarrying: in the matching process. Third parties influence is likely to work
education and urban residence. The DHS include little retrospec- against intermarriages (Spörlein, Schlueter, & van Tubergen, 2014).
tive information, so I cannot reconstruct the individual character-
istics at the time that the marriage started. Marriage decisions 3.2. DHS variables and the Kalmijn framework
are taken based on the characteristics of individuals but also on
expectations, such as joining a spouse in the city or being able to As indicated in Section 2, I use two variables that are likely to
graduate high school. I use characteristics at survey date to proxy influence the likelihood of intermarriage: education and urban res-
for past characteristics and expectations: current characteristics idence. These variables capture aspects of the types of factors listed
do not perfectly correspond to past characteristics but allow me above.
to take into account (realized) expectations. I use information on Education could affect individual preferences through several
the highest completed level of education at survey date, which channels. Education, especially secondary and higher, is in many
should be a good proxy of the level of education at the time that countries conducted in a vehicular language, thus helping to
the union was formed.7 I use urban residence at the survey date. remove language barriers in marriage markets. Additionally, by
Migration mostly takes place from rural to urban areas8: the current transmitting a common culture, education could switch prefer-
place of residence captures some unobserved characteristics of ences away from group identification and towards a national iden-
individuals that might also be correlated to their propensity to tification. Moreover, higher education takes place in (mixed) urban
settings: that corresponds to the diversity level factor. Educated
6
Depending on the countries, ethnic classifications became more or less detailed.
women might have more of a say in the choice of their spouse:
For instance, Akan subgroups are listed separately in Ghana in the older survey waves
but are only listed as ‘‘Akan” in the recent survey waves. However, the reverse
third parties may be less involved in the matching process.
phenomenon happened in Kenya, where groups listed together (Meru/Embu) are Urban areas are on average more mixed than rural ones: diver-
listed separately in more recent survey waves. sity levels are likely to be higher in cities than in the countryside,
7
It is unlikely that women can stay in school after getting married. Considering and marriage markets are likely to be less segregated. Social norms
only women in union who have attended primary school, I find that, under the
may be more accepting of intermarriages in cities than in rural
assumption that girls start school at age 8, 69% of women started their first cohabitant
union at least two years after completing their schooling, 21% around the same time areas.
as they completed their schooling, and 10% before that.
8
Among women for whom I have information on childhood place of residence, 3.3. How did marriage markets change?
12.6% of women who live in a rural area at the time of the survey grew up in an urban
area, whereas 41.3% of women living in an urban area grew up in a rural area. As this
sample consists of women belonging to earlier-born cohorts, these figures may be My main sample is made up of women born between 1955 and
even higher for later-born cohorts. 1989. During this period, several changes linked to the variables
listed above took place. Fig. 1 shows a visual representation of the In the case of ethnicity, I account for these differences by
changes that I can observe in the data. Its left panel shows that using linguistic distance measures (detailed methodology in
education levels as well as urbanization increased: these changes Appendix B). Computing the country-specific random linguistic
could lead to higher rates of intermarriages for later-born cohorts distance (dc ) is done according to the formula below:
than for earlier born cohorts.
X
n X
n
dc ¼
Moreover, another key change took place: the decrease in the
pwi pmj dij
religious diversity of the population. The right panel of Fig. 1 shows i¼1 j¼1
the decrease in the share of people identifying with faiths other
than Islam and Christianity and the associated decrease in religious where dij is the linguistic distance between group i and group j; pwi
diversity. Under the assumption that people meet on a national the share of married women who belong to the group i, and pmj the
marriage market and that no other factors affected interfaith mar- share of married men who belong to the group j. The linguistic dis-
riages, the decrease in religious diversity will mechanically result tance dii is set to 0. The linguistic distances dzi and diz , where z is the
in lower interfaith marriage shares. Under the same assumptions, group ‘‘other (ethnicity)”, are set to the median value of the linguis-
the share of interethnic marriages should remain stable, as there tic distance between pairs of ethnic groups listed in country c.
is no sizable change in ethnic diversity levels.9 In the case of religious differences, there is no standard way to
quantify the differences between groups. I study separately inter-
faith marriages – mixed marriages between Muslims, Christians,
4. Empirical strategy
and ‘‘other (faith)” – and Christian-Muslim marriages, the latter
arguably being the most distant kind of intermarriage along the
In this section, I present measures of ethnic and religious diver-
religious dimension in sub-Saharan Africa.
sity and introduce the specifications used to measure time trends.
4.2. Assessing time trends

4.1. Descriptive statistics: comparisons across countries and across
identity categories I design a simple additive model to estimate trends on inter-
marriages. My baseline specification is a linear probability model11
Cross-country comparisons must be done carefully in order to that I run first on the pooled sample (including country fixed-effects)
be meaningful: if intermarriage shares are low in a given country, and then at the country level. I present below the specifications used
it could be that, according to the classification used, there is not at the country level. I specify the model at the woman level12: indi-
much diversity in this country to begin with. As discussed in Sec- viduals (w) are women. This decision is due to the fact that 17% of
tion 2 in the case of Pentecostal churches, the classification chosen women appear with at least one co-wife in the dataset: a woman
in each survey cannot be considered as encompassing the same appears only one time in the dataset, but a man appears matched
reality across countries and identity categories. To neutralize this with all of his cohabiting wives.
classification effect, I compute the share of intermarriages that
we would observe if individuals were matched randomly in a 4.2.1. Model 1: time trends
national marriage market. This share measures how diverse each
country is. It is equivalent to computing an ethnic fractionalization
index (EF) and a religious fractionalization index (RF) for each Intermarriagew ¼ a þ b1 BirthYear w þ b2 Agew þ b3 Age2w þ w ð1Þ
(national) marriage market. More formally, this fractionalization
measure (F c ) corresponds to: Intermarriagew is an indicator variable that equals 100 if the
union is interethnic (interfaith), and 0 otherwise.13 I consider
X
n
unions to be interethnic (interfaith) if spouses do not belong to the
Fc ¼ 1 pwi pmi
same group. When both spouses belong to the group ‘‘other,” I con-
i¼1
sider them to be in an intraethnic (intrafaith) union. In Section 7,
In this equation, c denotes countries. n is the number of ethnic I test whether the results are robust to considering these unions as
(respectively religious) groups in the survey. Subscripts w denote interethnic (interfaith).
women, subscripts m denote men. pwi is the share of married BirthYearw is a continuous variable defined as the year of birth
women who belong to the group i; pmi the share of married men of each woman. It is the main variable of interest: if the coefficient
who belong to the group i. I assume that a person’s decision to associated to it is positive, it means that the share of intermar-
marry does not depend on the ethnic and religious identity of riages has increased over time. I use birth year rather than year
her/his matches. There are no additional entries on or exits from of first cohabitation to capture time trends for two reasons. First,
the marriage market compared to what we observe in the data: birth year is available for all women, while using cohabitation year
polygamous men are thus counted as many times as the number would restrict the sample to women in their first union, as I only
of their cohabiting wives. have information on the year of first cohabitation. Second, age at
Fractionalization indices do not include any information on how marriage or year of cohabitation is endogenous to educational
(dis) similar groups are10: they are useful to identify changes at the achievements and to the type of place of residence, while year of
extensive margin but do not allow for identification of changes at the birth is more exogenous to these individual characteristics.
intensive margin. I compare birth year to cohabitation year as a robustness check
in Section 7. Agew is the age at survey date.
9
A more thorough discussion of ethnic diversity at the country-level and
11
interethnic marriage patterns can be found in the online Appendix. The results do not change when using a logit model. Results available from the
10
Since the seminal paper by Greenberg (1956), adjusting measures of fractional- author.
12
ization by a measure of similarity between groups has been suggested to take into The results do not change if I collapse the dataset to keep one observation per
account the depth of cleavages: for instance, some ethnic groups speak the same man. In this case, I consider a man to be in an interethnic (interfaith) union if at least
language (e.g. in Zambia, Posner (2005)) while others are very distant according to the one of his wives is not a member of his ethnic (religious) group. Results available from
linguistic tree (e.g. the Somali and the Kikuyu, in Kenya). These differences are not the author.
13
taken into account by fractionalization measures: the distance between all groups is The indicator variable is set to 100 so that coefficients can be read as changes in
set to be the same. percentage points.
Fig. 1. Changes over time: Education, Urban residence, Diversity levels. Sample & data: Women in union, pooled sample. 95% confidence intervals included (except for
religious fractionalization). Left panel: Share of women living in urban areas, and share of women whose highest completed education level is primary/secondary school. The
share of women living in urban areas reflects current place of residence: the magnitude of the change would be higher using information on place of birth. Right panel:
Weighted average of religious fractionalization at country-level and share of women belonging to the group ‘‘other (faith)”.
4.2.2. Age and birth year effects control further for cohort composition effects, I add a dummy vari-
I add quadratic controls for age in the model to control for age able, Remarriedw , which takes the value 1 if the respondent has
effects. This ensures that the patterns that I identify in the data are remarried. I discuss alternative ways to measure the impact of
due to change across cohorts and not to age effects.14 I use surveys remarriage in Section 7.
implemented from 1992 to 2018: women born in 1955 were older Throughout the paper, I compare the coefficient associated with
than 35 in the first DHS survey of each country; women born in birth year between the specification 1 and the specification 2. The
1985 were younger than 35. As shown in Fig. 2, the timing of survey birth year coefficient in specification 1 measures time trends. The
waves can be seen in the age composition of cohorts: earlier-born birth year coefficient in specification 2 measures time trends that
cohorts are older at the survey date than later-born cohorts. cannot be explained by changes in education levels, in urbaniza-
Whether a woman is married or not is a function of age: differences tion, and in cohort composition due to remarriage. As such, it could
in age composition of cohorts (left panel) are mirrored by differences capture changes in preferences and in social norms. However, it
in the share of married and remarried women by cohort (right should be noted that several other variables may contribute to
panel). As women in earlier-born cohorts are older at the survey individuals’ likelihood to intermarry, such as parental education,
date, they are more likely to have married and more likely to have whether one’s parents intermarried, or whether both parents are
remarried, either after a divorce or being widowed. The same char- still alive at the time of the marriage decision: the coefficient in
acteristics are likely to drive both the type of marital status that I specification 2 may also capture some of these omitted factors.
observe (married/remarried/never married) and the type of marital
outcome that I observe (intermarried or not). For instance, if women
who marry young are more likely to marry within their group, then, 5. Results on pooled sample
without age controls, I would estimate time trends that are due to
the fact that cohorts differ with respect to their age composition. This section presents the results on the pooled sample. The
country-specific results are detailed in Section 6.
4.2.3. Model 2: time trends with controls
5.1. Descriptive statistics
Intermarriagew ¼ a þ b1 BirthYearw þ b2 Agew þ b3 Age2w
Table 1 displays the estimations of observed intermarriage
þ b4 Primaryw þ b5 Secondaryw þ b6 Urbanw
shares and contrasts them with the intermarriage shares that we
þ b7 Remarriedw þ w ð2Þ would have observed under random matching. Interethnic unions
are on average more frequent than interfaith unions: 20.4% of
I introduce additional variables in the model to test whether
women are married to a man who is not from the same ethnic
they explain changes in intermarriage shares. I assume these vari-
group as them, and 9.7% of women are married to a man who is
ables have a constant effect over time. I add dummies for the high-
not a member of the same religious group as them. However, the
est education level: Primaryw and Secondaryw , the reference
number of categories and the level of diversity differ depending
category being ‘‘no education”.15 Urbanw is a dummy that takes
on whether we consider ethnicity or faith: under random match-
the value 1 if the respondent lives in an urban area. Moreover, to
ing, we would observe around 80% of interethnic marriages and
14 around 33.8% of interfaith marriages. When we look at the ratio
This is possible as my main sample is made of up of countries for which I have at
least two survey waves. I hence observe birth cohorts at different ages. Thus, 82.7% of of the observed share of intermarriages to the random share of
women belong to birth cohorts that were sampled at least twice in their country. intermarriages, interfaith marriages and interethnic marriages
Given the quadratic function that I use to estimate age effects, I do not need all of the are roughly as common: between 25% and 30% of the random share
cohorts to have been sampled twice to estimate age effects and birth years effects
of intermarriages is realized.
separately.
15
DHS classification distinguishes between secondary and higher education. Only
I find that interethnic unions take place at a linguistic distance
2.3% of married women in my sample completed university, so I aggregate secondary that is similar to what we would observe under random matching.
education and higher education into a single category. In contrast with the fact that 28.7% of interfaith unions are realized,
Fig. 2. Survey effect: Marital status and age at survey date. Left panel: The sample includes only women in union at the time of the survey. 95% confidence intervals. The
‘‘maximum age” is the age of the oldest woman surveyed for each birth year. The ‘‘minimum age” is the age of the youngest woman surveyed for each birth year. These ages
depend on the timing of surveys. Women aged 15–49 are surveyed in DHS, hence the flat lines at these two ages. Right panel: The shares of married and of remarried women
are computed using all women surveyed. 95% confidence intervals. ‘‘Married women” are the women in union at the time of the survey, not the ever-married women.
Table 1
Average intermarriage shares and linguistic distance.
Interethnic marriages (%) Linguistic distance (nodes)

Observed Random Ratio N Observed Random Ratio N
20.4 80.0 25.5 97,111 3.29 3.25 1.01 21,704
Interfaith marriages (%) Muslim/Christian marriages (%)
a
Observed Random Ratio N Observed Random Ratio N
9.7 33.8 28.7 96,549 2.4 21.7 11.2 83,291
Data & sample: Women in union, pooled sample. Weighted data.

Interethnic, interfaith, and Muslim-Christian marriage shares: The observed share corresponds to the share observed in the population. The random share
corresponds to the share that we would observe if people currently in union had matched at random, under two assumptions. First, there is no exit from or
entry into the marriage market compared to what we observe. Second, polygamy decisions are independent from women’s ethnicity and religion:
polygamous men appear on the random market the same number of time as in the observed market. Random shares are computed for each country,
considering a national marriage market. The random share for the pooled sample is the weighted average of those national random shares. The ratio is
computed as the ratio of the observed share to the random share.
Linguistic distance: The random and the observed linguistic distances are computed considering a national marriage market, using information on
interethnically married couples. Linguistic distance between two spouses when only one of them belongs to the group ‘‘other (ethnicity)” is set to the
country-specific median linguistic distance, computed on distances for all pairs of ethnolinguistic groups.
Muslim-Christian marriage shares (observed and random) are computed using only marriages in which neither spouse is member of the group ‘‘other (faith)”.
a
One survey wave in Senegal does not include a question on religion.
Muslim-Christian marriages are rare16: they make up 2.4% of mar- Table 2 shows time trends for the three kinds of intermarriages
riages when considering only Muslim and Christian respondents, studied: interethnic, interfaith, and Muslim-Christian marriages.
and 2.1% otherwise. It is 11.2% of what we would observe under ran- Interethnic marriages are more frequent for women in later-born
dom matching. Most interfaith unions hence involve a spouse who cohorts than women in earlier-born cohorts: in column (1), the
identifies as Muslim or Christian and a spouse who belongs to the coefficient for birth year is positive and significant. After control-
group ‘‘other (faith)”. Individuals who are neither Muslim nor Chris- ling for education, urbanization, and remarriage (column (2)), the
tian are more likely to be in an interfaith union than Muslims and magnitude of the coefficient decreases, but the coefficient itself
Christians. Couples that include at least one follower of ‘‘other (faith)” remains positive and significant. This result indicates that other
make up 14% of the sample, but 79% of interfaith couples, most of factors, possibly changes in norms and preferences, contributed
them taking place between a Christian spouse and an ‘‘other (faith)” to the increase in the share of interethnic marriages.
spouse. The high propensity of ‘‘other (faith)” members to intermarry Interfaith marriages decreased over time (column (3)) as
is consistent with the fact that traditional religions are more tolerant expected from the decrease in religious diversity. The share of
of intermarriages than Islam and Christianity. It is also likely that the Muslim-Christian marriages (column (5)) did not vary over time:
conversion process from a traditional religion to Islam or to a Chris- it is likely that the norms and preferences concerning interfaith
tian denomination might not concern both spouses at the same time. marriages did not change over the period. Norms concerning reli-
gious intermarriages remaining the same is consistent with the
fact that the decrease in interfaith marriages is due to the decrease
5.2. Time trends
in religious diversity documented in Section 3 rather than to norms
shifting towards less tolerance of interfaith marriages.
5.2.1. Assessing time trends
Extrapolating the estimations from the model over 25 years
Fig. 3 shows the shares of each type of intermarriage over birth
(1960–1985), the model (columns (1) and (3)) estimates an
cohort of women, thus providing visual evidence on the magnitude
increase of 6.6 percentage points in interethnic marriage shares
of changes.
and a decrease in interfaith marriage shares by 4.5 percentage
16
66.5% of Muslim-Christian unions are unions between a Muslim man and a
points. Once education and urban residence are controlled for (col-
Christian woman. Most Islamic scholars consider that it is forbidden for Muslim umn (2)), the magnitude of the increase in interethnic marriage
women to marry non-Muslim men, but Muslim men can marry women who belong to shares is 4.2 percentage points. The fact that women in later-
other monotheist religions. Hence, this imbalance in the types of Muslim-Christian born cohorts are more educated and more likely to live in urban
unions indicates that such unions, while rare, are not only counted due to
areas than women in earlier-born cohorts explains one-third of
measurement errors.
Fig. 3. Intermarriage shares on the pooled sample. Sample & data: Women in union, pooled sample. 95% confidence intervals included. Observed share of intermarriages by
birth cohort of women.
Table 2
Trends – intermarriage shares
(1) (2) (3) (4) (5) (6)

Dependent variable Interethnic marriage Interfaith marriage Muslim-Christian marriage
Birth year 0.265*** 0.168*** 0.179*** 0.149*** 0.0143 0.00461
(0.0421) (0.0403) (0.0289) (0.0287) (0.0157) (0.0154)
Age 0.428** 0.152 0.306** 0.305** 0.0307 0.120
(0.199) (0.197) (0.141) (0.141) (0.0777) (0.0770)
Age squared 0.00515* 0.00202 0.00131 0.00134 0.000,225 0.00134
(0.00294) (0.00292) (0.00221) (0.00220) (0.00117) (0.00115)
Primary 2.925*** 0.568 0.568**
(0.637) (0.457) (0.222)
Secondary/Higher 7.914*** 1.512*** 1.339***
(0.787) (0.561) (0.306)
Urban 13.12*** 2.523*** 1.141***
(0.661) (0.414) (0.232)
Remarried 6.443*** 3.808*** 1.650***
(0.643) (0.502) (0.285)
Controls
Country-fixed effects U U U U U U
Observationsa 97,111 97,111 96,549 96,549 96,549 96,549
R-squared 0.247 0.272 0.134 0.139 0.029 0.033
Mean dependent variable (share) 20.4 9.7 2.1

Specification: OLS regression. Standard errors are clustered at the DHS-cluster level. As the specification includes country-fixed effects, there is no constant in the model.
Columns (1) and (2): the dependent variable is a variable that equals 0 if the union is an intraethnic one, 100 if it is an interethnic one.
Columns (3) and (4): the dependent variable is a variable that equals 0 if the union is an intrafaith one, 100 if it is an interfaith one. Three faith groups are defined: ‘‘Muslim”,
‘‘Christian”, and ‘‘other (faith)”.
Columns (5) and (6): the dependent variable is a variable that equals 100 if one spouse is Christian and the other one Muslim, and 0 otherwise.
Results for all columns should be read as changes in percentage points.
Significance levels are denoted as follows: * p < 0.10, ** p < 0.05, *** p < 0.01.
a
One survey wave in Senegal does not include a question on religion.
the trend on interethnic marriages. Once education and urban res- likelihood. Urban residence is also associated with an increase in
idence are controlled for (column (4)), the magnitude of the the likelihood of intermarriage.17 In the case of interfaith marriages,
increase in interfaith marriage shares is 3.75 percentage points: secondary education and urban residence are negatively correlated
these variables explain little of the trend observed, corroborating to the likelihood of being in an interfaith marriage, but these vari-
the idea that the decrease in interfaith marriages is mainly due ables capture the likelihood of belonging to the group ‘‘other (faith)”.
to declining levels of religious diversity. Members of this group are often followers of traditional religions,
and attendance in school and urban residence are negatively corre-
lated with the likelihood of being a member of this group.18
5.2.2. Assessing individual characteristics
Looking at interethnic marriages and at Muslim-Christian mar- 17
Using a sample of earlier-born cohorts for which I have information on childhood
riages, the coefficients on education and on urban residence are place of residence, I run the specifications from Table 2. The coefficient estimated for
consistent with what we would expect from variables capturing the current place of residence is slightly higher than the coefficient for the childhood
place of residence.
parts of the individual preferences factor and of the diversity level 18
76.6% of women who belong to the group ‘‘other (faith)” did not complete
factor. Completion of primary school rather than having no educa- primary school while 44.4% of Muslim and Christian women did not complete
tion is associated with a higher likelihood of intermarrying. Com- primary school. 12.8% of women who belong to the group ‘‘other (faith)” live in an
pletion of secondary school is associated with an even higher urban area while 28.5% of Muslim and Christian women live in an urban area.
Remarried women are more likely to be married outside of their shares of interethnic marriages higher than 25%. Two countries,
group than women who are still in their first union, whatever the Burkina Faso and Niger, are the only countries in the sample where
kind of intermarriage considered. Social norms may be different for the ethnic fractionalization index is lower than 70%. They both
women who marry for the first time and for women who remarry, have huge majority groups – the Mossi in Burkina Faso, the Hausa
as women have more freedom in choosing a spouse when they in Niger – which may mean that the context in which unions take
have already been married (Locoh & Thiriat, 1995). Similarly, place in these two countries is different from what happens in
earlier-born women may remarry under the same set of (more countries where there is no majority group in the demographic
accepting) norms as later-born women who enter their first union. sense. The exception is Cameroon: it has a positive but not signif-
Last, women who remarried may have different (unobserved) char- icant increase in interethnic marriage shares,19 while having the
acteristics that also lead them to marry outside of their group, same share of interethnic marriages as the average on the pooled
whether in their first union or in the subsequent ones. I discuss sample, and having no majority group.
these hypotheses about remarriage in Section 7. Turning to regression analysis, Table 3 lists the coefficient asso-
ciated with birth year for two sets of regressions, without and with
the following controls: education, urban place of residence, and
6. Results at country-level
remarriage. The share of interethnic marriages significantly
increased over time in Benin, Cote d’Ivoire, Ghana, Guinea, Kenya,
This section presents the descriptive statistics and results on
Mali, Senegal, Togo, and Uganda (Panel A). In terms of magnitudes,
time trends at the country level. For brevity, tables include only
estimates of a 0.3 increase in percentage points by birth year trans-
the coefficient associated with the variable BirthYear and show
late into an increase of 7.5 percentage points when extrapolating
results for the two main specifications (with and without controls).
over 25 years. Once I control for the individual characteristics cor-
The full results at the country level and the results with control
related with interethnic marriages, the trends remain positive and
variables introduced one by one are available in the online
significant in six countries out of nine: Benin, Cote d’Ivoire, Guinea,
Appendix.
Mali, Senegal, and Uganda. In Ghana, the coefficient turns insignif-
icant when introducing either education or urban residence to the
6.1. Descriptive statistics model. In Kenya, the coefficient drops when introducing the type of
place of residence to the model but only loses significance when all
The maps in Fig. 4 show the observed share of interethnic mar- of the control variables are introduced jointly. In the case of Togo,
riages and the ratio of the observed to random share of interethnic adding education levels to the model explains away the trend.
marriages. Striking differences between countries appear. In Looking at Panel B, the introduction of the control variables does
Congo-Brazzaville and in Zambia, more than 40% of married not change the results.
women are in an interethnic marriage, whereas this share is lower
than 10% in DRC, Kenya, Namibia, and Nigeria. The observed share
6.2.1.2. Individual characteristics
and the ratio of the observed share to the random share are similar.
Countries are strikingly similar with respect to correlates of
This is because countries have high random shares of interethnic
interethnic marriages, and there are no differences between coun-
marriages: there are only two countries where this random share
tries where interethnic marriage shares increased and countries
is lower than 75%.
where they did not increase. The results are consistent with what
The maps in Fig. 5 show the observed share of interfaith mar-
is found in the pooled sample: primary education, secondary edu-
riages and the ratio of the observed to random share of interfaith
cation, urban residence, and remarriage are all positively corre-
marriages. In stark contrast to interethnic marriage patterns, the
lated to the likelihood of being in an interethnic union. The two
share of interfaith marriages is low. The highest share of interfaith
exceptions are Uganda, where women who attended primary
marriages is 29.6% (Congo-Brazzaville), while the highest share of
school are less likely than their uneducated counterparts to marry
interethnic marriages is over 40%. However, the level of religious
outside of their ethnic group – when urban residence is controlled
fractionalization is much lower than the level of ethnic fractional-
for –, and Gabon, where urban residence is uncorrelated to
ization, hence the much darker shades of the map on the right
interethnic marriage. The share of Gabonese women living in an
panel. Countries are also more heterogeneous when considering
urban areas is above 80%, while the share of women living in urban
religious fractionalization rather than ethnic fractionalization, as
areas is lower than 45% in all of the other countries: urbanization
religious fractionalization ranges from 3.6% (Niger) to 64.1%
might stop being a mixing factor once urbanization levels are high.
(Benin).
On the pooled sample, ratios of observed to random shares are
6.2.2. Intensive margin: linguistic distance
similar for interethnic and interfaith marriages shares, but it is not
6.2.2.1. Descriptive statistics and time trends. Table 4 shows results
the case when looking at countries separately. Notably, the distri-
of the regression of the linguistic distance (conditional on being
bution of this ratio is wider when looking at interfaith marriages
in an interethnic union) on birth year, and on the two sets of
rather than at interethnic marriages: there is no country for which
additional variables. Changes at the extensive margin do not neces-
this ratio is higher than 60% when looking at interethnic marriages,
sarily correspond to changes at the intensive margin. The linguistic
but it is higher than 60% for interfaith marriages in Congo-
distance of interethnic marriages increased in three countries,
Brazzaville, Gabon, Namibia, Niger, and Zambia.
two countries where interethnic marriages became more frequent
– Benin and Togo – and one – Cameroon – where interethnic
6.2. Time trends on interethnic marriages marriages did not increase. The linguistic distance decreased in
Cote d’Ivoire, Kenya, and Senegal, all countries where the share
6.2.1. Extensive margin of interethnic marriage increased. The linguistic distance did not
6.2.1.1. Time trends. Fig. 6 presents a visual representation of the change in the nine other countries of the sample. Introducing indi-
change in the shares of interethnic marriages over time. Panel A vidual characteristics in the model changes the results only in
shows that trajectories of countries where interethnic marriages Uganda, where the trend turns negative. The ratio of the observed
became more frequent look similar. When looking at panel B, we
notice that out of six countries where interethnic marriage shares 19
While there seems to be a trend for Cameroon on Fig. 6, the trend is insignificant
did not increase, three – Zambia, Gabon, Malawi – already had when age controls are added.
Fig. 4. Shares of interethnic marriages. Data: Survey wave (DHS) conducted the closest to 2005. Sample: Women currently in union. Left panel: Share of interethnic marriages.
Right panel: Ratio of the observed share to the random share of interethnic marriages. Higher ratios mean that the share of interethnic marriages is closer to what would be
observed under random matching. The corresponding data can be found in the online Appendix.
Fig. 5. Shares of interfaith marriages. Data: Survey wave (DHS) conducted the closest to 2005. Sample: Women currently in union. Left panel: Share of interfaith marriages.
Right panel: Ratio of the observed share to the random share of interfaith marriages. Higher ratios mean that the share of interfaith marriages is closer to what would be
observed under random matching. The corresponding data can be found in the online Appendix.
linguistic distance to the random one (column (4)) is close to one For instance, comparing the cases of Benin and Kenya (linguistic
for all countries: conditional on being in an interethnic union, most trees for these two countries are depicted in Appendix B), the aver-
of the linguistic distance is realized. Moreover, in seven countries age linguistic distance of interethnic marriages decreased in Kenya
of the sample, this ratio is larger than one, indicating that intereth- and increased in Benin. In Benin, as interethnic marriages became
nic marriages are more distant than they would be if they were more frequent, all groups started intermarrying more, thus result-
formed at random (considering only intermarried people). There ing in a decrease of the share of Adja-Fon unions, whose distance is
is only one country, Senegal, where this ratio is lower than one one node, and an increase in the share of unions with a distance
and where the linguistic distance of interethnic marriages larger than four nodes (e.g. Yoruba-Peulh unions). Increasing lin-
decreased. guistic distances indicate that women in later-born cohorts marry
The evolution of linguistic distances depends on the type of further away from their group, and hence that some ethnic cleav-
interethnic marriages that are observed in the earlier-born cohorts. ages may have lost salience. In Kenya, the decrease in linguistic
Fig. 6. Observed interethnic marriage shares over birth cohorts. Sample & data: Women currently in union, weighted DHS data at country level. Panel A (left): Countries for
which the trend on interethnic marriages is significantly different from 0. Panel B (right): Countries for which the trend on interethnic marriages is not significantly different
from 0. Countries are sorted into these two panels according to regression results from Table 3. Countries appear in the legend in descending order with respect to the share of
interethnic marriages in the 1985–1889 cohort. BF: Burkina Faso; CI: Cote d’Ivoire.
Table 3
Trend – observed interethnic marriage shares.
(1) (2) (3) (4)

Dependent variable Interethnic marriage Share N
Birth year coefficient Each cell: coefficient from a separate regression
Panel A: Increase in interethnic marriage shares
Benin 0.242*** 0.176** 15.1 10,977
(0.0741) (0.0727)
Cote d’Ivoire 0.254* 0.262* 19.2 2677
(0.136) (0.134)
Ghana 0.262* 0.192 19.4 6487
(0.140) (0.140)
Guinea 0.728*** 0.694*** 14.0 4732
(0.200) (0.203)
Kenya 0.257*** 0.0820 10.5 9169
(0.0624) (0.0538)
Mali 0.253* 0.305** 30.6 8499
(0.130) (0.129)
Senegal 0.326*** 0.177* 23.5 8339
(0.0914) (0.0913)
Togo 0.338*** 0.164 14.4 3701
(0.113) (0.114)
Uganda 0.455*** 0.413*** 24.3 2465
(0.135) (0.123)
Panel B: No change in interethnic marriage shares
Burkina Faso 0.00631 0.0680 10.4 9170
(0.0856) (0.0849)
Cameroon 0.545 0.215 20.5 3066
(0.332) (0.317)
Gabon 0.364 0.410 38.0 2274
(0.289) (0.278)
Malawi 0.00730 0.156 31.8 9241
(0.121) (0.120)
Niger 0.154 0.114 12.7 5603
(0.120) (0.123)
Zambia 0.0675 0.0821 46.0 10,711
(0.132) (0.125)
Controls
Age & Age2 U U
Education U
Urban U
Remarried U
Sample & data: Women currently in union, weighted DHS data at country level. Specification: OLS regressions run separately for the 15 countries of the sample. Standard errors
are clustered at the DHS-cluster level. The dependent variable is a variable that equals 0 is the union is intraethnic, 100 if the union is interethnic.
Columns (1) and (2) report the coefficient associated to the birth year variable. Each cell corresponds to a separate regression. Column (3) reports the share of interethnic
marriages in the regression sample. Column (4) reports the number of observations for each country.
Results in columns (1) and (2) can be interpreted as changes in percentage points. Significance levels are denoted as follows: * p < 0.10, ** p < 0.05, *** p < 0.01.
Table 4
Trend – linguistic distance between spouses.
(1) (2) (3) (4) (5)

Dependent variable Linguistic distance Mean Ratio N
Increase in linguistic distance
Benin 0.0151** 0.0145** 3.4 0.8 1650
(0.00692) (0.00686)
Togo 0.0155** 0.0148* 3.4 1.5 459
(0.00746) (0.00790)
No change in linguistic distance
Ghana 0.0124 0.0102 3.2 1.0 1238
(0.00841) (0.00881)
Guinea 0.0230 0.0151 5.9 1.4 673
(0.0199) (0.0209)
Mali 0.00813 0.0105 6.6 1.3 2568
(0.00819) (0.00812)
Uganda 0.0170 0.0209* 1.7 0.7 605
(0.0117) (0.0126)
Decrease in linguistic distance
Cote d’Ivoire 0.0197** 0.0196** 4.0 1.0 515
(0.00941) (0.00876)
Kenya 0.0577*** 0.0460** 3.9 1.2 922
(0.0205) (0.0199)
Senegal 0.0277** 0.0283** 3.5 0.8 1813
(0.0126) (0.0123)
Increase in linguistic distance
Cameroon 0.0779*** 0.0566*** 3.0 0.7 653
(0.0225) (0.0198)
No change in linguistic distance
Burkina Faso 0.0169 0.0125 4.9 1.2 995
(0.0111) (0.0112)
Gabon 0.00515 0.00530 1.4 0.6 923
(0.00538) (0.00542)
Malawi 0.00144 0.000,873 1.8 1.1 3023
(0.00181) (0.00182)
Niger 0.0108 0.0120 4.4 1.3 794
(0.0102) (0.0102)
Zambia 0.00172 0.00172 1.6 0.9 4873
(0.00269) (0.00261)
Controls
Age & Age2 U U
Education U
Urban U
Remarried U
Sample & data: Women currently in an interethnic union, weighted DHS data at country level. Specification: OLS regression run separately for the 15 countries of the sample.
Standard errors are clustered at the DHS-cluster level. Dependent variable is the linguistic distance (measure defined in Appendix B) associated to each interethnic union.
Columns (1) and (2) report the coefficient associated to the birth year variable. Each cell corresponds to a separate regression. Column (3) reports the mean linguistic distance
for intermarried couples. Column (4) reports the ratio of the mean observed linguistic distance to the random linguistic distance, computed by randomly matching individuals
who married outside of their ethnic group. Column (5) reports the number of observations for each country.
distance stems mostly from the fact that in earlier-born cohorts 6.2.2.2. Individual characteristics
there are more interethnic couples in which at least one spouse Correlates of the linguistic distance of interethnic marriages are
is Kalenjin or Luo – the only two groups that belong to the Nilo- not the same as correlates of interethnic marriages: primary and
Saharan branch – than in later-born cohorts.20 As interethnic mar- secondary education are negatively correlated to the linguistic dis-
riages became more common, the share of such unions among inter- tance of marriage in most countries. Such a reversal between the
married people decreased, resulting in the decrease of the average extensive and the intensive margin can be explained by ‘over-
linguistic distance of interethnic unions. This result is still consistent selection’ of individuals. Higher education levels are correlated
with the fact that some ethnic barriers – not captured by linguistic with a higher likelihood of marrying outside of one’s ethnic group,
distances – are becoming less salient. so individuals who marry outside of their ethnic group and who
have also not attended school are likely to have unobserved char-
acteristics, such as being strong-willed, that also make them marry
further away from their group or marry without any consideration
20
Linguistic distance does not capture perfectly the cleavages between groups and, of group differences.
contrary to share of intermarriages, is sensitive to extreme values. For instance, in
Kenya, the distance between, on the one hand, the Luo and Kalenjin groups (Nilo-
Saharan branch) and, on the other hand, all other ethnic groups but the Somali is high 6.3. Time trends on interfaith and Muslim-Christian marriages
(over 7.5 nodes). Interethnic unions in Kenya have an average linguistic distance of
7.2 nodes if at least one of the spouses is Luo or Kalenjin, and a distance of 3.3 nodes
otherwise. Luo-Kalenjin unions themselves make up 1.6% of interethnic unions in
6.3.1. Time trends
Kenya, despite the fact the linguistic distance of the pair is lower than with other Fig. 7 presents a visual representation of the change in interfaith
groups, indicating that other factors than linguistic distance are also at play. marriage shares over time. Comparing the two panels, it appears
that countries seem to converge towards low levels of interfaith Table 6 displays results from the main specifications and from
marriages. Apart from Cameroon and Cote d’Ivoire, all of the coun- the regressions when varying the assumptions as mentioned
tries where interfaith marriages did not become less frequent are above. For brevity, the results are presented for the pooled sample.
countries where the share of interfaith marriages is lower than A full discussion of the country-level results can be found in the
10% for all of the cohorts. online Appendix.
Table 5 shows the coefficient associated with birth year for
country-specific regressions of the likelihood of being in an inter-
7.1. Testing for heterogeneity in the ‘‘other” group
faith union on birth year and on the two sets of additional vari-
ables. Without individual controls other than age, the share of
The categories ‘‘other ethnicity” and ‘‘other faith” are categories
interfaith marriages increased only in Cameroon. The share of
that are more heterogeneous than other categories. In the main
interfaith marriages decreased in Benin, Burkina Faso, Gabon,
specification, I assume that when both spouses belong to the group
Ghana, Kenya, Togo, and Zambia. Controlling for education levels,
‘‘other”, their union is in-group. Assuming that these unions are in
urban place of residence, and remarriage explains the trend in
fact out-group unions (‘‘other-other” assumption), more unions are
Benin, Gabon, and Togo. In Benin, the trend turns insignificant
now counted as intermarriages. Fig. 8 shows the comparison of
when introducing an indicator variable for remarried status. In
intermarriage shares using the main assumption and using the
Togo, the introduction of education variables as well as of urban
‘‘other-other” assumption: more unions appear as interethnic and
residence explains the trend. In Gabon, it is the joint effect of the
as interfaith, thus providing an upper bound on the share of such
three variables.
unions.
Muslim-Christian union shares only changed in three countries,
In Table 6, columns (1) and (2) are similar to columns (3) and
decreasing in Ghana and increasing (although to a small extent) in
(4): the results regarding the pooled sample are robust to counting
Togo and Uganda. The coefficients are not significant in all the
the ‘‘other”-‘‘other” unions as out-group unions. The absolute mag-
other countries, a finding that is consistent with the fact that there
nitude of the coefficient is higher under the ‘‘other-other” assump-
is no trend when looking at the pooled sample (Table 2).
tion, especially for interfaith marriages, which is consistent with
Crossing these two sets results with results on variation of the
the overall decrease in the share of ‘‘other (faith)”. At the country
share of ‘‘other (faith)” over time (the full results can be found in
level, the main results carry through. There is no country for which
the online Appendix), I find that the share of ‘‘other (faith)”
the share of interethnic marriages decreased. The trend on inter-
decreased in all of the countries where the share of interfaith mar-
faith marriages in Niger turns positive and significant, which is
riages decreased, with the exception of Zambia. Such share
consistent with the fact that the share of ‘‘other (faith)” increased
increased only in Senegal and Niger, where there was no decrease
over time in this country.
in the share of interfaith marriages. In keeping with results on the
pooled sample, decreasing interfaith marriage shares are likely to
be driven by the decline in the share of the group ‘‘other (faith)” 7.2. Testing the remarriage story
and the resulting decrease in the level of religious diversity. How-
ever, at the country-level, education and urbanization suffice to Women remarry either after a divorce or after being widowed.
explain the trend in Benin, Gabon, and Togo. In Ghana, where both As such, remarriage is a function of age – older women are more
interfaith marriages and Muslim-Christian marriages became less likely to be widows – and may also be a function of the character-
common, it is likely that social norms or preferences are moving istics of a woman’s first union – intermarriages may be more likely
away from tolerating interfaith marriages. to end by a divorce.21 The age controls do not capture the fact that
some unions may be more likely to end than others: cohort compo-
sition effects might bias the estimated time trends. Remarried
6.3.2. Individual characteristics
women are more likely to be in an interethnic (interfaith) marriage
Countries are heterogeneous with respect to correlates of inter-
than women in their first marriage, but it might be that they have
faith marriages, as education levels and urban residence also cap-
already been in a (first) marriage which was interethnic (interfaith).
ture the likelihood of being a member of ‘‘other (faith)” in most
To understand better how remarriage patterns affect my results, I
countries. Consistent with results on the pooled sample, urban res-
study bounds on the estimates and use a sub-sample analysis.
idence and secondary education are negatively correlated to the
First, I bound the estimates by making assumptions on the first
likelihood of being in an interfaith union in most countries, and
union of remarried women. Fig. 8 depicts how the different
signs on primary schooling differ across countries. Contrasting
assumptions affect intermarriage shares. The ‘‘higher bound”
with these results, education levels and urban residence are either
assumption assigns an interethnic union to all of the women who
insignificant or positively correlated with the likelihood of being in
have remarried. The ‘‘lower bound” assumption assigns an intraeth-
a Muslim-Christian marriage, thus mirroring results on individual
nic union to all of the women who have remarried. Three points
characteristics associated with marrying outside of one’s ethnic
must be noted on these assumptions. First, they are extreme
group.
assumptions: either 100% or 0% of remarried women are assumed
to have had an interethnic (interfaith) first marriage while 25.6% of
7. Robustness analysis them are in an interethnic marriage at the survey date and 13.1% in
an interfaith one. Second, the ‘‘higher bound” assumption on
I implement four robustness checks on my findings. First, I relax interethnic marriages is an extreme assumption as the ratio of
the assumption that a marriage is intraethnic or intrafaith when the share of remarried women who have married outside of their
both spouses belong to the group ‘‘other”. Second, I test whether group to the share of not-remarried women who have married out-
the results are robust to alternative assumptions on remarried side of their group changes over birth cohorts. This ratio is higher
women’s first unions and whether the trends are also found when
21
considering separately women in their first union and remarried Whether spouses belong to the same group might affect the likelihood of divorce.
women. Third, using only women in their first union, I test whether However, not all remarried women have divorced, some were widowed. For instance,
interethnic unions are associated to a lower age difference between spouses, to a
‘‘assimilation” and conversion take place over the length of a mar- higher likelihood to live in an urban area, and to higher education levels: it is likely
riage. Fourth, using only women in their first union, I compare time that interethnic marriages are less likely to be ended by the husband’s death than
trends measured using birth year and using cohabitation year. intraethnic marriages are.
Fig. 7. Observed interfaith marriage shares over birth cohorts. Sample & data: Women currently in union, weighted DHS data at country level. Panels A1 and B1 (left):
Countries for which the trend on interfaith marriages is negative and significantly different from 0. Panels A2, A3, and B2 (right): Countries for which the trend on interfaith
marriages is not significantly different from 0; Cameroon (A3) is the only country for which the share of interfaith marriages increased. Countries are sorted into these panels
according to the regression results from Table 5. Countries appear in the legend in descending order with respect to the share of interfaith marriages in the 1985–1889 cohort.
BF: Burkina Faso; CI: Cote d’Ivoire.
for earlier-born cohorts than for later-born cohorts for interethnic a positive and significant trend when considering remarriages,
marriages. Third, there is a trend on remarriage. When regressing even when education and urban residence are controlled for. These
the Remarried variable on the year of birth and quadratic age con- results are consistent with the hypothesis that women who
trols, there is a negative and significant trend: the share of remar- remarry have more of a say on whom they marry, and that social
ried women is higher in earlier-born cohorts than in later-born norms around interethnic marriages have relaxed over time. These
cohorts, even when age is controlled for.22 It means that the ‘‘higher results may also capture changes in the composition of the remar-
bound” assumption works against finding a positive trend and that ried sub-sample: among remarried women, the share of widows is
the ‘‘lower bound” assumption works in favor of finding a positive likely to be higher in earlier-born cohorts than in later-born
trend. cohorts, the share of divorcees is likely to be lower in earlier-
Table 6 (columns (5)–(8)) displays the coefficient on birth year born cohorts than in later-born cohorts. Divorced women might
under these two assumptions. Under the ‘‘lower bound” assumption, have different preferences from widows, and they may also be
the trend on interethnic marriages remains positive and signifi- more likely to choose to whom they remarry.
cant. However, under the ‘‘higher bound” assumption, there is no
trend on interethnic (first) marriages, and a significant negative
trend (of small magnitude) once education and urban residence 7.3. Testing the ‘‘assimilation”/conversion story
are controlled for. At the country-level, the time trends turn nega-
tive and significant in Benin, Senegal, and Togo, resulting on the Older women have spent more time in a union than younger
insignificant trend on the pooled sample. As the trend on intereth- women: as spouses spend a longer time in a union, their ethnic
nic marriages is barely negative and significant under the ‘‘higher or religious identity may change. I cannot test for explanations
bound” assumption that works in favor of finding a negative trend, about conversion or ‘‘assimilation” that take place before cohabita-
it is extremely unlikely that interethnic first marriages became less tion or marriage, but I can test these two channels during the time
frequent over time, and unlikely that their share remained con- in union. Exploiting the fact that there are at least two survey
stant. The trend on interfaith marriages remains negative and sig- waves for each country, I can study whether women who married
nificant under both assumptions. Nonetheless, the magnitude of for the first time the same year and were born the same year are
the coefficient drops under the ‘‘lower bound” assumption, as more (less) likely to report having the same ethnic (religious)
expected. The results on interfaith marriages at the country-level group as their husband when the length of union increases. How-
are robust to these bounds. ever, the identification ultimately rests on differences across sur-
Second, I test whether the trends that I observe come from vey waves, so this also captures any effect correlated to survey
remarried women or from women in their first union (columns waves and their spacing.
(9)–(11), and columns (13) and (14), Table 6). There are trends in Table 6 (column (11)) shows results from the regression of the
all of the sub-samples. Such trends are not found for all of the Intermarriage variables on birth year, length of union, and age at
countries, a result that may stem from differences between coun- first union. Comparing column (9) to column (11), I find that the
tries and sub-samples, or from the fact that the sample size is time trends are robust to controlling for length of union and age
too small in the remarried sub-sample. I find no sub-sample in at first union.
which the trend on interethnic marriages is negative and signifi- The coefficient of the number of years since cohabitation is not
cant. Regarding interfaith marriages, the trend is positive and sig- significant for interethnic marriages but is negative and significant
nificant only for remarried Nigerien women and for Cameroonian for interfaith marriages. Therefore, in the case of interethnic mar-
women. riages, ‘‘assimilation” and selective divorces seem unlikely. In the
Even if we believe that the results from the ‘‘higher bound” case of interfaith marriages, the longer the union, the more likely
assumption on interethnic first marriages are correct and hence it is that spouses have the same faith. This result is consistent with
that the share of interethnic first marriages did not change over the fact that the share of ‘‘other (faith)” decreased over time: con-
time, the results from the sub-sample analysis show that there is version during marriage may be one of the mechanisms that is
behind this decrease. Another hypothesis would be that interfaith
22
unions are more likely to break than intrafaith unions, but this
Two factors are likely to explain this trend on the share of remarried women.
First, widowhood being a less common experience due to changes in life expectancy.
hypothesis does not account for the decline in traditional religions.
Second, not remarrying may be an option that is accessible to a higher share of later- Concerning the age at first cohabitation, women who were
born women than to their earlier-born counterparts. older when they started cohabiting are more likely to be in an
Table 5
Trend – observed interfaith marriage shares.
(1) (2) (3) (4) (5) (6) (7)

Dependent variable Interfaith marriage Interfaith marriage
Muslim/Christian/other (faith) Share Muslim-Christian Share N

Panel A1: Decrease in interfaith marriage shares
Benin 0.150** 0.103 16.7 0.0430 0.0396 2.9 10,977
(0.0754) (0.0753) (0.0332) (0.0341)
Ghana 1.520*** 1.301*** 18.3 0.167*** 0.172*** 2.8 6487
(0.133) (0.131) (0.0581) (0.0585)
Kenya 0.189*** 0.123*** 6.4 0.0112 0.00311 1.0 9169
(0.0445) (0.0451) (0.0174) (0.0173)
Togo 0.214** 0.0945 18.9 0.0585** 0.0512* 1.2 3701
(0.109) (0.112) (0.0248) (0.0284)
Panel A2: No change in interfaith marriage shares
Cote d’Ivoire 0.122 0.148 19.3 0.0544 0.0616 2.6 2677
(0.134) (0.134) (0.0472) (0.0474)
Guinea 0.0423 0.0481 5.1 0.0167 0.00898 0.9 4732
(0.122) (0.122) (0.0268) (0.0271)
Mali 0.00585 0.000,175 6.2 0.0198 0.0201 1.1 8499
(0.0660) (0.0659) (0.0257) (0.0252)
Senegal 0.0399 0.0469 1.9 0.0222 0.0281 1.3 7777
(0.0594) (0.0585) (0.0516) (0.0500)
Uganda 0.0496 0.0140 5.7 0.0948* 0.0572 4.3 2465
(0.0570) (0.0603) (0.0501) (0.0507)
Panel A3: Increase in interfaith marriage shares
Cameroon 0.489** 0.523** 10.8 0.0439 0.0467 1.6 3066
(0.233) (0.236) (0.0840) (0.0870)
Panel B1: Decrease in interfaith marriage shares
Burkina Faso 0.251*** 0.237*** 12.1 0.00294 0.0213 3.6 9170
(0.0898) (0.0894) (0.0447) (0.0451)
Gabon 0.570** 0.402 18.6 0.0729 0.0402 4.0 2274
(0.239) (0.244) (0.118) (0.128)
Zambia 0.161*** 0.146*** 4.4 0.0123 0.0126 0.5 10,711
(0.0496) (0.0492) (0.0188) (0.0191)
Panel B2: No change in interfaith marriage shares
Malawi 0.123 0.0130 7.8 0.0250 0.0432 2.2 9241
(0.0786) (0.0816) (0.0370) (0.0383)
Niger 0.0912 0.0962* 1.8 0.00343 0.00439 0.4 5603
(0.0572) (0.0567) (0.0161) (0.0163)
Controls
Age & Age2 U U U U
Education U U
Urban U U
Remarried U U
Sample & data: Women currently in union, weighted DHS data at country level. Specification: OLS regression run separately for the 15 countries of the sample. Standard errors
are clustered at the DHS-cluster level.
Columns (1) to (2): Dependent variable is a variable that equals 0 is the union is intrafaith, 100 if the union is interfaith, considering three religious groups (Christian, Muslim,
other (faith)). Columns (1) and (2) report the coefficient associated to the birth year variable. Each cell corresponds to a separate regression. Column (3) reports the share of
interfaith marriages.
Columns (4) to (5): Dependent variable is a variable that equals 0 is the union is a Muslim-Christian union, 100 if the union is not. Columns (4) and (5) report the coefficient
associated to the birth year variable. Each cell corresponds to a separate regression. Column (6) reports the share of Muslim-Christian marriages. Column (7) reports the
number of observations for each country.
Results in columns (1), (2), (4) and (5) can be interpreted as changes in percentage points.
interethnic union, which is consistent with the fact that these married outside of their group. However, cohabitation year is less
women are more educated and more likely to live in an urban area exogenous than birth year, and age at marriage is higher for
than their counterparts, and that these characteristics are posi- later-born cohorts, leading these cohorts to start cohabiting at even
tively correlated to the likelihood of being in an interethnic union. later dates. In Table 6, comparing columns (10)–(12), the results
Older women at the time of their first cohabitation are less likely to are robust to using cohabitation year instead of birth year on the
be in an interfaith union, which is consistent with the fact that they sample of women who are still in their first union. At the
are less likely to belong to a traditional religion. country-level, the results are robust to using cohabitation year
instead of birth year. Cohabitation year appears to indeed be
7.4. Testing birth year v. cohabitation year endogenous to education and urban residence. When using cohab-
itation year rather than birth year, there is a trend in all of the
The start of cohabitation may be a better measure of norms at countries except Burkina Faso and Niger, but after controlling for
the time that the union started: it corresponds to what people education and urban residence, this effect remains significant in
may perceive, such as the fact that more (less) people are getting the countries where there was a trend using birth year and in
Table 6
Robustness checks on interethnic marriages and interfaith marriages – pooled sample.
Regression results – Dependent variable: Interethnic marriage – Each cell: birth year coefficient from a separate regression
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)
Sample All married women First union Remarried
Assumptions Main Main Bound Other Higher bound Lower bound Main Main Main Main Main Main
Dependent variable: Interethnic marriage
Birth year 0.265*** 0.168*** 0.296*** 0.228*** 0.00227 0.0808* 0.268*** 0.157*** 0.263*** 0.135*** 0.212*** 0.353*** 0.302***
(0.0421) (0.0403) (0.0514) (0.0501) (0.0457) (0.0445) (0.0369) (0.0351) (0.0437) (0.0415) (0.0433) (0.101) (0.0981)
Year of cohabitation 0.174***
(0.0357)
Number of years since cohabitation 0.0597
(0.0458)
Age at cohabitation 0.821***
(0.0799)
Share interethnic marriages 20.4 27.9 32.5 16.2 19.4 25.6
Dependent variable: Interfaith marriage
Birth year 0.179*** 0.149*** 0.402*** 0.299*** 0.354*** 0.304*** 0.0960*** 0.0789*** 0.146*** 0.123*** 0.146*** 0.273*** 0.268***
(0.0289) (0.0287) (0.0394) (0.0388) (0.0416) (0.0413) (0.0242) (0.0241) (0.0293) (0.0292) (0.0292) (0.0776) (0.0778)
J. Crespin-Boucaud / World Development 125 (2020) 104668

Year of cohabitation 0.0796***
(0.0252)
Number of years since cohabitation 0.221***
(0.0329)
Age at cohabitation 0.258***
(0.0545)
Share interfaith marriages 9.7 15.5 23.8 7.6 9.1 13.1
Controls
Country fixed-effects U U U U U U U U U U U U U U
Age & Age2 U U U U U U U U U U U U U
Education & Urban U U U U U U U
Remarried U

Specification: OLS regression. Standard errors are clustered at the DHS-cluster level. As the specification includes country-fixed effects, there is no constant in the model.
Dependent variable is a variable that equals 0 is the union is intraethnic (intrafaith), 100 if the union is interethnic (interfaith). Definition of the dependent variable varies in specifications (1) to (8). Results are estimated under the
main specification, but on two sub-samples in columns (9) to (14).
Columns (1) to (10) and (13) and (14) report the coefficient associated to the birth year variable.
Column (11) reports the coefficient associated to the birth year variable, as well as the number of years since first cohabitation and the age at first cohabitation.
Column (12) reports the coefficient associated to the year of first cohabitation variable.
(1), (2): Main specification, All women: Dependent variable: Interethnic (interfaith) marriages as observed in the data.
(3), (4): Other bound, All women: Dependent variable: Interethnic (interfaith) marriages, with ‘‘other”-‘‘other” unions counted as interethnic ones.
(5), (6): Higher bound, All women: Dependent variable: Interethnic (interfaith) marriages, with all women who remarried counted as being in an interethnic (intrafaith) union.
(7), (8): Lower bound, All women: Dependent variable: Interethnic (interfaith) marriages, with all women who remarried counted as being in an intraethnic (intrafaith) union.
(9), (10), (11), (12): Main specification, First union: Only women in their first union.
(13), (14): Main specification, Remarried: Only women who have remarried.
15
Fig. 8. Shares of interethnic/interfaith marriages over birth cohort. Sample & data: Women in union, pooled sample. 95% confidence intervals included. Left panel: Interethnic
marriages. Right panel: Interfaith marriages. Bound high: Shares under the assumption that all remarried women were in interethnic/interfaith first unions. Bound low: Shares
under the assumption that all remarried women were in intraethnic/intrafaith first unions. Bound other: Shares under the assumption that when both spouses belong to the
group ‘‘other”, they are in an interethnic/interfaith union.
Gabon. After controlling for education and urban residence, the Future research could aim at studying marriages that are intra-
trends estimated on interfaith marriages are different depending faith but interdenominational. An additional strand of research
on whether birth year or cohabitation year is used only in Cote could focus on better understanding the channels by which educa-
d’Ivoire and in Senegal. tion and urbanization impact the likelihood of marrying outside
one’s own ethnic group: Are cities different from rural areas
because they are more heterogeneous in terms of ethnicity? Or
8. Concluding remarks do people living in cities have more agency to choose a partner?
Are educated women more likely to marry outside of their ethnic
This paper documents patterns of interethnic and interfaith group because they have more agency in choosing a spouse or
marriages in sub-Saharan African countries. I use data from Demo- because they accessed more mixed markets by attending higher
graphic and Health Surveys that gather information on marital his- education institutions? Related work could investigate whether
tory, education and geographic location to build a sample of marrying outside one’s own ethnic group is the result of strategic
women born between 1955 and 1989. I find that the share of behavior (Luke & Munshi, 2006). Are urban-dwellers turning away
interethnic marriages varies between countries but that such from membership in their ethnic group to benefit from their mem-
unions are not uncommon: 20.4% of women are married to a bership in a religious group? Faith groups or networks related to
man who is not from the same ethnic group as them, contrasting attending the same church or mosque could also be an opportunity
with 9.7% of women who are married to someone who does not to access jobs and support. Deepening our understanding of
share their faith, and 2.1% of women in Muslim-Christian whether marriage decisions reinforce or change identity affilia-
marriages. tions would bring important contributions to the political economy
Studying marital outcomes of women born between 1955 and literature on conflict, as well as to the literature on networks.
1989, I find that interethnic marriages became more common in
half of the sample, and that their share remained constant in the
Declaration of Competing Interest
other half. This study concludes that higher educational achieve-
ments and widespread urbanization contributed to that increase,
The author declare that they have no known competing financial
but that these changes cannot explain all of the it, suggesting that
interests or personal relationships that could have appeared to
ethnic cleavages may lose salience in some parts of sub-Saharan
influence the work reported in this paper.
Africa. Linguistic distance for intermarried couples increased in
some countries, suggesting that all ethnic boundaries were less-
ened. In other countries, linguistic distance decreased: boundaries Acknowledgement
between groups that are close in terms of linguistic distance disap-
pear and boundaries with other groups may be reinforced. As I extend special thanks to Denis Cogneau for carefully reading
interethnic marriage shares are far from rare and are even increas- my paper several times and for suggesting numerous changes that
ing in half of the countries of the sample, using ethnicity as a proxy greatly improved it. I am grateful to Sylvie Lambert, Alexander
for a strong identification with one single group may be misguided. Moradi, Lisa Oberländer, and Oliver Vanden Eynde for insightful
In contrast, interfaith marriages are becoming less common, a fact discussions and comments on previous versions of this work. I
that can mostly be attributed to the conversion of former followers thank seminar participants at the Paris School of Economics (PSE)
of traditional religions to Islam or Christianity, thus resulting in and at the University of Sussex as well as two anonymous referees
lower levels of religious diversity. The share of Muslim-Christian for their constructive comments.
marriages remains low and has not changed in most countries.
However, new religious groups, especially new religious branches
within Christianity, are being separately counted in recent survey Appendix A. Data
waves: religious boundaries within faith groups may become more
salient as a result of the expansion of Christianity and Islam. See Table 7.
Table 7
Data description by survey wave.
Survey Ethnicity Religion Remarried Survey Ethnicity Religion Remarried

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Benin 1996 9 11 4.4 7 35.9 20.0 Malawi 2000 9 11 7.7 8 2.6 24.6
Benin 2001 9 11 4.4 7 28.2 18.7 Malawi 2004 9 9 7.6 8 1.5 19.9
Benin 2006 9 11 4.3 11 27.5 14.5 Malawi 2010 9 14 6.8 9 1.9 21.2
Benin 2011 9 10 1.8 10 22.0 9.7 Malawi 2015 9 11 5.6 8 0.5 23.6
Benin 2017 9 10 5.1 10 16.7 14.3 Mali 1995 10 12 10.5 4 8.3 12.7
Burkina Faso 1993 10 12 7.5 5 20.1 13.2 Mali 2000 10 11 4.5 5 4.4 12.5
Burkina Faso 2010 10 17 13.2 6 9.6 10.3 Niger 1992 7 12 2.5 4 0.3 29.5
Cameroon 1998 23 43 24.1 6 11.5 19.5 Niger 1998 7 11 1.7 4 0.7 24.9
Cameroon 2004 23 48 21.3 8 10.9 19.2 Niger 2006 7 9 1.1 4 1.3 17.7
Cote d’Ivoire 1994 14 44 44.3 5 29.0 16.8 Senegal 1992b 7 9 5.3 23.5
Cote d’Ivoire 2011 14 56 46.8 9 18.3 14.7 Senegal 2005 7 9 8.3 2 0.0 19.4
Gabon 2000 9 10 26.6 7 12.2 29.1 Senegal 2010 7 8 9.4 4 0.7 12.0
Gabon 2012 9 11 31.7 8 5.1 25.3 Senegal 2014 7 8 5.1 3 0.5 12.7
Ghana 1993 8 13 3.7 7 23.5 23.9 Senegal 2015 7 8 7.7 3 1.0 12.8
Ghana 1998 8 13 4.7 9 15.6 23.9 Senegal 2016 7 8 6.6 3 0.5 12.4
Ghana 2003 8 10 8.6 9 11.0 22.1 Senegal 2017 7 8 9.0 4 0.0 13.5
Ghana 2008 8 10 4.7 11 9.8 19.3 Togo 1998 6 7 8.6 7 48.5 22.1
Ghana 2014 8 9 3.3 10 6.4 21.1 Togo 2013 6 8 6.8 13 28.9 15.7
Guinea 1999 7 8 1.3 5 7.3 15.0 Uganda 1996 20 32 10.4 5 1.4 21.5
Guinea 2005 7 8 1.1 4 6.1 16.6 Uganda 2016 20 47 11.1 11 0.9 22.3
Guinea 2012 7 7 3.8 5 7.4 15.6 Zambia 1996 31 46 6.8 6 1.8 17.8
Kenya 1993 11 12 4.7 6 3.7 5.2 Zambia 2001 31 48 8.1 6 1.5 18.3
Kenya 1998a 10 11 4.8 6 3.0 4.5 Zambia 2007 31 52 6.2 5 1.2 16.0
Kenya 2003 11 15 7.1 6 2.7 5.8 Zambia 2013 31 54 5.3 5 1.0 17.9
Kenya 2008 11 14 4.5 5 2.7 4.9
Kenya 2014 11 23 6.2 5 1.6 7.2
Note: This table lists all survey waves included in the main sample, and presents results useful to understand recoding choices across waves.
Sample: Women in union at the time of the survey.
Ethnicity: Columns (1) and (7): Number of ethnic groups in the common classification (at least one married woman and one married man in each survey wave and each birth
cohort). Columns (2) and (8): Number of ethnic groups with at least one woman and one man listed in the DHS classification. Columns (3) and (9): Share of women belonging
to the ‘‘other (ethnicity)” group (includes foreigners) in the common classification.
Religion: Columns (4) and (10): Number of religious groups with at least one woman and one man listed in the DHS classification. Common classification is made up of three
groups for all waves, except for Senegal 2005, where Christian and ‘‘other (faith)” groups are pooled together. Columns (5) and (11): Share of women belonging to the ‘‘other
(faith)” group in the common classification.
Remarriage: Columns (6) and (12): Share of remarried women.
a
The only Somali people in this wave are two men: this wave is not representative of the north-east of Kenya.
b
Religious affiliation not included in the questionnaire. I use a different set of weights when not using this wave.
Appendix B. Linguistic distance measures guages (or language groups) and ethnic groups. First, I match each
group listed the recoded DHS classification to the corresponding
Gershman and Rivera (2018) describe the specificity of linguis- linguistic group according to the information listed in the Ethno-
tic trees in the case of ethnolinguistic groups in sub-Saharan Africa. logue dictionary (Simons & Fennig, 2017). Second, I compute the
While languages have not always been associated with ethnicity linguistic distance of each pair of linguistic groups within a
(Canut, 2002), there is currently a strong association between lan- country.
Fig. 9. Ethnolinguistic tree for groups listed in DHS Benin. The names in bold are the names of the ethnic groups listed in DHS Benin. The names in italic are the names of
linguistic groups the ethnic groups were matched to.
Fig. 10. Ethnolinguistic tree for groups listed in DHS Kenya. The names in bold are the names of the ethnic groups listed in DHS Kenya. The names in italic are the names of
linguistic groups the ethnic groups were matched to.
I define linguistic distance between two groups as the mean Desmet, K., Ortuño-Ortín, I., & Wacziarg, R. (2016). Linguistic cleavages and
economic development. In The Palgrave handbook of economics and language
number of nodes to their first common subfamily. The number of
(pp. 425–446). Springer.
nodes is computed from the linguistic group level. To exemplify Easterly, W., & Levine, R. (1997). Africa’s growth tragedy: Policies and ethnic
how linguistic distances are computed, I use the cases of Benin divisions. The Quarterly Journal of Economics, 112(4), 1203–1250.
(Fig. 9) and of Kenya (Fig. 10). The linguistic distance between Eifert, B., Miguel, E., & Posner, D. N. (2010). Political competition and ethnic
identification in Africa. American Journal of Political Science, 54(2), 494–510.
the ethnic groups Adja and Fon, in Benin, is 1 (one step needed Francois, P., Rainer, I., & Trebbi, F. (2015). How is power shared in Africa?
to go to the last common linguistic subfamily: Aja-Gbe (1) and Econometrica, 83(2), 465–503.
Fon-Gbe (1)). The linguistic distance between Kamba and Kikuyu Fryer, R. G. Jr., (2007). Guess who’s been coming to dinner? Trends in interracial
marriage over the 20th century. Journal of Economic Perspectives, 21(2), 71–90.
is 0 (they share a subfamily: Kikuyu-Kamba). The linguistic distance Furtado, D., & Theodoropoulos, N. (2011). Interethnic marriage: A choice between
between Kalenjin and Luo is 4.5 (the average of the distance Kalen- ethnic and educational similarities. Journal of Population Economics, 24(4),
jin-Nilotic (3) and Luo-Nilotic (6)). 1257–1279.
Gershman, B., & Rivera, D. (2018). Subnational diversity in Sub-Saharan Africa:
When computing the value of the observed linguistic distance, Insights from a new dataset. Journal of Development Economics, 133, 231–263.
if one spouse does not belong to any identified ethnic group, but Gisselquist, R. M., Leiderer, S., & Nino-Zarazua, M. (2016). Ethnic heterogeneity and
her/his partner does, I define the linguistic distance between them public goods provision in Zambia: Evidence of a subnational ‘diversity
dividend’. World Development, 78, 308–323.
as the mean linguistic distance of intermarried couples within the Gören, E. (2014). How ethnic diversity affects economic growth. World Development,
country. When computing the value of the random linguistic dis- 59, 275–297.
tance, I define the linguistic distance of couples in which only Greenberg, J. H. (1956). The measurement of linguistic diversity. Language, 32(1),
109–115.
one spouse is ‘‘other (ethnicity)” as the median value of the defined
Habyarimana, J., Humphreys, M., Posner, D. N., & Weinstein, J. M. (2007). Why does
distances at the language pair level. ethnic diversity undermine public goods provision? American Political Science
Review, 101(4), 709–725.
Kalmijn, M. (1998). Intermarriage and homogamy: Causes, patterns, trends. Annual
Review of Sociology, 24(1), 395–421.
References Kalmijn, M., & Van Tubergen, F. (2006). Ethnic intermarriage in the Netherlands:
Confirmations and refutations of accepted insights. European Journal of
Alesina, A., & La Ferrara, E. (2005). Ethnic diversity and economic performance. Population/Revue europeenne de demographie, 22(4), 371–397.
Journal of Economic Literature, 43(3), 762–800. Kanbur, R., Kumar Rajaram, P., & Varshney, A. (2011). Ethnic diversity and ethnic
Berge, L. I. O., Bjorvatn, K., Galle, S., Miguel, E., Posner, D. N., Tungodden, B., & Zhang, strife. An interdisciplinary perspective. World Development, 39(2), 147–158.
K. (2018). Ethnically biased? Experimental evidence from Kenya. Journal of the Locoh, T., & Thiriat, M.-P. (1995). Divorce et remariage des femmes en Afrique de
European Economic Association. l’Ouest. Le cas du Togo. Population, 61–93.
Bertrand-Dansereau, A., & Clark, S. (2016). Pragmatic tradition or romantic Lowes, S., Nunn, N., Robinson, J. A., & Weigel, J. (2015). Understanding ethnic
aspiration? The causes of impulsive marriage and early divorce among identity in Africa: Evidence from the implicit association test (IAT). American
women in rural Malawi. Demographic Research, 35, 47–80. Economic Review, 105(5), 340–345.
Bisin, A., & Verdier, T. (2000). ’Beyond the melting pot’: Cultural transmission, Luke, N., & Munshi, K. (2006). New roles for marriage in urban Africa: Kinship
marriage, and the evolution of ethnic and religious traits. The Quarterly Journal networks and the labor market in Kenya. The Review of Economics and Statistics,
of Economics, 115(3), 955–988. 88(2), 264–282.
Canut, C. (2002). Langues et filiation en Afrique. Les Temps Modernes, 4, 410–440. Mayrargue, C. (2004). Trajectoires et enjeux contemporains du pentecôtisme en
Chandra, K. (2006). What is ethnic identity and does it matter? Annual Review of Afrique de l’Ouest. Critique Internationale, 1, 95–109.
Political Science, 9, 397–424. McCauley, J. F. (2014). The political mobilization of ethnic and religious identities in
Churchill, S. A., & Smyth, R. (2017). Ethnic diversity and poverty. World Africa. American Political Science Review, 108(4), 801–816.
Development, 95, 285–302. Meyer, B. (2004). Christianity in Africa: From African independent to pentecostal-
Clark, S., Kabiru, C., & Mathur, R. (2010). Relationship transitions among youth in charismatic churches. Annual Review of Anthropology, 33, 447–474.
urban Kenya. Journal of Marriage and Family, 72(1), 73–88. Miles, W. F. S., & Rochefort, D. A. (1991). Nationalism versus ethnic identity in sub-
de la Cuesta, B., & Wantchekon, L. (2016). Is language destiny? The origins and Saharan Africa. American Political Science Review, 85(2), 393–403.
consequences of ethnolinguistic diversity in Sub-Saharan Africa. In The Palgrave Monden, C. W. S., & Smits, J. (2005). Ethnic intermarriage in times of social change:
handbook of economics and language (pp. 513–537). Springer. The case of Latvia. Demography, 42(2), 323–345.
Mozaffar, S., Scarritt, J. R., & Galaich, G. (2003). Electoral institutions, ethnopolitical Qian, Z., & Lichter, D. T. (2011). Changing patterns of interracial marriage in a
cleavages, and party systems in Africa’s emerging democracies. American multiracial society. Journal of Marriage and Family, 73(5), 1065–1084.
Political Science Review, 97(3), 379–390. Simons, G. F., & Fennig, C. D. (2017). Ethnologue: Languages of the world. SIL
Posner, D. N. (2005). Institutions and ethnic politics in Africa. Cambridge University International, 20.
Press. Simson, R. (2018). Ethnic (in)equality in the public services of Kenya and Uganda.
Qian, Z., & Lichter, D. T. (2007). Social boundaries and marital assimilation: African Affairs, 118(470), 75–100.
Interpreting trends in racial and ethnic intermarriage. American Sociological Spörlein, C., Schlueter, E., & van Tubergen, F. (2014). Ethnic intermarriage in
Review, 72(1), 68–94. longitudinal perspective: Testing structural and cultural explanations in the
United States, 1880–2011. Social Science Research, 43, 1–15.

1 s2.0 S0305750X1930316X Main

Uploaded by

Copyright:

Available Formats

1 s2.0 S0305750X1930316X Main

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 s2.0 S0305750X1930316X Main

Uploaded by

Copyright:

Available Formats

World Development 125 (2020) 104668

Contents lists available at ScienceDirect

Interethnic and interfaith marriages in sub-Saharan Africa

1. Introduction we would expect interethnic marriages to be rare in societies where

4.2. Assessing time trends

Interethnic marriages (%) Linguistic distance (nodes)

Data & sample: Women in union, pooled sample. Weighted data.

(1) (2) (3) (4) (5) (6)

Data & sample: Women in union, pooled sample. Weighted data.

(1) (2) (3) (4)

(1) (2) (3) (4) (5)

(1) (2) (3) (4) (5) (6) (7)

Panel A: Increase in interethnic marriage shares

J. Crespin-Boucaud / World Development 125 (2020) 104668

Data & sample: Women in union, pooled sample. Weighted data.

Survey Ethnicity Religion Remarried Survey Ethnicity Religion Remarried

You might also like