RELATIONSHIP TRANSITIONS AMONG YOUTH IN URBAN KENYA

Abstract

The process of courtship and marriage in sub-Saharan Africa has changed remarkably. These changes, however, have received scant attention, as recent research has focused on adolescent relationships’ links to HIV/AIDS rather than to marriage. Drawing on detailed reports of 1,365 romantic and sexual partnerships from youths in Kisumu, Kenya, we find that marital aspirations, school enrollment, emotional attraction, pregnancy, and independence from kin are all predictors of getting engaged or married. Furthermore, though men and women are much more likely to marry partners they believe are sexually exclusive, men who have multiple partners are actually more likely to get married. By focusing on the contemporary process of marriage, this paper offers an alternative portrayal of premarital relationships in sub-Saharan Africa.

Over the last half-century, a dramatic shift in the process of marriage has occurred throughout much of the developing world, particularly in Asia, the Middle East, and sub-Saharan Africa. Notable changes include a rise in the age of first marriage, especially for women (Mensch, Singh, & Casterline, 2005), decreasing rates of polygamy (Hetherington, 2001; Wilson, Ngige, & Trollinger, 2003), and greater individual choice of whom and when to marry (Parkin, 1966; Yan, 2002; Smith, 2007). These changes in the marriage process are best understood in the wider context of pronounced and ongoing cultural and social transformations including urbanization, globalization (especially exposure to western culture), increased formal education, greater migration, and expanded labor force participation (Grant & Furstenberg, 2007). As theorized by Goode (1963), urbanization, industrialization, and the adoption of western ideologies have moved societies away from rural extended family networks towards more urban, nuclear systems, characterized by greater individual autonomy in decision making and monogamous marital unions. Extensive exposure to popular western media, which frequently includes narratives of romantic love, has shaped ideals about both nonmarital and marital partnerships (Frederiksen, 2000). Urbanization also offers youths greater opportunities to socialize with peers of the opposite sex (Ghimire, Axinn, Yabiku, & Thornton, 2006). Such encounters further increase the likelihood that youths will want to and be able to find a potential spouse on their own. In countries as different as China, Nepal, and Turkey, studies have shown that the transition from arranged marriages to self-selected spouses has precipitated a greater reliance on dating and courtship as a means of finding suitable marriage partners (Fox, 1975; Ghimire et al., 2006; Yan, 2002).

Our study is based in Kisumu, the third largest city in Kenya. Kisumu is similar to other urban areas in the developing world in that it is experiencing both rapid urbanization and increased exposure to globalization. Like other countries in sub-Saharan Africa, Kenya’s urban population is growing rapidly, fueled primarily by rural-to-urban migration. Furthermore, since its independence in 1963, Kenya has witnessed periods of economic growth, substantial increases in formal education, and greater exposure to capitalist markets. The youth population in Kisumu is also typical of other urban adolescents and young adults in that relatively few are married, and interactions with members of the opposite sex, including dating and courtship, are becoming increasingly common. Unlike most other urban environments in the developing world, however, youth in Kisumu date and search for marital partners amidst one of the world’s worst AIDS epidemics. Nearly a quarter of Kisumu’s adult population is HIV-positive (Glynn et al., 2001).

The onset of the AIDS epidemic has brought increased attention to adolescent sexual relationships in sub-Saharan Africa, but it has also, in our opinion, distorted how these relationships are portrayed. Premarital relationships formed in the pursuit of love or marriage receive disproportionately little attention, although there is a small, but growing, body of literature that investigates the links between HIV/AIDS and entry into marriage (Clark, Poulin, & Kohler, 2009; Mukiza-Gapere & Ntozi, 1995; Magruder, 2007). More often adolescent relationships that are characterized as problematic and dangerous dominate the research agenda. To illustrate this point we conducted a systematic literature search for articles about adolescent relationships in JSTOR, Medline, and Google Scholar (full results available from authors upon request). This search reveals striking differences between studies conducted in the U.S. and those in sub-Saharan Africa. Far more articles on adolescent relationships in the U.S. focused on the role of love and romantic attachment (n=50), than on the exchange of money or gifts for sex (n=27). The reverse was found for sub-Saharan Africa, where 58 articles dealt with transactional sex and only 14 papers (which all relied on ethnographic or qualitative research) addressed the importance of romance and love in these relationships. The point, made persuasively by Poulin (2007) in her qualitative study of adolescent relationships in Malawi, that many of the exchanges of gifts and money are actually symbolic expressions of love and commitment rather than material compensation for sex is generally overlooked in other studies.

This paper aims to help offset this perceived imbalance by examining adolescent relationships from the perspective of searching for a suitable spouse. Exactly how youths search for and find a potential spouse is poorly understood, but undoubtedly it involves some degree of trial and error. Some relationships proceed almost immediately to marriage, others evolve steadily from a casual relationship into a more permanent union, and still others dissolve. In this paper, we identify relationships are most likely to transition into an engagement or marriage, those that are still on-going, and those that dissolve. To the extent that youths in urban Kenya have adopted so-called modern ideals about marriage, which include feelings of love and commitment, we would expect that relationships reflecting these characteristics would be more likely to evolve into marriage, whereas those that failed to meet these ideals would end. Furthermore, unlike much of the existing literature, we do not presume that some of these relationships are inherently riskier than others with respect to HIV/AIDS. Rather we explore differences in sexual behaviors, including having sex, using condoms, and being mutually sexually monogamous, within diverse and dynamic relationship contexts. Taking this relationship transitions perspective not only sheds light on the contemporary process of marriage in urban Kenya, but also helps reorient our research on HIV risk and premarital sex.

Changes in the Courtship and Marriage Process in Sub-Saharan Africa

Before evaluating whether youths in urban Kenya are following a so-called modern or western process of marriage, we identify some of the key characteristics of such marriages. “Modern marriage” is often contrasted with “traditional marriage,” which generally refers to all marriages that follow the customs and practices of past generations. However, there has been and continues to be tremendous regional, ethnic, and religious diversity in the marriage process across sub-Saharan Africa. Even within Kenya, for example, whereas young Kikuyu men and women (Kenyatta, 1959; Worthman & Whiting, 1987) and Luo males (Evans-Pritchard, 1950) had considerable say in the selection of their own spouses, mate selection among the Kamba was solely a decision of the parents (Wilson et al., 2003). Consolidating these different customs under the umbrella term of “traditional” is thus an oversimplification, as would be the assumption that these diverse cultures are all adopting a single so-called western or modern marital process (Harrel-Bond, 1976).

Nonetheless, anthropological research documents that an increasing number of youths, particularly those residing in urban centers, are embracing what Harrel-Bond (1976) refers to as the stereotype of western marriage. This newly emerging process is characterized by several distinctive features including self-selection of spouses, love and emotional attachment, sexual exclusivity, sexual activity prior to marriage, lower levels of homogamy (on observable traits), and greater independence from parents and kin prior to marriage. Of all these changes, perhaps the most salient is the transition from kin-selected to self-selected spouses. Historically, marriage in sub-Saharan Africa marked the union of two families or lineages. As such, involvement of parents and extended kin was both central and essential to the process. Over the last half-century, marriage in urban areas has become more widely seen as a union between two individuals with their preferences largely determining the ultimate outcome of the relationship. As an indicator of individuals’ preferences, we include a measure of the respondent’s marital aspirations at the beginning of the relationship.

Changes in who selects a spouse are directly related to why a particular spouse is chosen. Over the past 50 years, love and emotional compatibility have substantially displaced the family and kin concerns of economic well-being, social status, and the continuation of the lineage as ideal bases for entering marriage (Hetherington, 2001; Frederiksen, 2000; Smith, 2001; Smith, 2007; van der Vliet, 1991). We, therefore, examine whether reasons for entering a relationship (i.e. emotional, financial, or physical) are related to relationship outcomes. As notions of romantic love are often closely tied to mutual sexual exclusivity, measures of expectations of partner’s fidelity as well as reports of the respondent’s own sexual monogamy are investigated.

Although concurrent sexual partnerships are likely to reduce the probability of relationship progression into marriage, sexual activity within the partnership may indicate an important step in the evolution of some relationships. Historically in Kenya, attitudes towards premarital sexual activity, especially for women, varied greatly according to ethnicity and region (Worthman & Whiting, 1987). The Luo, for example, generally prohibited women from engaging in sexual intercourse prior to marriage, although some forms of physical contact were permitted (Evans-Pritchard, 1950). We measure sexual activity as having ever had sex with one’s partner. Attitudes, however, may be changing as young men and women become increasingly responsible for finding their spouses and the courtship process becomes elongated. Higher levels of premarital sexual activity are likely to be associated with higher premarital pregnancies. In cultures where premarital sex for women was discouraged, premarital pregnancies were also stigmatized and, in some instances, reduced the future marriageability of young women (Calves, 1999). Other cultures, however, sometimes required proof of a woman’s fertility before marriage (Wilson et al., 2003). Children born either inside or outside of marriage tend to be highly valued in many African societies (Bledsoe, 1990). A premarital pregnancy is therefore likely to increase the desirability of marriage, but failure to marry prior to the child’s birth may not be viewed in an entirely negative light.

Whether or not a relationship transitions into marriage may also depend on the level of similarity or homogamy between the partners. When searching for a marital partner, individuals usually select partners with similar levels of education, wealth, and social, ethnic, cultural and religious backgrounds (Fafchamps & Quisumbing, 2005). Traditionally, in an ideal match, husbands would be older and better educated than their wives, but they would belong to the same ethnic group and socioeconomic class. Today, homogamy with respect to observable characteristics (i.e. age, education, wealth, and ethnicity) may be less important than finding someone with similar (unobservable) personality traits and interests. Thus, although homogamy is still likely to be evident, its effect on relationship progression may be weak.

Lastly, as self-selected marriages become more common, we would expect to find higher levels of independence from parents and kin before marriage, especially for women. Independence is a broad term that encompasses not only an individual’s age, but also whether he or she has completed key transitions, including finishing schooling, establishing a separate residence away from parents and kin, and finding a job or other means of financial support. For the purposes of this paper, we measure independence as no longer relying on parents or kin for support. Fifty years ago, it was commonplace for women to enter into marriage soon after puberty and to transition directly from their natal home to the marital household. Today the median age of first marriage for women in Kenya is 19.7 years (CBS, MOH, & ORC-Macro, 2004). Thus, finding a significant association between independence and marriage for women would indicate a clear change from previous marital patterns. For men, the age of first marriage has not risen notably over the last 50 years and currently stands at age 25 (CBS, MOH, & ORC-Macro, 2004).

Implications for HIV/AIDS

These changes in the marriage process have direct implications for HIV risks associated with dating and courtship. Concerns about HIV may influence the nature of premarital relationships as well as the choice of a spouse. Moreover, searching for a marital partner may in some instances expose adolescents to considerable risk of acquiring HIV (Clark et al., 2009). When considering whether relationships that culminate in marriage carry greater or less risk with respect to HIV than other premarital relationships, several behaviors need to be considered. Chief among them are whether there is sexual activity within the relationship and, if so, whether condoms are consistently used. As noted above, sexual activity by women before marriage may be increasingly common. Several studies, however, suggest that condoms are more often used in casual and commercial partnerships than in more serious relationships, as the use of condoms may be interpreted as a lack of trust or fidelity (Muhwava, 2004). Expectations of sexual fidelity for both members of the dyad may also be higher in relationships that transition into marriage than in those that remain casual or dissolve. Indeed, the presence of HIV/AIDS in the community may heighten the importance attached to finding an HIV-negative spouse who will remain sexually faithful (Clark et al., 2009).

METHOD

Study Setting

As in many other urban centers in the developing world, youths in our study site, Kisumu, Kenya, are experiencing fundamental changes in their society as increasing western-style individualism, exposure to popular culture, widespread access to mass media, the Internet and mobile telephones, and mobility are transforming their lives, particularly with respect to gender relations and marital aspirations. A wide array of venues (community facilities, schools, night-clubs, etc.) provides young people numerous opportunities for interaction with members of the opposite sex. Situated on Lake Victoria, Kisumu serves as a major commercial and trading hub drawing both temporary and permanent immigrants from across Kenya as well as other parts of Africa. As a result, although the large majority of its residents belong to the Luo ethnic group, Kisumu is also home to many other ethnic communities. Precipitous declines in the local fishing industry and a staggering AIDS epidemic have undermined the local economy and left 60% of residents living in absolute poverty.

This paper draws on detailed life history data from a study conducted in the summer of 2007. The study employed a novel survey instrument called the “Relationship Histories Calendar” (RHC). The RHC is a modification of life history calendars, which have been successfully used in other studies to gather retrospective information on contraception use, births, migration, schooling, and employment (Axinn, Pearce, & Ghimire, 1999). Like other life history calendars, the RHC gathers retrospective information on monthly changes in residence, schooling, employment, and household composition. In addition, the RHC was specifically designed to capture the dynamic processes of youths’ romantic and sexual life histories. Respondents provided detailed information about each of their romantic and sexual partnerships over the last ten years, including their partners’ demographic characteristics, relationship characteristics (including why they began the relationship, the initial type of relationship, whether they wanted to marry their partner, and the duration of the relationship) and sexual behaviors in each relationship (use of condoms, frequency of sex, pregnancy and childbirth, concurrent sexual partnerships, and beliefs about whether their partners were sexually exclusive). Ethical approval was granted by all collaborating institutions.

Our sample was drawn by contacting every other household in 45 randomly selected urban enumeration areas. Men and women ages 18 to 24 in the selected households were eligible to be interviewed as index respondents. One index respondent was randomly chosen per household and he or she was randomly assigned to receive either the RHC or a more standard demographic survey. A comparison of the quality of the data gathered by each type of survey instrument found that overall the RHC collected equally consistent and comparable data relative to the standard survey with less social desirability bias found in the reporting of sexual behaviors (Luke, Clark, & Zulu, 2008). In the present study we only use data from the RHC. In total, 610 index respondents (288 women and 322 men) received the RHC. These respondents reported on a total of 1,588 sexual and romantic relationships. Because we are interested in exploring why some relationships transition to engagement or marriage when others do not, our unit of analysis is the relationship rather than the respondent.

We excluded 104 relationships that had little or no likelihood of transitioning into engagement or marriage such as “one-night-stands” or relationships with commercial sex workers. An additional 102 relationships were described as an engagement or marriage in the first month of the relationship and were also excluded. Overall, the survey had very little missing data, although about 2% of our sample was missing their partner’s age and education level. For these cases, we imputed the mean age and modal educational group for men’s and women’s partners. Seventeen other cases were missing a value for one of the other independent variables and were subsequently removed, yielding a final sample size of 1,365 relationships (534 female relationships and 831 male relationships).

Measures

Dependent Variable

For our dependent variable, we identified three distinct relationship outcomes. Relationships that ended were coded as “1,” those that transitioned into an engagement or marriage were coded as “2,” and those that were still on-going at the time of the survey were coded as “3.” The majority of relationships (63.3%) ended before the survey without resulting in either an engagement or marriage. About 10% ended in an engagement or marriage, and about a quarter were still on-going. Not surprisingly, women’s relationships were more likely than men’s to culminate in an engagement or marriage (21.0% vs. 4.9%). Our rationale for combining both engagement and marriage into one dependent variable was threefold. First, for men there were too few relationships transitioning into engagement or marriage to analyze these outcomes separately. Second, our analyses separating engagement and marriage for women yielded surprisingly few differences from those combining the two outcomes. Third, both engagement and marriage represent an important transition from a less serious relationship to a more serious one, with long-term future implications. Nonetheless, by combining these outcomes we do not intend to imply that engagement and marriage are identical states. In our study, we rely on self-reports of marital status and classify individuals who report having a fiancé(e) as being engaged.

Independent Variables

Respondent characteristics

Several respondent characteristics may influence the probability that a relationship dissolves, continues, or transitions to marriage. These include gender, age, religion, ethnicity, highest level of completed schooling, and whether or not the respondent is currently in school. Age, highest educational level attained, and school enrollment were all measured in the first month of the relationship. School enrollment was coded as a dichotomous variable and highest level of schooling completed was recoded into three categories; 1= none or primary (corresponding to Standards 1 to 8), 2= secondary (Forms 1 to 4), and 3= some college training. We created separate categories for the two most common ethnicities in this region (Luo and Luhya) and created an “other” category for all other ethnicities. Religion was also collapsed into the four most common categories (Catholic, Protestant, Pentecostal, African Traditional) and “other”, which encompassed Muslim (5.5%) and none (1.1%). Sample distributions of these variables by relationship outcomes are presented in Table 1. Table 1 also presents significance tests for each relationship outcome. Women, for example, are significantly more likely than men to have a relationship evolve towards marriage rather than either dissolve or continue. Men and women, however, are equally likely to have a relationship continue rather than end. Respondents who are in school are less likely to have relationships that are on-going or that result in an engagement or marriage.

Table 1.

Respondent and Partner Characteristics, by Relationship Outcome

	(1)	(2)	(3)
	Ended	Engaged/Married	On-Going	Sig. 1&2	Sig. 1&3	Sig. 2&3
	(n=864)	(n=153)	(n=348)
RESPONDENT CHARACTERISTICS
Gender				***		***
Women	36.2	73.2	31.3
Men	63.8	26.8	68.7
Age (fst mon)	16.5	17.6	18.8	***	***	***
Highest level of eduction (fst mon)					***	***
Primary	45.8	46.4	24.7
Secondary	49.7	45.8	61.8
Some college	4.5	7.8	13.5
In school (fst mon)	67.5	42.5	47.1	***	***
R ethnicity
Luo	75.4	76.5	79.3
Luhya	15.1	13.1	12.9
Other	9.6	10.5	7.8
R Religion						*
Catholic	25.8	34.6	25.9
Protestant	40.5	30.7	44.8
Pentecostal	16.3	19.0	16.1
African/Traditional	10.1	9.2	6.6
Muslims/Other/None	7.3	6.5	6.6
Independent	7.6	15.7	11.2	**	*
PARTNER CHARACTERISTICS & HOMOGAMY
Age (fst mon)	19.2	22.0	20.8	**	*
Age difference (male-female)	2.0	4.4	2.4	***		***
Highest level of eduction (fst mon)				***	***
None/Primary	50.6	37.9	36.5
Secondary	45.1	50.3	53.2
Some college	4.3	11.8	10.3
Education difference (male-female)				*	**
Male less than female	7.1	9.8	10.1
Male same as female	67.4	54.9	56.6
Male more than female	25.6	35.3	33.3
In school (fst mon)	67.7	33.3	62.6	***		***
Ethnicity
Luo	69.7	74.5	71.0
Luhya	11.7	9.8	10.9
Other	18.6	15.7	18.1
Ethnic difference	32.4	25.5	27.6
Economic status (fst mon)				***		***
Wealthy	24.5	9.8	21.0
Middle	64.0	68.6	70.7
Poor	11.5	21.6	8.3

Chi-squared tests for categorical variables and t-test for continuous variables.

R=Respondent, P=Partner

^*p < .05.

^**p < .01.

^***p < .001.

Although the RHC recorded monthly information about income and occupational status, it unfortunately only collected asset ownership measures (which are generally considered one of the best indicators of economic status in low-income countries) at the time of the survey. As current asset ownership is undoubtedly endogenous to relationship outcomes, we did not include it in our models. The other measures of economic status (income and occupation) showed no significant relationship to relationship outcomes and were removed from our final models. Lastly, respondents were asked who was primarily responsible for them in each month. If the respondent named him or herself or any other nonfamily members as their primary care-taker at any point prior to engagement or marriage, they were coded as “1.” Otherwise, they were coded as “0.” We use this variable as an indicator of relative independence from parents and natal kin.

Partner characteristics and homogamy

Partner characteristics reflect not only what type of partner makes the most desirable future spouse, but also what types of matches are most compatible. Table 1, therefore, reports both partner’s characteristics and measures of couple-level homogamy with respect to age, educational attainment, and ethnicity. Age differences, for example, are significantly larger in relationships ending in marriage or engagement than in other types of relationships. Respondents were asked to assess their partner’s economic status, classifying them as 1=wealthy, 2=middle, and 3=poor. The precise definition of these terms was left to the respondent. Our bivariate results show that relationships ending in engagement or marriage are less likely to involve a partner who is in school and more likely to involve a partner described by the respondent as poor. In our multivariate regression, we include all three measures of homogamy on observable characteristics (age, education and ethnic difference) as well as the partner’s enrollment in school and his or her perceived economic status.

Relationship characteristics and sexual behaviors

Central to our analysis is whether specific relationship characteristics are related to relationship outcomes (Table 2). We control for the type of relationship in the first month (1=serious, 2=casual, 3=dating). Relationship duration is measured as the number of months from the beginning of the relationship until the end (if it ended), engagement or marriage, or the date of the survey (if on-going). To capture the extent to which individuals’ own marital aspirations affect relationship outcomes (i.e. a rough indicator of individual preferences), respondents were asked whether they wanted to marry their partner in the first month of their relationship, and could respond “yes”, “no” and “never considered marriage.” Interestingly, although women were far more likely to be married by the time of the survey, they were only slightly more likely than men to say that they had initially wanted to marry their partners (26.0% vs. 21.7%, p=0.06). To explore the main motivations for each relationship, respondents were asked to give their most important reason for entering into the relationship. For our analyses, we classified liking someone’s personality and being in love as emotional reasons (31.8%), being physically attracted and wanting sex as physical reasons (35.3%), and receiving money or gifts as financial reasons (4.7%) for entering a relationship. A sizeable proportion of respondents indicated that they simply wanted to have a partner (14.1%), and within this group some explicitly mentioned that they wanted to find a spouse (3.9%). All other reasons, which primarily included social pressure and convenience, were categorized as other. Given the low proportion of men and women who entered a relationship for financial reasons, we combined financial and other into one category in the multivariate regression. Among relationships that eventually transitioned into an engagement or marriage, almost half were entered into for emotional reasons, suggesting an important connection between love and marriage in many premarital relationships.

Table 2.

Relationship Characteristics, by Relationship Outcome

	(1)	(2)	(3)
	Ended	Engaged/Married	On-Going	Sig. 1&2	Sig. 1&3	Sig. 2&3
	(n=864)	(n=153)	(n=348)
RELATIONSHIP CHARACTERISTICS
Relationship Type (fst mon)				***	***	**
Serious	15.1	28.8	26.2
Dating	53.0	58.8	49.4
Casual	31.9	12.4	24.4
Duration (in months)	14.5	18.2	19.0	**	***
Marital Aspirations (fst mon)				***	**	***
Does not want to marry	18.1	45.8	26.7
Want to marry	63.2	33.3	55.8
Never considered marriage	18.8	20.9	17.5
Reason for Entering Relationship (fst mon)				***	***	***
Physical	38.5	16.3	35.6
Emotional	26.4	49.0	37.6
Want partner/spouse	12.5	24.8	13.5
Financial	5.4	2.0	4.0
Other	17.1	7.8	9.2
Had Sex	75.8	89.5	65.2	***	***	***
Always use condoms	30.3	18.3	30.8	**		**
Pregnancy	9.8	63.4	9.2	***		***
P had other partners	46.9	17.0	21.6	***	***
R had other partners	35.2	22.2	45.4	**	**	***

Chi-squared tests for categorical variables and t-test for continuous variables.

^*p < .05.

^**p < .01.

^***p < .001.

Our last set of independent variables assesses sexual and reproductive aspects of the relationship, including potential HIV risk factors. We created a dummy variable to indicate whether sex occurred at any point up to and including the month before engagement or marriage. Relationships that progressed towards marriage were significantly more likely to be sexually active than relationships that ended (89.5% vs. 75.8%). Relationships that were still on-going were less likely than relationships that ended to be sexually active (65.2% vs. 75.8%), although these bivariate results do not control for relationship duration. Consistent condom use was also less frequently found in relationships that ended in engagement and marriage than other types of relationships. Given the high level of reported sexual activity and the low level of condom use, it is not surprising that pregnancies before either engagement or marriage were quite common (fully 63.4% of such relationships). In comparison, slightly less than 10% of relationships that had ended or were on-going experienced a pregnancy. These findings raise the interesting question whether sexually active relationships with low condom use, which may indicate higher levels of commitment and trust, are more likely to culminate in an engagement or marriage, or whether these relationships, which entail higher levels of unprotected sex, are more likely to involve a pregnancy and it is pregnancy that precipitates marriage or engagement. In our multivariate analyses, we further investigate this question. We note, however, that although a high proportion of engagements and marriages are preceded by pregnancies, not all pregnancies lead towards marriage. Of the 214 pregnancies reported in our sample, 45.3% preceded a marriage or engagement and another 15% occurred within on-going relationships. Nearly 40% of pregnancies, however, occurred in relationships that eventually broke up. Thus, it appears that although pregnancy may significantly increase the probability that a relationship transitions towards marriage, there remains a sizeable proportion of pregnancies in which the couple does not stay together. These relatively high levels of pregnancy, particularly those that do not precede marriage, stand in contrast to traditional Luo culture, where female virginity at the time of marriage was valued highly.

Respondents were also asked whether or not they thought their partner had had other sexual partners during the course of their relationship. Relationships that were still on-going and those that transitioned towards marriage were significantly less likely to involve partners who were suspected of infidelity. The significant association between relationships ending and suspected partner infidelity was found for both men and women. In contrast, there are interesting gender differences with respect to respondents’ own sexual exclusivity. Female respondents reported having other sexual partners in 14.3% of relationships that transitioned towards marriage compared to about 33% of relationships that ended or were on-going. Men, however, were most likely to report concurrent partnerships in their on-going relationships (51.1%) and their marital relationships (43.9%), compared to only 35.9% of their relationships that broke up (35.9%) (p<0.001).

Analytic Strategy

A multinomial logistic model was chosen to compare the predictors of the three different relationship outcomes: those that ended, those that transitioned to engagement or marriage, and those that were ongoing at the time of the survey. The largest category, containing relationships that ended, was chosen as the base category.

One of the main advantages of drawing on detailed retrospective life history data is that we are able to examine how each of these factors, measured either at the beginning of the relationship or at any point during the relationship, is related to the subsequent relationship outcome. As such, we are able to minimize concerns about endogeneity and reverse causality. Given the nature of our data, an alternative modeling strategy would be to use survival analysis with competing risks. Such models would allow us to capture greater detail with respect to some time-varying variables such as sexual activity, suspicion of partner infidelity, and condom use in each month. However, from a theoretical perspective we preferred to measure these variables as ever had sex, ever suspected a partner of infidelity, and always used a condom, which could be easily handled in our multinomial framework. Similarly we preferred to measure relationship characteristics such as reason for being in the relationship and marital aspirations in the first month only, rather than capture their monthly changes, as these measures are arguably less subject to historical reinterpretation and their temporal order is easier to discern.

Another advantage of these data is that we can look beyond individual predictors of marriage and include a rather large number of partnership and relationship characteristics. To better understand the implications of this additional information, we present a series of nested models with the first model displaying only individual characteristics, the second adding partner characteristics, and the third and fourth models containing respondent, partner, and relationship-specific characteristics. Pregnancy is added individually in our fourth model to better assess the direct and indirect effects of sexual activity and condom use on relationship outcomes. Because our analysis was at the relationship level and, therefore, each respondent could contribute more than one relationship to our sample, we clustered our regression results by respondent. Anticipating that the courtship and marriage process works differently for men and women, we initially ran separate models for men and women. After testing for interactions between gender and the independent variables, however, we found that only one of the interactions was significant at the 5% level. Our final model, therefore, combines men’s and women’s relationships, but includes an interaction between respondent’s sexual exclusivity and gender.

RESULTS

Before exploring which relationships end, continue, or transition using our multinomial model, we briefly examine the main reasons given in response to an open-ended question about why a relationship ended (Table 3). Here too the gender differences are quite modest. For both men and women, the most commonly cited reason for ending a relationship (constituting over 30% of all reasons) is that the respondent suspected or knew that his or her partner had other sexual partners. This similarity is striking, not only because women are presumed to be more monogamous than men, but also because multiple concurrent partnerships are believed to be more acceptable for men than for women. Further investigation into this finding showed that sexual infidelity is more often mentioned if the respondent initially wanted to marry his or her partner than if he or she did not wish to get married (39.1% vs. 30.9%, p<0.05, results not shown). Both male and female respondents are far less likely to mention their own other sexual partners as the main reason the relationship dissolved. These findings counter stereotypes that fidelity among partners is not highly valued in African societies, and suggest that sexual exclusivity may be an important criterion when choosing a potential life partner. Notably, the disapproval of friends and family is given as the main reason for ending a relationship in only 3.8% of cases, consistent with the notion of self-selected marital partners. Lastly, we find that although pregnancy is positively associated with engagement in Table 2, in a small percentage of cases (2.9% for women and 0.7% for men) pregnancy is given as the primary reason for ending a relationship.

Table 3.

Reasons Why Relationships Ended

	Women	Men	Sig.
	(n=313)	(n=551)
P has other partners	34.7	31.0
Migration/Distance	14.2	21.6	**
Lost Contact	12.5	14.3
R has other partners	6.1	6.5
Incompatibility	5.1	4.5
Family/friends disapproval	2.9	1.6	*
Married another/already married	3.9	2.4
R or P refused to have sex	3.2	1.1	*
R or P had pregnancy	2.9	0.7	*
R or P refused to marry	1.9	0.7
Schooling/studies	1.9	2.2
Other	10.6	13.3

Chi-squared tests for each dichotomous response category.

R=Respondent, P=Partner

^*p < .05.

^**p < .01.

^***p < .001.

Results from our multinomial nested models are shown in Tables 4 and 5. Model 1 in Table 4 shows the association between respondent characteristics and relationship outcomes. The first column under Model 1 presents the relative-risk ratios of relationships that transition into engagement or marriage to those that end. The second column under Model 1 displays the relative-risk ratios of relationships that are on-going to those that end. As expected, men in this age group are far less likely to get married or engaged than women, although men are equally likely to be in an on-going relationship relative to breaking-up. Also, the older the respondent is when he or she enters the relationship, the more likely the relationship is to transition or continue rather than end. Being enrolled in school in the beginning of a relationship generally reduces the risk of getting engaged or married by about half and is also associated with a lower probability of being in an on-going relationship. In Model 1, neither religious nor ethnic affiliations are found to be jointly significantly associated with relationship outcomes, but independence from one’s family is strongly associated with getting engaged or married, and less strongly associated with being in an on-going relationship.

Table 4.

Predictors of Relationship Outcomes Using Multinomial Logistic Regression (Basecategory=Relationship Ended), Clustered by Respondent

	Model 1				Model 2
	Engaged/Married		On-Going		Engaged/Married		On-Going
	RRR	Sig	RRR	Sig	RRR	Sig	RRR	Sig
	(N =1,365)				(N =1,365)
RESPONDENT CHARACTERISTICS
Male	0.18	***	1.00		0.24	***	0.97
R age (fst mon)	1.16	**	1.32	***	1.14	**	1.33	***
R highest edu (fst mon)
Primary (ref)	1.00		1.00		1.00		1.00
Secondary	0.84		1.68	**	0.98		1.66	**
Some college	1.31		2.28	*	2.08		2.40	*
R in school (fst mon)	0.52	**	0.75		0.58	*	0.69	*
R ethnicity
Luo	0.72		1.48		0.52		1.29
Luhya	0.47		1.03		0.47		1.07
Other (ref)	1.00		1.00		1.00		1.00
R Religion
Catholic (ref)	1.00		1.00		1.00		1.00
Protestant	0.61	*	1.11		0.57	*	1.09
Pentecostal	0.84		0.98		0.79		1.01
African/Traditional	0.76		1.00		0.76		1.00
Muslims/Other/None	0.77		1.15		0.88		1.21
R Independent	2.67	**	1.67	*	2.42	**	1.65	*
PARTNER CHARACTERISTICS & HOMOGAMY
Age difference (male-female)					1.04		1.01
Education difference (male-female)
Male less than female (ref)					1.00		1.00
Male same as female					1.15		0.95
Male more than female					1.31		1.04
P in school (fst mon)					0.59	*	1.37
Ethnic difference					0.63		0.65	*
P economic status (fst mon)
Poor (ref)					1.00		1.00
Middle					0.59	*	1.40
Wealthy					0.20	***	1.05
Pseudo R2	0.13				0.16
Log Pseudolikelihood	-1045.0				-1017.5

R=Respondent, P=Partner

^*p < .05.

^**p < .01.

^***p < .001.

Table 5.

Predictors of Relationship Outcomes Using Multinomial Logistic Regression (Basecategory=Relationship Ended), Clustered by Respondent

	Model 1				Model 2
	Engaged/Married		On-Going		Engaged/Married		On-Going
	RRR	Sig	RRR	Sig	RRR	Sig	RRR	Sig
	(N =1,365)				(N =1,365)
RESPONDENT CHARACTERISTICS
Male	0.16	***	0.70		0.18	***	0.72
R age (fst mon)	1.27	***	1.57	***	1.29	***	1.56	***
R highest edu (fst mon)
Primary (ref)	1.00		1.00		1.00		1.00
Secondary	0.94		1.47		1.30		1.44
Some college	1.76		2.34	*	2.09		2.25	*
R in school (fst mon)	0.52	*	0.52	**	0.49	*	0.53	**
R ethnicity
Luo	0.79		1.81		0.68		1.76
Luhya	0.66		1.29		0.44		1.30
Other (ref)	1.00		1.00		1.00		1.00
R Religion
Catholic (ref)	1.00		1.00		1.00		1.00
Protestant	0.43	**	0.84		0.39	**	0.85
Pentecostal	0.52	*	0.92		0.40	*	0.89
African/Traditional	0.56		1.10		0.70		1.11
Muslims/Other/None	1.46		1.39		1.33		1.32
R Independent	1.55		0.91		1.62		0.89
PARTNER CHARACTERISTICS & HOMOGAMY
Age difference (male-female)	1.03		1.00		1.02		1.00
Education difference (male-female)
Male less than female (ref)	1.00		1.00		1.00		1.00
Male same as female	1.28		1.02		1.58		1.02
Male more than female	1.15		1.05		1.43		1.08
P in school (fst mon)	0.52	**	1.14		0.65		1.15
Ethnic difference	0.81		0.68		0.74		0.69
P economic status (fst mon)
Poor (ref)	1.00		1.00		1.00		1.00
Middle	0.47	**	1.34		0.50	*	1.34
Wealthy	0.24	**	1.06		0.33	*	1.04
RELATIONSHIP CHARACTERISTICS
Relationship Type (fst mon)
Serious (ref)	1.00		1.00		1.00		1.00
Dating	0.83		0.50	**	1.12		0.50	**
Casual	0.46	*	0.52	**	0.55		0.51	**
Duration	1.04	***	1.06	***	1.04	***	1.05	***
Marital Aspirations (fst mon)
Does not want to marry (ref)	1.00		1.00		1.00		1.00
Want to marry	3.01	***	1.18		2.87	***	1.18
Never considered marriage	2.33	**	1.01		2.33	*	1.05
Reason for Entering Relationship (fst mon)
Physical (ref)	1.00		1.00		1.00		1.00
Emotional	2.91	**	1.23		2.48	**	1.22
Want partner/spouse	3.59	**	1.26		3.81	**	1.25
Other	1.19		0.80		0.89		0.81
Had sex	3.38	**	0.42	***	1.06		0.42	***
Always used condoms	0.44	**	1.19		1.41		1.18
Pregnancy					12.96	***	0.99
P had other partners	0.20	***	0.27	***	0.18	***	0.27	***
R had other partners	0.21	***	0.73		0.23	***	0.76
Interaction: R other partners* gender	9.26	***	3.02	**	8.60	***	2.90	**
Pseudo R2	0.31				0.35
Log Pseudolikelihood	-827.5				-785.5

R=Respondent, P=Partner

^*p < .05.

^**p < .01.

^***p < .001.

Model 2 in Table 4 adds variables relating to specific partner characteristics and measures of homogamy. Although our bivariate results show significantly larger age differences between partners who get engaged or married than among partners in other types of relationships, this difference is no longer statistically significant in the multivariate model. Differences with respect to educational level also do not appear to be associated with relationship outcomes, but having a partner who is in school at the beginning of the relationship decreases the risk of transitioning towards marriage relative to breaking up. Couples from different ethnic backgrounds (compared to those belonging to the same ethnic group) are more likely to end their relationships than to remain together. We also find that women who describe their partners as wealthy rather than poor are much less likely to get engaged or married relative to ending the relationship. Although we were initially surprised by this finding, further investigation revealed that women are far more likely to enter into partnerships with wealthy men for financial reasons. Over a fifth of women with wealthy partners stated that financial support was the main reason for entering into the relationship compared to only 6.3% of women with poor partners (p=<0.000). Wealthy men may therefore act as “sugar daddies,” offering little prospect of marriage.

In our final set of models, presented in Table 5, we include relationship characteristics as well as respondent and partner characteristics. To further investigate the direct effects of sexual intercourse and condom use as well as their indirect effects through pregnancy, we exclude pregnancy from Model 1, but include it in Model 2 of Table 5. We find that several relationship-specific characteristics are significantly related to relationship outcomes and that the inclusion of these variables in our analysis increases the pseudo-R-squared from 0.16 to over 0.30 in both of our models. Not surprisingly, casual partnerships are less likely than serious partnerships to be on-going and are more likely to end. In Model 1, casual partnerships are also less likely to evolve towards marriage compared to serious partnerships. Similarly, the longer a relationship has lasted, the more likely it is to continue or to transition towards marriage. Wanting to marry one’s partner at the beginning of a relationship is associated with a roughly 3 fold higher relative risk of getting engaged or married compared to breaking-up. However, marital aspirations have no effect on whether a relationship is on-going or ends. Interestingly, the reason for entering into a relationship is clearly associated with whether or not a relationship evolves towards marriage, but not with whether a relationship is currently on-going or has ended. Relationships founded on the basis of physical attraction are more likely to end, whereas those established for emotional reasons are more likely to evolve. Wanting to have a partner, including specifically entering the relationship to find a spouse, is also strongly associated with getting married or engaged.

Consistent with our bivariate results, we find a strong association between sexual behaviors and relationship outcomes in Model 1. Specifically, having sex has a strong positive effect and always using condoms has a strong negative effect on the risk of getting married relative to the risk of breaking-up. In Model 2, when we control for pregnancy, both these effects become insignificant. Model 2 shows that getting pregnant (or getting one’s partner pregnant) has a very large impact on relationship outcomes. Relationships in which a pregnancy occurred have an almost 13-fold increase in the relative risks of evolving either into engagement or marriage. These findings suggest that sexual activity and condom use are likely to have only an indirect effect on transitions towards marriage, primarily through pregnancy. In comparison, neither pregnancy nor condom use has a significant effect on whether the relationship is on-going or ended, although sexual activity is less common in on-going relationships relative to previous relationships.

We also find that for both men and women, suspecting one’s partner of having other sexual partners significantly and substantially increases the likelihood that the relationship dissolves relative to either making a transition towards marriage or continuing. The effect of the respondents’ own sexual infidelity on relationship outcomes, however, varies strikingly by gender. Men who report having concurrent sexual partners are actually more likely to have their relationships continue or to transition towards marriage relative to ending. In contrast, women who report having another sexual partner during their relationship are significantly less likely to get engaged or married.

Although men marry at older ages than women and, thus, controlling for age, fewer of their relationships progress towards marriage, we find no significant gender interactions except with respect to respondents’ concurrent partnerships. The absence of other significant interactions is interesting in its own right. In particular, we did not find men’s independence or their school enrollment to be stronger predictors of relationship outcomes than women’s. Similarly, women who want to get married are no more likely than their male counterparts to actually get married after controlling for age differences. Contrary to some stereotypes, we find that emotional as opposed to physical attraction is equally predictive of men’s and women’s likelihood of getting engaged or married. Perhaps most surprisingly, men and women are equally likely to view their partners’ sexual infidelity as detrimental to the prospect of getting engaged or married or even continuing the relationship.

DISCUSSION

Examining adolescent relationships and sexual behaviors from the perspective of looking for a suitable marriage partner casts these relationships in a different and – we would argue – more balanced light. Overall, our findings point to similarities, rather than differences, between the current dating processes in urban Kenya and the so-called western process of courtship and marriage. For example, we find that marital aspirations of both partners are strong predictors of the relationship outcome, providing clear evidence that young people take an active role in identifying whom they wish to marry. Interestingly, we do not find that women are more likely to want to get married, but only that they are able to achieve their marital aspirations at much younger ages than men. We also find that relationships that transition towards marriage are more likely to be founded on attraction to a partner’s personality and feelings of love rather than physical attraction. Establishing an emotional connection and falling in love is equally important for men and women in their choice of a future spouse. These findings about the importance of emotional attachment are broadly consistent with the ethnographic and qualitative literature on adolescent relationships, but mark the first time that the role of love and romantic relationships has been highlighted in a representative survey collecting quantitative data on adolescent relationships in sub-Saharan Africa.

Although feelings of love and compatibility matter, partner homogamy with respect to age and education is not important, although there is some evidence that couples from different ethnic backgrounds are more likely to break up. We find that both being independent and having finished schooling are important precursors to marriage for men, and, more surprisingly, for women. Although being independent becomes insignificant after adding controls for specific relationship characteristics, our findings suggest that it is not until after school and after gaining independence that young urban men and women are ready to enter into the types of relationships that may evolve into marriage.

Sexual activity before engagement or marriage is very common, although consistent condom use is not. As a result, pregnancy often preceded transitions towards marriage. Indeed, after controlling for pregnancy, neither sexual activity nor condom use is significantly associated whether a relationship transitions towards marriage. Premarital pregnancy is also believed to be an important factor in determining relationship outcomes in countries like the U.S., where despite 38.5% of all pregnancies occurring before marriage (Martin, et al., 2008), an unintended pregnancy may precipitate a marriage to avoid the stigma and perceived difficulties of having a child out of wedlock (Ellison, 2003). Lastly, we find that sexually fidelity from one’s partner is important for both men and women in determining which relationships last.

Our different analytic perspective has implications not only for how adolescent sexual relationships are portrayed, but also for identifying adolescents who are at risk of HIV and developing effective protection strategies. As discussed above, research on HIV risks among adolescents overwhelmingly focuses on transactional and commercial sexual relationships. In our study, we find that very few relationships of any type were entered into primarily for financial reasons. Even in our full sample (which includes one-night-stands and commercial sex partners) less than 5% of all relationships are initiated for financial reasons. Instead, most relationships are initiated because the respondent found the partner physically attractive, liked his or her personality or loved him or her, or simply wanted to have a partner. Even so, some may argue that focusing on transactional sex is justified in high HIV settings. Such an argument, however, rests on the unstated assumption that transactional sex is inherently risky, whereas sex as an expression of love or in the pursuit of finding a marriage partner is naturally protective.

We find little evidence to support this assumption. Indeed, our bivariate results show that relationships that transition towards marriage are most likely to be sexual. Nearly 90% of such relationships involved sexual intercourse before engagement or marriage. For more than a decade, HIV prevention messages have promoted abstinence before marriage. These data suggest that such messages have gone largely unheeded. Furthermore, condoms are used consistently in only 18.3% of relationships that eventually transition towards marriage compared to about 30% of relationships that are on-going or end. These two factors (higher rates of sexual activity and lower rates of condom use) not only put couples at a higher risk of pregnancy but they are also likely to elevate the risk of HIV transmission.

In contrast, expectations about partners’ sexual exclusivity are considerably higher in relationships that progress towards marriage or are on-going. The importance of partners’ sexual fidelity for both men and women in determining whom they will marry is quite striking. Not only is believing that one’s partner has been sexually exclusive a strong predictor of whether the couple gets engaged or married, but believing otherwise is the primary reason for breaking up. Women who report having concurrent sexual partners are also less likely to get engaged or married compared to women who are sexually exclusive. For men, however, the opposite is true. Men who have multiple concurrent partnerships are more likely to get married or engaged. This double standard with respect to men’s and women’s sexual exclusivity before marriage, as well as the gap between women’s expectations about their partners’ fidelity and his reported behaviors, has direct implications for HIV policies. Researchers at the World Bank have stated that finding ways to de-normalize multiple and concurrent partnerships is the key research priority in countries with generalized HIV/AIDS epidemics (Wilson & Halperin, 2008). Our findings suggest that multiple sexual partners are normatively unacceptable in a potential marriage partner to both men and women. Reinforcing these norms, and especially convincing young men to follow them, may prove to be a more realistic and ultimately feasible HIV protection strategy than either abstinence or consistent condom use, as it is more closely in line with current practices and ideals.

In conclusion, our investigation into the patterns of courtship, engagement and marriage provides a rare glimpse into the dynamics of contemporary partnerships among youth in urban Kenya. We find that despite the presence of an overwhelming AIDS epidemic, youth in Kisumu continue to wish to get married, and that this process of finding a suitable spouse clearly reflects the influences of globalization and urbanization. By examining premarital relationships within the broader context of these adolescents’ lives, particularly with respect to their marital aspirations, we offer not only a fresh and more positive depiction of these relationships, but we also challenge the current emphasis on ostensibly riskier relationships such as those involving transactional sex. As in every society, relationships among adolescents in urban Africa are complicated. No one perspective is likely to capture all this diversity, but an expansion of the types of studies conducted would likely produce a far richer and more nuanced understanding of relationship transitions, and perhaps uncover new ways of reducing risk associated with them.