An international transition away from familially arranged marriages toward participation in spouse choice has endured for decades and continues to spread through rural Asia today. Although we know that this transformation has important consequences for childbearing early in marriage, we know much less about longer-term consequences of this marital revolution. Drawing on theories of family and fertility change and a rural Asian panel study designed to measure changes in both marital and childbearing behaviors, this study seeks to investigate these long-term consequences. Controlling for social changes that shape both marital practices and childbearing behaviors, and explicitly considering multiple dimensions of marital processes, we find evidence consistent with an independent, long-standing association of participation in spouse choice with higher rates of contraception to terminate childbearing. These results add a new dimension to the evidence linking revolutions in marital behavior to long-term declines in fertility and suggest that new research should consider a broader range of long-term consequences of changing marital processes.
Social change worldwide is steadily eroding the practice of familially arranged marriage in Asia, Africa, and Latin America, but most recently in rural South Asia (Ahearn 2004; Allendorf 2009, 2012; Ghimire and Axinn 2006; Ghimire et al. 2006; Hart 2007; Malhotra 1991; Pasupathi 2002; Thornton and Lin 1994). Nearly three decades ago, demographers began providing evidence of wide-ranging consequences of this revolutionary transition in marital behavior. For example, the switch to participation in spouse choice speeds childbearing after marriage (Hong 2006; Rindfuss and Morgan 1983; Wang and Quanche 1996; Whyte 1990). Through the decades, we have learned more about the relationship between first birth and changing marital processes, but we have learned little about the longer-term consequences of this marital revolution. For example, although the transition away from arranged marriage speeds the first birth, we know little about how it shapes subsequent childbearing, especially the termination of childbearing. This question is important because the answer speaks to the potential of these worldwide revolutions in marital processes to produce widespread fertility declines through greater use of contraception. Here, we investigate this question and provide unique evidence of the independent influence of the transition away from arranged marriage on contraception to limit fertility.
A key obstacle to investigating the independent influences of marital arrangement on fertility limitation is the empirical evidence this investigation demands. First, many of the same social and economic changes that stimulate widespread transitions in marital behavior also shape fertility-limiting behaviors (Axinn and Yabiku 2001; Bloom and Reddy 1986; Ghimire et al. 2006; Hong 2006; Malhotra 1991; Thornton and Lin 1994). Therefore, investigating the independent consequences of the transition away from arranged marriage requires careful consideration of wide-ranging social and economic factors that may simultaneously shape both marital arrangement and contraception. Second, marital arrangements cannot be observed independent of marriage timing, which is also known to have important consequences for subsequent childbearing behavior (Bloom and Reddy 1986; Bongaarts 1978; Hong 2006; Suwal 2001; Wang and Quanche 1996). Thus, the empirical evidence needed to address these questions must combine detailed measures of marital arrangement with measures of marriage timing, as well as the social and economic factors that drive these dimensions of marital processes.
To guide our empirical investigation of these important issues, we construct a theoretical framework that combines (1) frameworks linking variations in marriage behavior to subsequent fertility with (2) frameworks linking the influence of social changes on those same marriage and fertility behaviors. The framework we construct acknowledges the multidimensional nature of marital processes and the need to formulate hypotheses in relation to specific dimensions of marriage, rather than assuming that marriage is a single, discrete event (Thornton et al. 1994). Additionally, this framework emphasizes the importance of changes in local community context for both marriage processes and fertility limitation (Axinn and Yabiku 2001; Brauner-Otto et al. 2007; Ghimire and Axinn 2010; Yabiku 2004). Finally, this framework is constructed on a strong life course foundation—a foundation predicting that early-life circumstances and experiences predict later-life choices (Elder 1974, 1985, 1994, 1998). This emphasis leads us to incorporate premarital nonfamily experiences and marital experiences early in the life course into models of contraception later in the life course.
To test hypotheses emerging from this framework, we use detailed data from Nepal that document local social and economic changes, parental family background, individual nonfamily experiences, marital experiences, and contraception for both husbands and wives. This study uses life history calendar (LHC) measures from 2,022 ever-married women to create measures of their premarital, marital, and contraception experiences matched with measures of community context and family background. Together, these measures provide a unique opportunity to document both the overall influence of marital processes on contraception to limit fertility and the independent long-term consequences of familially arranged marriage versus some degree of participation in spouse choice.
Nepal is an ideal setting for this research because, until recently, fertility limitation was not an option because of cultural factors as well as the limited availability of modern contraceptive methods (Bennett 1983; Fricke 1986; Tuladhar 1987, 1989). Additionally, virtually all marriages were arranged by parents and relatives at very young ages (Ahearn 2004; Choe et al. 2004; Ghimire et al. 2006; Morgan and Niraula 1995). However, recent changes in social, economic, and institutional contexts have stimulated a rapid increase in both contraception and age at first marriage, as well as the participation of individuals in the selection of their own spouse (Axinn and Yabiku 2001; Ghimire et al. 2006; Morgan and Niraula 1995). This setting thus provides a crucial opportunity to examine the relationship between multiple dimensions of marital processes and contraception for terminating childbearing.
Fertility transition has been such an important element of social change that numerous theorists have focused on aspects of fertility. Scholars have identified many individual-level factors that influence fertility, such as education, employment, media exposure, religion, individuals’ orientations about family, and family formation (Blossfeld and Huinink 1991; Brien and Lillard 1994; Caldwell 1982; Hirschman 1985; Hirschman and Rindfuss 1980; Notestein 1953). Also influencing fertility are family- and household-level factors, such as parents’ education, work, media exposure, and fertility behavior (Axinn and Yabiku 2001; Caldwell et al. 1983, 1988). Finally, fertility is affected by various dimensions of social context, such as the spread of nonfamily services (Axinn and Yabiku 2001), mass education (Axinn and Barber 2001), family planning policy (Entwisle and Mason 1985), and family planning programs (Brauner-Otto et al. 2007; Entwisle et al. 1997; Knodel 1987). This body of literature has provided numerous insights regarding the factors affecting dimensions of fertility behavior and has fueled theories designed to explain all or part of the transition from high fertility and no use of birth control to low fertility and widespread use of birth control.
Factors Promoting Change in Marital Processes
Research has also demonstrated the important influence of change and variation in individuals’ community context and nonfamily experiences on marital processes, including marital arrangement, age at marriage, and quality of the husband-wife relationship (Ahearn 2001, 2003; Ghimire et al. 2006; Hoelter et al. 2004; Yabiku 2004, 2005). Theory suggests that change and variation in a wide array of family behaviors, including marital processes, have been important vehicles of worldwide fertility change (Caldwell 1982; Chesnais 1992; Freedman 1979, 1987; Lesthaeghe 1983; Thornton 2001). However, the influence of marital processes on childbearing outcomes may not be independent of the community, family, and individual factors that shape both marital processes and childbearing behaviors.
A large body of literature has documented important contextual influences on marital and closely related behaviors (Axinn and Yabiku 2001; Brewster 1994; Crane 1991; Dyson and Moore 1983; Hirschman and Young 2000; Hogan and Kitagawa 1985; Rindfuss and Hirschman 1984; Yabiku 2004). Research in Asian settings demonstrates that new community services can have a substantial influence on marriage timing and marital relationships (Dyson and Moore 1983; Hirschman 1985; Hirschman and Rindfuss 1980; Rindfuss and Hirschman 1984; Thornton and Lin 1994; Yabiku 2004). In predominantly agrarian societies, in which most social and economic activities are organized by families, new access to nonfamily services in the local community can stimulate new marital expectations and behavior (Thornton and Fricke 1987; Thornton and Lin 1994). Proximity to nonfamily organizations (e.g., schools, employers, health services, and transportation services) encourages participation in spouse choice, later marriage, and stronger emotional bonds between husband and wife (Ghimire et al. 2006; Hoelter et al. 2004; Niraula 1994; Yabiku 2004, 2006).
Nonfamily experiences early in the life course also shape subsequent marital behaviors. Caldwell (1982) and Thornton (2005) argue that present-day schooling and mass media in many settings outside the West are crucial factors in changing cultures and values toward those in the West. This is likely to be particularly true in South Asia, where educational materials themselves often include British examples of family life and individual choice (Caldwell 1982; Caldwell et al. 1988). These materials are likely to include fewer positive attitudes toward arranged marriage, more positive attitudes toward later marriage, and greater emphasis on the emotional bond between husbands and wives (Ghimire et al. 2006; Hoelter et al. 2004; Yabiku 2004). In arranged-marriage societies of South Asia and elsewhere, nonfamily experiences among youth are likely to create greater independence of youth from their parents (Fox 1975; Ghimire et al. 2006; Thornton et al. 1994; Van Bavel and Kok 2009). Thus, new nonfamily experiences, exposure to new ideas about marriage that differ from historically common ideas in South Asia, and new independence from parents each work in the direction of less arranged marriage, later marriage, and more emphasis on the positive emotional bond between husbands and wives. In the following section, we discuss the mechanisms that are likely to link these same dimensions of marital experience to contraception to terminate childbearing.
Marital Processes Affecting Contraception
Many dimensions of marital processes may influence key childbearing behaviors, such as contraception. Perhaps the strongest and most obvious link between marital experiences and contraception to limit childbearing is marriage timing (i.e., age at first marriage). In settings like Nepal, where marriage remains the principal route to exposure to intercourse, changes toward later marriage have an immediate effect on childbearing by reducing fertility (Axinn and Yabiku 2001). Also linked to marital experiences is marriage duration (i.e., number of years since first marriage). Although both delayed age at marriage and shorter marital duration decrease exposure to the risk of childbearing by reducing exposure to sex, these same factors may also influence the use of contraception to limit childbearing. Recent studies of marital dynamics provide evidence that an increase in age at marriage has consequences for childbearing intentions and behaviors, including greater contraception to achieve childbearing intentions (Hoelter et al. 2004; Hong 2006; Mason and Smith 2000; Niraula 1994; Satayavada and Adamchak 2000; Schuler et al. 2006; Thornton and Lin 1994; Wang and Quanche 1996; Yabiku 2004, 2006). Based on this evidence, we expect that those who marry at older ages will be more likely to use contraception to terminate childbearing when their family size goal has been met. Longer marital durations provide more exposure to the risk of childbearing no matter what the family size goal, so we expect longer marital durations to increase rates of contraception.
A third potentially important dimension of marital processes is marital arrangement. Theories of family and demographic transition suggest that a shift from familially arranged marriage to participation in spouse choice can bring about a fundamental change in family dynamics and fertility (Ahearn 2001, 2003; Caldwell 1982; Fox 1975; Goode 1970, 1982; Macfarlane 1976, 1986; Mitchell 1971; Rindfuss and Morgan 1983; Shorter 1975). The effect of this change on fertility may be quite different for the initiation of childbearing than for the termination of childbearing. Rindfuss and Morgan’s (1983) influential study of four Asian countries suggests that a shift from arranged marriage to participation in spouse choice dramatically increases coital frequency early in marriage, leading to higher fertility. Other studies have found similar evidence in Nepal (Ahearn 2004; Fricke and Teachman 1993). On the other hand, courtship-driven marriage, or arranged marriages that involve some courtship period, may be characterized by higher levels of husband-wife communication and cooperation than arranged marriages (Adams 2000; Blood 1967; Xiaohe and Whyte 1990). This may make couples more efficient at reaching joint childbearing goals, obtaining contraceptives with the aim of meeting those goals, and using contraceptive methods consistently to achieve those goals. Again, given the context of new preferences for smaller families in Nepal, we expect individual participation in spouse choice to increase rates of contraception to terminate childbearing.
A fourth important dimension of marital processes is marital cohabitation. Unlike settings where cohabitation often precedes marriage, such as the United States (Bumpass and Sweet 1989; Bumpass et al. 1991; Smock 2000; Thornton et al. 2007), in a setting with early age at marriage, like Nepal, cohabitation often begins later than marriage. Particularly in very young marriages, delayed cohabitation may delay exposure to sex and the risk of pregnancy independent of age at marriage, and those who are not coresiding will have little motivation to contracept (Basu 1993; Bloom and Reddy 1986; McCarthy 1982). Even at older ages, temporary labor migration that physically separates spouses also reduces exposure to sex and the motivation to use contraception to avoid pregnancies. For these reasons, we expect that coresidence with a spouse will increase rates of contraception.
A fifth key dimension of marital processes is childbearing. In the vast majority of settings, marriage and childbearing are closely linked (Cherlin 1992; Thornton 2005; Thornton and Lin 1994); and regardless of family size goal, as the number of children ever born increases, the desire to limit fertility is expected to increase (Bulatao and Lee 1983; Easterlin and Crimmins 1985). In rural South Asia, the number of children ever born is known to be an especially strong predictor of the desire to terminate childbearing (Axinn and Yabiku 2001). Therefore, we expect rates of contraception to increase with an increasing number of children ever born.
Variations in both marital cohabitation and childbearing happen after the marital event, so they may be considered mechanisms through which marital arrangement and marriage timing shape subsequent fertility limitation. Because of this, we model the effects of marital arrangement and marriage timing on subsequent contraception both with and without measures of marital cohabitation and childbearing in order to determine the extent to which those marital events shape subsequent contraception independent of variations in marital cohabitation and childbearing.
Data and Methods
This study uses data collected by the Chitwan Valley Family Study (CVFS) in 1996. Data from the Chitwan Valley in rural Nepal provide a unique opportunity to test our theoretical framework. The data capture life history events during a historical period characterized by dramatic changes in marital processes and behavior within the lifetimes of respondents, including dramatic increases in youth participation in spouse selection and marriage timing (Axinn and Yabiku 2001; Ghimire et al. 2006; Mitchell 2010; Mitchell 1971; Yabiku 2004). The average age at first marriage increased from a mean of 13.5 years for those who married between 1950 and 1959 to 19 years for those who married between 1980 and 1989 (Ghimire 2003; Yabiku 2004). The proportion of individuals in each marriage cohort who participated in the choice of a spouse increased from virtually zero in the 1936–1945 marriage cohort to approximately 50 % in the 1986–1995 marriage cohort (Ghimire et al. 2006). The patterns of contraception are even more dramatic in Chitwan. Among women born between 1942 and 1951, less than 5 % used permanent methods of contraception before they reached age 25. However, among the women born between 1962 and 1971, more than 35 % used those methods by age 25 (Axinn and Barber 2001).
The CVFS selected a systematic probability sample of 171 neighborhoods in Western Chitwan and defined a neighborhood as a geographic cluster of 5 to 15 households. After a neighborhood was selected, all individuals aged 15–59 residing in the sampled neighborhood were interviewed. If any of the respondents had a spouse living elsewhere, that spouse was interviewed as well. A total of 5,271 individuals were interviewed, with a 97 % response rate. The CVFS provides rich retrospective measurement of the occurrence and timing of individual life events, including marital events collected using an LHC and linked measures of the characteristics of those events using a structured questionnaire. The LHC method provides more accurate retrospective measurement of life events than alternative measurement techniques (Belli 1998; Freedman et al. 1988). Moreover, the LHC used in the CVFS was specifically designed to use local events to help respondents recall the timing of personal events and to allow them to report their recall of marital events in a manner most consistent with local practices (Axinn et al. 1999).
For this study, we analyzed data gathered from 2,022 women1 in the CVFS who were aged 15–59 in 1996, were ever married, and had not used permanent contraception prior to marriage. We limit the sample to married women because premarital sex is extremely rare in this setting, making never-married women extremely unlikely to use contraceptive methods (Acharya and Bennett 1981; Axinn 1992; Tuladhar 1987, 1989).
Our main objective is to examine the influence of variation in marital processes on the transition from no contraception to widespread contraception to limit childbearing. Substantial ethnographic and survey research demonstrates that during this historical period Nepalese women used some reversible contraceptive methods (Norplant, Depo-Provera, and IUDs) to stop childbearing (Axinn 1992; Link 2011; Satayavada and Adamchak 2000; Sharan and Valente 2002; Stash 2005; Tuladhar 1987). For example, data from the CVFS reveal that 98.8 % of married women aged 25–54 who had at least one child and who had ever used any of these contraceptive methods said that they wanted no more children. Consequently, we consider Norplant, Depo-Provera, and IUDs, as well as sterilization, to be contraceptive methods used to stop childbearing in this setting and historical period. This specification has been used in previous studies that examined the influence of individual- and community-level factors on fertility limitation (Axinn and Barber 2001; Axinn and Yabiku 2001; Barber and Axinn 2004; Brauner-Otto et al. 2007; Link 2011).
The key dependent variable is beginning the use of any of five contraceptive methods: (1) injectables (e.g. Depo-Provera), (2) IUDs, (3) Norplant, (4) husband sterilization, or (5) wife sterilization, measured annually by the LHC. Using information from the LHC interview, we operationalize the timing of the transition from never having used any of the permanent contraceptive methods to ever having used any of those methods, ignoring distinction among methods. We code a time-varying dichotomous variable as 0 for all the years the respondent did not use any of the five contraceptive methods and as 1 for the first year that a respondent uses any of these methods. The variable coded 1 is used to estimate the hazard of any contraception to terminate childbearing. As shown in Table 1, 44 % of ever-married women used these contraceptive methods to limit childbearing.
We also investigate the use of any contraceptive methods. This broader definition includes temporary methods often used to delay births, such as condoms, spermicidal foam, or oral contraception. This definition also includes withdrawal to avoid pregnancy. We use this broad definition of contraception as a dependent variable to compare our results against the more narrow definition of contraceptive methods used to stop childbearing. Table 1 presents descriptive statistics for all measures used in these analyses. The distributions refer to the respondent’s last person-year contributed to the analysis; for women who used any contraceptive methods, this is the first year they used the method, and for women who did not use any contraceptive methods, this is the final year of data collection.
As discussed earlier, multiple dimensions of marital experiences may influence contraception. Here we describe the measurement of each of those theoretical dimensions.
Marital arrangement. We treat the process of transition from familially arranged marriage to participation in spouse choice as a continuum. We therefore conceptualize marital arrangement in terms of the level of respondent participation in the choice of her husband. The CVFS measured participation in spouse choice on an ordinal scale ranging from 1 to 5: 1 means that the respondent did not participate in the choice (i.e., arranged marriage), and 5 means that solely the respondent chose her spouse (i.e., participation in spouse choice; for a detailed description, see Ghimire et al. 2006). However, because there are few cases in categories 2 through 5, we combine first marriages into those in which the respondent had no choice (0, with an ordinal scale value of 1) and those in which the individual had at least some choice (1, with ordinal scale values of 2, 3, 4, and 5). Of the 2,022 ever-married women interviewed, 76 % reported that their first marriages were arranged solely by parents or relatives, and the remaining 24 % reported that they either participated to some degree or solely chose their husband.
Marriage timing. The LHC interview collected a complete history of each respondent’s marital experiences. The respondent was asked, “Have you ever married or lived as husband/wife with somebody even for a few days?”2 If the response was affirmative, the respondent was then asked, “In what year did you marry for the first time?” Using this information, we calculate respondent’s age at first marriage.
Marriage duration. The measure of marriage duration comes from the LHC data and is measured as the number of years since the beginning of marriage up to a year before the respondent used any method of contraception, got divorced, became widowed, or was interviewed; that is, divorce, widowhood, and interview are treated as censoring events. If a woman was married in the same year as the individual interview or married for less than six months, marriage duration is coded as 0. As shown in Table 1, marriage duration varies from 0 to 39 years, with a mean of 11.93 years.
Marital cohabitation. Because a vast majority of marriages in Nepal were arranged by parents at very young ages until very recently, the amount of time between the marriage being legally and socially contracted and the couple’s coresidence is often significant. It was quite common for a newly married young woman to spend a couple of years in her natal home before moving to her husband’s parents’ home to cohabit with her husband. Because this timing of cohabitation can vary across couples, it is important to ascertain both the timing of marriage and the timing of coresidence. In addition, because premarital and extramarital sex are strongly condemned and cohabitation is still the principal route to exposure to sex, marital cohabitation becomes an important proxy for exposure to sex and an important factor in our analysis of contraception.
Using the respondent’s complete history of living arrangements from the LHC, we code marital cohabitation (living with husband) as a time-varying dichotomous variable equal to 1 if a woman was living with her husband for more than six months in a particular year, and 0 if otherwise.
Childbearing Experience. Using the complete history of the respondent’s childbearing experiences, we calculate a time-varying interval-level measure of the number of children born.
To capture the multidimensional nature of nonfamily experiences, we focus on three nonfamily experiences: schooling, work, and media exposure. Because nonfamily experiences are likely to influence and be influenced by marital experiences, we limit our measures of nonfamily experiences to premarital experiences.
Schooling. The CVFS collected a complete history of respondents’ educational experiences using the LHC. The LHC recorded each year the respondent was in school or attended adult education (basic literacy programs for adults). Years of schooling is the cumulative total number of years a respondent spent in school or adult education up to one year before first marriage. The value for years of schooling ranges from 0 to 20, with a mean of 2.98 years (see Table 1).
Work. The measure of nonfamily work before marriage is the number of years that a respondent was employed outside the family up to one year before the first marriage. The value for years of nonfamily work before marriage ranges from 0 to 16, with a mean of 0.71, suggesting a very low level of premarital nonfamily work (see Table 1).
Media exposure. Our measure of media exposure is a dichotomous measure of whether the respondent ever watched a movie in a movie theater before her first marriage. As shown in Table 1, only 37 % of our sample had ever watched a movie before their first marriage.
Childhood Community Context
Guided by the mode of social organization framework (see Axinn and Yabiku 2001), we define childhood community context as childhood access to nonfamily services—such as schools, modern health services, employment centers, markets, and bus services—that are directly related to an individual’s daily social life. In 1996, individual respondents were asked a series of questions about whether there was a specific service within a one-hour walk from their place of residence at any time before age 12. From the responses to these questions, we construct dichotomous variables for whether each of these specific nonfamily services existed within a one-hour walk from the respondent’s residence at any time before age 12. Moreover, to avoid problems of multicollinearity, we sum these five variables to a scale, with values ranging from 0 to 5. These measures of childhood context have been tested for external validity and reliability using a series of ethnographic and archival techniques (Axinn and Pearce 2006) and have been extensively used in previous research (Axinn and Yabiku 2001; Axinn and Barber 2001; Barber and Axinn 2004; Yabiku 2004, 2005).3
The intergenerational transmission literature suggests that parents significantly influence child outcomes. Parental resources (both economic and human capital resources) during their children’s childhood are positively related to children’s psychological and behavioral outcomes later in life (Axinn et al. 1994; Bengtson 1975; Cooksey et al. 1997; Sewell et al. 1980). Because parental experiences may affect children’s nonfamily experiences and their contraception, we control for a number of parental characteristics in our multivariate models: parents’ education, contraception, and mother’s total number of children. The measure of parental education comes from the response to the question, “Did your father ever go to school?” A positive response is coded as 1, and a negative response is coded as 0. The same question was asked regarding the mother’s education, and responses are coded in the same way. To create the measure of parental education, we sum these two measures; values range from 0 to 2, where 0 means that neither parent went to school, 1 means that either parent went to school, and 2 means that both parents went to school. The measure of parents’ contraception comes from the response to the question, “Did your parents ever use any contraceptive methods before you were 12 years old?” We code positive responses as 1 and negative responses as 0. The total number of mother’s children comes from the response to the question, “How many children did your mother have?” The responses to this question are then coded as an interval-level variable.
Previous studies suggest significant differences between birth cohorts both in terms of marital behaviors and contraception (Axinn and Barber 2001; Axinn and Yabiku 2001; Ghimire et al. 2006; Thornton et al. 1994). Compared with older cohorts, younger cohorts participate more in the choice of their spouses, marry at older ages, and use contraception (Axinn and Yabiku 2001; de Jong et al. 2006; Ghimire et al. 2006). Therefore, we include a control for the respondent’s birth cohort. Respondent’s birth cohort is coded in four categories: (1) cohort 1 born between 1972 and 1981, (2) cohort 2 born between 1962 and 1971, (3) cohort 3 born between 1952 and 1961, and (4) cohort 4 born between 1937 and 1951. We code individuals as 1 if they were born in a given cohort and 0 if not; the oldest cohort (cohort 4, born between 1937 and 1951) is the reference group in our analyses.
Nepalese society comprises many ethnic and linguistic subgroups (Bista 1972; Dahal 1993; Gurung 1980, 1998). These subgroups differ in many respects that have important consequences for both marital experiences and childbearing behavior. Although ethnicity in Nepal is complex, scholars have often categorized ethnicity into five major groups for analytical purposes: Brahmin/Chhetri (high-caste Hindus), Dalit (low-caste Hindus), Hill Janajati (Hill indigenous), Newar, and Terai Janajati (Terai indigenous) (Axinn and Yabiku 2001; Blaikie et al. 1980). We adopt the same categories in this analysis (for details about ethnicity in Nepal see Bista 1972; Fricke 1986; Guneratne 1994; Gurung 1980; Macfarlane 1976). We code individuals as 1 if they are members of a given ethnic category and 0 if not; Brahmin/Chhetri is the reference group.
The breadth of the CVFS data allows us to simultaneously estimate the effects of multiple dimensions of marital behavior on contraception to stop childbearing. First, we estimate a model with basic controls for premarital factors (such as respondent’s nonfamily experiences, childhood community context, and parental experiences) and marriage duration. Second, we estimate the total effects of marital processes—arrangement and timing—controlling for premarital factors. In this second step, we add marital arrangement and timing simultaneously because they occur simultaneously (Ghimire et al. 2006). Finally, we add the measures of two postmarital experiences—marital cohabitation and number of children born—both as time-varying measures in a single model. This third step adds two of the most powerful influences on childbearing behavior, both of which occur temporally and causally after marital arrangement and timing.
We also consider multiple views of the life course after marriage. First, we investigate contraception at any time after marriage, focusing on the long-term consequences of marital processes. Second, we investigate contraception after marriage but before the first birth (treating the first birth as a censoring event), focusing on contraception in the first-birth interval. Third, we investigate contraception after the first birth (treating first birth as the start of the hazard) but before the second birth (treating second births as a censoring event), focusing on contraception in the second-birth interval. Finally, we investigate contraception after the first and second births (two different options for starting the hazard, now ignoring subsequent births as censoring events), focusing on long-term consequences of marital processes again, this time after births rather than in total. These alternative views largely yield identical results, although contraception in the first-birth interval is rare; we discuss this in more detail later in the interpretation of our findings.
Because the individuals in our study are clustered with others living in the same community who all have the same community characteristics, we estimate these models taking this data structure into account. Specifically, we use the GLIMMIX macro for SAS. The results presented in the tables have all been calculated using GLIMMIX and therefore properly specify the multilevel nature of the data. Estimating multilevel discrete-time hazard models depends on assumptions about modeling, conditional independence, noninformative covariates, and coarsening at random (for a detailed discussion, see Barber et al. 2000).
In Models 1–4 of Table 2, we present estimates of the effects of marital arrangement and marriage timing on contraception to stop childbearing. We transform the raw coefficients by exponentiating them. The coefficients we present are estimates of the multiplicative effects on the hazard of using contraception. A coefficient of 1.00 represents no effect, a coefficient greater than 1.00 represents a positive effect, and a coefficient less than 1.00 represents a negative effect. Because the frequency of events in any time interval is quite small, the odds of transition from never having used any of the contraception methods to using any method are similar to the rate of contraception (Axinn and Yabiku 2001; Brauner-Otto et al. 2007). We discuss our results in terms of rates.
In Model 1 (Table 2), we estimate a base model with a wide range of social and economic factors that may simultaneously shape marital arrangement, marriage timing, and contraception. These factors include respondent’s nonfamily experiences, childhood community context, parental experiences, birth cohort, and ethnicity. In general, the parameter estimates in our base model are consistent with findings from previous studies of this setting (Axinn and Yabiku 2001; Axinn and Barber 2001; Barber and Axinn 2004; Brauner-Otto et al. 2007; Link 2011).5
Marital Arrangement and Timing
Model 2 (Table 2) estimates the effects of marital arrangement and marriage timing on contraception. Because neither participation in spouse choice nor age at first marriage can be observed without the occurrence of the other, we added the measures simultaneously. Both participation in spouse choice and age at first marriage have strong, statistically significant effects on the rate of permanent contraception independent of other factors in the model. Row 1 of Model 2 shows the estimated effect of participation in spouse choice. The odds multiplier of 1.28 suggests that those women who had some participation in choosing a spouse use contraception to limit childbearing at rates 28 % higher than women who did not participate in their spouse selection. As shown in Row 2 for Model 2, age at first marriage has a strong, positive, and statistically significant effect on the rate of using contraception to limit childbearing. The odds multiplier of 1.06 for age at first marriage means that a one-year increase in age at first marriage increases the rate of contraception by 6 %. Thus, an age at first marriage that is five years later would increase the long-term hazard of contraception by 34 %. Finally, these two strong and statistically significant effects of marriage are also independent, meaning that both the level of participation in spouse choice and the timing of marriage shape long-term contraception behaviors.
These two dimensions of variation in marital processes also improve the overall model fit. We use a –2 log-likelihood ratio as a measure of model fit. This statistic increased from 172,664 for Model 1 to 172,845 for Model 2—an increase of 181 points with 2 degrees of freedom. This is a statistically significant and substantively important improvement in model fit. Similarly, the model deviance value decreased from 6,681 for Model 1 to 6,646 for Model 2—a decrease of 35 points with 2 degrees of freedom.
Reestimating Models 1 and 2 using any contraception rather than contraception to end childbearing produces similar results (not shown in tables). For example, when we switch to the hazard of any contraception for Model 2, the estimated effect for participation in spouse choice is 1.24 (a 24 % increase in the hazard of contraception) and the estimated effect for age at first marriage is 1.08 (an 8 % increase in the hazard of contraception for each year of age). Once again, both effects are statistically significant, substantively important, and independent. With this alternative specification model fit, statistics also demonstrate these two measures improve our overall model—an increase in the –2 log-likelihood of 181 for an increase of 2 degrees of freedom (not shown in tables). Thus, both approaches to measuring contraception after marriage yield the same substantive conclusions: both more involvement in spouse choice and older ages at first marriage increase the use of contraception to avert births, and these two influences are independent of each other.
In addition to contraception at any time after marriage, we also investigate contraception after marriage but before the first birth (treating first births as a censoring event), focusing on contraception in the first-birth interval. In contrast to the results presented in Table 2, in the first-birth interval, participation in spouse choice does not have any significant effect on contraception (not shown in tables). Next, we investigate contraception after the first birth (treating first birth as the start of the hazard), but before the second birth (treating second birth as a censoring event), focusing on contraception in the second-birth interval. In the second-birth interval, both participation in spouse choice and age at first marriage increase the hazards of contraception, and the effects are even larger than those presented in Table 2 (not shown in tables). Finally, we investigate contraception after the first and second births (two different options for starting the hazard, now ignoring subsequent births as censoring events), focusing on long-term consequences of marital processes again but now after births rather than in total. These alternative views largely yield identical substantive results, with empirical estimates of the effects of participation in spouse choice and age at first marriage in the same direction but somewhat smaller than those shown in Table 2 (not shown in tables). Although contraception to avert births in the first-birth interval is rare, after the first-birth interval, the long-term consequences of participation in spouse choice and age at first marriage are similar to those presented in Table 2 in each of these later portions of the marital life course.
Postmarital Cohabitation and Childbearing
We investigate the role of postmarital cohabitation and childbearing in two steps. Before we estimate our final model, we estimate a model of these postmarital experiences without including participation in spouse choice and age at first marriage. To accomplish this, we add time-varying measures of marital cohabitation and childbearing experience to Model 1. The results, shown in the column for Model 3 in Table 2, show that cohabitation with husband and number of children born both have a strong, statistically significant effect of increasing the rate of contraceptive use to limit childbearing. The odds multiplier of 1.60 for cohabitation with husband means that the rate of contraceptive use is 60 % higher for women who live with their husband than for women who do not live with their husband. Likewise, the odds multiplier of 1.65 for number of children born means that having an additional child increases the rate of contraception by 65 %. To get a sense of the magnitude of this effect, this means that a woman with four children is 4.5 times more likely to use contraception to limit her future childbearing than a woman with one child. Thus, as expected, both of these postmarital dimensions of marital experiences have substantial consequences for contraception.
In this setting, it is not at all surprising that both marital cohabitation and number of children born drive contraception to avoid births. The contribution of these two factors to model fit is substantial.
Relative to Model 1, the Model 3 –2 log-likelihood ratio increased from 172,664 to 179,462—an increase of 6,798 points with 2 degrees of freedom. This substantial improvement in model fit is quite a bit larger than the overall contribution of participation in spouse choice and age at first marriage.
Of course, both participation in spouse choice and age at marriage not only come first but also shape the subsequent marital life course of these couples. In Model 4 of Table 2, we present estimates of the effects of both marital arrangement and marriage timing, now including marital cohabitation and childbearing. As expected, the effects of both participation in spouse choice and age at first marriage are somewhat reduced as important indirect effects of these two dimensions of marriage shape contraception via marital cohabitation and childbearing. The reduction in the odds multiplier from 1.28 (Model 2) to 1.20 (Model 4) for participation in spouse choice suggests that a substantial portion—almost 29 %—of the effect of participation in spouse choice works through the two postmarital experiences. Likewise, the reduction in the odds multiplier for age at first marriage from 1.06 (Model 2) to 1.03 (Model 4) again indicates that almost half of the effect of age at first marriage works through the two postmarital experiences.
Also important is that the remaining effects of variation in participation in spouse choice and age at first marriage are independent of the large effects of marital cohabitation and childbearing. This is somewhat surprising. Marital cohabitation and childbearing after marriage are more proximate to the decision to use contraception than marital arrangement and timing, but clearly some portion of the total effects of marital arrangement and timing shape decisions to contracept through other mechanisms. These independent effects of marital arrangement and timing mean that these features of marital experiences have long-term implications for contraception to avert births in addition to the long-term consequences for marital cohabitation and childbearing. Relative to Model 3, the Model 4 –2 log-likelihood ratio increases from 179,462 to 179,539—an increase of 77 points with 2 degrees of freedom. This improvement in model fit is statistically significant with 2 degrees of freedom.
The widespread erosion of familially arranged marriage probably does have a significant independent influence on the widespread increase in the use of contraception to limit fertility, at least in settings like Nepal. Demographers have provided decades of evidence that this marital revolution shapes the timing of first births (Feyisetan and Bankole 1991; Fricke and Teachman 1993; Hong 2006; Rindfuss and Morgan 1983; Wang and Quanche 1996). Because the measurement demands have limited what we know about the longer-term consequences of this marital revolution, however, relatively little evidence exists about how the erosion of arranged marriage has affected the use of contraception. The unique evidence provided in the current study is based on a long-term South Asian panel study of couples, their families, and their communities to demonstrate the possibility of an independent influence of the transition away from arranged marriage on the use of contraception.
A key issue is that the same set of community, family, and individual social changes that promote changes in marital processes are also expected to alter childbearing behavior, increasing contraception. Because the Chitwan Valley Family Study (CVFS) was specifically designed to investigate community, family, and individual influences on both marital processes and contraception, it provides an unusually rich set of measures of these factors corresponding to a portion of the life course before marriage. These rich measures include those presented in Tables 1 and 2, as well as other measures examined over more than a decade of investigation with these data but not included here (Axinn and Barber 2001; Axinn and Yabiku 2001; Barber and Axinn 2004; Barber et al. 2002; Brauner-Otto et al. 2007; Ghimire et al. 2006; Link 2011; Yabiku 2004, 2005, 2006). Together, this set of measures produces some of the most comprehensive data available for studying the influences of marital processes on fertility-limiting behaviors.
Another important aspect of this study is our recognition that it is impossible to observe marital arrangements without also observing the timing of marriage, but that the timing of marriage is expected to influence contraception to limit childbearing through mechanisms that are independent of marital arrangement (Bongaarts 1978; Ghimire et al. 2006; Hong 2006; Suwal 2001; Thornton and Lin 1994; Wang and Quanche 1996). Fortunately, the CVFS measured both marital arrangement and marriage timing. In addition, the rural Nepalese setting in which these data were collected included high variance in both dimensions of marriage, such that all combinations of arrangement and timing can be observed. Given that there is, of course, no random assignment of individuals to marriage arrangements or ages in any data set, the measurement and design of the CVFS provides an unusually powerful tool for investigating this topic.
The findings are simple to report: participation in spouse choice is associated with subsequent higher rates of contraception to limit childbearing, often many years later. This strong, statistically significant association is independent of many other key associations. First, older ages at marriage and longer marital durations are associated with higher rates of contraception, but these factors do not diminish the association with marital arrangement. Second, both marital cohabitation and the number of children born have exceptionally strong associations with use of permanent contraception, as expected in this setting. However, although these events occur after marriage, and are therefore subsequent to the observation of marital arrangement, these strong associations do not diminish the association between participation in spouse choice and subsequent contraception either. Third, other known sources of variation in subsequent marital and childbearing behaviors, including premarital nonfamily experiences, childhood community context, parental experiences, birth cohort, and ethnicity, also shape contraception as expected, but they do not remove the association of participation in spouse choice with subsequent childbearing behavior. The strong, statistically significant association between the decline in marital arrangement and the increase in marital contraception appears to be independent of other key factors.
Although the findings are straightforward, their implications are wide ranging. First, documentation of this important relationship between marital arrangement and contraception means that other dimensions of marital processes may also be important both as predictors of fertility behavior and as mechanisms linking nonfamily changes to fertility behaviors. Detailed investigation demonstrates that the spread of education in schools and mass media promote a transition away from arranged marriage (Ghimire et al. 2006; Thornton and Lin 1994). If greater participation in spouse choice promotes contraception, then greater participation in spouse choice quite likely links education and media exposure to use of contraception (Axinn and Barber 2001; Barber and Axinn 2004). A wealth of previous research demonstrates that high levels of communication between husbands and wives increase contraception to limit childbearing (Beckman 1983; Hill et al. 1959; Link 2011; Sharan and Valente 2002). Communication between spouses may work like marital arrangement to connect other nonfamily factors to contraception via marital processes. Other marital processes may also work to shape fertility-limitation behaviors. Both positive and negative dimensions of the emotional bond between husbands and wives are identified by fertility theory as potential influences on fertility limitation (Caldwell 1982). Empirical evidence suggests that premarital nonfamily experiences, especially education in Western-oriented schools, increase positive dimensions of this emotional bond and reduce criticisms and conflict (Hoelter et al. 2004). Potentially, many different dimensions of the marital relationship may influence contraception and act as mechanisms linking other nonfamily changes to fertility limitation. The potential for future research to investigate these possibilities is extremely high.
The association between participation in spouse choice and contraception documented here also implies a high potential for future research on other long-term consequences of change or variation in marital processes. Beginning with Rindfuss and Morgan’s (1983) breakthrough study, demographic research on the revolution away from arranged marriage has focused on relatively short-term outcomes, such as first-birth timing. Use of contraception to limit childbearing is a much longer-term outcome usually reserved for the very end of childbearing in a setting like rural Nepal (Axinn and Yabiku 2001). Other longer-term outcomes—including divorce, labor force participation, migration, retirement, child-rearing practices, health-related behaviors, morbidity, and mortality—all deserve attention. The evidence presented here is consistent with the possibility that revolutionary changes in marital processes, such as the move away from arranged marriage and the transition toward older age at marriage, can be associated with other substantial long-term changes in demographic behaviors.
This research was jointly supported by generous grants from the National Institute of Child Health and Human Development (R03HD055976, R01HD032912, and R24HD041028) and by a grant from the Fogarty International Center to the University of Michigan’s Population Studies Center. We thank Cathy Sun at the Population Studies Center for her assistance with creating analysis files, constructing measures and conducting analyses; the staff at the Institute for Social and Environmental Research–Nepal for data collection; and the Western Chitwan Valley residents for their valuable contributions to this research. The authors alone remain responsible for any errors or omissions.
Although the vast majority of research on fertility has focused on women only, a small and growing body of research conceptualizes childbearing behavior at the couple level (Axinn and Barber 2001; Thomson 1997). The empirical evidence from this work demonstrates that husbands’ and wives’ characteristics have separate, independent effects on the couples’ fertility and contraception. Because wives’ characteristics maintain separate and independent effects, we follow the majority of the literature on childbearing transitions and focus on women only.
Living together as husband and wife in Western culture may not necessarily be considered marriage. However, Hindu society places a very high value on female virginity. In Nepal, if a woman spends a night with a man and this is publicly known, she becomes impure and is considered to be married to that man. Therefore, we use a very restrictive definition of marriage (Bennett 1983; Gray 1991, 1995; Majupuria and Majupuria 1989).
The CVFS made a special investment in designing contextually appropriate measures of social change. Because Nepal as a whole, and the study area more specifically, had very few nonfamily services in the early period of its settlement history, the measure of access to childhood nonfamily services is services being within a one-hour walk = 1, and no services within a one-hour walk = 0. Additionally, in a rural context like Nepal, with its rugged topography and almost no access to transportation services, the measure of access to nonfamily services in terms of walking time to the service is more relevant than spatial distance. Therefore, access to nonfamily services is measured in walking time rather than distance.
Although it may appear that the discrete-time method of creating multiple person-years for each individual inflates the sample size resulting in artificially deflated standard errors, this is not the case (Allison 1982, 1984; Petersen 1986, 1991). In fact, the estimated standard errors are consistent estimators of the true standard errors (Allison 1982).
Although we parameterized marriage duration (the baseline hazard) in quadratic form, we also tested three other functional forms: a log function, a linear function, and a series of six-month-increment dichotomous indicators. The results vary only slightly across these four alternative functional forms. We chose the quadratic functional form because it provided the strongest overall model fit. This form is also consistent with previously published discrete-time models of contraception from this study setting (Axinn and Barber 2001; Axinn and Yabiku 2001; Barber and Axinn 2004; Brauner-Otto et al. 2007).