Pergamon PII: S0277-9536(96)00385-1
Soc. Sol. Med. Vol. 45, No. 5, pp. 677-687, 1997 t'i 1997 Elsevier Science Ltd All rights reserved. Printed in Great Britain 0277-9536/97 $17.00 + 0.00
SOCIAL NETWORK ASSOCIATIONS WITH CONTRACEPTIVE USE A M O N G CAMEROONIAN WOMEN 1N VOLUNTARY ASSOCIATIONS T H O M A S W. V A L E N T E , ' * S U S A N C. W A T K I N S , -~ M I R I A M N. JATO, 3 A R I A N E V A N D E R S T R A T E N ~ and L O U I S - P H I L I P P E M. T S I T S O L 4 'School of Public Health, The Johns Hopkins University, Baltimore, MD, U.S.A., :Population Studies Center, University of Pennsylvania, Philadelphia, PA,U.S.A., ~Center for Communication Programs, The John Hopkins University, Baltimore, MD, U.S.A. and 4Ministry of Public Health, Yaound& Cameroon Abstract This paper examines the association between social networks and contraceptive use. Using data from a survey of women belonging to voluntary associations in Yaound+, Cameroon, we find that the behavior and characteristics of the members of a respondent's personal networks are associated with her contraceptive use, over and above a set of her own individual characteristics that are usually found to be important. Respondents who report that their network partners approve of contraception, use it, and encourage the respondent to use are more likely to use contraception themselves; the association with encouragement is particularly strong. Moreover, there is a strong association between the specific methods of contraception used by a respondent and those used by her network partners, suggesting that members of personal networks exchange and evaluate specific methods. Because most of the respondent's network partners were interviewed, we are able to compare the respondent's perceptions of contraceptive use by her network partners with the network partner's actual use. We find that it is perceptions of use that matter, even if those perception are incorrect. ,~ 1997 Elsevier Science Ltd Key words" social networks, diffusion of innovations, family planning, women's groups, contraceptive use. Cameroon
INTRODUCTION
Change in behavior is usually studied by examining associations between individual attributes and indicators of behavioral change. Yet individuals are not isolated from others in their society, and it is likely that social interaction is also associated with behavior. In this paper, we consider the example of reproduction, specifically the use of contraception. Many analyses have shown that an individual woman's economic or educational characteristics are correlated with her contraceptive use. The present study considers whether we might gain additional understanding of women's reproductive behavior by taking social interaction into account. Using cross-sectional survey data from women who belong to voluntary associations, or tontines, in Cameroon, we ask whether (1) a woman's contraceptive use is associated with the contraceptive attitudes and behavior of the members of her personal network, and (2) whether specific attitudes, behaviors or characteristics of a woman's network influence her contraceptive use. *Address for correspondence: Department of Population Dynamics, School of Public Health, The Johns Hopkins University, Baltimore, MD 21205, U.S.A. m ,s:s-B
The motivation for this study comes from accumulated evidence that the deliberate control of fertility within marriage may be an innovation that diffuses, either from a central source such as the media or through interpersonal contacts, thus influencing the widespread transitions to lower fertility rates (Rogers, 1995; Bhatia et al., 1980; Rosenfield et al., 1973; Cleland and Wilson, 1987). The most influential models of fertility behavior have emphasized the associations between aggregate socioeconomic structures or individual socioeconomic characteristics and the costs and benefits of children (Becker, 1960, 1981; Schultz, 1976). Yet attempts to account for fertility declines using these models have been disappointing, and patterns in the timing and pace of fertility declines in Europe and developing countries have led some analysts to propose a more prominent role for social interaction (Pollack and Watkins, 1993; Watkins, 1994). Social interaction may provide a venue for social learning and/ or social influence; with others, women may exchange information about a new contraceptive method such as the I U D or the advantages and disadvantages of more or fewer children, evaluate the relevance of the use of these methods in their own lives, and/or assess their peers' approval or disap677
678
Thomas W. Valente et al.
proval of innovative behavior, thus facilitating or constraining fertility control. There has been a limited amount of research on social interaction and fertility change. Some of this research considers modern contraceptive methods as innovations, and uses models from the literature on the diffusion of innovation (Rogers, 1995; Valente, 1993; Valente and Rogers, 1995; Stycos and Back, 1964; Palmore, 1968; Retherford and Palmore, 1983; Montgomery and Chung, 1994). A second approach draws on network theories, which assume that social interaction is not random but patterned, and that who talks with whom is important. Network techniques have been developed to measure the interpersonal communication structure of a community (Valente, 1996; Scott, 1991; Marsden, 1990; Wellman, 1988; Knoke and Kuklinski, 1982; Bongaarts and Watkins, 1995), and have proven useful in the study of the spread of HIV/AIDS through sexual contact networks (Klovdahl, 1985; Klovdahl et al., 1994; Obbo, 1993). This paper draws on both approaches, focusing on diffusion networks (Valente, 1995). We begin by establishing that there are bivariate associations between a respondent's contraceptive use and her perceptions of her networks: whether they approve of contraception, use it themselves, and encourage her to use. We then take advantage of an unusual feature of our data set--the inclusion of the respondent's network partners as members of our sample. We ask whether respondents are likely to use the same methods of contraception as their network partners, and whether it is the respondent's perceptions that matter, and whether these perceptions need to be correct. Finally, we examine other characteristics of the women's networks, and use multivariate techniques to compare the associations between a woman's contraceptive use, her individual characteristics and the characteristics of her networks.
DATA AND METHODS
Voluntary associations
The data for this study come from a survey of members of nine voluntary associations of women in Cameroon, known as tontines. Tontines are often formed by migrants who seek the company of others with common customs and language; many groups thus have formed along ethnic lines, and consist of family and friends from the same province. Some were founded around specific issues-for example, one group (Groupe de Soutien des Meres de la Briqueteri) was started by a nurse who wanted recent mothers to have information about breastfeeding. Tontines exist across the socioeconomic spectrum and range from loose affiliations of friends to formal political and economic organizations. Some groups have male counterparts, and
most (probably all) men know of and approve of these associations. The tontines tend to be well organized and to meet regularly. Many function as rotating credit associations: women contribute money on a regular basis, and the funds are distributed to each member, either in turn or according to specific demands (e.g. for a wedding or funeral) (March and Taqqu, 1986; Geertz, 1962; Ardener, 1964; Anderson, 1966; Kurtz, 1973). Some groups raise funds in other ways and for other purposes: for example the Metta-Women's Cultural and Development Association has conducted fundraising drives for school construction and other capital projects. Many invite experts to give presentations about health, education, financial concerns or issues of community interest: oral rehydration therapy, HIV/ AIDS, and family planning have all been agenda items for some groups in the past year. These informal associations have been used in the past to aid family planning programs (Hammerslough, 1991; Misch and Margolin, 1975; Park et al., 1974; Park et al., 1976) but the diffusion of innovation within these groups has never been assessed. Key informants helped us (1) to develop a list of accessible women's groups; (2) to select 13 groups that were cohesive and dynamic, had more than 30 and fewer than 200 members, were ethnically distinctive and diverse (each group was to be relatively homogeneous with respect to ethnicity, and together they would cover the 10 ethnically distinct provinces of Cameroon), and had access to contraceptives; and (3) to approach the groups to solicit their participation in the study. Of the 13 selected, 10 agreed to participate, but one was omitted from this analysis since it was located outside of Yaound~ (Jato et al., 1994, 1995 for more details). Data collection
Interviews were conducted from mid-November to mid-December 1993 by locally recruited and trained interviewers. The members of three groups were interviewed privately at their homes; response rates for these three groups were greater than 90%. Resource limitations prohibited interviewing respondents from every group in their homes, however. The other six groups were therefore interviewed during their regularly scheduled meetings. Interviews conducted during the meetings allowed less time to solicit respondents and were restricted to those in attendance, hence the response rates for these, averaging 58%, was considerably lower. Successful interviews were conducted with 580 women, ranging from 35 to 120 per group. Respondents over 45 years old were omitted from the data analysis, leaving a final sample size of 495. The survey instrument covered standard demo-. graphic characteristics, contraceptive awareness and use, and network questions. Respondents were asked whether they had ever used contraception,
Social networks and contraceptive use in Cameroon
679
Table 1. Response rates and demographic characteristics of women's groups Name
Total
N Age
Average Wealth b
Educ?'
Past use Modern Trad.
AFEBAM (N) Muslim (L) MEWOCUDA (N) GSMB (L) BIABIA (L) KONGADZEM (N) PONGO (R) NGON A BEKE (R) ADESFO (N)
87 76 170 55 69 149 67 104 64
80 70 120 35 39 48 47 81 60
29.5 33.8 29.4 27.7 32.9 33.5 38,3 36.3 38.0
2.5 1.5 2.8 3,0 2.9 3.6 3.2 2.8 2.6
3.3 3.9 3.9 3.2 3.8 4.6 5.2 3.6 5.2
45 39 71 42 51 79 52 47 73
75 36 73 64 57 91 31 47 31
Total DHS (Yaounde)
841
580
32.5
2.7
4.(I
57 12
60 26
Note: (N) National, (R) Regional, and (L) Local. %cale 1-6: (1) none (2) Koran (3) primary (4) secondary 1 (5) secondary 2 (6) university. bindex 1 7: sum of house, car, radio, TV, gas stove, refrigerator, and running water.
whether they were currently using it, and the specific methods they knew and used (both past and current). Then they were asked about their networks. Network data are often collected by asking respondents to name others with whom they talk, either in general or about some specific issue. Data on the characteristics of those named--their network partners--are then collected, using one of two approaches. In the ego-centric approach, the respondent provides information about her network partners; in the census or full-network approach, all members of a bounded community are interviewed so that characteristics of the network partners are provided by the network partners themselves. This study is unusual in using both approaches, permitting us to compare the respondent's perception of her network partners' contraceptive behavior with their own reported contraceptive behavior. As a result, the 580 women in our sample include both respondents and their network partners. Respondents were asked to name five women ("nominations") in their group with whom they talked most frequently during the preceding six months (see network questions in Appendix A). Most respondents named five individuals, although a few (5%) of those named were not in their group; these nominations were discarded. Tontine members could be nominated by more than one respondent; about 55% of the network partners were named by three or more respondents. We then asked each respondent for information about her ties with each of the network partners she had nominated: their relationship (e.g. family, friend), the duration of that relationship, and the type of advice (e.g. financial, child care) or help (e.g. material, emotional) they received from the network partner during the past year. We then took the next step (often neglected in network studies) of asking the respondent to state whether she thought each woman named: (1) approves of family planning, (2) uses contraceptives, (3) uses a "modern" or "traditional" method, and (4) encouraged the respondent to use contraception.
Table 1 presents some of the characteristics of the nine tontines. We computed the average personal network scores for age, education, and wealth (measured as the sum of the number of possessions owned divided by the number of network members). Tontine members over age 45 who were nominated as network partners were included in these and other network measures; although they are unlikely to be candidates for contraceptive use, they may well be influential. On average the group members are in their early 30s, and have some modern possessions. The average level of education for most groups is some primary school (three have an average of some secondary school), and few have members who have had no schooling or university schooling. The average educational level is about the same as that obtained in a recent representative sample of women of reproductive age interviewed in Yaound6 (Bal6pa et al., 1992). Members of a particular tontine are rather similar to one another: indeed, the tontines may be taken as examples of the adage "birds of a feather flock together." A one-way analysis of variance (ANOVA) of age, education, and wealth shows that differences across tontines were greater than differences within them; the differences in the variances across tontines were statistically significant only for education. The groups are also relatively homogeneous with respect to ethnicity, language and religion (not shown): one is 98% Muslim, six are over 90% Christian, two are predominantly Englishspeaking, five are predominantly French-speaking. Despite the within-group homogeneity and the across-group differences; however, in the analyses that follow we combined the data from the nine groups. Our primary justification for doing this is ease of presentation and interpretation. We are comfortable in doing this because multivariate analyses (described later) that included dummies for the tontines showed that they were rarely statistically significant. We note that these tontines cannot be taken as representative of the provincial populations from which they come, or even representative of the
Thomas W. Valente et al.
680
Table 2. Bivariate odds ratios for respondent's contraceptive status and perceptions of support by network partners (NWPs) Respondent's contraceptive status Ever-used a method Any Clinic Non-clinic
Proportion of N WPs perceived to: Approve of contraception Have used contraception Have encouraged respondent to use
4.3*** 7.6*** 16.6"**
3.6*** 4.6*** 5.3***
1.1 1.3 1.0
*p < 0.05; **p < 0.01; ***p < 0.001. N = 477.
urban population of Cameroon for three reasons: (1) migrants are probably different from those who remain in the provinces, (2) people who join tontines may differ from those who do not in ways that are relevant for their contraceptive use, and (3) the tontines that were selected by our key informants may differ from other tontines in Yaound& That our sample is not representative of Cameroon inhibits generalization; the sample does, however, offer an unusual opportunity to explore the process by which social interaction may be associated with contraceptive use.
FINDINGS
Perception o f networks partners approval, use, and encouragement of family planning It is useful to begin with a brief description of the networks of the tontine members. Most relevant is that these networks are composed of women whom the respondent knows well and with whom she converses often, a result similar to that found for networks of Kenyan women (Watkins et al., 1995). More than 80% of the network partners were women whom the respondents had known for three years or more and women with whom the respondent talked with at least once a week. Since most of the tontines meet monthly, this suggests considerable interaction outside the meetings. Family members constituted 32% of the network partners; women of common ethnicity but not family members 26%; and friends but not family or of common ethnicity 35%. The respondent and her network partners were not as similar to one another as we had expected: the correlation between the respondent's age, education, and wealth and the average age, education, and wealth of her network members were -0.02, 0.42 and 0.33, respectively. Nearly half of the sample was nominated as a net*Here (and in the following table) we report only bivariate or multivariate odds ratios; readers interested in obtaining the percentages, or similar tables (with similar results) for current contraceptive use may do so by contacting the senior author. tOf the respondents who reported ever-use of contraception, the most popular methods were the pill (27%), the condom (21%), rhythm (17%) and abstinence (9%); all other methods had 5% or fewer users.
work partner by two or fewer respondents, but about 4% were nominated 10 times or more--these women, it appears, are frequently mentioned as conversation partners. Most respondents claimed to know the contraceptive attitudes and behavior of their network partners. Respondents believed that, of the women in their personal network, 73% had heard of contraceptives and 4% had not heard of contraceptives (for 23% they did not whether they had or had not heard). Respondents also believed that, of the women in their personal network, 68% approved of contraceptives and 6% did not approve of contraceptives (for 26% they did not know whether they approved or not). If discussing one's own contraceptive use is a more private issue than simply exchanging knowledge or generalized expressions of approval, one would expect that the respondents were more likely to say they knew about their partnet's knowledge and approval than their use. This was not, however, the case. For 61% of the network partners, the respondent said that the partners did or did not use contraception, and fully 90% of these proceeded to report whether the network partner used "modern" or "traditional" methods, suggesting that they had discussed specific methods. In Table 2, we regressed the respondent's contraceptive use on her perception of (1) whether or not the members of her network approved of contraceptive use, (2) their use of contraception, and (3) whether or not her network partners encouraged the respondent to use contraception.* Our dependent variables are whether the respondent ever-used a contraceptive (i.e. the respondent has used it, but may or may not be using it currently) and whether she ever-used a clinic-based method (primarily pill use) and whether she used a non-clinic based method (primarily condoms and the calendar method).t Consider first the respondents" ever-use of any form of contraception (column 1). There was clearly an association between the respondent's perceptions of the attitudes and behavior of her network partners, and her own contraceptive use. All three measures were statistically significant for respondents who ever-used any contraceptive method, and for those who ever-used clinic-based methods. Networks do not appear to matter, however, for the non-clinic based methods. We might expect that
Social networks and contraceptive use in Cameroon example is more influential than exhortation, that is, that the perceived use of contraception by one's networks has more power than their words. It is then somewhat surprising that the strongest association was between the respondent's ever-use of contraception and whether or not her network partners encouraged her to use. The likelihood of having ever used a contraceptive method was 7.6 times greater for a respondent who perceived that her network partners used contraception (compared to perceiving that the partner did not use contraception or when the respondent did not know the contraceptive status of her network partners), but it was 16.6 times greater for a respondent who was encouraged to use by her network partners (compared to respondents who said that their network partners had not encouraged them to use). The network partners" perceived use of contraception and their encouragement to use are particularly strongly associated with the respondent's use of clinic-based methods; none of the associations with the respondent's use of non-clinic methods were statistically significant. The respondent's use of contraception may have begun long ago. It is thus possible that associations with current contraceptive use were even stronger (although conversations with network partners may also have occurred in the past, since many respondents had known their network partners for several years, and it is not clear when the encouragement occurred). If we confine our analysis to current contraceptive use (not shown), the patterns are quite similar, although the odds ratio for respondent's perception of her network partner's use is larger than that for her report of the network partner's encouragement (both are statistically significant at p < 0.001).
Choice oJ contraceptive methods We considered, for those respondents who had ever-used contraception, whether the respondent's choice of methods was similar to that of her network, using the network partner's own reports of their contraceptive use. Unlike the previous table, here we use not the perception of the respondent, but self-reports from the network partners themselves. Table 3 shows considerable clustering in the distribution of methods used. Individuals who used a particular method were more likely to be surrounded by others who reported knowing of and using that same method. For example, IUD users were 3.1 times as likely as women who used other methods to have a network composed of people aware of the IUD and 2.1 times as likely to have a network composed of 1UD users. The largest odds ratios were for condom awareness and use, followed by that of the diaphragm and withdrawal. It is likely that women's conversations about specific methods were not confined to the five
681
Table 3. Bivariate odds ratios for respondent's knowledge and ever-use of specific contraceptive methods and the proportion of network partners who knew and used the same method Contraceptive method IUD Pill Diaphragm Spermicide Condom Rhythm Withdrawal Abstinence
Odds ratio for knew same method
Odds ratio for used same method
3.1' 2.5* 13.4"* 5.1"* 16.3"** 7.7*** 10.0"** 2.7**
2.1 1.5 NA ~' 12.5'* 12.1"** 6.2*** 8.4*** 6.3***
*p < 0.05: **p < 0.01: ***p < 0.001. N - 477. ~Only one respondent reported use of a diaphragm.
women they named as the ones they talked with most frequently. We do not examine this directly here, but we did estimate whether the nine tontines differed significantly from one another in the specific methods that predominated among the members of that tontine. For the 13 specific methods, a one-way ANOVA showed some significant differences, particularly in the use of non-clinic based methods.
Correct knowledge ~f network's contraceptive behavior Many scholars hold that perceptions are important: things that are real to the actor are real in their consequences. Prior analysis of network data in Korea (Montgomery and Chung, 1994) found that perceptions were more strongly associated with the respondent's contraceptive use than was the network partner's actual behavior. This result is somewhat puzzling, since it seems plausible that those women who correctly knew that their network partners used contraceptives would be more likely to imitate that behavior, perhaps because there was a direct transfer of information about contraceptive methods (how it works, the side effects), perhaps because they were reassured that their own use would be acceptable to people who matter to them. Nonetheless, our results also show that certain perceptions may be more important than behavior. There is likely to be considerable "noise" in the comparison of a network partner's contraceptive use as reported by the respondent and as reported by the network partner (when she herself is a respondent). We asked respondents whether they thought their network partner uses a family planning method, implying current use; some respondents may have interpreted this as ever-use. In addition, a conversation about use may have occurred some time ago: since women often discontinue contraceptive use, a respondent may be unaware that her network partner has discontinued. Despite these sources of inaccuracy, respondents were right about as often as they were wrong in reporting on their network partner's contraceptive status: 31% were correct, 28% were incorrect (the
Thomas W. Valente et al.
682
Table 4. Bivariate odds ratios for respondent's contraceptive status and her knowledge of the contraceptive status of her network partners (NWPs) Respondent's knowledge of NWPs contraceptive status (%) Correct (31%) NWPs correctly thought used NWPs correctly thought did not use Incorrect (28%) NWPs incorrectlythought used NWPs incorrectlythought did not use Did not know (42%) NWPs respondents did not know NWPs not in sample
Respondent's contraceptivestatus Ever-used a method Any Clinic Non-clinic
22 9
24.1*** 0~8
5.3"** 0.6
1.4 0.7
22 6
18.2"** 0.1"*
9.0*** 0.2*
1.2 0.4
39 3
0.3** 0.3
0.3** 0.6
0.9 2.2
*p < 0.05; **p < 0.01; ***p < 0.001.
N - 477. rest said they didn't know). In response to the question as to whether the network partner uses " m o d e r n " or "traditional" methods, respondents were likely to over-estimate the use of modern methods by their network partners. Table 4 shows that it is not that the correctness of the respondents' knowledge about her network partners use of fertility control, but rather that they perceive that the network partners are using, whether or not they are correct in this attribution. This applies, however, only to clinic-based methods. Those who claimed they did not know the contraceptive status of their network partners were less likely to use than those who claimed they did, correctly or incorrectly.
Network characteristics If contraceptive use is at least in part a function of interaction within a personal network, the next question is: what are the characteristics of personal networks that are most strongly associated with the use of contraception? Following the network literature, we focus on the heterogeneity/homogeneity of networks (Granovetter, 1973; Liu and Duff, 1972; Marsden, 1987). Heterogeneous networks are likely to provide their members with more new information than are homogeneous networks that are composed of individuals who are very similar to one another (Valente, 1995). On the other hand, when networks are homogeneous, their members may be perceived to be more credible, and the information and opinions exchanged among members more trustworthy (Rutenberg and Watkins, 1996). Heterogeneity/homogeneity is calculated using the standard deviation of the scores for a respondent's network. We also calculated the average scores, under the assumption that women whose networks were, for example, more highly educated may be more likely to use contraception. (Wealth and education were not highly correlated in this data set (r = 0.24).) The education and wealth heterogeneity of the respondents' networks were not associated with contraceptive use. Age heterogeneity was negatively associated with contraceptive use indicating that
those respondents with personal networks of similar ages were more likely to have ever-used contraception. In terms of familial status, we found a slight tendency for respondents with personal networks of many family members were more likely to have ever-used contraception.
Multivariate analysis Do the bivariate associations that we have found between a respondent's contraceptive use, her perception of her network partners attitudes, contraceptive behavior and encouragement, the accuracy of her perceptions and the characteristics of her networks hold in a multivariate context? In Table 5 we show a series of logit regression models (Aldrich and Nelson, 1984), where the dependent variable is the most general of the measures of contraception that we have been discussing: the respondent's everuse of any contraceptive method. We begin in Model 1 with individual characteristics often found to be associated with contraceptive use: the respondent's age, education, and wealth. Only wealth was statistically significant: each additional possession was associated with a 30% increase in the likelihood that the respondent used contraception. Model 2 adds the characteristics of the respondent's networks: whether they are heterogeneous/ homogeneous with respect to age, education, and wealth. As noted earlier, network variance is theoretically important. (In preliminary multivariate analyses we also included the average age, educational level and wealth of the networks in the full model, but these measures were not statistically significant). Although adding these variables does not significantly improve the fit of the model, we found that a narrow range of ages of the respondent's network partners was associated with a greater likelihood that she used contraception. In Model 3 we added the respondenfs perception of her network partners' support for contraception: whether they approve and encourage her to use. Adding these variables improved the fit of the model significantls,. As was the case in the bivariate analysis presented earlier, the encouragement of network partners was statistically significant and substantively large (the
Social n e t w o r k s a n d c o n t r a c e p t i v e use in C a m e r o o n
683
Table 5. Multivariate odds ratios for respondent's contraceptive status, her individual characteristics, the characteristics of her network partners (NWPs), her perceptions of her NWPs, and her knowledge of her NWPs contraceptive status Respondent ever-used any contraception Predictors Respondent's: Age Education Wealth Standard deviation of NWPs': Age Education Wealth Respondent's perception of NWPs: Approval Encouraged use Respondent's knowledge of NWPs contraceptive status: Correctly thought used Correctly thought did not use Incorrectly thought used Incorrectly thought did not use Log liklihood Pseudo R 2
I
2
3
4
0.9 1.0 1.3"**
1.0 1.0 1.3'**
1.0 1.0 1.1
1.0 1.0 1.1
0.8** 2.9 1.8
0.8** 2.7 1.2
0.8** 3.2 1.1
1.6 11,2"**
1.2 8.7***
2.5 4.6 1,9 0.3 -172 5%
-167 7%
-145 I7%
-138 20%
*p < 0.05; **p < 0.01; ***p < 0.001. N = 477. Note: All four equations ha',e significant chi-square values (p < 0.001).
odds ratio was 11.2, meaning that respondents whose networks were entirely composed of those who encouraged use were 11 times more likely to have used contraception than those whose networks were entirely composed of those who did not encourage use). When the network partner's approval and encouragement were added, the importance of the respondent's wealth disappeared. Lastly, in Model 4, we asked whether taking into account whether the respondent was correct in her assessment of her network partner's contraceptive status mattered, over and above her perception. Recall that the bivariate analysis had shown that
respondent's who perceived their network partners were using contraception were significantly more likely to use themselves, regardless of the correctness of this perception. In the multivariate context, this no longer held: the respondent's perception was not significant, regardless of whether she thought they were using (either correctly or incorrectly). This change appears to be due to the inclusion of the encouragement variable: when encouragement was excluded from the analysis, the respondent's incorrect perception that her network partners were using became statistically significant (odds ratio 4.4,
p < 0.05).
Table 6. Schematic depiction of Model 4 results for respondent's contraceptive status, her individual characteristics, the characteristics of her network partners (NWPs), her perceptions of her NWPs, and her knowledge of her NWPs contraceptive status
Predictors Respondent's: Age Education Wealth Standard deviation of NWPs': Age Education Wealth Respondent's perception of NWPs: Approval Encouraged use Respondent's knowledge of NWPs contraceptive status: Correctly thought used Correctly thought did not use Incorrectly thought used Incorrectly thought not used
Any
Respondent's contraceptive status Method ever-used Method current use Clinic Non-clinic Any Clinic Non-clinic
/
¢"
,/
,/
¢"
,/
¢'
,/
,/
N = 477. Note: All six equations have significant chi-square values (p < 0.001).
d" ¢,
¢
¢" ¢,
¢,
684
Thomas W. Valente et al.
We did the same analyses using as dependent variables current use of any contraceptive method, ever-use and current use of clinic-based methods, and ever-use and current use of non-clinic based methods. We show the results schematically in Table 6, in order to emphasize the patterns rather than the specific coefficients; using Model 4, we indicate only whether the odds ratios were significant at the .05 level or better. There are several similarities across all or most of the analyses: (1) when network measures were included, the only individual characteristic that mattered was wealth; (2) encouragement to use contraception was important for the respondent's use of contraception; (3) a small variance in the age of the network partners was important; and (4) when the contraceptive use of the network partners was important, it was her incorrect perception (she thought they used but they did not). There were, however, interesting differences in the patterns for the respondent's use of clinic based and non-clinic-based methods. Respondent's wealth was important for the former but not the latter, whereas the educational heterogeneity of the respondent's network partners was important for the latter (the greater the heterogeneity, the more likely the respondent was to use contraception) but not the former. We examined whether including dummy variables for the tontines changed any of these results. Although some of the tontines were significantly different from the others, the significance of none of the odds ratios for the predictor variables was changed. When we performed the analyses excluding the encouragement variable, we again found that the respondents' perceptions that their network partners used (although it might be incorrect) mattered for the respondent's use of any method or clinic-based methods, but not for non-clinic based methods. DISCUSSION
It is clear that the respondents in our sample interact with others in ways that are relevant for understanding their contraceptive use. When asked to name at least five women with whom they talked frequently in the past six months, most women responded with five names, and most said they were women whom they had known for several years and with whom they talked frequently. Although many of their conversations surely did not include family planning, it is likely that some did: 55% of the sample reported that at least one network partner encouraged them to use contraceptives. Women's contraceptive use was associated with the characteristics of her network partners. What is it about these networks that matters? In these data, most important is that the respondent is encouraged by her network partners to use contraception (if encouragement is omitted from the analysis, the respondent's perception that her network partners
use contraception is a weak substitute). We suspect that Cameroonian women are at least somewhat ambivalent about contraceptive use: numerous surveys in African countries and elsewhere have shown women to be quite concerned about the side effects of modern family planning, and qualitative studies in Senegal and Kenya show that women turn not to clinics but to other women like themselves to evaluate these side effects. Some of this evaluation may occur by observation: thus, women may be influenced by the example of those whom they perceive to be using (correctly or incorrectly), since they may interpret this as an indication that the side effects are acceptable. But they are likely--as we have found to be even more convinced by women who directly encourage them. We also find that the variance in the age of networks matters--more precisely, if the network partners are similar in age, the respondent is more likely to use contraception. We suspect that women are likely to talk about family planning with women whose circumstances are similar to their own (Watkins et al., 1995); age is probably a marker for a variety of these circumstances (e.g. a similar number of children). That networks are important does not mean that individual characteristics are not. Our multivariate analyses showed that the respondent's wealth was important for the use of clinic-based methods, although, somewhat surprisingly, the respondent's age and education were not. Wealth was measured by modern possessions and was only weakly correlated with education (0.24), but it may be a proxy for the monetary costs that make clinic attendance more or less likely (e.g. bus fares, someone to take care of the children while the woman goes to the clinic). This is consistent with our finding that wealth does not matter for the use of non-clinic based methods. It may also be that wealth provides contacts (for example, the ability to participate in a tontine). An alternative explanation for our findings about the importance of networks in the use of contraception is that "birds of a feather flock together", that women cluster in tontines or in networks with others like themselves. Thus, they may independently decide to use contraception to control their childbearing, and what appears to be influence either reciprocal or directed may simply be selectivity. Women who belong to these tontines are clearly different: they are more educated and undoubtedly wealthier than other women of Yaound6. The relevant question in this study of networks is not the selectivity of tontines but the selectivity of networks, i.e. whether women selected their network partners from the other tontine members with respect to characteristics that would make them more or less likely to use contraception. The selectivity of networks can best be examined with longitudinal data, which are not available here. Two of our findings, however, suggest that the as-
Social networks and contraceptive use in Cameroon sociation between individual respondents and their network partners is not simply due to selectivity. First is that women who report that their network partners encourage them to use contraception are particularly likely to use it themselves. While the perception that a network partner uses contraception may be a projection of the individual's desires women may attribute to others behavior in which they are engaged (Billy and Udry, 1 9 8 5 ) encouragement seems less likely to be imagined. Thus, that encouragement is important suggests that there is a direct influence from the network partners to the individual respondent. Second is our finding that women who use the same specific contraceptive method flock together. It is unlikely that women cluster together because they use the same contraceptive method, more likely that they choose a condom over withdrawal because they discussed its advantages and disadvantages with the members of their network. Thus, that women encourage each other to use, and are more likely than not to use the same methods as those in their networks, provides support for an interpretation of networks as providing venues for the diffusion of information about specific contraceptives or for social influence. CONCLUSIONS In sum, we have examined the degree to which contraceptive use among members of informal women's associations in Yaound~, Cameroon was associated with a woman's set of strong network ties. Informal associations provide natural boundaries for studying networks, and they enable the application of sophisticated network models. The tontines of Yaound6 provide a rare example for the study of interpersonal communication networks and fertility behavior. What have we learned that might be useful for family planning programs? One implication is that research on the determinants of the use of family planning should not be confined to collecting data on individual characteristics, but should also include information on the respondent's network partners. If, as we have found, their attitudes and behavior are associated with those of the respondent, we should know more about the women with w h o m respondents interact. Second, women appear to be getting some of their information about family planning from their friends and relatives: the clustering of the use of specific methods suggests that they talked about these methods. Yet the method a friend uses may not be medically appropriate for an individual. Thus, discussion by clinic personnel with their clients of the full range of available methods would be useful, and would be likely to spread from informed clients to the members of her networks. We also infer from the importance of a network partner's encouragement that it is likely that a considerable amount of these
685
informal conversations concern side effects. This suggests that family planning programs might do more to provide accurate information on the full range of methods available and the expected side effects associated with each one, so that women have correct information rather than relying only on the experience of their friends and relatives. Lastly, we have found that the perceptions that women have about the behavior of other women is important for their own behavior: perceptions that their friends use, and even more, encourage them to use. This suggests that program planners might take advantage of "satisfied users" to talk to women who remain ambivalent about the use of family planning. Even more effective would be the use of network measurement techniques to determine which "satisfied users" had the largest networks, for example, the 4% of our respondents who were nominated by 10 or more other tontine members. This, we propose, would be a cost-effective way of applying diffusion network approaches to achieve program objectives. Acknowledgements--Our thanks to the women from the women's groups who participated in this study and to Regina Dennis, Richard Greene, Mrs. Kenfack, Damaris Mbass, and Rose Njoh for their assistance in carrying out this project and Rebecca Davis who made helpful comments on prior drafts. This research was supported by USAID cooperative agreement #DPE-3052-A-00-0014-00, Johns Hopkins University, Population Communication Services.
REFERENCES
Aldrich, J. H. and Nelson, F. D. (1984) Linear Probability. Logit and Probit Models. Sage, Newbury Park, CA. Billy, J. O. and Udry, J. R. (1985) Patterns of adolescent friendship and effects on sexual behavior. Social Psychology Quarterly 48, 27-41. Ardener, S. (1964) The comparative study of rotating credit associations. Journal of the Royal Anthropological Institute 94, 201 229. Anderson, R. T. (1966) Rotating credit association in India. Economic Development and Cultural Change 14, 334-339. Bal6pa, M., Fotso, M. and Barr+re, B. (1992) Enqu~te D~mographique et de Sant~ Cameroun." 1991. Macro International Inc., Calverton, MD. Bhatia, S., Faruque, A. S. G. and Chakraborty, J. (1980) Peer pressures and the use of contraceptive sterilization in rural Bangladesh. International Family Planning Perspectives 3, 107-109. Becker, G. S. (1960) An economic analysis of fertility. In Demographic and Economic Change in Developing Countries, pp. 209-231. National Bureau of Economic Research, Princeton University Press, Princeton, NJ. Becket, G. S. (1981) A Treatise on the family. Harvard University Press, Cambridge, MA. Bongaarts, J. and Watkins S. (1995) Social interactions and contemporary fertility transitions. Unpublished manuscript, the Population Council, New York. Cleland, J. and Wilson, C. (1987) Demand theories of the fertility transition: an iconoclastic view. Population Studies 41, 5-30.
686
Thomas W. Valente et al.
Geertz, C. (1962) The rotating credit association: a "middle rung" in development. Economic Development and Cultural Change 10, 241-263. Granovetter, M. (1973) The strength of weak ties. American Journal of Sociology 78, 1360-1380. Hammerslough, C. R. (1991) Women's groups and contraceptive use in rural Kenya. Unpublished manuscript. Jato, M., van der Straten, A., Kumah, O. M., Vondrasek, C. and Tsitsol, L. (1994) Using focus group discussions to explore the role of women's groups (tontines) in family planning information dissemination in Yaound+, Cameroon. Health Transition Review 4, 90-94. Jato, M., van der Straten, A., Kumah, O. M., Vondrasek, C., Tsitsol, L., Seukap, R. and Effiom, C. (1995) Women's tontines in Yaounde, Cameroon: using networks for family planning communication. Project report, Center for Communication Programs, Johns Hopkins School of Public Health, Baltimore, MD. Klovdahl, A. S., Potterat, J. J., Woodhouse, D. E., Muth, J. B., Muth, S. Q and Darrow, W. W. (1994) Social networks and infectious disease: the Colorado Springs study. Social Science & Medicine 38, 79 88. Klovdahl. A. S. (19853 Social networks and the spread of infectious diseases: the AIDS example. Social Science & Medicine 21, 1203-1216. Knoke, D. and Kuklinski, J. H. (1982) Network Analysis. Sage, Newbury Park, CA. Kurtz, D. V. (1973) The rotating credit association: an adaptation to poverty. Human Organization 32, 49-58. Liu, W. T. and Duff, R. W. (1972) The strength in weak ties. Public' Opinion Quarterly 36, 361 366. March, K. S. and Taqqu, R. L. (1986) Women's" lnJbrmal Associations in Developing Countries: Catalysts .[br Change? Westview Press, Boulder, CO. Marsden, P. V. (1987) Core discussion networks among Americans. Annual Review of Sociology 52, 122 131. Marsden, P. V. (1990) Network data and measurement. Annual Review o["Sociology 16, 435 463. Misch, M. R. and Margolin, J. B. (1975) Rural Women's Groups as Potential Change Agents." A Stuclv of Colombia, Korea and the Philippines. George Washington University, Program of Policy Studies in Science and Technology, Washington, DC. Montgomery, M. R. and Chung, W. (1994) Social Networks and the Diffusion of Fertility Control: The Korean Case. Paper presented at the Seminar on Values and Fertility Change, IUSSP, Sion, Switzerland, February. Obbo, C. (1993) HIV transmission through social and geographical networks in Uganda. Social Science & Medicine 36, 949 955. Palmore, J. A. (1968) Awareness sources and stages in the adoption of specific contraceptives. Demography 5, 960 972. Park, H. J., Chang, K. K., Han, D. S. and Lee, S. B. (1974) Mothers' Clubs and Family Planning in Korea. School of Public Health, Seoul National University, Seoul, Korea. Park, H. J., Kincaid, D. L., Chung, K. K., Han, D. S. and Lee, S. B. (1976) The Korean Mothers" Club Program. Studies in Family Planning 7, 275-283. Pollack, R. A. and Watkins, S. C. (19933 Cultural and economic approaches to fertility: proper marriage or mesalliance? Population and Development Review 19, 467 496. Retherford, R. D. and Palmore, J. A. (1983) Diffusion processes affecting fertility regulation. In Determinants O/' Fertility in Developing Countries, eds R. A. Bulatao and R. D. Lee, Vol. 2, pp. 295-337. Academic Press, New York. Rogers, E. M. (1995) Diffusion of Innovations, 4th edn. Free Press, New York.
Rosenfield, A. G., Asavasena, W. and Mikhanorn, J. (1973) Person-to-person communication in Thailand. Studies in Family Planning 14, 145-149. Rutenberg, N. and Watkins, S. C. (1996) Conversation and Contraception in Nyanza Province, Kenya. Paper presented at the annual meeting of the Population Association of America, New Orleans, LA, May. Schultz, T. P. (1976) Determinants of fertility: a microeconomic model of choice. In Economic Factors in Population Growth, ed. A. J. Coale, pp. 89-124. Wiley, New York. Scott, J. (1991) Social Network Analysis." A Handbook. Sage, Newbury Park, CA. Stycos, J. M. and Back, K. W. (19643 The Control oJ Human Fertility in Jamaica. Cornell University Press, Ithaca, NY. Valente, T. W. (1993) Diffusion of innovations and policy decision-making. Journal of Communication 43, 30-45. Valente, T. W. (1995) Network Models" of the Diffusion o[ Innovations. Hampton Press, Cresskill, NJ. Valente, T. W. (1996) Social network thresholds in the diffusion of innovations. Social Networks 18, 69 89. Valente, T. W. and Rogers, E. M. (19953 The origins and development of the diffusion of innovations paradigm as an example of scientific growth. Science Communication: An Interdisciplinary Social Science Journal 16, 238-269. Watkins, S. C. (1994) Social Interaction and Fertility Change. Paper prepared of the "Situating fertility" workshop sponsored by the Social Sciences Research Council, the Johns Hopkins University, Baltimore, MD. Watkins, S. C., Rutenberg, N. and Green, S. (19953 Diffi~sion and Debate." Reproductive Change in Nyanza Province. Kenya. Paper presented at the annual meeting of the Population Association of America, San Francisco, CA, April. Wellman, B. (1988) Structural analysis: from method and metaphor to theory and substance. In Social Structures: A Network Approach, eds B. Wellman and S. D. Berkowitz, pp. 19 61. Cambridge University Press, Cambridge.
APPENDIX A
Network Questions" Included in the Questionnaire (1) Please tell me, within the last six months, the names (complete names) of five people in your group that you talked to most often in the past six months? __.[Questions b through I of asked all five nominations] (2) Is a member of 4 = your group, 3 = another women's group, 2 belongs to no women's group, or 1 -: you don't know'? (3) Do you speak with _ _ 4 = almost every day, 3-at least once a week, 2 = at least once a month, or 1 - less than once a month'? (4) How are you related to ': 5 = lhmily member 4 = tribal member, 3 = coworker 2 = friend or neighbor 1 - other (5) What type of advice have you received from during the past year'? (CIRCLE ALL THAT APPLY): 5 = Financial 4 - Health/child care 3 = Housewifery 2 - Professional 1 Other (6) What type of help have you received from during the past year? (CIRCLE ALL THAT APPLY): 5 = Material 4 = Financial 3 = Labor/ service 2 = Emotional 1 = Other (7) For how long have you known ? For 5 = 3
Social n e t w o r k s a n d c o n t r a c e p t i v e use in C a m e r o o n y e a r s o r m o r e , 4 = 1 - 2 y e a r s , 3 = six m o n t h s to a y e a r , 2 = 3 - 6 m o n t h s , 1 = less t h a n 3 m o n t h s (8) H a s _ _ heard of family planning? 0 = no 1 = yes 2 = D K / N A (9) D o e s approve of family planning? 0 = no 1 = yes2 = DK/NA (10) D o y o u t h i n k _ _ uses a f a m i l y p l a n n i n g
687
m e t h o d ? 0 = n o 1 = yes 2 = D K / N A (11) I f yes, Do you think uses a 0 = traditional or a 1 = modern method? (12) H a s _ _ e v e r e n c o u r a g e d y o u to p r a c t i c e f a m i l y p l a n n i n g ? 0 = n o 1 = yes 2 = D K / N A