This paper was presented at the Association for Education in Journalism
and Mass Communication in Toronto, Canada, August 2004.
If you have questions about this paper, please contact the author
directly. If you have questions about the archives, email
[log in to unmask] For an explanation of the subject line, send email to
[log in to unmask] with just the four words, "get help info aejmc," in the
body (drop the "").
(Oct 2004)
Thank you.
Elliott Parker
************************************************************************
The Contextual Effects of Gender Norms, Communication, and Social Capital
on Family Planning Behaviors in Uganda: A Multi-Level Approach
Byoungkwan Lee ([log in to unmask])
Hye-Jin Paek ([log in to unmask])
Charles T. Salmon ([log in to unmask])
Kim Witte ([log in to unmask])
Paper submitted to the International Communication Division of the
Association for Education in Journalism and Mass Communication, August 4-7,
2004, in Toronto, Canada
Author Note: Byoungkwan Lee (M.A., Michigan State University) is a doctoral
candidate in the Mass Media Ph.D. program at Michigan State University.
Hye-Jin Paek (M.A., University of Wisconsin-Madison) is a doctoral student
in the School of Journalism & Mass Communication at the University of
Wisconsin-Madison. Charles T. Salmon (Ph.D., University of Minnesota) is
Ellis N. Brandt Professor of Public Relations at Michigan State University.
Kim Witte (Ph.D., University of California) is a Professor in the
Department of Communication at Michigan State University.
* Please send all correspondence to the first author at the Department of
Advertising, 569 Communication Arts and Science Bldg., Michigan State
University, East Lansing, MI 48824-1212. Electronic mail:
[log in to unmask], phone: (517) 355- 7581.
Contextual Effects 5
The Contextual Effects of Gender Norms, Communication, and Social Capital
on Family Planning Behaviors in Uganda: A Multi-Level Approach
ABSTRACT
This study hypothesized a multi-level model to examine the contextual
effects of gender norms, exposure to health-related radio programs,
interpersonal communication, and social capital on family planning behavior
in Uganda. The results of HLM showed that all of the four variables were
marginally significant predictors of family planning behavior. We found
that gender norms as a contextual factor significantly interacted with the
individual-level perceived benefit. The significant cross-level interaction
between the contextual variable of exposure to a health-related radio
program and the individual-level variable of interpersonal communication
was also found.
The Contextual Effects of Gender Norms, Communication, and Social Capital
on Family Planning Behaviors in Uganda: A Multi-Level Approach
Issues relating to human reproduction have steadily shaped an important
public health agenda that includes national and international programs.
These programs have intended to address the health and demographic impacts
of high fertility, reduce teenage pregnancies, promote safe motherhood, to
increase child survival, to halt the spread of HIV/AIDS and other sexually
transmitted diseases, and to avert domestic violence against women and
children . In particular, the effectiveness of family planning services on
reducing fertility in developing countries has received increasing academic
attention over the past two decades. Although numerous studies have
examined the determinants of family planning behaviors, especially
contraceptive use, most studies have focused on micro-level (or
individual-level) factors. More specifically, most of these studies have
found the influences of individuals' socio-demographic factors (i.e.,
gender, age, race, education, income, marital status, and religiosity),
family structure (i.e., number in household and number of living children),
or social-psychological variables (i.e., self-efficacy, attitudes, and
beliefs) on family planning behaviors.
However, given that personal values, beliefs and behaviors are always
situated within and shaped by the social context of relationships among
people who share the experience of belonging to a community , understanding
individuals' social surroundings and social factors (contextual effects)
might be important to predict their health-related behavior as well as
considering variables at the individual level. Indeed, many studies have
emphasized and demonstrated the importance of contextual effects in
understanding health outcomes and health-related behavior: mental health ,
sexual behavior , binge drinking among adolescents , youth smoking , and
adolescent drug use . Recently, a number of studies in the area of
reproduction have paid attention to the contextual effects on family
planning behaviors (e.g., .
Despite the studies that have examined the impact of the social context on
health-related behavior, most of the studies have focused on socioeconomic
(i.e., community SES and family structure) or physical and structural
environments (i.e., community size, distance from family planning clinic,
and residence type) as community-level variables. However, the current
study focuses on the effects of contextual factors beyond socioeconomic and
physical environments on family planning behavior. Our interest concerns
how shared values and norms, social network, and social relations within a
community influence individuals' decision and practice regarding family
planning behavior. The purpose of this study is to examine the contextual
effects of gender norms, media use, interpersonal communication, and social
capital on family planning behavior in Uganda. Considering that many public
service announcements (PSAs) and media campaigns focusing on changes at the
individual-level fail, we expect that our findings on contextual effects
will advance the body of health communication literature and provide
campaign practitioners with a much broader and richer understanding of how
social surroundings matter in order to promote media health campaigns.
Social Psychological Predictors of Family Planning Behavior
A substantial body of literature has examined the influences of demographic
and
socioeconomic factors on family planning behavior. Many previous studies
have found levels of education and number of living children as strong
predictors of family planning behavior. For example, women with higher
levels of education are significantly more likely to practice modern
contraceptive methods , and women with at least five living children are
more likely than those with no living children to use contraceptives . In
addition, religion, women's age, and marital status have also been found as
significant predictors of family planning behavior although their findings
are not consistent. For example, the findings showed that single or
formerly married women , younger or middle aged women , and women who are
Catholic are significantly more likely to utilize contraceptives.
In order to examine the effects of individual-level factors on family
planning behavior, as noted earlier, this study focuses on social
psychological variables instead of demographic and socioeconomic variables.
Individuals' beliefs that they can motivate themselves and regulate their
own behavior plays an important role in the process of behavior change in
whether they will consider changing habits that are detrimental to their
health . The conceptualization of perceived self-efficacy in the social
learning theory is an important construct predicting health behavior. In
the context of health communication, perceived self-efficacy can be defined
as "people's beliefs that they can exert control over their own motivation,
thought processes, emotional states, and patterns of behavior" . Since
behavior change depends on one's perceived capability to cope with stress
and boredom and to mobilize one's resources and courses of action required
to meet the situational demands, efficacy beliefs affect the intention to
change risk behavior, the amount of effort expended to attain this goal,
and the persistence to continue striving in spite of barriers and setbacks
that may undermine motivation . Many studies have examined and supported
the hypothesis that individuals' perceived self-efficacy is positively
related to contraceptive behavior (e.g., . The findings of these studies
showed that perceived self-efficacy with contraceptive methods is
positively associated with effective use of contraceptives.
This study also expects beliefs regarding family planning practices to be
an important predictor of family planning behavior; perceived barriers to
and expected benefits (or response efficacy in other literature) of
adopting family planning methods. According to the Health Belief Model
(HBM), a social-psychological model based on value-expectancy theory,
perceived barriers refer to the potential negative aspects of a particular
health action and act as impediments to undertaking the recommended
behavior . In contrast, perceived benefits refer to the perceptions that
the perceived risk would be substantially reduced by taking a specific
action . Perceived benefits is the result of beliefs about the benefits
gained by a particular action weighed against the costs of or barriers to
action . While there have been inconsistent findings on the relationship
between perceived benefits and condom use (e.g., , perceived barriers have
received relatively consistent support as a significant negative predictor
of safer sex behaviors (e.g., .
The Effects of Contextual Factors on Family Planning Behavior
Most behaviors that impact public health are exquisitely interwoven with
social values
and norms that determine healthy or unhealthy practices. Blau (1960) views
social values and norms as "common orientations toward social conduct that
prevail in a society or group" (p. 179). He argued that the common values
and norms in a group have influences on the conduct of its members through
two kinds of fear that an individual feels: fear of his/her conscience and
fear of social sanctions. In other words, individuals conform to prevailing
social values and norms not only to avoid feeling guilty if they did not
but also to gain social approval and to avoid disapproval by doing so . The
psychological mechanism concerning individuals' conformity to group and
authority has been well evidenced in two classic experiments .
Authoritarian social values and norms prevailing in a community can
function as external factors that constrain individuals' own attitudes and
behaviors. An individual's perception about how the community views their
actions affects their health-related behaviors . As such, community beliefs
and norms relating to health-related behaviors affect individuals'
health-related attitudes and behaviors.
Social norms related to gender roles strongly affect sexual and
reproductive health (see for studies and findings on the relations between
sexual and reproductive health and women's empowerment in detail). Gender
roles that are socially constructed and socialized determine differences
between women and men in access to economic resources and decision-making.
Traditional gender norms in African countries, especially, southern African
countries, support male authority and allow men, rather than women, to
access economic resources. Despite some progress in gender equity and/or
equality, most African countries still represent patriarchal, patrilineal
and male-dominant societies. In many patriarchal and male-dominant
societies, culture dictates that 'good' women are ignorant about sex and
passive in sexual interactions . An organization of the United Nations,
United Nations Population Fund (UNFPA), indicates that this makes it
difficult for women to inform themselves about risk reduction, and even
more difficult, even if they are informed, for them to negotiate safer sex
or the use of condoms . Varga (2003) argued that the relationship between
gender norms and sexual risk taking was reflected in adolescents' sexual
negotiation dynamics. Varga (2003) explained that, "a girl's respectability
is gained by her being sexually available to her partner, allowing him
sexual decisionmaking authority, exhibiting coyness and resistance to his
sexual advances, being sexually faithful, and avoiding pregnancy" (p. 163).
Many studies have evidenced a positive relationship between women's
egalitarian beliefs and ability to practice fertility or family planning
behaviors (see . Reciprocally, the use of family planning can also promote
a woman's perception of herself as empowered, resulting in more egalitarian
beliefs in general . Gender norms in a community have been considered as a
contextual factor that affects individuals' reproductive behaviors as well
as sexual behaviors (e.g., . Contextual factors such as gender norms affect
the ability to control one's fertility. For instance, Greenwell (1996)
found that community beliefs concerning childbearing preferences and
reproductive health behaviors have a strong influence on individual
attitudes toward family planning and fertility preferences. Waszak et al.
(2003) argued that traditional gender norms "limit a women's ability to use
family planning if she perceives herself as being bound to cultural
expectations or the will of her husband" (p. 197). As such, it is expected
that the individuals' psychological benefits of practicing family planning
behavior might be mitigated by patriarchal and masculine gender norms
shared in a community.
Another important concern in this study is the influences of mass and
interpersonal communication on family planning behavior. Communication at
all levels personal, family, community, and mass media plays a major
role in the field of reproductive health individual decision-making .
Piotrow et al. (1997) view the role of communication as "the key process
underlying changes in knowledge of the means of contraception, in attitudes
toward fertility control and use of contraceptives, in norms regarding
ideal family size, and in the openness of local cultures to new ideas and
aspirations and new health behavior" (p. 2). Exploring the influence of
communication on reproductive health has been considered one of the
important perspectives to promote people to practice family planning.
Valante and Saba (1998) argue that communication campaigns that disseminate
information about family planning methods and services will increase the
demand for family planning services and eventually lead to reduced fertility.
Many studies have documented that individuals' exposure to mass media
affects their family planning behavior (e.g., . Thus, the findings of these
studies show that exposure to family planning messages in the media
increase the likelihood of practicing contraception. It should be noted,
however, that the effect of mass media on behavioral outcomes might be
mediated by interpersonal communication among individuals. According to a
traditional hypothesis of media effects , the mass media are effective at
changing awareness, knowledge, and attitude, but interpersonal
communication is often necessary for behavioral change. Several studies
have shown that mass media campaigns have an indirect effect on health
behavior through their effects on interpersonal communication, attitudes,
or social norms regarding health-related outcomes (e.g., . For example,
examining the relative influence of mass media and interpersonal
communication on a reproductive health behavior in Bolivia, the findings of
Valante and Saba (1998) indicate that the mass media influence
information-related steps to behavior change such as family planning method
awareness and knowledge, whereas the personal network exposure is more
strongly associated with all behavior change steps. Rimal et al. (1999)
showed that exposure to media campaign messages not only has a direct
effect on overall health behaviors regarding cardiovascular disease, but it
also affects health information seeking and interpersonal communication,
which in turn affect the health behaviors. Storey et al. (1999) also found
that the Radio Communication Project (RCP) in Nepal had a significant
indirect effect on modern family planning, through its effects on
interpersonal communication about family planning.
Social capital is another important contextual factor that affects family
planning behavior. The concept of social capital is defined and used to
explain various forms of social behavior. Social capital studies have
focused on the influences of social factors (i.e., attitudes, norms, and
social networks) on the performance of social, political and economic
institutions. Coleman (1990) argues that social capital contains
multi-faceted entities rather than a single entity - all of which comprises
some aspect of a social structure and facilitates certain actions of
individuals within the structure. Accordingly, he posits that social
capital is productive, making possible the achievement of certain ends that
would not be attainable in its absence. Coleman (1990) suggested three
forms of social capital: 1) obligations and expectations, 2) information,
and 3) norms existing between and among social relations. In a similar but
extended line of argument, Putnam (2000) defines social capital as
"connections among individuals - social networks and the norms of
reciprocity and trustworthiness that arises from them" (p. 19). Thus, he
includes not only networks themselves but also shared norms and mutual
trust, which facilitate coordination and cooperation for mutual benefit.
It should be noted, however, that social capital is "a community-level
variable whose counterpart at the individual level is measured by a
person's social networks" . When we consider social capital as 'community
cohesion' that results from positive aspects of community life, which are
associated with the positive community norms of trust and reciprocity
between community members and with a positive local identity , social
capital can be defined as positive community networks and relationships
between community members. These positive community networks and
relationships enable a community to produce and maintain positive health
outcomes, as these "serve as a buffer to health-damaging stress" .
Indeed, many studies have examined and supported a hypothesis that high
levels of social capital are associated with positive health outcomes: for
example, binge drinking , mortality , HIV infection , and self-rated health
. Kawachi et al. (1997) identified four social capital indicators (social
trust, perceived lack of fairness, perceived helpfulness of others, and
membership in groups) and found that each of these four measures was
significantly associated with income inequality and mortality. Their path
analysis revealed that social capital plays a mediating role in the effect
of income inequality on mortality. Examining the links between HIV-related
sexual health and social capital in a South African mining community,
Campbell et al. (2002) found significant associations between nine
organizational memberships operationalized as social capital and HIV
infection, as well as three risk factors for HIV infection - casual
partners, condom use with casual partners, and alcohol consumption.
Weitzman and Kawachi (2000) also found a positive relationship between the
levels of social capital on college campuses and individual risks of binge
drinking. It is expected that differences in social capital between
communities affect individuals' family planning behavior and moderate the
effects of social psychological variables at the individual level on family
planning behavior. Social capital, as discussed earlier, might play a role
in mitigating and preventing various risk behavior caused by unwanted
pregnancy in communities where positive community networks and
relationships between community members are active.
Based on the above discussion, we suggest a hypothesized multi-level model
to investigate the contextual effects of gender norms, communication
variables, and social capital on family planning behavior. First, at the
individual level, we predict that perceived benefits, self-efficacy, and
social capital would be positively related to family planning behavior, and
that perceived barrier and gender norms (i.e., male dominance) will be
negatively related to family planning behavior. Second, we examine the
effects of four contextual variables (gender norms, listening to a
health-related radio program, interpersonal communication about family
planning, and social capital) measured at the aggregate level. In other
words, whether the four contextual variables measured at the
aggregate-level affect the levels of family planning behavior across
communities is examined. Finally, we explore the possibility of cross-level
interaction, by examining whether the four contextual variables measured at
the aggregate-level moderate the effects of individual-level variables on
family planning behavior. Figure 1 illustrates the hypothesized model.
Figure 1 about here
Method
The present study is based on a secondary analysis of the community survey
data that was collected in Uganda as part of The Delivery of Improved
Services for Health (DISH) project. The project was facilitated through a
bilateral agreement between the Government of Uganda (GOU) and United
States Agency for International Development (USAID), in order to make good
quality reproductive, maternal, and child health services more widely
available and accepted in Uganda; a country that has one of the highest
total fertility and mortality rates in Africa (see for more detailed
information).
A random household sample was drawn from all parishes and villages within a
5-km radius of the health centre. Within each village, systematic random
sampling techniques were used to select households where appropriate adults
older than 18 years old were interviewed face-to-face. The hierarchically
nested data involving individual respondents nested within villages are
suitable for the purpose of our study that explores relationships between
individual-level and contextual-level factors for family planning behavior.
Thus, the villages become our aggregate-level unit of analysis, while
individuals within each village are the individual-level unit of analysis.
Out of the total number of 415 valid respondents within 39 villages,
villages with less than ten respondents were excluded, leading to a total
sample size of 350 individual respondents within 30 villages. Several
scholars argue different criteria for appropriate sample sizes; for
instance, Paterson and Goldstein (1992) argue that a minimum of 25
individuals in each of 25 groups is needed to do useful work. claim that a
larger sample size at the aggregate-level than at the individual-level is
better to see random effects across communities. Although our sample size
at the individual-level is rather small, aggregate-level sample size is
sufficient enough to achieve our goal of examining contextual effects.
Measurement
Five demographic variables serve as control variables: age, gender,
education level, religion, and number of living children. Respondents have
mean age of 31.65 (SD=10.27) with a slightly higher proportion of females
(57.4%) to males (42.0%). Surveyed on a 7-point ordinal scale ranging from
"illiterate" to "tertiary", the sample showed a mean education level of
3.92 that falls between "below primary" and "primary" (SD=1.14). For the
seven religious categories, Catholics (40.3%) and Protestants (43.4 %) were
dominant; thus, each was created as a dummy variable (Catholic =1, other=0;
Protestant=1, other=0). Finally, 59 percent of respondents were married (SD
= .49) with an average number of living children of 3.78 (SD= 2.61). The
six demographic variables including the two dummy variables of religion
were residualized in our model for two reasons: 1) to achieve both control
and model parsimony simultaneously and 2) to take care of the centering
issue with regards to sensible interpretation by using residualized scores
as values of the variables (see for a detailed discussion on the centering
issue).
Independent variables. We include social psychological and communication
variables as independent variables. Except for exposure to health-related
radio program, all of the variables were measured with multiple items;
thus, the multiple items for each variable were averaged or summed to
create each index. Confirmatory factor analyses (CFA) using LISREL 8.30
program were performed in order to confirm the factor structure of each
variable. Appendix B shows detailed item wordings and factor loadings of
CFA for each index with multiple goodness-of-fit indices that were
recommended by structural equation modeling scholars (. Finally, Cronbach's
alpha reliability coefficients were computed to assess internal consistency.
For the psychological variables, perceived barrier was measured by
averaging four items such as "I am worried about side effects from family
planning methods," "Family planning methods are inconvenient," "Family
planning services are hard to get," and "It is embarrassing to get family
planning service." These items were measured with a 5-point Likert scale
ranging from "1" (strongly disagree) to "5" (strongly agree). Cronbach's
alpha coefficients were .63, and factor loadings in the CFA ranged from .56
to .97 with an excellent fit (?2 (2) = 1.17, p = .56, RMSEA = .00, SRMR =
.01, CFI = 1.00).
An index of perceived benefit was constructed by averaging four items using
the same 5-point Likert scale as above: "The pill is effective in
preventing pregnancy," "IUD's are effective in preventing pregnancy,"
"Condoms are effective in preventing pregnancy," and "Implant/injections
are effective in preventing pregnancy" (alpha= .80). The CFA factor
loadings ranged from .67 to .81 with a good fit (?2 (2)= 1.53, p = .46,
RMSEA = .00, SRMR = .01, CFI = 1.00). Likewise, self-efficacy was an
average index of four items with regards to perceived capability or
easiness to use different family planning methods, using the same Likert
scale above (alpha= .68). The CFA factor loadings ranged from .43 to .82
with an excellent fit (?2 (2)= 1.41, p = .49, RMSEA = .00, SRMR = .02, CFI
= 1.00).
Social capital and gender norms indices were created by averaging multiple
items using a 5-point Likert scale ranging from "1" (strongly disagree) to
"5" (strongly agree). Six items of social capital included community
tightness, mutual trust, reciprocity and obligations, and social
participation (see Appendix A for the specific item wordings). Cronbach's
alpha reliability was .76 and the CFA factor loadings ranged from .38 to
.76 with a reasonably good fit (?2 (8)= 32.08, p = .00, RMSEA = .08, SRMR =
.04, CFI = .95). Five items of gender norms indicate traditional sex role
and male preference such as "Boys should help with housework the way girls
do" and "Having male children increases a family's prestige" (alpha= .80).
The CFA factor loadings ranged from .55 to .84 (goodness-of-fit indices; ?2
(5)= 13.89, p = .02, RMSEA = .07, SRMR = .04, CFI = .95). It should be
noted that these two variables also served as contextual variables, which
were measured as an aggregation of their individual-level measures across
villages.
For the communication variables, exposure to health-related radio program
and interpersonal communication regarding family planning behavior served
as both individual-level and contextual-level factors. Exposure to a
health-related radio program was measured with multiplication of the two
measures: "do you have the habit of listening to radio?" (5-point ordinal
scale ranging from "1"(don't listen to radio) to "5" (listen daily)) and
"whether listened to health programs" (1=No, 2=Yes). Interpersonal
communication was an additive index of three binary items (1=yes, 2=no): I
have talked about family planning with 1) spouse, 2) friends, and 3)
siblings (alpha = .71). The CFA factor loadings ranged from .57 to .74.
Since one factor model with three items is just identified, a
goodness-of-fit test showed a perfect fit.
Family planning behavior served as our dependent variable, using a 5-point
scale single item with the wording of "I currently use a family planning
method" (1=never, 5=every time). Missing values contained in all the
variables used here were listwise deleted.
Analysis
Hierarchical linear modeling (HLM) was performed with an estimation of
restricted maximum likelihood across the two levels (individuals nested in
villages). The HLM is often more appropriate than ordinary least squares
regression methods (OLS) because the former acknowledges a unique error
structure at each level, which the OLS procedure does not automatically do .
In order to examine whether a multi-level model is appropriate i.e.,
whether any variance is detected at the multi-level structure intra-class
correlation was computed from the empty model (or full unconditional model,
termed by Raudenbush and Bryk (2002)), which has no variable introduced
with only random error allowed to be free.1 The intra-class correlation
captures the proportion of variance that lies between level-2 units, which
is .15 in this study.2 This finding indicates that about 15 percent of the
variation is accounted for at group level. This amount of variation
explained at group level is moderate, as Snijders and Bosker (1999) note
that intra-class correlations in previous educational research with values
between .05 and .20 are common. Further, the finding indicates that the
variance component of intercept (U0) was about .38 with statistical
significance (?2 = 82.97, df = 29, p < .001), implying that there is a
significant variability in the mean score of family planning behavior
across the 30 villages.
Considering the sensitive and complex nature of the multi-level modeling,
this study starts from an empty model to a random coefficient model step by
step, introducing level-1 (individual) and level-2 (contextual) variables.
This procedure is useful to build the most parsimonious and best-fitting
model and to quantify variance explained in multilevel models with
sense-making interpretation . At each step, the results should be inspected
to see which parameters are significant, and how much residual error is
left at the two distinct levels for optimal hypothesis testing. One venue
of maximum likelihood estimation is to allow for the likelihood ratio test
(or deviance test) between the two models, so as to examine whether more
complex models involving more parameters have a better fit than less
complex models. In this way, we can avoid model over-fitting and the
resulting misinterpretation that accompanies the inclusion of inappropriate
variables.
Results
With the step-by-step procedures described above, Table 1 shows the
development of our model from its simplest form (empty model) to our final
model.3 Our model comprises two components; (1) a fixed effects component
(Gamma coefficients denoted as Gij or _ij, where "i" indicates number of
indicators at individual-level and "j" at aggregate-level) and (2) a random
effects component (tau coefficients denoted as Uij) that tells variability
still remaining unexplained at contextual-level. The fixed effects, again,
comprise individual-level effects in which individual-level independent
variables predict the dependent variable, mean-as-outcome effects in which
level-2 predictors explain mean level of the dependent variable, and
slope-as-outcome or cross-level interaction effects in which the level-2
variables influence the slope coefficients of the individual variables.
Table 1 shows fixed effects coefficients at both levels and Table 2 reports
random components at the aggregate level.
Individual-level Effects
As predicted, perceived barrier was a significantly negative predictor of
the dependent variable (_10 = -.28, p < .01) and perceived benefit was
positively related to family planning behavior (_20 = .30, both at p <
.001). The more people think that family planning behavior is difficult or
inconvenient to adopt, the less likely they are to use a family planning
method. In contrast, people's perceptions about positive consequences and
benefits of using various kinds of family planning methods led them to use
such methods. However, self-efficacy did not significantly predict the
dependent variable. This suggests that people's beliefs of whether they
themselves or their partners were easily able to use family planning
methods did not affect their actual family planning behavior.
The roles of communication variables in family planning behavior were one
of our main interests. The results show that interpersonal communication
was positively (and strongly) related to the family planning behavior (_60
= .32, p < .001), while exposure to a health-related radio program was not
a significant predictor of the dependent variable (_70 = -.01, p = ns).
Thus, the more people discuss family planning methods with their siblings,
friends, and partners, the more likely that they adopted any of the family
planning methods; however, the extent to which they listen to a
health-related radio program did not affect their family planning behavior.
However, no effects of gender norms and social capital, which were measured
at the individual level, on family planning behavior were found.
Table 1 about here
Aggregate-level (Contextual) Effects
Given the effects of the individual-level variables, investigating the role
of contextual factors provides insights about the social environments in
which family planning behavior takes place. This approach is based on the
assumption that associations between individual-level predictors and the
outcome variable differ across the communities. Variability of the
associations across communities was detected with the empty model that has
variance component without any predictor. We found the presence of
variability in the data, which is a rationale for our multi-level analysis.
First of all, we hypothesized that our four contextual variables gender
norms, exposure to health-related radio program, interpersonal
communication about family planning, and social capital influence the
mean level of the family planning behavior (see the second column in
Table1). In our random-intercept model that only allows the intercept to be
random across communities, all of the four variables were marginally
significant predictors of the individual-level dependent variable. In other
words, the overall mean of family planning behavior was lower in
communities which are tightly-knit and people trust and help each other
more (_01= -.63, p < .10); where there were less male preference or
traditional sex roles (_02= -.89, p < .10); where more community members
talk about family planning methods (_03= .37, p < .10); and where more
members listen to health-related radio programs than any other community
(_04= .17, p < .10).
Although all of the four contextual variables were significant in the
random-intercept model, our final model shows that the effects of social
capital and gender norms became nonsignificant after entering cross-level
interactions. Accordingly, the effects of the two communication variables
only remained marginally significant in our final model (_03= .37, p < .10
for interpersonal communication; _04= .21, p < .10 for exposure to
health-related radio program).
Cross-level Interactions
Given the fact that levels of family planning behavior differ across
communities, influenced by the contextual factors, we are compelled to
question whether any contextual-level variable influences the strength of
the effects of individual-level predictors on the dependent variable. Since
there is little multi-level research devoted to exploration of cross-level
interaction and little theoretical background to identify the relationship
between specific level-1 and level-2 variables, we relied upon our
theoretical rationale and exploratory analysis to finalize our model. In so
doing, first, four individual-level variables --perceived benefit,
self-efficacy, social capital, gender norms, and exposure to health-related
radio program -- were excluded (or fixed at level-1), because they were not
significant predictors of the dependent variable and showed no random
effects at the contextual level in a preliminary random-coefficient model
that allows all the individual- level variables allowed to be random.
Second, HLM exploratory analysis was performed to see which contextual
variable was a potential candidate as a moderator of the relationship
between individual-level variables and the dependent variable. When there
are no strong theories concerning which variables play significant roles as
aggregate-level predictors, Hox (1995) recommends performing the procedure
by starting with the simplest possible model and including the various
types of parameters step by step. In addition, we had some ideas, although
not strong, that gender norms would be an aggregate-level moderator of
social psychological variables such as perceived barrier and perceived
benefit and the contextual variable of exposure to health-related radio
program would interact with the interpersonal communication variable. The
third cell in Table 1 indicates our final model that contains the
cross-level interaction.
The model shows that, indeed, gender norms as a contextual factor
significantly interacted with the individual-level perceived benefit
variable (_21 = -96, p < .01). It implies that perceived benefit is a
positive and strong predictor of family planning behavior, but its effect
became weaker in communities where strong male-centered norms are
prevalent. However, the cross-level interaction between gender norms and
perceived barrier was not statistically significant, unlike our expectations.
The most interesting finding lies in the significant cross-level
interaction between the contextual variable of exposure to a health-related
radio program and the individual-level variable of interpersonal
communication (_61 = .16, p < .05). Talking about family planning with
people to whom you can easily approach was a strong predictor of adopting a
family planning method at individual-level; then, such effects of the
variable became even stronger in communities where people listen to
health-related radio programs more in general.
Random Effects: Still Variance Unexplained?
After entering all the hypothesized individual-level and
contextual-variables, some variance still remained across communities, as
shown in Table 2. It implies that there are still contextual variables that
are unknown and unexplained in our model, which makes variability of mean
level of family planning behavior (U0 = .17, p < .01) and that explains
variability of the effects of perceived barrier across communities (U1 =
.40, p < .05).
Table 2 about here
Model Fit Test
Finally, we should be able to justify why our final model that further
fixed the random component of perceived benefit (U2) and of interpersonal
talk (U6) is optimal. In so doing, deviance (or likelihood-ratio) tests
were performed to determine that our final model is not over-fitted and is
indeed best-fitted (and parsimonious) among available models (see Table 2).
When compared to the random intercept model that allows only the intercept
coefficient to be random, our model shows a significant fit improvement (?2
difference test; _2D (2) = 16.65, p < .001). In addition, the more complex
model, which allows U2 and U6 variance components to be random across
communities, fails to show a significant fit improvement in comparison to
ours (_2D (7) = 0.97, p = ns).
Discussion and Conclusions
The purpose of this study was to examine the contextual effects of gender
norms, media use, interpersonal communication, and social capital on family
planning behavior in Uganda. For this purpose, we tested three hypotheses:
1) the effects of several social psychological variables as well as gender
norms, mass and interpersonal communication, and social capital, which were
measured at the individual level, 2) the effects of the four contextual
variables measured at the aggregate level, and 3) cross-level interactions
between the individual level and contextual variables.
As hypothesized, we found two social psychological variables such as
perceived benefit and perceived barrier as significant predictors of family
planning behavior, which is consistent with previous findings in the HBM
literature (e.g., , although we could not identify self-efficacy as a
significant predictor.
Interestingly, and as was somewhat expected, social capital and gender
norms measured at the individual level were not significant predictors of
family planning behavior. As we argued earlier, these two variables have an
inherently more collective and aggregate nature; therefore it is
appropriate for them to be treated as contextual variables. As such, it
appears that these findings might provide partial evidence of our argument.
According to the results, gender norms as a contextual variable affect the
levels of family planning behavior across communities, even after
controlling for individual level variables. In other words, the more norms
in a village emphasize traditional gender roles (i.e., male dominance in
decisionmaking and boy preferences), the less likely village members are to
adopt a family planning method. It should be noted, however, that there was
no effect of gender norms on family planning behavior at the individual
level. These findings suggest that traditional gender norms as a contextual
factor function as a social mechanism to limit individuals' ability to use
family planning. This reasoning might be supported by the findings on
cross-level interaction between the contextual variable of gender norms and
perceived benefit. The effect of perceived benefit on family planning
behavior was significantly moderated by gender norms at the aggregate
level. Thus, as male-centered norms in a village weaken individuals'
expectation of benefits gained from utilizing a family planning method,
individuals' intention and ability to adopt a family planning method might
be diminished. These findings can be explained by Zaltman and Duncan's
(1977) theory regarding resistance to change. Zaltman and Duncan (1977)
argued that conformity to social norms is one of the strong forces for
resistance to changes or adopting innovations. They noted, "norms provide
stability and behavioral guidelines that define what individuals can expect
from one another
any change that is incompatible with existing norms will
tend to be resisted by most members of the social system" (Zaltman &
Duncan, 1977, p. 74).
At the individual level, while the effect of exposure to a health-related
media program on family planning behavior was not found, interpersonal
communication significantly affected family planning behavior. It seems, as
discussed earlier, that these findings support a notion that while
interpersonal communication is a direct predictor of behavioral change, the
mass media channel is more effective at changing awareness, knowledge, or
attitude. In other words, the findings on the cross-level interaction
between the contextual variable of exposure to a health-related radio
program and the individual-level variable of interpersonal communication
seem to support this argument. The effects of interpersonal communication
became even stronger in communities where people listened to health-related
radio programs more in general. The amount of mass media use at the
aggregate-level reflects the amount of information exchanged and consumed
by members. If this is the case, exposure to a health-related media program
at the aggregate-level might determine the degree of interpersonal
communication regarding family planning behavior. As Chaffee noted (1982),
"the more people talk with other people about information from the mass
media, the greater is the total impact of the media on social action"
(p.76). In these regards, we need to recall the findings of Cammack and
Heaton (2001), which indicated that higher exposure at the regional level
is associated with greater family planning program success beyond exposure
to the mass media at the individual level.
It should be noted, however, that although the effects of the other three
contextual variables were in the expected positive directions, the
contextual effect of social capital on the mean level of family planning
behavior was in a negative direction. This result can be explained in line
with the role of gender norms in practicing family planning behavior. When
we view social capital as obligations and expectations, information, and
norms existing between and among social relations (Coleman, 1990), we
cannot completely rule out the possibility that social capital can
negatively function depending on what the dominant values within a society
are. For example, in a traditional society like Uganda, patriarchal and
male-dominant values are regarded as an important value to be maintained
and conformed to by the members. In fact, Zaltman and Duncan (1977)
indicated that group solidarity might be a strong force for resistance "if
the subsystem that is the object of change satisfies important needs of
functions in many other parts of the social system" . As such, social
capital in Uganda might result in reinforcing the existing values or norms
regarding biased gender roles. Campbell et al. (2002) argued that even
though social capital might be associated with beneficial outcomes in some
contexts, this might not always be the case. They indicated that the main
current of social capital research has failed to explain the fact that
community networks and relationships can negatively function. Their
findings provided empirical evidence on the concept of 'negative social
capital'. For instance, they found that stokvel (voluntary savings clubs
accompanied by social festivities) membership was associated with increased
sexual health risks.
Through an extension of the theoretical perspectives on health
communication, this study has several significant practical implications in
the context of family planning communication campaigns. First, the findings
of this study suggest that communication campaigns for family planning
should be designed to change community norms that prevent health behavior
as well as individuals' behavior. In fact, community-based health campaigns
as a new approach for successful campaign management have garnered recent
attention . Second, campaign designers might need to be concerned about the
different roles of mass and interpersonal communication. Thus, while a mass
media campaign might play an important role at the early stages in behavior
change (i.e., awareness, knowledge, or persuasion), the interpersonal
network may be effective at the later stages (i.e., decision, action, or
maintenance). For example, through interpersonal networks, women might
decide whether or not to continue using the contraceptive method that they
adopted. In these regards, Rudy, Tabbutt-Henry, Schaefer, and McQuide
(2003) emphasize the client's role in maintaining the adoption of a family
planning method. In addition, the findings in this study imply that family
planning practices should not be simply considered as individual reactions
or mere reactions to the social structure, but should be understood as an
interaction of people's everyday activities with their surroundings and
resources . As such, for an efficient and successful campaign, its plan and
implementation need to be connected with various social organizations, such
as medical or political organizations.
Despite these implications, several limitations of this study should be
noted. Due to the complex nature of multi-level analysis, inclusion of one
additional variable often makes the entire model much more complicated and
harder to interpret. Therefore, all demographic variables in this study
were residualized for the purpose of achieving model parsimony and control
simultaneously. However, some may be interested in the effects of the
demographic variables on family planning behavior. The preliminary
regression analysis showed that the younger, the more educated, and the
more children they have, the more likely they are to adopt family planning
behavior, consistent with those of previous studies.
Although multi-level analysis is a valuable approach to capture contextual
effects on individuals' health-related actions in Uganda (a country where a
primary society or gemeinschaft still exists), when we apply our findings
to a more urbanized and developed society like the United States, a
fundamental question can be raised concerning the actual boundary of
contextual effects or conceptualization of community. As such, future
replication using data collected from an urbanized and developed country is
suggested.
Contextual Variables as Predictors
Level-1
Individual-level
Level-2
Aggregate-level
(contextual
variables)
Figure 1. A Hypothesized Model of Multi-Level Approach to Family Planning
Behavior
Gender Norms
Mass Media Exposure
Interpersonal Communication
Social Capital
Cross Level Interaction
Family Planning Behavior
Self-Efficacy
Perceived Barriers
Perceived Benefits
Mass Media Exposure
Interpersonal Communication
Gender Norms
Social Capital
Table 1. Random-Coefficient Regression Model of Individual and Contextual
Predictors of Family Planning Behavior
Empty
Model
(full unconditional)
Random
Intercept
Model
(conditional)
Final
Model
(full conditional)
Coefficient
S.E.
Coefficient
S.E.
Coefficient
S.E.
Mean-as-outcome
Intercept (_00)
2.38***
.14
2.36***
.10
2.33***
.10
SOCIAL CAPITAL (_01)
-.63#
.37
-.42
.32
GENDER NORM (_02)
-.89#
.47
-.71
.44
TALK ABOUT FP (_03)
.37#
.21
.35#
.20
LISTENING TO RADIO (_04)
.17#
.10
.21#
.10
Slope coefficient
(Individual-level predictors)
Perceived barrier (_10)
-.22*
.11
-.28*
.11
Perceived benefit (_20)
.31**
.08
.30***
.07
Self-efficacy (_30)
-.05
.08
-.06
.08
Social capital (_40)
.06
.12
.05
.11
Gender norm (_50)
-.14
.12
-.14
.11
Talk about family planning (_60)
.28***
.07
.32***
.07
Listening to health-related radio (_70)
-.00
.03
-.01
.03
Cross-level interaction
(Slope-as-outcome)
Perceived barrier x
GENDER NORM (_11)
.13
.72
Perceived benefit x
GENDER NORM (_21)
-.96**
.34
Talk about family planning x
LISTENING TO RADIO (_61)
.16*
.07
1. Significant test for fixed effects follows t-ratio distribution; # p <
.10, * p < .05, ** p < .01, *** p < .001
3. Variables with capital letters indicate aggregate-level predictors
(aggregated score of individual-level variables)
Table 2. Random Effects across Communities
Empty
Model
Random
Intercept
Model
Random Intercept
-Random Slope
Model
Variance component
SD
Variance component
SD
Variance component
SD
Intercept (U0)
.38***
.62
.44***
.20
.17**
.17
Perceived barrier (U1)
.40*
.16
Level-1 effect (R)
2.22
1.49
1.98
1.41
1.36
1.84
Intra-class correlation
.15
Deviance
1238.31
1206.51
1191.86
# of parameter
2
2
4
_2D test (dfD)
14.65 (2)***
1. Chi-square distribution; * p < .05, ** p < .01, *** p < .001
3. Restricted Maximum Likelihood estimation
4. _2D test is performed for comparison between random intercept model and
random slope model (including all the explanatory variables at two levels
Notes
1 The formula is as follows:
Level-1 Model
Y = B0 + R
Level-2 Model
B0 = G00 + U0
In the level-1 model equation, Y is the individual-level dependent variable
(family planning behavior), B0 is the mean score (intercept) for the group
(villages), and R is a random error per each individual that is normally
distributed with mean 0 and variance s2.
In the level-2 model equation, G00 is the grand mean of the dependent
variable and U0 is a random error term that is normally distributed with
mean 0 and variance t2.
2 The intraclass correlation coefficient _I can be defined as:
Population variance between
macro-units _2
_I
= ------------------------------------------------------------- =
--------------
total
variance _2 + _2
3 The equation for our final model is as follows:
Level-1 (individual-level) Model
Y = B0 + B1*(perceived barrier) + B2*(perceived benefit) +
B3*(self-efficacy) + B4*(social capital) + B5*(gender norm) +
B6*(interpersonal talk) + B7*(exposure to radio) + R
Level-2 (aggregate-level) Model
B0 = G00 + G01*(SOCIAL CAPITAL) + G02*(GENDER NORM) + G03*(INTERPERONSL
TALK) + G04*(EXPOSURE TO RADIO) + U0
B1 = G10 + G11*(GENDER NORM) + U1
B2 = G20 + G21*(GENDER NORM)
B3 = G30
B4 = G40
B5 = G50
B6 = G60 + G61*(EXPOSURE TO RADIO)
B7 = G70
Here, B0 indicates mean intercept at individual level, which is an outcome
of the average intercept G00 and fixed coefficients G01 through G04, and
random component, U0 at community-level. B1 to B7 are regression
coefficients at individual-level, which also serve depend on
aggregate-level regression coefficients, G11, G21, and G61 at community level.
Finally, R is variance component (or residual) at level-1, while U0 to U8
are random components at community-level, indicating variability across
communities.
Appendix A. Descriptive Statistics of the Variables
Variables b
Individual level
(n = 350)a
Aggregate level
(N=30)
Mean
SD
Mean
SD
Age
31.65
10.27
--
--
Gender (2=female)
1.58
.49
--
--
Married
.59
.49
--
--
Education level
3.92
1.14
--
--
Number of living child
3.78
2.61
--
--
Protestant (dummy)
.44
.50
--
--
Catholic (dummy)
.41
.49
--
--
Perceived barrier
2.47
.84
--
--
Response efficacy
3.63
1.01
--
--
Self-efficacy
3.13
1.08
--
--
Family planning behavior
2.33
1.60
--
--
Social capital c
3.57
.82
3.57
.31
Gender norm c
3.58
.86
3.57
.32
Listening to health-related radio c
6.22
2.89
6.25
1.04
Talk about family planning behavior c
4.72
1.28
4.74
.47
a There were a number of missing variables across the variables, which were
listwise deleted in the analysis.
b Dependent variable is family planning behavior at the individual level.
Demographic variables were residualized onto independent variables, thus,
were not included in our final model.
c The four variables were aggregated and served as contextual variables in
our analysis.
Appendix B. Factor Loadings and Goodness-of-fit indices of Confirmatory
Factor Analysis a
Perceived
Barrier
Perceived Benefits
Self-efficacy
Social Capital b
Gender
Norm b
Talk c
I am worried about side effects from family planning methods.
.97
Family planning methods are inconvenient.
.59
Family planning services are hard to get.
.63
It is embarrassing to get family planning service.
.56
The pill is effective in preventing pregnancy.
.68
IUD's are effective in preventing pregnancy.
.81
Condoms are effective in preventing pregnancy.
.67
Implant/injections are effective in preventing pregnancy.
.69
I/my partner (is easily able to) use the pill to prevent pregnancy.
.82
I/my partner (is easily able to) use Norplant/injectables to prevent pregnancy.
.63
I/my partner (is easily able to) get a tubal ligation to prevent pregnancy.
.48
I/my partner (is easily able to) use condoms to prevent pregnancy.
.43
This is a close-knit community.
.76
People in this community can be trusted.
.58
People in this community will intervene if someone's children were engaging
in delinquent behavior.
.46
In this community, if you perform a favor for one of your neighbors they
will likely return the favor at some future date.
.54
One of the norms in this community is that people help one another.
.76
In this community, it is common for ordinary people to participate in one
or more social, business, or community organizations.
.38
Boys should help with housework the way girls do.
.55
The family line continues only through male children.
.57
Having male children increases a family's prestige.
.84
Having a son is essential for performing the last rites of parents.
.67
Sons can provide economic security in the parent's old age but not daughters.
.77
I have talked with spouse about family planning.
.57
I have talked with friends about family planning.
.74
I have talked with siblings about family planning.
.71
Goodness-of-fit indices
?2 (df), p-value
1.17 (2), p= .56
1.53(2),
p= .46
1.41(2),
p= .49
32.08(8),
p= .00
13.89(5),
p= .02
.00(0),
p=1.00
RMSEA
.00
.00
.00
.08
.07
--
SRMR
.01
.01
.02
.04
.04
--
CFI
1.00
1.00
1.00
.95
.95
--
a A series of the CFA analyses were performed for each variable, because
the main goal was to confirm the factor structure of each variable rather
than to perform structural equation modeling. Factor loadings are reported
from completely standardized solution.
b Although the Chi-square statistic shows that the two models are not a
good fit, the other goodness-of-fit indices indicated that the models were
reasonably a good fit.
c The CFA with three items just identify the model, as the goodness-of-fit
indices indicate the perfect fit. Thus, the model does not enable us to
gauge the overall fit test, however, it can produce the factor loadings of
the model.
|
