Professional Documents
Culture Documents
Cult Med Psychiatry. 1997. v. 21, N. 3. P. 303-35. (22956)
Cult Med Psychiatry. 1997. v. 21, N. 3. P. 303-35. (22956)
ABSTRACT. The association of social support and health outcomes has received consider-
able attention in recent years, but the cultural dimension of social support has not been
extensively investigated. In this paper, using data collected in a Brazilian city, we present
results indicating that those individuals whose reported access to social support more
closely approximates an ideal cultural model of access to social support have lower blood
pressure and report fewer depressive symptoms and lower levels of perceived stress. The
cultural model of social support is derived using a combination of participant observa-
tion, semi-structured interviews, and the systematic ethnographic technique of cultural
consensus modelling. These results are then used to develop a measure of an individual’s
approximation to that model of social support in a survey of four diverse neighborhoods in
the city (n = 250). We call this approximation to the ideal cultural model of social support
“cultural consonance” in social support. The association of health outcomes with cultural
consonance in social support is independent of individual differences in the reporting of
social support, and of standard covariates. In the case of blood pressure and perceived stress,
it is independent of diet, and other socioeconomic and psychosocial variables. The asso-
ciation with depressive symptoms is not independent of other psychosocial variables. The
implications of these results are discussed with respect to research on cultural dimensions
of the distribution of disease.
INTRODUCTION
that context. This process can often remain mysterious; going from general
cultural patterns to specific measurements can require considerable inter-
pretation, and hence can be easily contested. For example, in the case of the
African American community, the generational differences in the effects
of social support could be attributed to lifespan developmental processes
in the formation of social relationships, as opposed to the sociocultural-
processual model proposed by Dressler (1991a). Or, rather than having
identified a cultural dimension of social support, it might be argued that
kin and nonkin are merely providing special resources relevant to the
specific stressors faced by the young and old.
It would be useful, in other words, to systematize the inferential steps
leading to the derivation of measures of social support in such a way as
to clearly specify the cultural dimension of social support, and to contrast
this dimension of social support with other ways of looking at the process
(e.g., an individual difference model). Our aim in this paper is to do just
that, employing culture theory and related methods.
tation, the most abstract organizing scheme for social support systems is
the set of definitions, expectations, and strategies for social interaction that
can be referred to as social structure. Social structure defines with whom
interaction is permitted, prohibited or preferred, and what sorts of trans-
actions within interactions can be expected. These rules and expectations
will only be incompletely realized, however, due to various sorts of social
environmental constraints. The observable realization of social structural
ideals then is referred to as social organization. Stepping down another
level of abstraction, the kinds of social relationships centered on the indi-
vidual can be described as the ego-centered social network. It is to this
network that the individual will turn for help during times of felt need; the
belief or perception that such help is (or would be) forthcoming in such
times is what will be referred to as social support.
The question then becomes, how can the cultural construction of social
support within a particular setting be examined in such a way that this
information can in turn be used to study the influence of cultural defini-
tions of support on the distribution of health status? What is required is
some means of defining a shared model of social support, and in turn deter-
mining to what extent individuals in a sample adhere to or approximate this
shared model. The way in which health status is distributed relative to indi-
vidual approximation to this cultural model of social support can then be
examined. In the remainder of this paper, these issues will be examined in
the context of research in a Brazilian city, applying the cultural consensus
model of Romney, Batchelder and Weller (1986) to the study of cultural
models of social support, and integrating the use of that model with more
conventional ethnographic methods and traditional survey techniques.
The research on which this study is based was carried out in the city of
Ribeirão Preto, a city of approximately 500,000 population in the state
of São Paulo. Ribeirão Preto sits in the middle of the richest agricultural
region in Brazil. Much of the land surrounding the city is devoted to
the cultivation of sugar cane, which is refined into sugar and used in the
production of alcohol fuel for automobiles. Significant amounts of land
are also devoted to the cultivation of coffee and citrus. The city was
founded in the late nineteenth century as a market center for the large
farms (fazendas) in the region. Later, it became important as a rail center,
and for light manufacturing such as the brewing of beer. The city grew
dramatically in the years following the second world war as a financial
service center and a regional center for health care. Today it is among the
most affluent cities in Brazil.
One of the challenges of this research was to adequately sample the full
range of socioeconomic variation in the city. To do so, we first identified
four neighborhoods in the city that spanned that range. Households were
then sampled within those neighborhoods for more intensive study. The
poorest neighborhood was a favela on the edge of the city. Favelas are semi-
legal squatter settlements made up both of migrants seeking to partake of
the affluence of the region, and of the socially marginal who have not been
able to attain economic stability. They are semi-legal in the sense that the
building of houses on the site has not been approved by the municipality,
but the residents nevertheless pay rent on the house sites to the owner of
the land. In fact, upon leaving the favela, residents will even sell their
houses to new arrivals. Houses in a favela vary enormously, ranging from
ragtag affairs hammered together from castoff lumber and corrugated tin,
to concrete block and multi-roomed houses with electricity siphoned from
a passing power line. People in the favela tend to be unstably employed as
unskilled laborers and domestics.
The second neighborhood sampled was a conjunto habitacional. A
conjunto is a kind of public housing project. In partnership with the local
city government, a builder will develop houses on a tract of land on the edge
of the city; the conjunto studied here was begun in about 1988 and consisted
of several hundred 3–4 room concrete houses. The houses are then sold
to buyers who qualify for low-cost loans on the basis of being stably
employed. Persons in the conjunto tend to be employed in semi-skilled
jobs such as driving a bus or in construction or low-status service jobs. The
key, however, is the stability of the employment. A conjunto passes from
nondescript sameness to variation in housing styles very quickly, as people
add rooms (sometimes stories), garden walls, garages, and architectural
details to their basic houses. As the conjunto evolves, more basic services
are added, such as small grocery stores, drug stores, and other shops. In this
way, the conjunto takes on the feeling of a small independent community
attached to the larger city.
The third neighborhood sampled was a traditionally lower middle class
area of the city. Early in the century this was a new neighborhood created
by the influx of Italian and Spanish immigrants to Ribeirão Preto; now it
is an old neighborhood completely engulfed by the city proper. The streets
are cobbled (rather than paved) and the houses present seamless walls to
the street. There is a large praça or square with a church in the center
of the neighborhood, and many shops, bars, and restaurants. In short, this
neighborhood has a much more urban feel to it, although it is not in the city
center proper. Houses here are substantially larger than in a conjunto, and
although the economic participation of residents tends to be quite varied,
incomes and occupational statuses are substantially higher. People tend to
be in business, or to work as lower level managers in factories or the public
sector.
The fourth and last neighborhood studied is an upper middle class
area made up primarily of recently constructed houses. The residents are
upper level managers, prosperous businesspersons, and professionals. This
happens to be an attractive neighborhood for physicians and related profes-
sionals, some of whom have studied and taken graduate and postgraduate
degrees in the United States and Europe. The houses verge on being enor-
mous with extensive and well-tended gardens. Although we did not trace
such connections precisely, it would not be unlikely for favelados to be
serving as domestics or tending the gardens in this neighborhood.
Three sets of data were collected in the four neighborhoods that were
sampled. First, there was a survey of a representative sample of households.
In the favela, a complete enumeration of households was obtained from
a community association; forty households were chosen at random from
that listing. For the other three neighborhoods, detailed maps of occupied
house sites were obtained from the city government, and forty households
were chosen at random from each of the neighborhoods. Households were
contacted and invited to participate in the research. We asked the head of
household, spouse (if present), and one child over the age of 18 (if present)
to participate in the research. If a household refused to participate (rarely
did only one individual refuse), another was substituted at random. Overall,
68.5% of the households originally contacted agreed to participate. This
resulted in a final sample of 304 individuals. Because of sampling by
households, there is a potential problem of a lack of independence among
cases when data are analyzed by individual; also, many of the variables
in which we are interested are more important for older and economically
established persons. Therefore, in subsequent analyses we delete dependent
children, resulting in a sample of 250 male and female household heads.
Data collection in the survey required four separate interviews with
each respondent. In the first interview, social and psychological data were
collected by a trained psychologist. The second and third interviews were
24-hour dietary recalls, conducted by trained nutritionists. One 24-hour
recall was always conducted on a Monday, in order to sample dietary
habits conventionally restricted to weekends. The second 24-hour recall
was conducted on any other weekday indifferently. The fourth interview
was conducted by a nurse who collected blood pressures, serum samples
for blood chemistry, bioelectric impedance measures of body composition,
and a brief health history. When data are analyzed that require only the
information obtained in the initial interview, the sample size is 250. There
were of course dropouts who did not complete the more extensive round of
data collection that included diet and biological measures; when analyses
are carried out that require the biological measurements, the sample size
drops to 208.
Second, intensive structured ethnographic interviews were conducted
with a set of 20 informants, 5 from each neighborhood. To be described
more fully below, these interviews generated data on social support that
were then analyzed using the cultural consensus model of Romney, Weller,
and Batchelder (1986). The key informants for these interviews were
selected by the interviewer (MCB) from the survey component of the
research on the basis of the “typical” household in that neighborhood. The
selection of key informants was guided by a knowledge of the range of
variation of a subcommunity and effort was made to arrive at representa-
tiveness.
Third, unstructured ethnographic interviews were conducted with eight
households, two from each of the neighborhoods. Again, these households
were selected in part as being typical of these areas, but of course in part
The interview used to generate data for the cultural consensus model
was actually first suggested by observing respondents replying to questions
about social support in the survey component of the research. Questions
about social support consisted of a series of probes regarding common
problems, such as needing to borrow money, or needing help with a
personal problem. Then, respondents were asked, in turn, about relatives,
friends, neighbors, and compadres as potential sources of support in
response to those problems. In observing survey respondents, it seemed that
they replied to these questions with a clear sense of hierarchy of resort to
different sources of social support. It seemed as if, for most people in most
instances, the sequence of “relative ! friend ! neighbor ! compadre”
was precisely the pattern adopted in response to most questions.
Cultural consensus analysis was chosen to test this observation. The
hypothesis guiding this test was that this pattern of responses would be
shared by a set of key informants at levels consistent with what Romney et
al. refer to as the “one culture” hypothesis. In cultural consensus modelling
the one culture hypothesis refers in this instance to a shared and culturally
constructed way of thinking and talking about seeking access in a social
support system among this set of informants. In the cultural consensus
interview, respondents were again presented with the same series of five
problems as were used in the survey component of the research. They
were then presented with the four potential sources of social support, plus
a fifth alternative of “some other person.” They were asked who the most
important person in that set would be to ask for help for that particular
problem. They were then asked who the next most important person would
be, the next, and so on, until all alternatives were exhausted. This provided
a ranking of alternative sources of social support, from first through fifth,
for all five of the potential problems. It should also be pointed out here that
the order of the problems, and the order in which the potential supporters
were asked about, were varied from the order used in the survey component
of the research.
As noted above, 20 key informants, five from each of the sampled neigh-
borhoods, participated in this task; none had any difficulty with it. This
resulted in a 20 informants-by-25 problem/supporter matrix of ratings2
that were analyzed using cultural consensus analysis as programmed in
ANTHROPAC 4.7 (Borgatti 1992). In essence, the first step in conducting
a consensus analysis is to examine common factors in a matrix (as in factor
analysis) taking the informants as the variables. If there is a single initial
factor that accounts for most of the variation relative to the other factors
(i.e., if all informants cluster on a single factor), then it is reasonable to
assume that all of the informants are drawing on a single, underlying,
Several things are worth noting here about this consensus model. First,
the model clearly has substantive meaning. It conforms to what would be
anticipated on the basis of ethnographic analyses of social relationships in
Brazil, as well as bearing out the observation made in the survey component
of the current study. Second, the danger existed that what was observed
in the survey component of the research was merely a function of the
order of presentation of items. However, the fact that the informants in
the consensus analysis went against the ordering of the items on the fifth
problem, ranking friends over relatives as the preferred source of support,
is clear evidence that these responses are substantively meaningful, and
not a function of how the questions were asked. Third, beyond friends
and relatives as sources of support, it is clear that there is considerable
flexibility in who would be chosen next, since the consensus scores for
these other sources of support are quite close in magnitude.
Romney et al. (1986) are quite clear in their derivation of consensus
theory regarding the generalizability of findings such as these. Given a
reasonable sampling of key informants, and the failure to reject the one
culture hypothesis, it can be assumed that the information provided by
the consensus ranking of the items generalizes to the social context being
studied. In other words, Table 2 contains a best estimate of a shared cultural
model of how to access social support in response to a variety of common
problems in this Brazilian city, one that generalizes across diverse socioe-
conomic groupings. In this model, kin are the preferred source of social
support for most problems, followed by friends, neighbors, compadres,
and other persons. Only in response to job-related problems are nonkin
(friends) rated as the preferred source of support.3
The quantitative evidence provided above for the hierarchy of resort with
respect to social support can be compared to the third set of data collected,
the open-ended interviews. A middle-aged, married woman from the favela
discussed social support in the following terms.
I think that people have to first turn to God for everything, and afterwards I think that I
would go to the side of my husband. Then I go to the side of my sister. I always seek her
because we are the closest in the family. We always look to each other for help. If I see
something that is beyond us, I try to get the help of some other person. I always ask my
sister first, and I think that everybody must do this, if it is mother, father, husband, wife.
While this pattern was reported in most interviews, there was one
interview in which the helpfulness of the family was explicitly rejected.
This interview was conducted with a husband and wife in the traditionally
middle class neighborhood.
I: Who would be the most important person to ask for help when someone is confronting a
stressful situation?
R (husband): A good friend. At times they can be much more helpful than family.
R (wife): Family have too many opinions. At times you can count on many more things
from a person from outside than you can with the family. At times your family is the cause
of stress. Then, a friend is better.
neighbor ! compadre” pattern; for the fifth question, weights were used
that reflected the “friend ! family ! neighbor ! compadre” pattern.
Applying these weights to the survey sample enabled us to compare
individual patterns of response to the consensus model. For example, if the
person responded that he/she would request support from a relative, then
any or all of the remaining responses would also be valid in terms of the
hierarchy of resort. If he/she did not name a relative, however, the hierarchy
would be maintained only if he/she did name a friend. If he/she did not
name a friend, then the hierarchy would be maintained only if he/she
named a neighbor. If the respondent named only a compadre, however,
then there would be as little evidence of use of the hierarchical model as
if he/she had named no one. Therefore, the individual actually has three
opportunities to respond in accordance with the hierarchy of resort on each
question. The respondent also has a number of different ways of deviating
from that hierarchy of resort. If only a compadre is named, as noted above,
that would be a deviation. If, however, the respondent named a relative
and a neighbor, the hierarchy of resort would be less well maintained than
if he/she had also named a friend. While this solution is not perfect, it does
enable us to compare individual patterns of responses with the consensus
model, and it does not require that the respondent always name the most
highly ranked source of support in order to maintain adherence to the
hierarchy. Finally, in that these patterns are being examined across five
different questions, we can detect a consistency of response on the part of
the individual.
Each respondent has three chances to answer each question in a way
consistent with the consensus model. We required that they use at least one
of those chances in order for the responses to that question to be regarded as
consistent with the shared cultural model derived from consensus analysis.
When this was done, fully half of the respondents, 54.4%, answered all
five questions in a way that was consistent with the consensus model.
Another 19.6% of the respondents answered four of the five questions in
a way consistent with the consensus model. The remaining 26.0% of the
respondents answered three or fewer of the questions in a way consistent
with the consensus model. We will refer to these groups as “consensus
group 1 (or CG1); consensus group 2 (or CG2); and, consensus group 3
(or CG3)” respectively. CG1 is highest in its consonance with the cultural
consensus model, and CG3 is lowest in its consonance with the cultural
consensus model.4
The next step in the analysis is to determine what, if any, association there
is between an individual’s approximation to the cultural ideal of social
support and that individual’s health status as measured by blood pressure,
depressive symptoms, and perceived stress. Blood pressure measurements
were made using a DINAMAP Vital Signs Monitor Model 845XT. This
is an automated system for the monitoring of blood pressure that virtually
insures no inter-observer bias. Five blood pressure measurements were
taken, and the results reported here employ the mean of all five of those
readings. In addition to systolic and diastolic blood pressure, pulse pressure
will be used as a dependent variable; this is the difference between systolic
and diastolic blood pressure. (Descriptive statistics for all variables used
in the analyses are reported in Appendix Table A1.)
Depressive symptoms are measured with a 10-item scale, adapted from
one used in research in the USA (Dressler 1991a). Respondents were asked
to report how often during the past month they had experienced common
symptoms of depression, such as sadness, hopelessness, helplessness, and
fatigue. They reported the frequency of experience of the symptoms on a
four-point scale. The scale has acceptable internal consistency reliability
in this sample (Cronbach’s alpha = 0.79). Perceived stress is measured
with a shortened form of the scale developed by Cohen and associates
(Cohen, Kamarck, and Mermelstein 1983). This scale assesses the extent
to which individuals perceive themselves to be burdened and under strain
from forces outside of their control. Additional items concern their self-
perceived abilities to make and to manage changes in their lives. Four of
the original fourteen items in this scale proved to be unreliable, in the sense
that the item-remainder correlations for those items were small; the remain-
ing ten items have acceptable internal consistency reliability (Cronbach’s
alpha = 0.80). All of the scales were translated from English by bilin-
gual members of the research team working in a group, and nuances of
meaning were resolved through discussion. These scales are very similar
to ones widely employed in research in Brazil (Almeida-Filho 1987);
furthermore, general ethnographic work indicated that persons from a
variety of socioeconomic settings talked regularly about stress and depres-
sion in just these terms.
The association of cultural consonance in social support with systolic
and diastolic blood pressure, pulse pressure, depressive symptoms, and
perceived stress, is shown in Table 3. In each of these analyses, statistical
effects and adjusted health outcome values were derived using a general
linear model with age, sex, and body mass index as control variables in
Numbers in the cells of the table are adjusted values. For blood pressure, values are
adjusted for age, sex, and the body mass index. For reported symptoms, values are
adjusted for age and sex.
relation to blood pressure, and age and sex as control variables in relation
to depressive symptoms and perceived stress. In each case, the highest (and
hence least favorable) health outcome values are observed among persons
in CG3, that is, among those persons who least approximate the ideal
cultural model of social support. Health outcome values are also higher
among the intermediate group, but not as high. The overall statistical
effects of cultural consonance in social support are significant for all of the
variables, except for diastolic blood pressure.5
These analyses demonstrate that, net of control variables, individuals
who most closely approximate the cultural ideal of social support in their
responses to questions about social support also have relatively better
health status.
ences in the perception of social support, and that the results reported above
are an artifact of these individual differences.
A new variable, “perceived support,” was formed by counting the total
number of potential sources of social support named by the respondent.
Then, the relationship of this variable to each of the outcomes, controlling
for the covariates named above, was examined using multiple regression
analysis. In the case of systolic blood pressure ( = 0.15, p < 0.02),
diastolic blood pressure ( = 0.13, p < 0.03), and perceived stress (
= 0.14, p < 0.03), perceived support was associated with lower (and
hence more favorable) levels of the outcome variables. For pulse pressure
and depression, these effects were also inverse but of borderline statistical
significance (p 0.10). Therefore, total perceived support is associated
with better health status, consistent with an individual difference model.
Table 4 shows these same multiple regression analyses, but with cultural
consonance in social support added to the equation. In this case, what is
referred to as “dummy variable” regression is being used. Cultural conso-
nance is converted to two dichotomous variables: one indicates whether
or not an individual falls into CG2 (= 1 in dummy variable coding) versus
either CG3 or CG1 (= 0 in dummy variable coding); the other indicates
whether or not an individual falls into CG3 (= 1 in dummy variable cod-
ing) versus CG2 or CG1 (= 0 in dummy variable coding). No code is
introduced for falling into CG1, since including both of the other dum-
my variables completely accounts for all cases. These two variables can
be interpreted as any other independent variables. The results in Table 4
indicate that there is virtually no change in the statistical effects of cultur-
al consonance in social support when perceived support is controlled for.
This can be seen directly in Table 4, since metric regression coefficients are
reported. These can be interpreted as the adjusted category deviations from
CG1 for CG2 and CG3, and hence can be compared directly to category
differences in Table 3. When cultural consonance is introduced into the
equation, the statistical effects of perceived support drop to nonsignifi-
cance. Given that there is a rather large correlation between perceived
social support and cultural consonance in social support, several parame-
ters of the regression models in Table 4 were scrutinized rather closely,
including the tolerances associated with each variable, the variance
inflation factor for each variable, and the condition index for each model.
Each of these can be used to detect the problem of multicollinearity; there
was no indication from any of these parameters that collinearity influenced
the results presented here.
The substantive implications of the results in Table 4 are that the results
of the analyses relating cultural consonance in social support and health
DISCUSSION
The aim of this paper has been to examine the cultural dimension of
social support in Brazil and to determine if this cultural dimension of
social support is associated with health outcomes. The approach has been
to compare the association of the cultural dimension of social support
with health outcomes to the association of a measure assessing purely
individual differences in perceived support, as well as other competing
explanatory variables. Overall, the results suggest that the extent to which
an individual’s reported access to social support approximates the cultural
ideal of social support, or what we have referred to as cultural consonance
in social support, is a much stronger correlate of health outcomes than
perceived social support conceived of as an individual difference variable.
Furthermore, cultural consonance in social support retains its association
with these health outcomes after competing explanatory variables have
been controlled for.
The orientation guiding this research is novel, or at least somewhat
unusual, for research in biocultural medical anthropology. As others have
noted, especially Ware and Kleinman (1992), efforts to link culture with
psychophysiologic outcome variables have not been particularly produc-
tive. All too often “the cultural,” to borrow Borofsky’s (1994) term, has
been reduced to ethnic or racial categories, linguistic affiliations, or values
and attitudes that end up being indistinguishable from inventories of
personality traits. Despite the rich and fertile ground provided by culture
theory, there has been little progress in using that theory to examine the
psychophysiologic differences between individuals. Instead, researchers
have been forced into a kind of methodological solipsism (Sabini and
Schulkin 1994), examining only those social and psychological factors
that are consistent with what we have referred to as an individual differ-
ence model of social support.
What is limiting about such an approach is that it virtually forces the
researcher to regard as important only that which can be regarded as
consistent with individual psychological processes. In the case of research
involving social support, it reduces down to an examination of the varying
kinds and amounts of support that the individual believes or perceives are
available to him or her. Yet, culture theory and other kinds of social and
cultural constructivism (Gaines 1992; Sabini and Schulkin 1994) suggest
that the mind must be seen in a context larger than the individual. And
to be fair, researchers in behavioral and psychosomatic medicine have
also felt this need. No doubt, it is this comprehension that has provided
much of the energy for research on social support. Including the social
along with the individual-perceptual has proven valuable in extending
toms and more perceived stress, than those persons whose responses were
more consistent with cultural constructions of access to social support. This
was not a function of the overall support perceived by individuals, nor was
it a function of other social, psychological, dietary, or anthropometric vari-
ables (except in one instance). In short, the ways in which individuals
approximate the ideal cultural model of social support is strongly related
to their mental and physical well-being. Put this way, this seems almost
a trivial statement. Yet being able to demonstrate this relationship empir-
ically, and in a way in which the cultural can be disaggregated from the
social-psychological, is not trivial.
The only instance in which another explanatory variable compromised
the association of cultural consonance in social support with an outcome
variable was when depressive symptoms were examined. In that case,
the inclusion of locus of control in the analysis reduced the association
of cultural consonance and depressive symptoms to nonsignificance. Not
surprisingly, cultural consonance in social support is associated with locus
of control. Individuals less consonant with the cultural model of social
support score significantly in the direction of an external locus of control
(F = 5.2; df = 2,247; p = 0.006); that is, individuals less consonant with
the cultural model of social support express a view of themselves as under
the control of forces external to themselves. Conventional discussions of
resources for coping with stressors often oppose social resources, such
as social support, and “personal” resources, such as the beliefs regarding
self-efficacy described by the construct of locus of control (Dressler 1991a:
18–22). The results regarding depressive symptoms could be interpreted
as demonstrating the superior empirical efficacy of locus of control with
respect to depression; this interpretation, however, may require too much
faith in the power of statistical techniques to disentangle causal relation-
ships. It is equally plausible that a particular view of oneself as embedded
in a culturally appropriate nexus of supportive relationships is conducive
to a view of oneself as able to influence events and circumstances. In this
sense, constructs describing the self may mediate the association between
cultural constructs and symptoms of distress.
The empirical demonstration of the importance of cultural consonance
here is not without its shortcomings. It should be clear from this presen-
tation that these data were collected using what might be thought of as an
emergent research design. That is, over a period of three years insights were
generated and research operations to examine those insights were devel-
oped. This resulted, for example, in a lack of functional correspondence
between the phrasing of questions to develop the consensus model of social
support and the phrasing of questions regarding access to social support in
the survey component of the research. This led to a somewhat ungainly, and
not entirely satisfactory, process of assessing individual approximation to
the cultural model of social support. Nevertheless, the pattern, consistency,
and robustness of the results lend credence to the research operations. Some
of these continuing methodological problems can be addressed in future
research.
These results pose additional interesting questions, one of which
concerns the influences on cultural consonance in social support. That
is, why do some people not report access to social support in a way that is
consistent with their culture? First, do they not know their own culture? It is
important to emphasize here that phrasing this question in this way is based
on a chain of inference combining ethnography and survey research. We
are assuming that what we learn from key informants can be generalized
to represent the knowledge generally shared within a community. And of
course, this has always been the rationale behind the use of key informants
in anthropological research; namely, that intensive interviewing with a few
well-chosen key informants can provide some kinds of cultural informa-
tion that holds true for a community. The consensus model systematizes
this general idea by, in essence, measuring the reliability of key infor-
mants. If informants are measurably consistent with one another, then it
can be assumed that the information they provide can be generalized to the
community. Furthermore, taking into account potential intracultural diver-
sity, as we did, contributes to our confidence that, in this case, the cultural
model of social support generalizes across diverse contexts. Therefore,
the likely answer to this question is that the divergence of individuals’
reporting of their preferences in social support is not the result of their
not knowing the model, because ethnographic operations indicate that the
model is generally shared.
Second, are there obstacles preventing them from accessing social
support in a culturally preferred way? One way of approaching this issue is
to examine the correlates of cultural consonance. The association of other
variables with cultural consonance in social support is neither extensive
nor strong. For example, the percentage of persons who fall into CG3,
the group that least well approximates the cultural ideal, is highest in the
favela (41.8%), and progressively declines in the conjunto (24.6%), the
traditionally middle class neighborhood (19.3%), and the upper middle
class neighborhood (16.4%). This difference is statistically significant
(2 = 16.12, df = 6, p = 0.013). Beyond what appear to be these obvi-
ous wealth differences, there are few correlates of cultural consonance in
social support. This association needs to be understood in relation to the
fact that key informants in the favela show no difference in their cultural
competence with respect to the model of social support in the cultural con-
sensus analysis. This would suggest that the favelados share the cultural
model, but that they do not use it, which is consistent with other obser-
vations of lower class social organization in Brazil. As others have noted
(Kottak 1967; Sarti 1995), the demands of maintaining the cultural ideal of
the dense social network for the Brazilian poor may be more than the sys-
tem can handle. At the same time, however, the deviation from that ideal
shows up in higher blood pressure and more expressed distress. This may
then be a part of the way in which the poor suffer the insults to their health
so widely observed in epidemiologic studies, by not having the resources
sufficient to live life in accordance with prevailing cultural standards.
These points lead also to a consideration of the psychophysiologic
processes by which a lack of cultural consonance produces its effects.
Research on mechanisms mediating social and psychological processes
has progressed considerably, and in most such models individual percep-
tion or appraisal sits squarely in the middle of the process (McEwen and
Stellar 1993). In research on social support, it is generally assumed that the
perception that social support is available in turn reduces the likelihood that
events or circumstances will be perceived as stressful, thus reducing the
probability of disease (Cohen, Kaplan, and Manuck 1994: 204). The results
presented here can be interpreted as at least complicating this perspective.
As noted earlier in the analysis, persons in the cultural consonance group
that least approximates the ideal model report less overall support, but it
is not that lower reported support that is most strongly associated with
poorer health outcomes; rather, it is the lack of consonance with the cultur-
ally shared pattern of reporting social support that is associated with poorer
health outcomes. The individuals may or may not perceive themselves to
have little support. This presents a challenge to the standard model of
stress appraisal. Either appraisal is not all that important, or the concept
of appraisal must be modified to take into account the central role of the
cultural construction of those appraisals.
A slightly different view of the process, one that explicitly takes culture
into account, was suggested by Cassel and his colleagues (Cassel, Patrick,
and Jenkins 1960; Henry and Cassel 1969) some years ago. In this view,
under certain conditions, there may develop a kind of cultural incongru-
ence, or a discrepancy between the cultural meaning system of a community
and the perceptions of individuals. Such a discrepancy could be generated
by social change, or by the movement of migrants into a host community.
By either pathway, the conflict between prevailing systems of meaning and
the ideas or behaviors of individuals could lead to confusion and uncer-
tainty in social interaction, which in turn could be stressful and lead to
poor health status. Such a process might involve conscious appraisal of the
discrepancy. Or, it might be more akin to a cognitive dissonance process,
in which individual behaviors are subject to competing demands. In either
event, these results suggest that individual appraisals must be explicitly
viewed in the context of cultural meaning systems, and not as purely
individual perceptions. For our purposes here, however, it is probably
most important to emphasize that these results, if they stand up to further
empirical scrutiny, may contribute to a revision of prevailing theories of
psychosomatic processes.
Needless to say, it is precisely that further empirical scrutiny that is now
needed. These results provide an interesting and potentially fruitful avenue
for examining the overlap of human biological processes and cultural
processes, something that has long intrigued students of culture, but which
has proven difficult to realize.
ACKNOWLEDGEMENTS
The research on which this study is based was funded by a grant from
the National Science Foundation (BNS-9020786) and with research funds
provided by The University of Alabama. Dr Kathryn S. Oths has provided
guidance, assistance and critique throughout the entire project.
NOTES
1. The point we are making here is relevant regardless of whether or not one accepts the
hypothesis that perception of support is the only important aspect of support. The same
argument would apply if one were to insist on studying received support.
2. Cultural consensus modelling was originally developed for use with dichotomous
items, although its use has been extended to rankings and interval data (Romney and
Batchelder 1987; Weller 1987). As a way of checking the results obtained here, the
highly ranked responses were treated as the “correct” answers in the task, and the lower
ranked responses were treated as “incorrect” responses, generating a matrix of yes-no
reposes. When this was used as input, essentially the same results were obtained in the
consensus analysis, lending support to our use of the rankings as scores.
3. As a way of assessing the reasonableness of this conclusion, and as a way of determining
the association of this cultural model with reported behaviors, the data given in Table 2
were examined in relation to aggregate reported social support behaviors in the survey
sample. Two columns of numbers were generated. The first was simply the ratings from
the cultural consensus model given in Table 2, only ordered as a column in which each
row was a problem/supporter combination. The second column of numbers was the
percentage of respondents in the survey research who reported selecting a relative in
response to needing a loan, followed by the percentage who reported selecting a friend
Descriptive statistics
REFERENCES
Almeida-Filho, Naomar de
1987 Migration and Mental Health in Bahia, Brazil. Salvador, Brazil: Caja de Ahorros
de la Inmaculada de Aragon.
Azevedo, Thales de
1963 Social Change in Brazil. Gainesville, FL: University of Florida Press.
Berkman, Lisa F., and S. Leonard Syme
1979 Social Networks, Host Resistance, and Mortality. American Journal of Epidemi-
ology 109: 186–204.
Blumethal, J.A., M.M. Burg, J. Barefoot, R.B. Williams, T. Harvey, and G. Zinet
1987 Social Support, Type A Behavior, and Coronary Artery Disease. Psychosomatic
Medicine 49: 331–340.
Borgatti, Stephen P.
1992 ANTHROPAC 4.05. Columbia, SC: Analytic Technologies.
Borofsky, Robert
1994 On the Knowledge and Knowing of Cultural Activities. In Robert Borofsky, ed.,
Assessing Cultural Anthropology, pp. 331–347. New York: McGraw Hill.
Gaines, Atwood D.
1992 Ethnophychiatry: The Cultural Construction of Psychiatries. In Ethnopsychia-
try: The Cultural Construction of Professional and Folk Psychiatries, pp. 3–49.
Albany, NY: State University of New York Press.
Gerin, William, D’vorah Milner, Shalinee Chawla, and Thomas G. Pickering
1995 Social Support as a Moderator of Cardiovascular Reactivity in Women. Psycho-
somatic Medicine 57: 16–22.
Griffith, Linda S., Beverly J. Field, and Patrick J. Lustman
1990 Life Stress and Social Support in Diabetes: Association with Glycemic Control.
International Journal of Psychiatry in Medicine 20: 365–372.
Henry, James P., and John Cassel
1969 Psychosocial Factors in Essential Hypertension. American Journal of Epidemiol-
ogy 90: 171–200.
Hess, David J.
1995 Introduction. In David J. Hess and Roberto A. DaMatta, eds., The Brazilian
Puzzle: Culture on the Borderlands of the Western World, pp. 1–27. New York:
Columbia University Press.
Hirdes, John P., and William F. Forbes
1992 The Importance of Social Relationships, Socioeconomic Status, and Health Prac-
tices with Respect to Mortality Among Healthy Ontario Males. Journal of Clinical
Epidemiology 45: 175–182.
Janes, Craig R.
1990 Migration, Social Change, and Health: A Samoan Community in Urban California.
Stanford, CA: Stanford University Press.
Kessler, Ronald C., and Jane D. McLeod
1985 Social Support and Mental Health in Community Samples. In Sheldon Cohen and
S. Leonard Syme eds., Social Support and Health, pp. 219–240. Orlando, FL:
Academic Press.
Kottak, Conrad
1967 Kinship and Class in Brazil. Ethnology 6: 427–443.
Matta, Roberto da
1987 A Casa e a Rua, 2nd ed. Rio de Janeiro: Guanbara.
1991 Carnivals, Rogues, and Heroes. Translated by John Drury. Notre Dame, IN:
University of Notre Dame Press.
McEwen, Bruce S., and Eliot Stellar
1993 Stress and the Individual: Mechanisms Leading to Disease. Archives of Internal
Medicine 153: 2093–2101.
Miller, Charlotte I.
1979 The Function of Middle-Class Extended Family Networks in Brazilian Urban
Society. In Maxine L. Margolis and William E. Carter, eds., Brazil: Anthropolog-
ical Perspectives, pp. 305–316. New York: Columbia University Press.
Orth-Gomer, Kristina, and Jeffrey V. Johnson
1987 Social Network Interaction and Mortality. Journal of Chronic Diseases 40: 949–
957.
Romney, A. Kimball, Susan C. Weller, and William H. Batchelder
1986 Culture as Consensus: A Theory of Culture and Informant Accuracy. American
Anthropologist 88: 313–338.
Romney, A. Kimball, William H. Batchelder, and Susan C. Weller
1987 Recent Applications of Cultural Consensus Theory. American Behavioral Scien-
tist 31: 163–177.
WILLIAM W. DRESSLER, MAURO CAMPOS BALIEIRO AND JOSE ERNESTO DOS SANTOS