Field methods
Data Collection Period I. The first field data collection period took place in October–November 2007. This collection period, which established the basis for the culture consensus questions, was implemented in two phases. In the first phase, women (n = 11) were interviewed about infant illness and treatment in their community. Women in this sample were mothers from the village of Karare who were selected because they had varying degrees of health knowledge but are not necessarily representative of women in the community. Questions were posed in semi-structured interview form ( Bernard, 2005) and were followed by probing questions when necessary.
The women’s answers were compiled into a list that included the frequency of each response. All infant illnesses mentioned by the women were included in the second phase of data collection along with three symptoms mentioned most frequently by the women for each illness. Western medical treatments mentioned at a frequency greater than one were included, as well as all traditional treatments mentioned for each illness. In addition, since some women stressed the importance of using different treatments depending on the kind and severity of illness, information about treatment options was included in the list of responses. These responses were converted into true-false questions (e.g. “Silalei is used to treat pneumonia in babies”; “It is better to treat malaria at home before going to the hospital”). In total, women’s interview responses generated 111 true-false questions relating to infant illness and treatment in Ariaal society.
The second phase of data collection combined this list of 111 questions with an additional 9 questions that related to basic characteristics of each woman, including number of children, self-reported poverty, latrine use, village, attendance at an NGO health seminar, use of medical facilities, and regularly boiling water for their infant. Each illness mentioned by the women – pneumonia, common cold, diarrhea, malaria, measles, ntingadu (joint pain, possibly brucellosis), eye infections, ear infections, worms, and an unspecified illness caused by a tick – also had questions relating to symptoms, treatment decisions, and Western and traditional treatments for illness. Thirty women took part in the second phase of data collection, answering true or false for all 111 questions. Women were recruited using non-probability quota sampling of interested women from two different villages. Fifteen women were from the village of Karare and fifteen were located in the village of Parkishon, located about 10 km away from Karare and the nearest dispensary.
Data Collection Period II. The second data collection period took place during November 2008–January 2009 and was part of a larger study involving mother-infant health, breastfeeding, and immunity. Questions from the traditional medicine subdomain (total = 46) were administered alongside a detailed questionnaire and mother and infant anthropometric measurements. The traditional medicine questions were chosen because knowledge in this subdomain may be less dependent on access to Western medical facilities and may better represent potential knowledge across all villages. For this data collection period, a door-to-door survey was made of the villages of Karare, Kituruni, and Parkishon of lactating women; all breastfeeding women in these communities were invited to participate. While exact participation percentages are not available due to the mobility of some participants and change in breastfeeding status in others, 251 breastfeeding mother-infant pairs (infant ages range from 2 weeks to 25 months) from the villages of Karare, Kituruni, and Parkishon were ultimately recruited into this phase of data collection.
Mothers were administered a questionnaire that included questions relating to maternal and infant characteristics, family and household composition, socioeconomic status, recent infant illness, and hygiene behaviors. These questionnaires were translated orally into Samburu by two trained female research assistants. Infant age was calculated from the date of birth on their dispensary-issued health cards; mothers whose infants were too young to have been given their BCG vaccination reported infant age instead (5% of sample).
Anthropometric measurements were taken from both mothers and infants. Height (or recumbent length), weight, mid-upper arm circumference and triceps skinfold were measured using standard anthropometric techniques (Frisancho, 2008). Mothers and infants were weighed together and the weight of the mother subtracted to obtain infant weight. The weights of women who were wearing traditional jewelry were adjusted downward by 2.2 kg, the mean weight of 5 jewelry pieces (SD = 1 kg).
Statistical methods
Data Collection Period I. In order to perform the cultural consensus analysis, quantitative data from the first data collection period, second phase, was organized into a large matrix, with participant on one axis and question on the other axis, and the true-false answer coded as either 0 or 1. The matrix was recompiled into four submatricies based on the relevant domain: illness knowledge, treatment decision making, knowledge of Western medicine, and knowledge of traditional medicine. In interviews, women indicated that these areas of knowledge were independent, and separating these questions into submatricies produced a better fit than when combined into one single matrix. Submatricies were analyzed using the cultural consensus process in ANTHROPAC v. 4.98 ( Borgatti, 2006). This method uses factor analysis to determine the “correct” answer to each question, based on consensus, as well as the knowledge of each individual. It can obtain significant results with a small sample size ( Romney et al., 1986).
Cultural consensus analyses begin by assuming individuals in the same group share one cultural model. To determine the best fit of this cultural model, it uses a least squares factor analysis with the minimum residual method. This procedure estimates and compares the cultural knowledge of each woman as well as the relative correctness of each answer. This process generates several factors, or eigenvalues, that can account for the variation found in the consensus matrix. The first eigenvalue represents the variance in the matrix due to sharing one cultural model, while the second eigenvalue represents variance due to other factors (Smith et al., 2004). In order for a submatrix to be considered a likely cultural domain, the ratio of the first eigenvalue to the second eigenvalue should be at least 1:3 with a ratio of 1:10 providing strong support for a “true” cultural domain (Borgatti, 1996 and Romney, 1999). This indicates that the greatest amount of variance in the data is due to shared cultural knowledge rather than some other effect, supporting the assumption of one cultural model. The ratio of eigenvalues for domains in this analysis range from 1:6.2 to 1:30.2, indicating a high degree of consistency of responses in each set of questions and confirming likely domains of Ariaal health knowledge.
Consensus analysis yields three useful results: 1) it determines the “correct” answer to each question by assessing the majority’s answers, 2) it assesses the level of knowledge of each individual, and 3) it provides information about how well each question fits with other questions in the domain through comparison of eigenvalues. Since not all women are equally knowledgeable about infant health, individual knowledge levels of each health domain are the variables of interest for this study. Individual knowledge of a given domain is calculated based on the relationship of the woman’s answers to the best fit consensus model, adjusted for guessing. It is determined on a scale from 0 to 1, with a higher number indicating higher levels of knowledge as assessed by cultural consensus.
Next, in order to determine what factors influence Ariaal women’s level of health knowledge, women’s knowledge scores in each of the four subdomains were associated with their community and individual characteristics. Number of children, village, latrine use, attendance at an NGO workshop on maternal/infant health, and self-described socioeconomic status (“poor” or “not poor”) were statistically associated with women’s knowledge in each subdomain using either a Pearson’s correlation or independent two-sample t-test for populations with unequal variance. Significance was assessed at α = 0.05.
Data Collection Period II. Traditional medicine questionnaire data were organized into a matrix and a cultural consensus analysis was performed as described above. The ratio of the first eigenvalue to the second eigenvalue for the traditional medicine subdomain is 1:9.2, indicating that the set of tr