The Positive Outcomes for Orphans (POFO) study is a longitudinal cross-cultural research study designed to identify characteristics of care, across diverse care structures and cultures, associated with better child outcomes. Two-stage random sampling survey methodology identified a sample of 1,480 OAC ages 6-12 living in family setting in 309 randomly selected geographical community clusters in six study sites across five LMICs: Cambodia, India (two sites), Kenya, Tanzania, and Ethiopia. Additional study details have been published elsewhere [17, 18].
From a group of 13 countries in which members of the research team had existing relationships with grassroots organizations with an interest in the proposed research, five countries were selected that were culturally, historically, ethnically, religiously, politically, and geographically diverse from each other. In India, two sites, Hyderabad and Nagaland, were chosen due to their vastly different populations represented in terms of religion, geography, income levels and political histories. Political boundaries were used to define the study area at each site [17, 18].
Community Sampling Area Selection
The primary community sampling aim was to select an unbiased sample of family-based care settings. In each of the six study areas, 50 sampling areas ("clusters") were randomly selected; in 4 of the 6 sites clusters were equally split between rural and urban areas, while in 2 the entire site was urban. At each of the six study sites, clusters were defined as villages, streets, or similar geographic units that were randomly selected from the respective next larger politically or administratively defined area (e.g., wards, kebeles, or other areas comparable to counties or zip codes in the United States). These areas in turn were randomly selected from all such areas within the boundaries of each study site. The number of clusters per study site ranged from 47 to 58 for a total of 311 clusters.
Selection of Family-dwelling OAC
A family-dwelling OAC was defined as a single or double orphan or a child abandoned by both parents who was living in a family situation as opposed to living in an institution or on the streets. In each community sampling area, up to five eligible children were selected, either randomly from available lists or through a house-to-house census conducted until five households with age-eligible children were identified. In households with multiple age-eligible children, the child whose first name started with the earliest letter in the alphabet was selected for study inclusion. In total, 1,480 OAC living in families were enrolled in the study.
The children's (self-identified) primary caregivers were asked to respond to surveys about themselves and the children. In total, 1,480 community-based caregivers participated in the assessments.
One local male and female interviewer and a lead investigator from each site were trained on study protocol and procedures. A week-long training took place at a central location with all interviewers and primary investigators present. Interviewers were certified after repeated direct observation or video taping of interviews with local non-study children. Cross-site training and site visits with interviewer observation ensured accuracy and consistency across interviewers and study sites.
Data collection protocol
Baseline data collection started between May 2006 and April 2007, depending on the study site, and continued for approximately six months. We have no reason to believe that the variation in the enrolment timeframe across study sites biased the results. Informed consent was obtained from each participating caregiver. Assent was given by all participating children. Interviews were conducted with children and caregivers using their native language in the child's residence. Baseline instruments collected information about: 1) children aged 6 to 12 residing in family settings that had at least one parent who had died or had been abandoned by both parents; and 2) the children's primary caregivers. Age inclusion criteria were based on survey instrument validity and pilot testing. Ethical approval was provided by the Duke University Institutional Review Board (IRB) and by local and national IRBs in all participating countries.
Baseline child-level characteristics were collected from caregivers during the interview. Caregivers reported on child demographics (gender, age); school attendance; relationship to caregiver (including biological child, other relative, or non-relative) and the child's general health (using one item from the Medical Outcomes Study Short Form ), with response options of "very good," "good," "fair," "poor," "very poor". Caregivers also reported if children worked ("During the past week, did the child do any kind of work for someone who is not a member of this household;" responses: "yes, paid;" "yes, unpaid," "no"), engaged in household chores ("During the past week, did the child do any household chores, such as farming, childcare, or other housework;" responses: "yes," "no," "don't know"), and approximately how many hours per week the child spent in each activity.
During the same baseline interviews, caregivers reported on their own demographics (gender, age, marital status); education (highest grade completed); general health (as above); whether they earn an income ("Do you earn an income;" "yes" or "no"); and the number of children they care for. A wealth index comparable to that used in the Demographic and Health Surveys was constructed to summarize assets and physical characteristics of each participating household .
Work and Labour variable construction
The outcome variables for this analysis were child work and child labour. Children were coded as having engaged in any work if their caregiver reported that they engaged in either work outside the home or in household chores. Hours spent working were categorized as either <21 hours, 21 to 27 hours; or child labour, defined according the UNICEF definitions of 28 hours or more of any work, youth younger than age 12 working for financial compensation, or youth aged 12 or older working 14 or more hours per week for pay. The POFO study did not assess hazardous work, which is included in UNICEF's definition of child labour, making labour estimates from this study conservative. The cut-off of 21 hours was included in the analyses because ILO suggested this as an alternative limit for the definition of child labour.
Logistic regression was used to predict the dichotomous outcome of any work vs. no work; multinomial logistic regression models were used to identify associations with the four work and labour categories: no work, <21 hours per week, 21-27 hours per week, and child labour, as defined above. Covariates alternatively included child characteristics, caregiver characteristics, or child and caregiver characteristics. To account for the stratified sampling and clustering of children within sites, logistic models were estimated with randomly-distributed site-specific intercepts; multinomial logistic regression models included site fixed effects and were estimated with robust standard errors. Analyses were conducted using Stata version 11.1.