Surveys about attended births appear to be deceptive in CAR: are the population saying what they think NGO’s want to hear?

BACKGROUND: Non-governmental organizations (NGOs) and donors often promote certain practices to a community, such as in-facility births and then evaluate the efficacy of those interventions, in part, by surveying those populations. METHODS: A project to assess the accuracy of birth and death monitoring by local community-based monitors was undertaken with a partner health agency in areas (pop. 94,000) where they supported medical facilities. Thirty clusters of 30 households each were selected at random, probability proportional to size. Half of those households were enrolled for a monthly visitation surveillance process. To gain insights into the effects of the agency’s services, an additional 240 households were selected at random and interviewed from 8 nearby “matched villages” not serviced by any NGO as a comparison sample. RESULTS: The 896 households with 4243 living residents within the NGO service area were interviewed about household births and deaths within the past 8 months. They reported an annualized birth rate of 5.6% (95% CI: 4.5–6.7) with only 3% of those births occurring at home. The reported death rate was 4.2/1000/month (95% CI: 3.3–5.0). In the “matched villages,” the population reported a similar birth and death rate, but they reported 29% of births occurring within the home. The monthly surveillance data found over the year that followed that 32% of births occurred at the home. Clinic and hospital birth attendance data suggested an attended annual birth rate of only 2.8%, consistent with the surveillance data implication that a huge fraction of births occur at home. CONCLUSION: It is believed that because the baseline interviews occurred with a stranger, this induced interviewees to say what they thought the interviewers wanted to hear. This calls into question the validity of household surveys when agencies have a known agenda or position, and highlights the need for external validation or triangulation of survey findings.