Alive & Thrive Nigeria Phase 1 Impact Evaluation Health Provider Survey

These baseline and endline surveys were part of a cluster-randomized evaluation of the impact of the Alive & Thrive (A&T) Nigeria program on infant and young child feeding (IYCF) practices. Data were collected in 23 local government areas (LGAs) in Kaduna State and 16 local government areas in Lagos State. Part of A&T’s intervention was training and support for health providers and traditional birth attendants (TBAs) and was designed to improve their counseling and support for IYCF practices. Cross-sectional baseline and endline surveys with facility-based providers and TBAs were conducted to measure their knowledge, attitudes, and practices in the intervention LGAs compared with the control LGAs. Separate survey questionnaires were developed for facility-based providers and TBAs. Stata MP, Version 16 Sample and sampling procedures: // Facility-based providers: The target population for the facility-based provider survey included facility-based providers involved in IYCF promotion in Lagos and Kaduna states. Lists of facilities in Lagos and Kaduna were provided by the State Ministries of Health to serve as a sampling frame. A stratified random sample of roughly 130 facilities was selected in each state. Allocation was roughly proportional across strata defined by local government area (LGA) and facility type (e.g., faith-based, private, primary, secondary, tertiary). Upon arriving at sampled facilities, the interviewers identified the appropriate staff member to complete the survey, generally either the midwife-in-charge or the officer-in-charge; thus, a single questionnaire was completed at each sampled facility with the exception of secondary and tertiary facilities. In secondary and tertiary facilities, separate interviews were conducted with the midwife-in-charge and the nutritionist (two interviews per facility). At endline, additional eligibility criteria for the facilities were: o Facilities in intervention and control LGAs were required to have at least 5 antenatal care clients, 2 deliveries, and provision of immunizations to 10 children per month. o Facilities in intervention LGAs had received A&T facility-based IYCF training. // Traditional birth attendants: The target population for the TBA survey was TBAs working in the Lagos and Kaduna States. Lists of TBAs in Lagos and Kaduna were not available to serve as a sampling frame. Consequently, the household survey component of the impact evaluation was used as a scaffold to design a random sample of TBAs. The sample of primary sampling units (PSUs, or 1-km2 grid cells) from the household survey was used as a pseudo sampling frame for the TBA survey. Two PSUs per LGA were randomly selected with probability proportional to population size, and field enumeration was used to find and list all TBAs in the area. The lists from field enumeration were then used to select a simple random sample of two TBAs per LGA to complete the survey. At endline, additional eligibility criteria for TBAs were: o TBAs in Lagos had to be registered with the Traditional Medicine Board and provide maternity services. o TBAs in intervention LGAs in both states were trained on IYCF by A&T. // Weighting: • Facility-based providers: For facility-based providers, a stratified simple random sample was selected from a list of facilities supplied by the Nigeria Ministry of Health. Thus, the design-based sampling weights represent the inverse probabilities of selection for the sample. Because it was possible to have two respondents from secondary and tertiary facilities, we applied an adjustment factor to the design weights to account for the possibility of multiple respondents per facility. If a facility had a single respondent, the adjustment factor was 1 and the design weight was unaffected. To obtain the final analysis weights, we performed a simple ratio adjustment to reallocate the weights for the non-respondents to the respondents within each state and type of facility. This was done to ensure that the sum of the final analysis weights provides an estimate of the number of facility-based providers in the population. The weights could not be calibrated to control totals, because known population totals are not available for health care providers in Nigeria. // Data users are advised to produce estimates and standard errors for the facility-based provider survey using the final analysis weight (W_FINAL). It is recommended that standard errors be produced assuming a two-stage “with replacement” design. Strata are defined by VARSTRAT, and first-stage clusters are defined by VARCLUST. Some sampling strata have been collapsed to ensure a minimum of two clusters per stratum for estimation of within-stratum variance. // Traditional birth attendants: Design-based sampling weights were calculated for each TBA as the product of three inverse probabilities of selection: (1) the inverse probability of selection of the PSU for the household survey, (2) the inverse probability of selection of the PSU for field enumeration, and (3) the inverse probability of selection of the provider from the roster created during enumeration. Because full list frames and control totals were not available, we did not perform nonresponse or poststratification adjustments on the design-based weights. Thus, the final analysis weights were simply the design-based weights trimmed to the 95th percentile value. Weight trimming was implemented to reduce the impact of large observed differences between the weighted and unweighted estimates. // Although analysis weights were calculated and have been provided for the TBA survey (W_FINAL), they are highly variable due to the small number of respondents and the atypical sample design required to overcome the lack of a sampling frame. Consequently, we believe it is reasonable for data users to produce unweighted estimates with the understanding that the results are not necessarily representative of all TBAs in the target population. Standard errors that account for the clustering in the sample design can be produced using the stratum and cluster indicators (VARSTRAT; VARCLUST).