Methods of estimating contraceptive prevalence rates for small areas: applications to the Dominican Republic and Kenya.

The suitability of the synthetic estimation procedure and the regression-based procedure for generating estimates for small areas compared with the direct estimation procedure was examined in their application in 30 provinces of the Dominican Republic and 32 districts of Kenya. The Dominican sample included 396 primary sampling units (PSUs). The provinces in the Dominican Republic and the districts in Kenya were designed as small areas. The contraceptive prevalence rate (CPR) was calculated for married women aged 15-49. Using the direct procedure the CPR was estimated for 7 regions and 30 provinces of the Dominican Republic and 8 provinces and 32 districts of Kenya. The synthetic estimation method showed that the CPR increased with an increase in the womans age particularly from age groups 15-19 and 35-39. Logistic regression analysis also showed that the log odds of contraceptive use were higher for women aged 25-34 and 35-49 than for those aged 15-24. The estimates provided by these three methods are consistent (the larger the number of PSUs in the area the better the estimate). The range of both synthetic and regression estimates for provinces (districts) within each region in the Dominican Republic (province in Kenya) is shorter than the one for the direct estimate. For the Dominican Republic the correlation value between the direct and synthetic provincial estimates was 0.90 it was 0.88 between the direct and regression provincial estimates and it was 0.95 between the synthetic and regression provincial estimates. For the 30 districts in Kenya (excluding West Pokot and Kakamega) the correlation values between district estimates were 0.10 0.90 and 0.85 respectively. Restricting the analysis for provinces (districts) having at least 200 cases slightly higher correlation values were found supporting the fact of consistency property. In general the correlation values were slightly higher for the Dominican Republic than for Kenya. Of the three methods used to estimate the prevalence rates the regression approach was more suitable.