Small Area Estimation (SAE) techniques have received much attention in recent times due to increasing demand
for micro-level official statistics. Because of small domain-specific sample sizes (even zero size), direct
estimation may lead to estimates with large sampling variability (Rao, 2003). The basic idea of SAE method
is to link the variable of interest with auxiliary information (e.g., Census and Administrative data) in a random
effects model. SAE method is broadly classified into two methods - Unit level SAE and Area level SAE. When
unit level survey data is not available, area level SAE is utilized for small area estimates. One of the basic area
level models is the Fay-Herriot (FH) model (1979) which relates small area direct survey estimates to area specific
covariates. When multiple dependent variables are considered correlated, multivariate Fay-Herriot model
may produce better results than univariate FH model (Rao, 2003), but these models have received relatively
little attention.
The child nutrition status in Bangladesh are based only on surveys which produce national or regional level
estimates. However, national level indicators of child undernutrition often hide the real spatial distribution
across the country. The standard direct estimation methods cannot be used due to small sample size for a
significant number of administrative units such as districts. As a result, district level child nutrition indicators
have not previously been calculated. The recent Bangladesh Demographic and Health Survey (BDHS) 2011
covers all districts but only includes a small sample of children in each district thereby making consistent estimates
of malnourished children are difficult to obtain. The purpose of this presentation is to develop univariate
and bivariate Fay-Herriot models to estimate district level proportion of malnourished children with greater
efficiency.