Project's title:
Evaluation of statistical methods for missing data analysis

Background:
Demographic data collected from surveys are subject total non-response and partial response. The former is unit non-response and the later is item non-response. These non-responses are collectively referred to as missing data and total ignorance of missing data can lead to unrealistic estimates. It is therefore it is important to give statistical treatment to missing data to arrive at acceptable estimates and error margin.

• To examine missing mechanism, i.e. ignorable (random) or non-ignorable (non-random) in the case of sensitive and non-sensitive information;
• To provide estimation of demographic indicators under statistical treatment of unit and item non-responses;
• To examine the consequence of incorporating non-responses in the estimation procedure in terms of variance structure and design effect.

Methodology:
Unit non-responses are statistically treated by integrating response rate with the design weight. Item non-responses are recovered by imputation methods, such as, hot deck imputation and multiple imputations. Extended estimation and variance formulae are studied and adopted.

Source of Data:
Design weight and response rates for household, women and children from DLHS-RCH, Round-II.