Treatment of Missing Data in Multivariate Analysis (From Survey Design and Analysis - Current Issues, P 91-116, 1977, Duane F Alwin, ed. - See NCJ-74033)

• NCJ Number: 74038
• Author(s): J Kim; J Curry
• Date Published: 1977
• Length: 26 pages
• Sale Source: Sage Publications, Inc
• Publication Type: Statistics
• Language: English
• Country: United States of America

Annotation

Procedures for treating missing data in the statistical analysis of survey data are reviewed, with emphasis on the practical needs of survey researchers with a relatively complex analysis problem but little statistical sophistication.

Abstract

Discussion emphasizes the situation in which variables are measured at least on an interval scale and are considered to be random variables. The most important practical question is when and under what conditions the problem of missing information can be considered to be trivial. A discussion of assessing the nature of missing data identifies types of missing data, presents a test for the randomness of missing values, and discusses the use of indicator variables. The advantages and disadvantages of listwise and pairwise deletion of missing cases are also examined. Methods for using maximum information to estimate parameters and missing values are presented, using both iterative and noniterative procedures. It is concluded that in choosing a particular method of handling missing values, convenience and feasibility must be considered along with the nature of the method. When the sample size is over 1,000, the choice may not make much difference. Survey researchers should be aware of the complications arising from missing data and of the fact that they may be using a less than optimal solution to the problem. Situations involving genuinely fixed independent variables or categorical variables are briefly discussed. Tables, notes, and 52 references are included.