PARAMETER ESTIMATION AND RESTRICTION OF RANGE SELECTIVITY ASSUMPTIONS: MODELLING THE MISSING DATA.
Item
-
Title
-
PARAMETER ESTIMATION AND RESTRICTION OF RANGE SELECTIVITY ASSUMPTIONS: MODELLING THE MISSING DATA.
-
Identifier
-
AAI8708306
-
identifier
-
8708306
-
Creator
-
MCGANNEY, MARY LOU.
-
Contributor
-
Alan L. Gross
-
Date
-
1987
-
Language
-
English
-
Publisher
-
City University of New York.
-
Subject
-
Education, Educational Psychology
-
Abstract
-
In test validation studies, the psychometrician must often work with variables on which there are missing data. In a typical case one has full data on a predictor variable, but only partial data on a criterion variable for the subgroup of selected persons. The missing data can be assumed to be missing (1) at random with respect to all of the variables, or (2) as a function of the full data or predictor variable, or (3) as a function of the missing data variable. The third of these is referred to as the nonignorable case, while the others are referred to as ignorable cases.;It has been shown that one can obtain maximum likelihood estimates of the parameters underlying the predictor and criterion variables in a straightforward way if the data are missing according to certain simple patterns and if the missing data processes are ignorable. Where the missing data patterns are not simple, the likelihood function becomes complex and maximization is not straightforward. Where the data are missing as a function of a missing data variable, the missing data process itself must be modelled and included in the likelihood function. In both cases, an iterative maximization procedure must be used to maximize the likelihood function.;In order to investigate the importance of the missing data assumptions and to explore the use of estimation procedures for the nonstraightforward cases, this work analyzed a single real data set with three variables (with missing data on two) under three different sets of missing data assumptions or models: two ignorable models and one nonignorable. The methodology for obtaining maximum likelihood estimates in the nonstraightforward cases is presented. In addition, for each missing data assumption, maximum likelihood parameter estimates are reported and a discussion of the problems encountered in using the methodology is included.;As a result of intractable computational difficulties, marginal rather than simultaneous maximum likelihood parameter estimates were obtained in the nonignorable case. Contrary to expectations, the estimates under each set of assumptions were approximately the same, strongly suggesting that the missing data were missing at random.
-
Type
-
dissertation
-
Source
-
PQT Legacy CUNY.xlsx
-
degree
-
Ph.D.
-
Program
-
Educational Psychology
