nach oben

2012 | Buch

Kapitel lesen Erstes Kapitel lesen

Missing Data

Analysis and Design

verfasst von: John W. Graham

Verlag: Springer New York

Buchreihe : Statistics for Social and Behavioral Sciences

Enthalten in: Springer Professional "Wirtschaft+Technik" , Springer Professional "Wirtschaft" , Springer Professional "Technik"

Einloggen, um Zugang zu erhalten

Über dieses Buch

Missing data have long plagued those conducting applied research in the social, behavioral, and health sciences. Good missing data analysis solutions are available, but practical information about implementation of these solutions has been lacking. The objective of Missing Data: Analysis and Design is to enable investigators who are non-statisticians to implement modern missing data procedures properly in their research, and reap the benefits in terms of improved accuracy and statistical power.

Missing Data: Analysis and Design contains essential information for both beginners and advanced readers. For researchers with limited missing data analysis experience, this book offers an easy-to-read introduction to the theoretical underpinnings of analysis of missing data; provides clear, step-by-step instructions for performing state-of-the-art multiple imputation analyses; and offers practical advice, based on over 20 years' experience, for avoiding and troubleshooting problems. For more advanced readers, unique discussions of attrition, non-Monte-Carlo techniques for simulations involving missing data, evaluation of the benefits of auxiliary variables, and highly cost-effective planned missing data designs are provided.

The author lays out missing data theory in a plain English style that is accessible and precise. Most analysis described in the book are conducted using the well-known statistical software packages SAS and SPSS, supplemented by Norm 2.03 and associated Java-based automation utilities. A related web site contains free downloads of the supplementary software, as well as sample empirical data sets and a variety of practical exercises described in the book to enhance and reinforce the reader’s learning experience. Missing Data: Analysis and Design and its web site work together to enable beginners to gain confidence in their ability to conduct missing data analysis, and more advanced readers to expand their skill set.

Inhaltsverzeichnis

Frontmatter

Missing Data Theory

Frontmatter

Chapter 1. Missing Data Theory

Abstract

In this first chapter, I accomplish several goals. First, building on my 20+ years of work on missing data analysis, I outline a nomenclature or system for talking about the theory underlying the modern analysis of missing data. I intend for this nomenclature to be in plain English, but nevertheless to be an accurate representation of statistical theory relating to missing data analysis. Second, I describe many of the main components of missing data theory, including the causes or mechanisms of missingness. Two general methods for handling missing data, in particular multiple imputation (MI) and maximum-likelihood (ML) methods, have developed out of the missing data theory I describe here. And as will be clear from reading this book, I fully endorse these methods. For the remainder of this chapter, I challenge some of the commonly held beliefs relating to missing data theory and missing data analysis, and make a case that the MI and ML procedures, which have started to become mainstream in statistical analysis with missing data, are applicable in a much larger range of contexts that typically believed.

John W. Graham

Chapter 2. Analysis of Missing Data

Abstract

In this chapter, I present older methods for handling missing data. I then turn to the major new approaches for handling missing data. In this chapter, I present methods that make the MAR assumption. Included in this introduction are the EM algorithm for covariance matrices, normal-model multiple imputation (MI), and what I will refer to as FIML (full information maximum likelihood) methods. Before getting to these methods, however, I talk about the goals of analysis.

John W. Graham

Multiple Imputation and Basic Analysis

Frontmatter

Chapter 3. Multiple Imputation with Norm 2.03

Abstract

In this chapter, I provide step-by-step instructions for performing multiple imputation with Schafer’s (1997) NORM 2.03 program. Although these instructions apply most directly to NORM, most of the concepts apply to other MI programs as well.

John W. Graham

Chapter 4. Analysis with SPSS (Versions Without MI Module) Following Multiple Imputation with Norm 2.03

Abstract

In this chapter, I cover analyses with SPSS (v. 16 or lower) following multiple imputation with Norm 2.03. This chapter also applies to newer versions of SPSS that do not have the MI module installed. The chapter is split into three parts: (a) preliminary analyses for testing reliability, including exploratory factor analysis; (b) hypothesis testing analyses with single-level, multiple linear regression; and (c) hypothesis testing with single-level, multiple (binary) logistic regression (I also touch on other hypothesis testing analysis, such as multilevel regression with the Mixed routine).

John W. Graham

Chapter 5. Multiple Imputation and Analysis with SPSS 17-20

Abstract

In this chapter, I provide step-by-step instructions for performing multiple imputation and analysis with SPSS 17-19. I encourage you to read Chap. 3 before reading this chapter.

John W. Graham

Chapter 6. Multiple Imputation and Analysis with Multilevel (Cluster) Data

Abstract

In this chapter, I provide a little theory about multilevel data analysis and some basic imputation strategies that match up with the desired analysis. I then describe the automation utility for performing multilevel (mixed model) analysis with SPSS 15/16 and SPSS 17-19 based on Norm-imputed data. Finally, I describe the automation utility for using HLM 6/7 with Norm-imputed data.

John W. Graham

Chapter 7. Multiple Imputation and Analysis with SAS

Abstract

In this chapter, I provide step-by-step instructions for performing multiple imputation and analysis with SAS version 9. I describe the use of PROC MI for multiple imputation but also touch on two other ways to make use of PROC MI for handling missing data when hypothesis testing is not the issue: (a) direct use of the EM algorithm for input into certain analysis programs, and (b) generating a single data set imputed from EM parameters.

John W. Graham

Practical Issues in Missing Data Analysis

Frontmatter

Chapter 8. Practical Issues Relating to Analysis with Missing Data: Avoiding and Troubleshooting Problems

Abstract

If you follow the advice I have given in previous chapters, the chances are good that the results of your multiple imputation and analysis will be good. However, unforeseen things happen. Also, if you happen to be helping another person with these analyses, the material in this chapter will give some strategies for working through the problems.

John W. Graham

Chapter 9. Dealing with the Problem of Having Too Many Variables in the Imputation Model

Abstract

One of the most difficult problems with performing multiple imputation relates to having too many variables in the imputation model. In many instances, the problems associated with having too many variables in the model can be avoided by using the strategies suggested in the previous chapter. Still, situations arise in which more variables need to be included in the model than can feasibly be handled by the current software. In this chapter, we reiterate the “Think FIML” approach to multiple imputation, which will help you avoid many pitfalls in this regard. Also, for situations in which the Think FIML approach is not enough, we describe two other strategies for dealing with this problem. The first strategy involves reducing the number of variables by imputing whole scales rather than the individual items making up the scales. The second strategy involves dividing up the variables into two or more sets that can be imputed separately with minimal bias.

John W. Graham, M. Lee Van Horn, Bonnie J. Taylor

Chapter 10. Simulations with Missing Data

Abstract

If you have some experience with simulation work, then much of what I say here in the early part of this chapter should be a review. However, even if you do have prior experience with this topic, I believe it will be good to see my take on the more traditional Monte Carlo approach to missing data. Also important is that having a good sense of the traditional Monte Carlo approach to simulations will be a good setup for the non-Monte Carlo simulations I describe toward the end of this chapter.

John W. Graham

Chapter 11. Using Modern Missing Data Methods with Auxiliary Variables to Mitigate the Effects of Attrition on Statistical Power

Abstract

Missing data in a field experiment may arise from a number of sources. Participants may skip over questions inadvertently or refuse to answer them; they may offer an illegible response; they may fail to complete a questionnaire; or they may be absent from an entire measurement session in a longitudinal study. The last is often called wave nonresponse. Many participants who are unavailable for one or more occasions of measurement are available at later occasions. We define attrition is a special case of wave nonresponse in which a participant drops out of a study after a certain time and is no longer available at any subsequent wave of data collection.

John W. Graham, Linda M. Collins

Planned Missing Data Design

Frontmatter

Chapter 12. Planned Missing Data Designs I: The 3-Form Design

Abstract

This chapter draws heavily on the material covered in the article, “Planned missing data designs in psychological research” published in the journal, Psychological Methods, by Graham et al. (2006). Early in that article, we made this statement:

John W. Graham

Chapter 13. Planned Missing Data Design 2: Two-Method Measurement

Abstract

In the early stages of developing measures of any construct, the primary objective is to develop a reasonable measure of the construct. Thus, it is common for the early measures of a construct to be a measure that is readily available, often involving a simple self-report, or a relatively noninvasive physical, physiological, or biological approach to measurement. However, the early measures often have construct validity problems that become obvious only as the science of measurement matures for that construct. Overtime, researchers develop better measures of the construct, that is, measures with considerably improved construct validity.

John W. Graham, Allison E. Shevock

Titel: Missing Data
verfasst von: John W. Graham
Verlag: Springer New York
Electronic ISBN: 978-1-4614-4018-5
Print ISBN: 978-1-4614-4017-8
DOI: https://doi.org/10.1007/978-1-4614-4018-5