Adaptive Survey Design

Barry Schouten, Andy Peytchev, James Wagner

• Publisher: Chapman and Hall/CRC
• Year: 2017
• Print ISBN: 9780367735982
• eBook ISBN: 9781351650014
• Edition: 1
• Topic: Mathematics
• Subtopic: Probability & Statistics
• Index: Mathematics

Adaptive survey designs (ASDs) provide a framework for data-driven tailoring of data collection procedures to different sample members, often for cost and bias reduction. People vary in how likely they are to respond and in how they respond. This variation leads to opportunities to selectively deploy design features in order to control both nonresponse and measurement errors. ASD aims at the optimal matching of design features and the characteristics of respondents given the survey budget. Such a goal is sensible, but ASD requires investment in more advanced technical systems and management infrastructure and asks for the collection of relevant auxiliary data. So what are current best practices in ASD? And is ASD worthwhile when the same auxiliary data are employed in the estimation afterwards? In this book, the authors provide answers to these questions, and much more.

Section IV

Advanced Features of Adaptive Survey Design

9

Indicators to Support Prioritization and Optimization

9.1Introduction

ASD found its origin in attempts to reduce the impact of nonresponse bias on survey statistics. Since nonresponse bias is unknown for the survey outcome variables of interest, the survey literature has put forward various indicators that are employed as surrogates for nonresponse bias. ASD may consider such indicators as quality criteria in making decisions or optimizing the allocation of resources. In this chapter, we present an overview of the indicators that are used most often and we discuss their statistical properties. The indicators come from different survey contexts and from different modeling paradigms. As a consequence, they have rarely been discussed in a single document.

The indicators typically depend on auxiliary information, and without specifying what auxiliary variables are included and how, the indicator values are meaningless. Given that the values of the indicators are inseparable from the auxiliary variables that are included, their values mostly have a relative rather than an absolute meaning. That is, while they may give an indication of issues of performance relative to auxiliary variables, they do not provide an absolute estimate of the bias. In this chapter, we will not discuss the search for sensible auxiliary variables, which is the topic of Chapter 3. However, since ASD is essentially a form of nonresponse adjustment by design, just like traditional adjustment, useful auxiliary variables need to relate both to response behavior and to the survey variables of interest.

The indicators also typically need to be estimated using samples drawn from a population. In this chapter, we will only briefly mention the estimation of indicators and refer to Chapter 5 for details about modeling nonresponse. Since indicators are estimated from samples drawn from a larger population, they have a certain imprecision. This imprecision is often neglected, but for surveys with small samples this imprecision plays a dominant role; moderate to large samples are needed to allow ...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Contents
• Acknowledgments
• Authors
• Section I: Introduction to Adaptive Survey Design
• Section II: Preparing an Adaptive Survey Design
• Section III: Implementing an Adaptive Survey Design
• Section IV: Advanced Features of Adaptive Survey Design
• Section V: The Future of Adaptive Survey Design
• References
• Index