Survey Methodology and Missing Data (eBook)

Seppo Laaksonen is a professor of statistics at the University of Helsinki, Finland, and has worked at various survey institutes including Statistics Finland, Eurostat and The Finnish Center for Social and Health Research. The former scientific secretary (2001-2003) and vice president of the International Association of Survey Statisticians (2007-2009), he has been a member of the sampling expert team of the European Social Survey since 2001. He has also been involved in a number of European research projects and is a consultant for surveys in Moldova, Ethiopia, Slovenia, the United Kingdom and Hungary.

Preface 5
Contents 8
1: Introduction 12
References 15
2: Concept of Survey and Key Survey Terms 16
2.1 What Is a Survey? 16
2.2 Five Populations in Surveys 17
A Multiframe Example 19
2.3 The Purpose of Populations 20
2.4 Cross-Sectional Survey Micro Data 21
2.4.1 Specific Examples of Problems in the Data File 22
2.5 X Variables-Auxiliary Variables in More Detail 26
2.6 Summary of the Terms and the Symbols in Chap. 2 29
2.7 Transformations 29
Example 2.1 Summary Variable with Linear Transformations 30
Example 2.2 Summary/Compound Variable Using Exploratory Factor Analysis and Factor Scores 33
References 37
3: Designing a Questionnaire and Survey Modes 38
3.1 What Is Questionnaire Design? 39
3.2 One or More Modes in One Survey? 41
Examples of Mixed-Mode Surveys 42
Estonian Pilot Mixed-Mode Survey 2012 for the ESS 42
Mode Effects in Estimates 43
3.3 Questionnaire and Questioning 44
3.4 Designing Questions for the Questionnaire 46
3.5 Developing Questions for the Survey 47
Example 3.1 Instance in Which the Scale Was Kept Similar to Earlier Social Surveys 49
Example 3.2 Screening Example of the Finnish Security Survey 50
3.6 Satisficing 51
3.7 Straightlining 53
Example 3.3 Textual Versus Coded Categories 54
3.8 Examples of Questions and Scales 55
Example 3.4 Two Alternative Lengths of Scales 55
Example 3.5 Different Scales for `Happiness´ in the Two Questionnaires 56
References 58
4: Sampling Principles, Missingness Mechanisms, and Design Weighting 59
4.1 Basic Concepts for Both Probability and Nonprobability Sampling 60
4.2 Missingness Mechanisms 62
4.3 Nonprobability Sampling Cases 63
4.4 Probability Sampling Framework 68
4.5 Sampling and Inclusion Probabilities 68
Implicit Stratification 71
PPS with Replacement, with a Valid Inclusion Probability 72
Example 4.1 ESS Sampling of Dwellings 74
Example 4.2 Inclusion Probabilities and Weights of the Test Data with Three-Stage Cluster Design 76
4.6 Illustration of Stratified Three-Stage Sampling 78
4.7 Basic Weights of Stratified Three-Stage Sampling 78
Example 4.3 Basic Weights of the Test Data for the Cluster Domain (see Sect. 6.2) 80
4.8 Two Types of Sampling Weights 81
Example 4.4 The Weights of the 2012 PISA Survey 82
References 86
5: Design Effects at the Sampling Phase 87
5.1 DEFF Because of Clustering, DEFFc 89
5.2 DEFF Because of Varying Inclusion Probabilities, DEFFp 92
Example 5.1 Design Effects Because of Unequal Weights in the Test Data, by Eight Strata 92
Example 5.2 Design Effects Because of Unequal Weights Based on the Design Weights in Some Countries of the ESS, Round 6 Count...
5.3 The Entire Design Effect: DEFF and Gross Sample Size 93
5.4 How Should the Sample Size Be Decided, and How Should the Gross Sample Be Allocated into Strata? 94
Example 5.3 Components of the Design Effect for the Variable `Plausible Value of Science Literacy´ in PISA, 2015 96
References 99
6: Sampling Design Data File 100
6.1 Principles of the Sampling Design Data File 101
6.2 Test Data Used in Several Examples in this Book 103
References 106
7: Missingness, Its Reasons and Treatment 107
7.1 Reasons for Unit Non-response 109
7.2 Coding of Item Non-responses 110
7.3 Missingness Indicator and Missingness Rate 110
7.4 Response Propensity Models 114
Example 7.1 Propensity Model for Item Response 115
Example 7.2 Response Propensity Probit Model of the Finnish Security Survey 116
References 118
8: Weighting Adjustments Because of Unit Non-response 119
Advance Reading 119
8.1 Actions of Weighting and Reweighting 120
8.2 Introduction to Reweighting Methods 120
8.3 Post-stratification 121
Example 8.1 Post-stratification in the Test Data of the SRS Domain 124
8.4 Response Propensity Weighting 125
Example 8.2 The Response Propensity Weighting of the Test SDDF Data 128
8.5 Comparisons of Weights in Other Surveys 130
8.6 Linear Calibration 132
Example 8.3 From the Basic Weights to Linear Calibration in the Test Data (Continued from Example 8.2) 133
8.7 Non-linear Calibration 135
Example 8.4 Comparison of Four Weights in Simulated Data 136
8.8 Summary of All the Weights 139
References 141
9: Special Cases in Weighting 143
9.1 Sampling of Individuals and Estimates for Clusters Such as Households 144
9.2 Cases Where Only Analysis Weights Are Available Although Proper Weights Are Required 145
9.3 Sampling and Weights for Households and Estimates for Individuals or Other Subordinate Levels 145
9.4 Panel Over Two Years 146
Example 9.1 Income Changes in a Two-year Panel 147
Reference 148
10: Statistical Editing 149
10.1 Edit Rules and Ordinary Checks 150
10.2 Some Other Edit Checks 152
10.3 Satisficing in Editing 153
10.4 Selective Editing 153
10.5 Graphical Editing 154
10.6 Tabular Editing 155
10.7 Handling Screening Data during Editing 155
10.8 Editing of Data for Public Use 155
Example 10.1 Cross-Tabulation of the Two Categorical Variables for Logical Checking (Two-Dimensional Edit Rule) 157
References 161
11: Introduction to Statistical Imputation 162
Advance Reading 163
11.1 Imputation and Its Purpose 164
11.2 Targets for Imputation Should Be Clearly Specified 166
11.3 What Can Be Imputed as a Result of Missingness? 167
11.4 `Aggregate Imputation´ 167
11.5 The Most Common Tools for Handling Missing Items Without Proper Imputation 169
Example 11.1 Multivariate Linear Regression for `Happiness by age´ Using the ESS 170
11.6 Several Imputations for the Same Micro Data 173
Example 11.2 Possible Imputation Strategies in the Case of Item Non-responses of Five Variables 173
References 176
12: Imputation Methods for Single Variables 177
12.1 Imputation Process 178
12.2 The Imputation Model 179
12.3 Imputation Tasks 181
12.4 Nearness Metrics for Real-Donor Methods 183
12.5 Possible Editing After the Model-Donor Method 184
12.6 Single and Multiple Imputation 185
Example 12.1 PISA `Multiple Imputation´ 187
12.7 Examples of Deterministic Imputation Methods for a Continuous Variable 188
Special Cases and an Example on Real Donors 191
12.8 Examples of Deterministic Imputation Methods for a Binary Variable 196
12.9 Example for a Continuous Variable When the Imputation Model Is Poor 197
12.10 Interval Estimates 199
References 200
13: Summary and Key Survey Data-Collection and Cleaning Tasks 202
14: Basic Survey Data Analysis 206
14.1 `Survey Instruments´ in the Analysis 207
14.2 Simple and Demanding Examples 208
14.2.1 Sampling Weights That Vary Greatly 208
14.2.2 Current Feeling About Household Income, with Two Types of Weights 209
14.2.3 Examples Based on the Test Data 210
14.2.4 Example Using Sampling Weights for Cross-Country Survey Data Without Country Results 213
14.2.5 The PISA Literacy Scores 214
14.2.6 Multivariate Linear Regression with Survey Instruments 216
14.2.7 A Binary Regression Model with a Logit Link 219
14.3 Concluding Remarks About Results Based on Simple and Complex Methodology 221
References 222
Further Reading 223
Journals Much Focused on Surveys 223
Survey Textbooks 223
Research Articles Dealing With Surveys: Calibration and Weighting 224
Research Articles Dealing With Surveys: Editing and Imputation 225
Research Articles Dealing With Surveys: Other Literature 225
Index 227