Modelling Spatial and Spatial-Temporal Data: A Bayesian Approach
Size: | 44 MB (45901948 bytes) | Extension: |
Author(s): Robert P. Haining, Guangquan Li
Series: Chapman & Hall/CRC Statistics in the Social and Behavioral Sciences Publisher: Chapman and Hall/CRC, Year: 2020 ISBN: 1482237423,9781482237429 Description:
Modelling Spatial and Spatial-Temporal Data: A Bayesian Approach is aimed at statisticians and quantitative social, economic and public health students and researchers who work with small-area spatial and spatial-temporal data. It assumes a grounding in statistical theory up to the standard linear regression model. The book compares both hierarchical and spatial econometric modelling, providing both a reference and a teaching text with exercises in each chapter. The book provides a fully Bayesian, self-contained, treatment of the underlying statistical theory, with chapters dedicated to substantive applications. The book includes WinBUGS code and R code and all datasets are available online. Part I covers fundamental issues arising when modelling spatial and spatial-temporal data. Part II focuses on modelling cross-sectional spatial data and begins by describing exploratory methods that help guide the modelling process. There are then two theoretical chapters on Bayesian models and a chapter of applications. Two chapters follow on spatial econometric modelling, one describing different models, the other substantive applications. Part III discusses modelling spatial-temporal data, first introducing models for time series data. Exploratory methods for detecting different types of space-time interaction are presented, followed by two chapters on the theory of space-time separable (without space-time interaction) and inseparable (with space-time interaction) models. An applications chapter includes: the evaluation of a policy intervention; analysing the temporal dynamics of crime hotspots; chronic disease surveillance; and testing for evidence of spatial spillovers in the spread of an infectious disease. A final chapter suggests some future directions and challenges. Robert Haining is Emeritus Professor in Human Geography, University of Cambridge, England. He is the author of Spatial Data Analysis in the Social and Environmental Sciences (1990) and Spatial Data Analysis: Theory and Practice (2003). He is a Fellow of the RGS-IBG and of the Academy of Social Sciences. Guangquan Li is Senior Lecturer in Statistics in the Department of Mathematics, Physics and Electrical Engineering, Northumbria University, Newcastle, England. His research includes the development and application of Bayesian methods in the social and health sciences. He is a Fellow of the Royal Statistical Society.
|
Table of contents : Cover Half Title Series Page Title Page Copyright Page Dedication Contents Preface What Are the Aims of the Book? What Are the Key Features of the Book? The Structure of the Book Acknowledgements Part I Fundamentals for Modelling Spatial and Spatial-Temporal Data 1 Challenges and Opportunities Analysing Spatial and Spatial-Temporal Data 1.1 Introduction 1.2 Four Main Challenges When Analysing Spatial and Spatial-Temporal Data 1.2.1 Dependency 1.2.2 Heterogeneity 1.2.3 Data Sparsity 1.2.4 Uncertainty 1.2.4.1 Data Uncertainty 1.2.4.2 Model (or Process) Uncertainty 1.2.4.3 Parameter Uncertainty 1.3 Opportunities Arising from Modelling Spatial and Spatial-Temporal Data 1.3.1 Improving Statistical Precision 1.3.2 Explaining Variation in Space and Time 1.3.2.1 Example 1: Modelling Exposure-Outcome Relationships 1.3.2.2 Example 2: Testing a Conceptual Model at the Small Area Level 1.3.2.3 Example 3: Testing for Spatial Spillover (Local Competition) Effects 1.3.2.4 Example 4: Assessing the Effects of an Intervention 1.3.3 Investigating Space-Time Dynamics 1.4 Spatial and Spatial-Temporal Models: Bridging between Challenges and Opportunities 1.4.1 Statistical Thinking in Analysing Spatial and Spatial-Temporal Data: The Big Picture 1.4.2 Bayesian Thinking in a Statistical Analysis 1.4.3 Bayesian Hierarchical Models 1.4.3.1 Thinking Hierarchically 1.4.3.2 Incorporating Spatial and Spatial-Temporal Dependence Structures in a Bayesian Hierarchical Model Using Random Effects 1.4.3.3 Information Sharing in a Bayesian Hierarchical Model through Random Effects 1.4.4 Bayesian Spatial Econometrics 1.5 Concluding Remarks 1.6 The Datasets Used in the Book 1.7 Exercises 2 Concepts for Modelling Spatial and Spatial-Temporal Data: An Introduction to “Spatial Thinking” 2.1 Introduction 2.2 Mapping Data and Why It Matters 2.3 Thinking Spatially 2.3.1 Explaining Spatial Variation 2.3.2 Spatial Interpolation and Small Area Estimation 2.4 Thinking Spatially and Temporally 2.4.1 Explaining Space-Time Variation 2.4.2 Estimating Parameters for Spatial-Temporal Units 2.5 Concluding Remarks 2.6 Exercises Appendix: Geographic Information Systems 3 The Nature of Spatial and Spatial-Temporal Attribute Data 3.1 Introduction 3.2 Data Collection Processes in the Social Sciences 3.2.1 Natural Experiments 3.2.2 Quasi-Experiments 3.2.3 Non-Experimental Observational Studies 3.3 Spatial and Spatial-Temporal Data: Properties 3.3.1 From Geographical Reality to the Spatial Database 3.3.2 Fundamental Properties of Spatial and Spatial-Temporal Data 3.3.2.1 Spatial and Temporal Dependence 3.3.2.2 Spatial and Temporal Heterogeneity 3.3.3 Properties Induced by Representational Choices 3.3.4 Properties Induced by Measurement Processes 3.4 Concluding Remarks 3.5 Exercises 4 Specifying Spatial Relationships on the Map: The Weights Matrix 4.1 Introduction 4.2 Specifying Weights Based on Contiguity 4.3 Specifying Weights Based on Geographical Distance 4.4 Specifying Weights Based on the Graph Structure Associated with a Set of Points 4.5 Specifying Weights Based on Attribute Values 4.6 Specifying Weights Based on Evidence about Interactions 4.7 Row Standardisation 4.8 Higher Order Weights Matrices 4.9 Choice of W and Statistical Implications 4.9.1 Implications for Small Area Estimation 4.9.2 Implications for Spatial Econometric Modelling 4.9.3 Implications for Estimating the Effects of Observable Covariates on the Outcome 4.10 Estimating the W Matrix 4.11 Concluding Remarks 4.12 Exercises 4.13 Appendices Appendix 4.13.1 Building a Geodatabase in R Appendix 4.13.2 Constructing the W Matrix and Accessing Data Stored in a Shapefile 5 Introduction to the Bayesian Approach to Regression Modelling with Spatial and Spatial-Temporal Data 5.1 Introduction 5.2 Introducing Bayesian Analysis 5.2.1 Prior, Likelihood and Posterior: What Do These Terms Refer To? 5.2.2 Example: Modelling High-Intensity Crime Areas 5.3 Bayesian Computation 5.3.1 Summarising the Posterior Distribution 5.3.2 Integration and Monte Carlo Integration 5.3.3 Markov Chain Monte Carlo with Gibbs Sampling 5.3.4 Introduction to WinBUGS 5.3.5 Practical Considerations when Fitting Models in WinBUGS 5.3.5.1 Setting the Initial Values 5.3.5.2 Checking Convergence 5.3.5.3 Checking Efficiency 5.4 Bayesian Regression Models 5.4.1 Example I: Modelling Household-Level Income 5.4.2 Example II: Modelling Annual Burglary Rates in Small Areas 5.5 Bayesian Model Comparison and Model Evaluation 5.6 Prior Specifications 5.6.1 When We Have Little Prior Information 5.6.2 Towards More Informative Priors for Modelling Spatial and Spatial-Temporal Data 5.7 Concluding Remarks 5.8 Exercises Part II Modelling Spatial Data 6 Exploratory Analysis of Spatial Data 6.1 Introduction 6.2 Techniques for the Exploratory Analysis of Univariate Spatial Data 6.2.1 Mapping 6.2.2 Checking for Spatial Trend 6.2.3 Checking for Spatial Heterogeneity in the Mean 6.2.3.1 Count Data 6.2.3.2 A Monte Carlo Test 6.2.3.3 Continuous-Valued Data 6.2.4 Checking for Global Spatial Dependence (Spatial Autocorrelation) 6.2.4.1 The Moran Scatterplot 6.2.4.2 The Global Moran’s I Statistic 6.2.4.3 Other Tests for Assessing Global Spatial Autocorrelation 6.2.4.4 The Global Moran’s I Applied to Regression Residuals 6.2.4.5 The Join-Count Test for Categorical Data 6.2.5 Checking for Spatial Heterogeneity in the Spatial Dependence Structure: Detecting Local Spatial Clusters 6.2.5.1 The Local Moran’s I 6.2.5.2 The Multiple Testing Problem When Using Local Moran’s I 6.2.5.3 Kulldorff’s Spatial Scan Statistic 6.3 Exploring Relationships between Variables 6.3.1 Scatterplots and the Bivariate Moran Scatterplot 6.3.2 Quantifying Bivariate Association 6.3.2.1 The Clifford-Richardson Test of Bivariate Correlation in the Presence of Spatial Autocorrelation 6.3.2.2 Testing for Association “At a Distance” and the Global Bivariate Moran’s I 6.3.3 Checking for Spatial Heterogeneity in the Outcome-Covariate Relationship: Geographically Weighted Regression (GWR) 6.4 Overdispersion and Zero-Inflation in Spatial Count Data 6.4.1 Testing for Overdispersion 6.4.2 Testing for Zero-Inflation 6.5 Concluding Remarks 6.6 Exercises Appendix. An R Function to Perform the Zero-Inflation Test by van Den Broek (1995). See more detail in Section 6.4.2. 7 Bayesian Models for Spatial Data I: Non-Hierarchical and Exchangeable Hierarchical Models 7.1 Introduction 7.2 Estimating Small Area Income: A Motivating Example and Different Modelling Strategies 7.2.1 Modelling the 109 Parameters Non-Hierarchically 7.2.2 Modelling the 109 Parameters Hierarchically 7.3 Modelling the Newcastle Income Data Using Non-Hierarchical Models 7.3.1 An Identical Parameter Model Based on Strategy 1 7.3.2 An Independent Parameters Model Based on Strategy 2 7.4 An Exchangeable Hierarchical Model Based on Strategy 3 7.4.1 The Logic of Information Borrowing and Shrinkage 7.4.2 Explaining the Nature of Global Smoothing Due to Exchangeability 7.4.3 The Variance Partition Coefficient (VPC) 7.4.4 Applying an Exchangeable Hierarchical Model to the Newcastle Income Data 7.5 Concluding Remarks 7.6 Exercises 7.7 Appendix: Obtaining the Simulated Household Income Data 8 Bayesian Models for Spatial Data II: Hierarchical Models with Spatial Dependence 8.1 Introduction 8.2 The Intrinsic Conditional Autoregressive (ICAR) Model 8.2.1 The ICAR Model Using a Spatial Weights Matrix with Binary Entries 8.2.1.1 The WinBUGS Implementation of the ICAR Model 8.2.1.2 Applying the ICAR Model Using Spatial Contiguity to the Newcastle Income Data 8.2.1.3 Results 8.2.1.4 A Summary of the Properties of the ICAR Model Using a Binary Spatial Weights Matrix 8.2.2 The ICAR Model with a General Weights Matrix 8.2.2.1 Expressing the ICAR Model as a Joint Distribution and the Implied Restriction on W 8.2.2.2 The Sum-to-Zero Constraint 8.2.2.3 Applying the ICAR Model Using General Weights to the Newcastle Income Data 8.2.2.4 Results 8.3 The Proper CAR (pCAR) Model 8.3.1 Prior Choice for ρ 8.3.2 ICAR or pCAR? 8.3.3 Applying the pCAR Model to the Newcastle Income Data 8.3.4 Results 8.4 Locally Adaptive Models 8.4.1 Choosing an Optimal W Matrix from All Possible Specifications 8.4.2 Modelling the Elements in the W Matrix 8.4.3 Applying Some of the Locally Adaptive Spatial Models to a Subset of the Newcastle Income Data 8.5 The Besag, York and Mollié (BYM) Model 8.5.1 Two Remarks on Applying the BYM Model in Practice 8.5.2 Applying the BYM Model to the Newcastle Income Data 8.6 Comparing the Fits of Different Bayesian Spatial Models 8.6.1 DIC Comparison 8.6.2 Model Comparison Based on the Quality of the MSOA-Level Average Income Estimates 8.7 Concluding Remarks 8.8 Exercises 9 Bayesian Hierarchical Models for Spatial Data: Applications 9.1 Introduction 9.2 Application 1: Modelling the Distribution of High Intensity Crime Areas in a City 9.2.1 Background 9.2.2 Data and Exploratory Analysis 9.2.3 Methods Discussed in Haining and Law (2007) to Combine the PHIA and EHIA Maps 9.2.4 A Joint Analysis of the PHIA and EHIA Data Using the MVCAR Model 9.2.5 Results 9.2.6 Another Specification of the MVCAR Model and a Limitation of the MVCAR Approach 9.2.7 Conclusion and Discussion 9.3 Application 2: Modelling the Association Between Air Pollution and Stroke Mortality 9.3.1 Background and Data 9.3.2 Modelling 9.3.3 Interpreting the Statistical Results 9.3.4 Conclusion and Discussion 9.4 Application 3: Modelling the Village-Level Incidence of Malaria in a Small Region of India 9.4.1 Background 9.4.2 Data and Exploratory Analysis 9.4.3 Model I: A Poisson Regression Model with Random Effects 9.4.4 Model II: A Two-Component Poisson Mixture Model 9.4.5 Model III: A Two-Component Poisson Mixture Model with Zero-Inflation 9.4.6 Results 9.4.7 Conclusion and Model Extensions 9.5 Application 4: Modelling the Small Area Count of Cases of Rape in Stockholm, Sweden 9.5.1 Background and Data 9.5.2 Modelling 9.5.2.1 A “Whole-Map” Analysis Using Poisson Regression 9.5.2.2 A “Localised” Analysis Using Bayesian Profile Regression 9.5.3 Results 9.5.3.1 “Whole Map” Associations for the Risk Factors 9.5.3.2 “Local” Associations for the Risk Factors 9.5.4 Conclusions 9.6 Exercises 10 Spatial Econometric Models 10.1 Introduction 10.2 Spatial Econometric Models 10.2.1 Three Forms of Spatial Spillover 10.2.2 The Spatial Lag Model (SLM) 10.2.2.1 Formulating the Model 10.2.2.2 An Example of the SLM 10.2.2.3 The Reduced Form of the SLM and the Constraint on δ 10.2.2.4 Specification of the Spatial Weights Matrix 10.2.2.5 Issues with Model Fitting and Interpreting Coefficients 10.2.3 The Spatially-Lagged Covariates Model (SLX) 10.2.3.1 Formulating the Model 10.2.3.2 An Example of the SLX Model 10.2.4 The Spatial Error Model (SEM) 10.2.5 The Spatial Durbin Model (SDM) 10.2.5.1 Formulating the Model 10.2.5.2 Relating the SDM Model to the Other Three Spatial Econometric Models 10.2.6 Prior Specifications 10.2.7 An Example: Modelling Cigarette Sales in 46 US States 10.2.7.1 Data Description, Exploratory Analysis and Model Specifications 10.2.7.2 Results 10.3 Interpreting Covariate Effects 10.3.1 Definitions of the Direct, Indirect and Total Effects of a Covariate 10.3.2 Measuring Direct and Indirect Effects without the SAR Structure on the Outcome Variables 10.3.2.1 For the LM and SEM Models 10.3.2.2 For the SLX Model 10.3.3 Measuring Direct and Indirect Effects When the Outcome Variables are Modelled by the SAR Structure 10.3.3.1 Understanding Direct and Indirect Effects in the Presence of Spatial Feedback 10.3.3.2 Calculating the Direct and Indirect Effects in the Presence of Spatial Feedback 10.3.3.3 Some Properties of Direct and Indirect Effects 10.3.3.4 A Property (Limitation) of the Average Direct and Average Indirect Effects Under the SLM Model 10.3.3.5 Summary 10.3.4 The Estimated Effects from the Cigarette Sales Data 10.4 Model Fitting in WinBUGS 10.4.1 Derivation of the Likelihood Function 10.4.2 Simplifications to the Likelihood Computation 10.4.3 The Zeros-Trick in WinBUGS 10.4.4 Calculating the Covariate Effects in WinBUGS 10.5 Concluding Remarks 10.5.1 Other Spatial Econometric Models and the Two Problems of Identifiability 10.5.2 Comparing the Hierarchical Modelling Approach and the Spatial Econometric Approach: A Summary 10.6 Exercises 11 Spatial Econometric Modelling: Applications 11.1 Application 1: Modelling the Voting Outcomes at the Local Authority District Level in England from the 2016 EU Referendum 11.1.1 Introduction 11.1.2 Data 11.1.3 Exploratory Data Analysis 11.1.4 Modelling Using Spatial Econometric Models 11.1.5 Results 11.1.6 Conclusion and Discussion 11.2 Application 2: Modelling Price Competition Between Petrol Retail Outlets in a Large City 11.2.1 Introduction 11.2.2 Data 11.2.3 Exploratory Data Analysis 11.2.4 Spatial Econometric Modelling and Results 11.2.5 A Spatial Hierarchical Model with t4 Likelihood 11.2.6 Conclusion and Discussion 11.3 Final Remarks on Spatial Econometric Modelling of Spatial Data 11.4 Exercises Appendix: Petrol Retail Price Data Part III Modelling Spatial-Temporal Data 12 Modelling Spatial-Temporal Data: An Introduction 12.1 Introduction 12.2 Modelling Annual Counts of Burglary Cases at the Small Area Level: A Motivating Example and Frameworks for Modelling Spatial-Temporal Data 12.3 Modelling Small Area Temporal Data 12.3.1 Issues to Consider When Modelling Temporal Patterns in the Small Area Setting 12.3.1.1 Issues Relating to Temporal Dependence 12.3.1.2 Issues Relating to Temporal Heterogeneity and Spatial Heterogeneity in Modelling Small Area Temporal Patterns 12.3.1.3 Issues Relating to Flexibility of a Temporal Model 12.3.2 Modelling Small Area Temporal Patterns: Setting the Scene 12.3.3 A Linear Time Trend Model 12.3.3.1 Model Formulations 12.3.3.2 Modelling Trends in the Peterborough Burglary Data 12.3.4 Random Walk Models 12.3.4.1 Model Formulations 12.3.4.2 The RW1 Model: Its Formulation Via the Full Conditionals and Its Properties 12.3.4.3 WinBUGS Implementation of the RW1 Model 12.3.4.4 Example: Modelling Burglary Trends Using the Peterborough Data 12.3.4.5 The Random Walk Model of Order 2 12.3.5 Interrupted Time Series (ITS) Models 12.3.5.1 Quasi-Experimental Designs and the Purpose of ITS Modelling 12.3.5.2 Model Formulations 12.3.5.3 WinBUGS Implementation 12.3.5.4 Results 12.4 Concluding Remarks 12.5 Exercises Appendix: Three Different Forms for Specifying the Impact Function f. 13 Exploratory Analysis of Spatial-Temporal Data 13.1 Introduction 13.2 Patterns of Spatial-Temporal Data 13.3 Visualising Spatial-Temporal Data 13.4 Tests of Space-Time Interaction 13.4.1 The Knox Test 13.4.1.1 An Instructive Example of the Knox Test and Different Methods to Derive a p-Value 13.4.1.2 Applying the Knox Test to the Malaria Data 13.4.2 Kulldorff’s Space-Time Scan Statistic 13.4.2.1 Application: The Simulated Small Area COPD Mortality Data 13.4.3 Assessing Space-Time Interaction in the Form of Varying Local Time Trend Patterns 13.4.3.1 Exploratory Analysis of the Local Trends in the Peterborough Burglary Data 13.4.3.2 Exploratory Analysis of the Local Time Trends in the England COPD Mortality Data 13.5 Concluding Remarks 13.6 Exercises 14 Bayesian Hierarchical Models for Spatial-Temporal Data I: Space-Time Separable Models 14.1 Introduction 14.2 Estimating Small Area Burglary Rates Over Time: Setting the Scene 14.3 The Space-Time Separable Modelling Framework 14.3.1 Model Formulations 14.3.2 Do We Combine the Space and Time Components Additively or Multiplicatively? 14.3.3 Analysing the Peterborough Burglary Data Using a Space-Time Separable Model 14.3.4 Results 14.4 Concluding Remarks 14.5 Exercises 15 Bayesian Hierarchical Models for Spatial-Temporal Data II: Space-Time Inseparable Models 15.1 Introduction 15.2 From Space-Time Separability to Space-Time Inseparability: The Big Picture 15.3 Type I Space-Time Interaction 15.3.1 Example: A Space-Time Model with Type I Space-Time Interaction 15.3.2 WinBUGS Implementation 15.4 Type II Space-Time Interaction 15.4.1 Example: Two Space-Time Models with Type II Space-Time Interaction 15.4.2 WinBUGS Implementation 15.5 Type III Space-Time Interaction 15.5.1 Example: A Space-Time Model with Type III Space-Time Interaction 15.5.2 WinBUGS Implementation 15.6 Results from Analysing the Peterborough Burglary Data 15.7 Type IV Space-Time Interaction 15.7.1 Strategy 1: Extending Type II to Type IV 15.7.2 Strategy 2: Extending Type III to Type IV 15.7.2.1 Examples of Strategy 2 15.7.3 Strategy 3: Clayton’s Rule 15.7.3.1 Structure Matrices and Gaussian Markov Random Fields 15.7.3.2 Taking the Kronecker Product 15.7.3.3 Exploring the Induced Space-Time Dependence Structure via the Full Conditionals 15.7.4 Summary on Type IV Space-Time Interaction 15.8 Concluding Remarks 15.9 Exercises 16 Applications in Modelling Spatial-Temporal Data 16.1 Introduction 16.2 Application 1: Evaluating a Targeted Crime Reduction Intervention 16.2.1 Background and Data 16.2.2 Constructing Different Control Groups 16.2.3 Evaluation Using ITS 16.2.4 WinBUGS Implementation 16.2.5 Results 16.2.6 Some Remarks 16.3 Application 2: Assessing the Stability of Risk in Space and Time 16.3.1 Studying the Temporal Dynamics of Crime Hotspots and Coldspots: Background, Data and the Modelling Idea 16.3.2 Model Formulations 16.3.3 Classification of Areas 16.3.4 Model Implementation and Area Classification 16.3.5 Interpreting the Statistical Results 16.4 Application 3: Detecting Unusual Local Time Patterns in Small Area Data 16.4.1 Small Area Disease Surveillance: Background and Modelling Idea 16.4.2 Model Formulation 16.4.3 Detecting Unusual Areas with a Control of the False Discovery Rate 16.4.4 Fitting BaySTDetect in WinBUGS 16.4.5 A Simulated Dataset to Illustrate the Use of BaySTDetect 16.4.6 Results from the Simulated Dataset 16.4.7 General Results from Li et al. (2012) and an Extension of BaySTDetect 16.5 Application 4: Investigating the Presence of Spatial-Temporal Spillover Effects on Village-Level Malaria Risk in Kalaburagi, Karnataka, India 16.5.1 Background and Study Objective 16.5.2 Data 16.5.3 Modelling 16.5.4 Results 16.5.5 Concluding Remarks 16.6 Conclusions 16.7 Exercises Part IV Addendum 17 Modelling Spatial and Spatial-Temporal Data: Future Agendas? 17.1 Topic 1: Modelling Multiple Related Outcomes Over Space and Time 17.2 Topic 2: Joint Modelling of Georeferenced Longitudinal and Time-to-Event Data 17.3 Topic 3: Multiscale Modelling 17.4 Topic 4: Using Survey Data for Small Area Estimation 17.5 Topic 5: Combining Data at Both Aggregate and Individual Levels to Improve Ecological Inference 17.6 Topic 6: Geostatistical Modelling 17.6.1 Spatial Dependence 17.6.2 Mapping to Reduce Visual Bias 17.6.3 Modelling Scale Effects 17.7 Topic 7: Modelling Count Data in Spatial Econometrics 17.8 Topic 8: Computation References Index |