Noel A Cressie

Statistics for Spatial Data, Revised Edition.

Spatial statistics — analyzing spatial data through statistical models — has proven exceptionally versatile, encompassing problems ranging from the microscopic to the astronomic. However, for the scientist and engineer faced only with scattered and uneven treatments of the subject in the scientific literature, learning how to make practical use of spatial statistics in day-to-day analytical work is very difficult.

Goodness-of-fit statistics for discrete multivariate data

1 Introduction to the Power-Divergence Statistic.- 1.1 A Unified Approach to Model Testing.- 1.2 The Power-Divergence Statistic.- 1.3 Outline of the Chapters.- 2 Defining and Testing Models: Concepts and Examples.- 2.1 Modeling Discrete Multivariate Data.- 2.2 Testing the Fit of a Model.- 2.3 An Example: Time Passage and Memory Recall.- 2.4 Applying the Power-Divergence Statistic.- 2.5 Power-Divergence Measures in Visual Perception.- 3 Modeling Cross-Classified Categorical Data.- 3.1 Association Models and Contingency Tables.- 3.2 Two-Dimensional Tables: Independence and Homogeneity.- 3.3 Loglinear Models for Two and Three Dimensions.- 3.4 Parameter Estimation Methods: Minimum Distance Estimation.- 3.5 Model Generation: A Characterization of the Loglinear, Linear, and Other Models through Minimum Distance Estimation.- 3.6 Model Selection and Testing Strategy for Loglinear Models.- 4 Testing the Models: Large-Sample Results.- 4.1 Significance Levels under the Classical (Fixed-Cells) Assumptions.- 4.2 Efficiency under the Classical (Fixed-Cells) Assumptions.- 4.3 Significance Levels and Efficiency under Sparseness Assumptions.- 4.4 A Summary Comparison of the Power-Divergence Family Members.- 4.5 Which Test Statistic?.- 5 Improving the Accuracy of Tests with Small Sample Size.- 5.1 Improved Accuracy through More Accurate Moments.- 5.2 A Second-Order Correction Term Applied Directly to the Asymptotic Distribution.- 5.3 Four Approximations to the Exact Significance Level: How Do They Compare?.- 5.4 Exact Power Comparisons.- 5.5 Which Test Statistic?.- 6 Comparing the Sensitivity of the Test Statistics.- 6.1 Relative Deviations between Observed and Expected Cell Frequencies.- 6.2 Minimum Magnitude of the Power-Divergence Test Statistic.- 6.3 Further Insights into the Accuracy of Large-Sample Approximations.- 6.4 Three Illustrations.- 6.5 Transforming for Closer Asymptotic Approximations in Contingency Tables with Some Small Expected Cell Frequencies.- 6.6 A Geometric Interpretation of the Power-Divergence Statistic.- 6.7 Which Test Statistic?.- 7 Links with Other Test Statistics and Measures of Divergence.- 7.1 Test Statistics Based on Quantiles and Spacings.- 7.2 A Continuous Analogue to the Discrete Test Statistic.- 7.3 Comparisons of Discrete and Continuous Test Statistics.- 7.4 Diversity and Divergence Measures from Information Theory.- 8 Future Directions.- 8.1 Hypothesis Testing and Parameter Estimation under Sparseness Assumptions.- 8.2 The Parameter ? as a Transformation.- 8.3 A Generalization of Akaikes Information Criterion.- 8.4 The Power-Divergence Statistic as a Measure of Loss and a Criterion for General Parameter Estimation.- 8.5 Generalizing the Multinomial Distribution.- Historical Perspective: Pearsons X2 and the Loglikelihood Ratio Statistic G2.- 1. Small-Sample Comparisons of X2 and G2 under the Classical (Fixed-Cells) Assumptions.- 2. Comparing X2 and G2 under Sparseness Assumptions.- 3. Efficiency Comparisons.- 4. Modified Assumptions and Their Impact.- Appendix: Proofs of Important Results.- A1. Some Results on Rao Second-Order Efficiency and Hodges-Lehmann Deficiency (Section 3.4).- A2. Characterization of the Generalized Minimum Power-Divergence Estimate (Section 3.5).- A3. Characterization of the Lancaster-Additive Model (Section 3.5).- A4. Proof of Results (i), (ii), and (iii) (Section 4.1).- A5. Statement of Birchs Regularity Conditions and Proof that the Minimum Power-Divergence Estimator Is BAN (Section 4.1).- A6. Proof of Results (i*), (ii*), and (iii*) (Section 4.1).- A7. The Power-Divergence Generalization of the Chernoff-Lehmann Statistic: An Outline (Section 4.1).- A8. Derivation of the Asymptotic Noncentral Chi-Squared Distribution for the Power-Divergence Statistic under Local Alternative Models (Section 4.2).- A9. Derivation of the Mean and Variance of the Power-Divergence Statistic for ? > -1 under a Nonlocal Alternative Model (Section 4.2).- A10. Proof of the Asymptotic Normality of the Power-Divergence Statistic under Sparseness Assumptions (Section 4.3).- A12. Derivation of the Second-Order Terms for the Distribution Function of the Power-Divergence Statistic under the Classical (Fixed-Cells) Assumptions (Section 5.2).- A13. Derivation of the Minimum Asymptotic Value of the Power-Divergence Statistic (Section 6.2).- A14. Limiting Form of the Power-Divergence Statistic as the Parameter ? ? +- ? (Section 6.2).- Author Index.

The origins of kriging

In this article, kriging is equated with spatial optimal linear prediction, where the unknown random-process mean is estimated with the best linear unbiased estimator. This allows early appearances of (spatial) prediction techniques to be assessed in terms of how close they came to kriging.

Fitting variogram models by weighted least squares

The method of weighted least squares is shown to be an appropriate way of fitting variogram models. The weighting scheme automatically gives most weight to early lags and down-weights those lags with a small number of pairs. Although weights are derived assuming the data are Gaussian (normal), they are shown to be still appropriate in the setting where data are a (smooth) transform of the Gaussian case. The method of (iterated) generalized least squares, which takes into account correlation between variogram estimators at different lags, offer more statistical efficiency at the price of more complexity. Weighted least squares for the robust estimator, based on square root differences, is less of a compromise.

Robust estimation of the variogram: I

It is a matter of common experience that ore values often do not follow the normal (or lognormal) distributions assumed for them, but, instead, follow some other heavier-tailed distribution. In this paper we discuss the robust estimation of the variogram when the distribution is normal-like in the central region but heavier than normal in the tails. It is shown that the use of a fourth-root transformation with or without the use of M-estimation yields stable robust estimates of the variogram.

Classes of Nonseparable, Spatio-Temporal Stationary Covariance Functions

Abstract Suppose that a random process Z(s;t), indexed in space and time, has spatio-temporal stationary covariance C(h;u), where h ∈ ℝd (d ≥ 1) is a spatial lag and u ∈ ℝ is a temporal lag. Separable spatio-temporal covariances have the property that they can be written as a product of a purely spatial covariance and a purely temporal covariance. Their ease of definition is counterbalanced by the rather limited class of random processes to which they correspond. In this article we derive a new approach that allows one to obtain many classes of nonseparable, spatio-temporal stationary covariance functions and fit several such classes to spatio-temporal data on wind speed over a region in the tropical western Pacific ocean.

Accounting for uncertainty in ecological analysis: the strengths and limitations of hierarchical statistical modeling

Analyses of ecological data should account for the uncertainty in the process(es) that generated the data. However, accounting for these uncertainties is a difficult task, since ecology is known for its complexity. Measurement and/or process errors are often the only sources of uncertainty modeled when addressing complex ecological problems, yet analyses should also account for uncertainty in sampling design, in model specification, in parameters governing the specified model, and in initial and boundary conditions. Only then can we be confident in the scientific inferences and forecasts made from an analysis. Probability and statistics provide a framework that accounts for multiple sources of uncertainty. Given the complexities of ecological studies, the hierarchical statistical model is an invaluable tool. This approach is not new in ecology, and there are many examples (both Bayesian and non-Bayesian) in the literature illustrating the benefits of this approach. In this article, we provide a baseline for concepts, notation, and methods, from which discussion on hierarchical statistical modeling in ecology can proceed. We have also planted some seeds for discussion and tried to show where the practical difficulties lie. Our thesis is that hierarchical statistical modeling is a powerful way of approaching ecological analysis in the presence of inevitable but quantifiable uncertainties, even if practical issues sometimes require pragmatic compromises.

Kriging Nonstationary Data

Abstract Spatial data modeled to have come from a random function with a nonstationary mean are considered. The spatial prediction method known as kriging exploits second-order spatial correlation structure to obtain minimum variance predictions of certain average values of the random function. But to do so, it must be assumed that either the mean function (the drift) is known up to a constant or the second-order structure (the variogram) is known exactly. Knowledge of the drift allows the (stationary) variogram to be estimated and leads to ordinary kriging. Knowledge of the variogram allows the drift to be estimated and leads to universal kriging. More usually, neither is known. This article shows how median polish of gridded spatial data provides a resistant and relatively bias-free way of kriging in the presence of drift, yet yields results as good as the mathematically optimal (but operationally difficult) universal kriging. Comparisons are performed on two data sets.

Spatial Modeling of Regional Variables

Abstract In this article, accumulated sudden infant death syndrome (SIDS) data, from 1974–1978 and 1979–1984 for the counties of North Carolina, are analyzed. After a spatial exploratory data analysis, Markov random-field models are fit to the data. The (spatial) trend is meant to capture the large-scale variation in the data, and the variance and spatial dependence are meant to capture the small-scale variation. The trend could be a function of other explanatory variables or could simply be modeled as a function of spatial location. Both models are fit and compared. The results give an excellent illustration of a phenomenon already well-known in time series, that autocorrelation in data can be due to an undiscovered explanatory variable. Indeed, for 1974–1978 we confirm a dependence of SIDS rate on proportion of nonwhite babies born, along with insignificant spatial correlation. Without this regressor variable, however, the spatial correlation is significant. In 1979–1984, perhaps due to reporting bias o...

Characterizing the manifest probabilities of latent trait models

The problem of characterizing the manifest probabilities of a latent trait model is considered. The item characteristic curve is transformed to the item passing-odds curve and a corresponding transformation is made on the distribution of ability. This results in a useful expression for the manifest probabilities of any latent trait model. The result is then applied to give a characterization of the Rasch model as a log-linear model for a 2J-contingency table. Partial results are also obtained for other models. The question of the identifiability of “guessing” parameters is also discussed.