Pub. Date:
Springer New York
Metric Methods for Analyzing Partially Ranked Data / Edition 1

Metric Methods for Analyzing Partially Ranked Data / Edition 1

by Douglas E CritchlowDouglas E Critchlow


Current price is , Original price is $169.0. You

Temporarily Out of Stock Online

Please check back later for updated availability.


A full ranking of n items is simply an ordering of all these items, of the form: first choice, second choice, •. . , n-th choice. If two judges each rank the same n items, statisticians have used various metrics to measure the closeness of the two rankings, including Ken­ dall's tau, Spearman's rho, Spearman's footrule, Ulam's metric, Hal1l11ing distance, and Cayley distance. These metrics have been em­ ployed in many contexts, in many applied statistical and scientific problems. Thi s monograph presents genera 1 methods for extendi ng these metri cs to partially ranked data. Here "partially ranked data" refers, for instance, to the situation in which there are n distinct items, but each judge specifies only his first through k-th choices, where k < n. More complex types of partially ranked data are also investigated. Group theory is an important tool for extending the metrics. Full rankings are identified with elements of the permutation group, whereas partial rankings are identified with points in a coset space of the permutation group. The problem thus becomes one of ex­ tending metrics on the permutation group to metrics on a coset space of the permutation group. To carry out the extens"ions, two novel methods -- the so-called Hausdorff and fixed vector methods -- are introduced and implemented, which exploit this group-theoretic structure. Various data-analytic applications of metrics on fully ranked data have been presented in the statistical literature.

Product Details

ISBN-13: 9780387962887
Publisher: Springer New York
Publication date: 01/06/1986
Series: Lecture Notes in Statistics , #34
Edition description: 1985
Pages: 216
Product dimensions: 6.10(w) x 9.25(h) x 0.36(d)

Table of Contents

I. Introduction and Outline.- II. Metrics on Fully Ranked Data.- A. Permutations: Some Important Conventions.- B. Metrics on Permutations: Discussion and Exampl es.- C. The Requirement of Right—Invariance.- III. Metrics on Partially Ranked Data: The Case where Each Judge Lists His k Favorite Items Out of n.- A. The Coset Space Sn/Sn-k.- B. The Hausdorff Metrics on Sn/Sn-k.- C. The Fixed Vector Metrics on Sn/Sn-k.- IV. Metrics on Other Types of Partially Ranked Data.- A. The Coset Space Sn/S, Where S = Sn1 ×Sn2 × …×Snr.- B. The Hausdorff Metrics on Sn/S.- C. The Fixed Vector Metrics on Sn/S.- D. Hausdorff Distances between Different Types of Partially Ranked Data: A Complete Proof of the Main Theorem.- E. The Tied Ranks Approach to Metrizing Partially Ranked Data.- 1. A Description of the Tied Ranks Approach.- 2. Relations among the Tied Ranks, Hausdorff, and Fixed Vector Metrics.- 3. Limitations of the Tied Ranks Approach.- V. Distributional Properties of the Metrics.- A. Exact Distributions.- B. Asymptotic Distributions.- VI. Data Analysis, Using the Metrics.- A. Fitting Probability Models to Partially Ranked Data.- 1. Mallows’ Model for Fully Ranked Data.- 2. The Extension of Mallows’ Model to Partially Ranked Data.- 3. A Likelihood Ratio Interpretation of the Triangle Inequality.- 4. Maximum Likelihood Estimation for the Model.- 5. A Goodness-of-Fit Result.- 6. An Example: The Educational Testing Service Word Association Data.- B. Multidimensional Scaling for Partially Ranked Data.- 1. An Example, Using Leann Lipps Birch’s Cracker Preference Data.- C. Two Sample Problems for Partially Ranked Data.- 1. A Two-Sample Test Based on the Minimal Spanni ng Tree.- 2. A Two-Sample Test Based on the Nearest Neighbors Graph.- Appendix A — The Existence Of Fixed Vectors.- Appendix C — Fortran Subroutines For Fitting Mallows’ Model To Partially Ranked Data.- Appendix E — Comparison Of Exact And Asymptotic Distributions.- Index Of Notation.

Customer Reviews