Sequential Rank Agreement Methods for Comparison of Ranked Lists

• Activity Number: 692
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Claus Ekstrøm* and Thomas Gerds
• Companies: University of Copenhagen and University of Copenhagen
• Keywords: ranks ; agreement ; non-parametric ; ordered lists ; consensus ; sequential

Abstract

Ranked lists of predictors are common occurrences as part of statistical analyses and they frequently appear in the results section of scientific publications. Often, the aim is to find a consensus set of shared predictors from the top of these ranked lists. For example, in high-dimensional data situations like genomics, genes may be ranked differently based on several different ranking methods or they may be ranked slightly differently under varying conditions.

In this work we show how sequential rank agreement can be used to gauge the similarity among ranked lists such that it is possible to make inference about how far the lists agree on the ranking which enables the investigator to make an improved decision on the consensus set of predictors that are of interest. Our method provides both an intuitive interpretation and can be applied to any number of lists even if these are censored. To demonstrate the performance of our approach we show results from a both a simulation study and an application to genomics data, and we illustrate how sequential rank agreement can combine and improve ranking results from both parametric and non-parametric statistical learning algorithms.