|
Activity Number:
|
20
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, July 31, 2016 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistical Learning and Data Science
|
|
Abstract #320137
|
|
|
Title:
|
A Distribution Function Approach for Signal Reconstruction from Ranking Data
|
|
Author(s):
|
Michael Schimek* and Vendula Svendova
|
|
Companies:
|
Medical University of Graz and Medical University of Graz
|
|
Keywords:
|
Bootstrap ;
distribution function ;
estimation ;
indirect inference ;
ranking data
|
|
Abstract:
|
The analysis of Big Data is challenging and requires new statistical strategies. Promising is the rank order representation of such data. We assume a set of items of arbitrary size ordered by different ranking mechanisms, resulting in ranked lists. Hall and Schimek (2012) as well as Sampath and Verducci (2014) proposed methods that can identify top ranked highly informative subsets of items. Further assume that the underlying signals or decision processes are unobserved. What we aim at is the signal reconstruction from informative ranked sublists. We assume a simple statistical model of the unobserved multiple measurements (those that have produced the rankings), consisting of a signal component and a normal error component. For the evaluation of the model based on the empirical matrix of ranks we apply indirect inference. A distribution function approach in combination with a numerical optimization technique allows us to estimate a generic (signal) parameter for each item. Under the empirical distribution function we can run a non-parametric Bootstrap to obtain the standard errors of these parameters. The behaviour of the proposed approach is studied by means of simulation.
|
Authors who are presenting talks have a * after their name.
