CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Estimating the effects of match-object covariates in a generalized Bradley--Terry model Authors:  Muhammad Hilmi Bin Abdul Majid - University of Warwick (United Kingdom) [presenting]
Chenlei Leng - Warwick (United Kingdom)
Abstract: In paired comparison data, pairs of object within a set are compared and for each comparison, there will be a winner and a loser. This is a special case of network data whereby it is known that there is a directed edge (match) between two objects but the direction of the edge (match outcome) is unknown. A generalized Bradley--Terry model that includes match-object specific covariates will be introduced. Due to the incidental parameter problem, the maximum likelihood estimate for the covariates are biased and statistical inference will be invalid. A conditional likelihood estimate using simple cycles of length three can be used to remove the bias and is proven to be consistent and asymptotically normal.