Research Seminar Series in Statistics and Mathematics

Wirtschaftsuniversität Wien, Departments 4 D4.4.00809:00 - 10:30

Art Vortrag/Diskussion
Vortragende/rIoannis Kosmidis (Department of Statistics, University of Warwick)
Veranstalter Institut für Statistik und Mathematik
Kontakt katrin.artner@wu.ac.at

Ioannis Kosmidis (Depart­ment of Statis­tics, Univer­sity of Warwick) about “Loca­ti­on-­ad­justed Wald statis­tics”

The Insti­tute for Statis­tics and Mathe­ma­tics (Depart­ment of Finance, Accoun­ting and Statis­tics) cordi­ally invites ever­yone inte­rested to attend the talks in our Rese­arch Seminar Series, where inter­na­tio­nally renowned scho­lars from leading univer­si­ties present and discuss their (working) papers.

The list of talks for the summer term 2018 is avail­able via the follo­wing link:
Summer Term 2018


Infe­rence on a scalar para­meter of inte­rest is commonly constructed using a Wald statistic, on the grounds of the vali­dity of the stan­dard normal appro­xi­ma­tion to its fini­te-­s­ample distri­bu­tion and compu­ta­tional conve­ni­ence. A promi­nent example are the indi­vi­dual Wald tests for regres­sion para­me­ters that are reported by default in regres­sion output in the majo­rity of statis­tical compu­ting envi­ron­ments. The normal appro­xi­ma­tion can, though, be inade­quate, espe­cially when the sample size is small or moderate rela­tive to the number of para­me­ters. In this talk, the Wald statistic is viewed as an esti­mate of a trans­for­ma­tion of the model para­me­ters and is appro­pria­tely adjusted so that its null expec­ta­tion is asym­pto­ti­cally closer to zero. The bias adjust­ment depends on the expected infor­ma­tion matrix, the firs­t-order term in the bias expan­sion of the maximum likelihood esti­mator, and the deri­va­tives of the trans­for­ma­tion, all of which are either readily avail­able or easily obtainable in stan­dard soft­ware for a wealth of well-used models. The fini­te-­s­ample perfor­mance of the loca­ti­on-­ad­justed Wald statistic is examined analy­ti­cally in simple models and via simu­la­tion in a series of more realistic model­ling frame­works, inclu­ding gene­ra­lized linear models, meta-­re­gres­sion and beta regres­sion. The loca­ti­on-­ad­justed Wald statistic is found able to deliver signi­fi­cant impro­ve­ments in infe­ren­tial perfor­mance over the stan­dard Wald statistic, without sacri­fi­cing any of its compu­ta­tional simpli­city.

zurück zur Übersicht