A0244
Title: Assessing dispersion in a two-way contingency table under profile transformations and reciprocal averaging
Authors: Ting-Wu Wang - University of Newcastle (Australia) [presenting]
Eric Beh - University of Wollongong (Australia)
Rosaria Lombardo - University of Campania (Italy)
Abstract: Analysing the association between the categorical variables of a two-way contingency table can often be problematic due to dispersion issues that arise from the assumption that the cell frequencies are a Poisson random variable. Despite such an assumption requiring parity between the variance and expectation, the variance is typically larger than its expectation so that over-dispersion exists in the data. Applying a power transformation to the Poisson random variable has been a popular technique to overcome the presence of over-dispersion for nearly 100 years. However, detecting the presence of over-dispersion has not been examined when using reciprocal averaging; a method used to determine scores that best discriminate the categories of the contingency table while maximising the association between its variables. Nor has the power transformation of the data been examined for reciprocal averaging. Therefore, we shall be considering an index of dispersion that monitors for any dispersion issues that exist in the contingency table when applying a power transformation to the data. We shall also be assessing how variations in the power impact on the row and column scores obtained when applying a reciprocal averaging to the transformed cells of the contingency table.
