-Data-sets derived from subjective experiments are often exploited to construct objective quality models using parametric statistics such as MOS and multiple regression. In this paper, using data type and normality tests, we verify that non-normally distributed user opinion scores in nominal or ordinal responses should not be analysed using parametric statistics. The paper introduces a number of non-parametric statistics for valid model building and parameter estimation based on user opinion scores. A set of modelling results are also presented to demonstrate the effectiveness of non-parametric statistics.
|Title of host publication||2012 IEEE Consumer Communications and Networking Conference, CCNC'2012|
|Number of pages||6|
|Publication status||Published - 2012|
|Name||2012 IEEE Consumer Communications and Networking Conference, CCNC'2012|