On the properties of subjective ratings in video quality experiments

Subjective video quality experiments are classical statistical measurements, and as such the mathematical tools for their analysis are well understood. However, there remain certain practical aspects that are rarely discussed, yet are essential to the design of efficient experiments. Using numerical simulations as well as real-life data from a multitude of subjective experiments, this paper attempts to shed more light on the distribution and variability of subjective ratings, the effects of discrete rating scales, and the number of subjects needed.