Modelling readership correlations

One of the conclusions of the paper entitled “Revisiting mediaplanning models assumptions� that we gave at the Cambridge
Readership symposium in 2003, was the necessity to develop ways to provide relevant models accounting for the correlations
which exist between different issues readership.
The motivation behind this demand was to fill a major gap of mediaplanning models that ignore such correlations at personal
probabilities level and weaken reach and frequency calculations.
Also if one could solve the problem in the simple case of print media it would open a way for a better handling of crossmediaplanning.
The difficulty was three folds:
· A lack of adequate bivariate distributions to describe those correlations
· The complexity of their integration within existing mediaplanning models
· Critical data processing speed issues
Since 2003 major statistical theory advances have happened allowing for a better understanding and theorisation of bivariate
beta distributions and several research works are available now.

Download Paper (PDF)