Understanding and measuring advertising and buyer loyalty have both been on marketers' agendas for many decades. However, the act of bringing the two agendas together has been limited due to the data requirements associated with doing so appropriately. While Project Apollo has ceased, it has left behind an incredibly valuable pure single source data set that allows us to explore individual's loyalty and buying behaviour in relation to advertising exposure within different windows. We demonstrate a new approach to analysis, which focuses on a key parameter (Phi) from the Beta Binomial Distribution, which can be used to better understand the relationship between advertising and purchase probabilities (a measure of Latent Loyalty). The new approach demonstrates that we have much to learn about advertising and that such data is incredibly powerful in terms of its ability to measure relationships that we could only guess at before this Single Source data was available.
Most television programmes can be classified into two main programme types, Information and Entertainment. Information programmes tend to get smaller audiences but higher appreciation scores than do Entertainment programmes. Between these two main programme types, the correlation of audience appreciation with audience size is therefore negative. But for different programmes of the same type the correlation is positive, though low. Higher appreciation scores tend to go to the programmes with the larger audiences. A theoretical interpretation is that the more demanding a pro- gramme is, the more interesting and/or enjoyable it has to be before people will watch it.
The paper shows how different versions of factor analysis applied to the same data can lead to different results. In contrast, factor analyses of data which differ can lead to the same results. These are two major limitations of the technique. Two alternative analysis procedures are described. Firstly, any clusters in the data can readily be seen from the correlation matrix if the correlations are rounded to one or two digits and the variables are ordered according to the average size of their correlations. Secondly, since correlations only reflect the strength of a relationship and not its nature, a further alternative is to establish the actual relationships. This shows up the same kinds of clustering as factor or correlational analysis do at their best, but also adds real hut simple quantification. It leads to a depth of understanding and relative ease of communication which seem impressive.
In this article we show how the effects of many marketing actions can now be evaluated in some detail. The basic question considered is not whether there was any positive effect at all (eg on total sales) but what kind of effect it was (eg extra penetration or heavier purchasing from existing customers). To make such evaluations, we need to compare what actually happened with what would have happened without the marketing action. Research in the last decade has facilitated such comparisons without having to run controlled experiments. Instead of having to measure directly what would have happened without the marketing action, it is now possible to predict such norms successfully. The case-history described in this article is one where a controlled experiment could in fact have been mounted, if it based on a study for the J. Walter Thompson Company had been planned in time. The special lesson therefore is to show how an evaluation could actually be carried out after the event, and how ibis was in any case much cheaper. The approach adopted has already been applied in evaluating many marketing situations, such as price-changes, new brands, relaunches, seasonal trends, life-cycle assessments, private label brands, and various kinds of consumer and retail promotions. As a specific example we describe here a case-history which involved a consumer promotion or deal'.
In this document we describe the basic structure of buyer behaviour for a certain frequently-bought branded product. The purpose is to illustrate, in terms of a real-life but coded example, the analysis of brand-switching, of heavy buyers, of penetration growth and of repeat-buying loyalty which are provided by our "Loyalty Reports" service.
The number of people who buy two particular brands in a given time period does not depend on the brands as such but only on the numbers of people who buy each brand and on a general constant. The constant varies by product-field and by the length of the analysis-period, the correlation between buying any two brands being positive in relatively long periods and negative for shorter periods.
In the United Kingdom some 85% of homes have a television set capable of receiving at least two different channels. The results to be described in this paper concern the use these viewers make of the choice available to them. There has been much speculation in the past on the ways viewers use television, the extent to which they identify with particular channels or stations, the lengths they go to to seek out particular programmes and so on. . This paper presents quantitative results on actual viewing behaviour.
In this paper I want to consider a new way of analysing consumer purchasing data. In brief the method consists of classifying consumers not by the usual demographic characteristics, but in terms of their purchasing behaviour in some previous period. We shall be considering the way in which a comprehensive mathematical model of consumer purchasing behaviour is useful, firstly in suggesting the form of analysis referred to above, and secondly in providing the necessary background for a proper interpretation of the analysis when carried out. The mathematical model referred to is the negative binomial model of consumer purchasing and the work to be described in this paper is part of a long-term programme of research and can be regarded as a development of the work previously reported to this conference in September 1962.