numbers and values of customers can be tracked over time. So, for example, the effect of advertising can readily be monitored – after a campaign, the number of new customers would be expected to rise, the frequency of well-known customers might increase, the mean spend of certain frequency groups might change and so forth. Simply asking every ‘nth’ customer at the till about his or her frequency of visit and entering this data into the till could transform the understanding of the sales performance of a retail chain. Opportunities of this type are many, and is one of the reasons that quantitative research is going into a new period of importance.
It is probably true to say that those companies that embrace these ideas both fully and formally and who focus their attention on building up good pictures from internal data and augmenting it with market research are likely to be the winners in the future. One might even give it a name: ‘the new quantitative research’.
So we can expect to see increasing attention being paid to the grow- ing amount of non-classical internal data, and the analysis of it being subsumed into the insight departments where it can be put alongside classical market research and market analysis and its value augmented by consumer calibration. Achieving this will be a real battle for the future – and a bloody and political one it will be too.
CHARACTERISTICS OF QUANTITATIVE
Strengths of quantitative market research data
Quantitative research is very helpful in many ways: it is good at show- ing market sizes, structure and dynamics in terms of both the products that are sold and the segmentation in consumer terms. This informa- tion is clearly important for the company to be able to put its overall strategy in context. Quantitative data is also good in terms of the eval- uation of products, especially in their overall attractiveness, and to a certain extent data can help with pack evaluation, for example, in speed of brand recognition. In terms of brands, they are useful for mea- suring the extent of usage, who uses them, how well they are rated and what people think about them in terms of image statements. The output of a quantitative survey can also be expressed in bite-sized statements, such as ‘the market grew by 15 per cent last year’, and such informa- tion is easily retained and used to put decisions in context.
Quantitative research is therefore helpful in nailing down certain ‘facts’
in a way that is generally accepted as true. This reduces the range of debate that takes place in a company, and this is a positive thing as it helps the company to focus better and not to become cluttered up with
‘red herrings’.
Potential weaknesses in quantitative data
Analysis tends to be non-holistic
A feature of ad hoc quantitative studies are that they tend to be anal- ysed question-by-question – such and such a proportion of people said
‘yes’ to this question and so forth. Analysis takes the form of seeing sig- nificance in these straight counts. The disadvantage of this is that the holistic nature of an individual is essentially being denied. Analysis is also carried out by comparing the counts of one group of respondents with another, and this can begin to flesh out the meaning, and help to give it a human face, though only in a crude stereotypical way. The sorts of comparisons that are made include ‘usage’ (compare frequent to infrequent users or to non-users), demographic (compare ABs to DEs) and, less commonly, attitude.
There are techniques that help in looking at an individual in terms of his or her answers to a set of questions. Typically this is called a ‘seg- mentation’ exercise, in which respondents are allocated to some type of group. These classification methods generally are based on taking a set
of attitudinal and behavioural data scales, transforming them onto a common scale by, for example, standardizing them, pruning the redun- dant scales (that is highly inter-correlated scales) by examination of the inter-correlation matrix or by factor analysis, and then submit- ting the whole to cluster analysis and deciding (by inspection) how many clusters are present. The process is very subjective and is hugely dependent on the scales used. However, this can be helpful in getting into segmentations of the market (assuming that there really are such groups), but can fall down when a company seeks to identify the groups in order to market to them.
An alternative method is to apply similar methods to census and related data to identify and apply common social structures in small areas, and attribute some form of consequential market behaviour to them. This method, called geodemographics, has the advantage of not only characterizing and classifying the market segments, but telling a marketer wherepotential customers actually live. This can be of great value for direct marketing activity (direct mail, leaflet drops, poster sites and so on) and for deciding on store locations for retailers.
A more direct method (Callingham and Baker, 2002) is to use the idea of demographic clusters. This is simply a three-way table of age, sex and social class with the cells populated with whatever information is relevant. This takes into account that the standard methods of analysis, which analyse by sex, age and wealth (social class) separately, fail to recognize that a young man is different from a young woman. The problem of using demographic clusters is one of economy: that is, depending upon how the tables are constructed, there could be over a hundred cells. This would need a survey of over 10,000 to populate the cells, and this would normally be too expensive. However, with the recent advent of Internet survey methods, large sample sizes are becoming more of a possibility.
The use of such a system has many advantages. First, the groups are understandable as they are the groups that we normally use in every- day life to interpret people. Second, the major media surveys are of a size that mean that they can be analysed by these groups, and third, in principle it would be possible to obtain counts of these groups in small geographical areas from the census for direct marketing and retail purposes.
Another way in which combinations of questions are used together is to form a mathematical relationship between the questions at the
individual level, for example the response to price. This can enable an expected response to a variety of situations that were not formally asked about in the questionnaire to be calculated for each individual, and then summed across them all to get at an overall population response.
This type of approach accepts the individual nature of a respondent, but of course assumes that the basic mathematical model is sound.
Quantitative research is not really objective
One of the implicit claims of quantitative research is that it has some form of objectivity about it, which gives it an aura of greater truth than qualitative research. However, it is well known that the way that a respondent will answer a question will vary depending upon how it is asked and the context in which it is embedded. In addition, respon- dents, who are not objects but thinking beings, will try to work out the purpose of the survey, and this can influence how they answer: for example, with questions on pay in an employee survey. Furthermore, respondents, being polite social entities, will try their best to answer the questions asked, no matter how silly they are, and of course cannot answer questions that are not asked, even if they would have been of great importance.
Halo effects can be rife
Much of the investigations conducted in market research are actually on topics that are not of great importance to the respondent, and which they have not thought about much. Respondents can therefore hear many of the questions simply in terms of ‘is it good or is it bad?’ and answer accordingly. This results in there being considerable correlation between the question answers, which produces a ‘halo effect’, and can make interpretation more difficult and uncertain.
This can be particularly evident in questions about how ‘important’
a series of aspects of a product or concept are – a respondent may say that they are all important! Attempts can be made to counter this by asking questions in a series of ways so that the importance of each part of the concept or brand can be estimated for each respondent. Such questions may seek to force the respondent to ‘trade off’ one thing against another, and results in a series of derived variables that charac- terize a person, and that can then be subject to further analysis (Baker and McDonald, 1999). There is an implicit assumption here that the
differences in the importance of the various aspects of the product are of sufficient meaningfulness to be worth measuring, and also that the choices are made in a linear way.
Other ways of judging importance include the emerging ‘latent’
methods. In these, regression-type equations are derived and the coeffi- cients of the independent variables used as the basis of measures of their importance. Increasingly, analysis of this type is moving from a ‘global’
one (where one equation is obtained from the whole data set) to ‘local’, where a number of equations are derived from subsets of the data.