Editor’s note: Kevin Gray is president of Cannon Gray, a marketing science and analytics consultancy. Koen Pauwels is professor of marketing at Northeastern University, Boston and BI Norwegian Business School, Oslo, Norway.
Kevin and Koen may buy the same brand for the same reasons. On the other hand, they may buy the same brand for different reasons, buy different brands for the same reasons or even buy different brands for different reasons. The brands they purchase and the reasons why may vary by occasion.
What is quantitative research?
Quantitative research has been defined in various ways. Here is one definition from the University of Southern California:
Quantitative methods emphasize objective measurements and the statistical, mathematical or numerical analysis of data collected through polls, questionnaires and surveys, or by manipulating pre-existing statistical data using computational techniques. Quantitative research focuses on gathering numerical data and generalizing it across groups of people or to predict or explain a particular phenomenon.
In marketing research, quant has historically meant consumer surveys. Analysis of consumer survey data has typically been limited to reporting numbers, perhaps broken down by age group, gender and a few other respondent groups of interest. The emphasis is mainly on the who, what, when, where and how, though segmentation, conjoint, key driver and other analyses that delve into the why are occasionally conducted with consumer survey data. Marketing mix modeling and predictive analytics are two forms of quant that do not require consumer surveys and usually draw upon other data sources such as sales data. Survey data may be integrated with these data or used for reference.
By and large, detailed exploration of the motivations underlying the behavior of consumers has been left to qualitative methods such as focus groups, in-depth interviews or, more recently, insights communities and social media analytics. This contrasts with other disciplines, such as psychology, where quantitative research is frequently the primary tool used to understand possible causes of behavior. Note the last four words in the definition of quantitative research given earlier.
We feel quant is used too narrowly in marketing research, especially in light of today’s fast computers and rich array of statistical methods for analyzing data. Owing to small sample sizes and the high measurement error associated with coding verbatim comments, qualitative research can suggest hypotheses but cannot explore them beyond a certain point. Quantitative methods can also be used to develop hypotheses, not simply to test them. Many marketing researchers seem to have forgotten this in the rush to embrace new technologies which may or may not actually be useful.
Attitudinal measurement
In recent years an urban legend regarding attitudinal measurement seems to have surfaced, namely that attitudes can only be measured by traditional five-point agree/disagree scales. Max-diff is sometimes offered as a solution though the value of attitudinal measurement to marketers is often questioned.
There are many kinds of attitudes relevant to marketers that can be measured in several ways, five-point agree/disagree scales and max-diff are just two. Attitudinal measurement is utilized extensively by scholars in many disciplines, including marketing. For examples, see the Journal of Marketing Research, the Journal of Marketing and Journal of International Marketing. Textbooks have been written on this subject, and two popular ones are Marketing Scales Handbook (Bruner) and Handbook of Marketing Scales (Bearden et al). Geo-demographics and life stages usually only partly explain these attitudes.[1]
What is generally true, however, is that fundamental values such as religious beliefs and political orientation are not useful for marketers in most product and service categories. Attitudes relating to variety-seeking, hedonism, materialism, uncertainty avoidance, price sensitivity, locus of control and country of origin, to name just a few, are highly pertinent to many categories. Marketingscales.com provides thousands of scales that have been used by marketing scholars and marketing researchers.
The relative importance of various aspects of functionality, such as ease of use and cleans well, can also shed light on consumer behavior and suggest clues for new product development.[2] Furthermore, utilities derived from conjoint analysis and max-diff can be used in various kinds of multivariate analysis, not just ordinary segmentation.
Attitudes, behaviors (claimed behaviors in the case of surveys) and demographics can be tied together in segmentation and key driver analysis in ways that were not possible until recently. Structural equation mixture modeling, for example, combines cluster analysis, factor analysis and regression.[3] There are many other recently-developed procedures in addition to older methods that have not diffused into marketing research.
Consumer surveys are much more than simple tracking and the ill-timed customer satisfaction surveys that annoy us when our credit card company has fouled up again. Much of what we need to better understand why consumers do what they do and what they want is already collected in many usage, attitude and tracking studies. However, these data are frequently merely reported and not analyzed beyond simple crosstabs. Savvy marketing research veterans have long known that what is contained in management summaries of qualitative research can often be tested and explored in more depth in a follow-up quant study designed for this purpose. We can now do this faster, for less money and better than ever.
Consumer surveys are just one kind of quantitative research. But they are data rich and, furthermore, primary research has many advantages compared to analyzing data that have been collected for other reasons.[4] Designed and conducted appropriately, consumer surveys can tie together the who, what, when, where, how and why. This is very useful for new product development and for communicating with consumers.
Customers can buy most products from a wide variety of outlets. Many use cash and are not members of loyalty programs. There are other limitations as well. A bank, for instance, will have data on its customers from a certain period but not prior to that, and no data about its customers’ transactions with other banks. Moreover, light users are an important source of volume in most fast-moving consumer goods categories because there are so many of them. Data on light users can be sparse and is non-existent for people who do not use a category, a critical group for new varieties of products. Lastly, hard data such as these are seldom error-free even when reasonably comprehensive.
Most secondary data are incomplete but this does not mean they are never useful. Increasingly, marketing researchers are combining data from various sources, an example being a bank including a sample of its customers in a usage and attitudes survey of banking customers. The data the bank has in its customer records can help it interpret responses of its customers and, in some instances, those of the total sample. As a recent academic example, Behice Ilhan and co-authors used qualitative research to identify Facebook users who attack a brand on its own page. Next, machine learning and time series analysis showed the prominence of such behavior, how it motivates brand fans to defend the brand and to what extent this full chain of events hurts or benefits the brand’s engagement.
Some examples
Here are a few examples of how quantitative research can help us better understand consumers, what they are seeking that we are not offering them and how to effectively communicate with them:
A financial services company undertook a segmentation study among general consumers and a sample of its own customers. Internal customer data were leveraged to enrich the survey data and flesh out the segments, and to help management better understand the why underlying the what for their own (and possibly competitor) customers. For example, one key finding was that risk acceptance/aversion, general financial sophistication and other attitudes predicted interest in new investment vehicle concepts beyond that explained by life stage, demographics and historical behavior.
An online retailer wanted to reallocate marketing across tens of options. Suspicious of last-click attribution, it commissioned vector autoregressive modeling that considered long-term effects and interactions of not just bringing prospects to the retailer but also of increasing check-out and revenue. It found a much higher revenue impact of content-integrated marketing actions (e.g., affiliates and price comparison sites) vs. content-separated actions (e.g., e-mails, retargeting) and increased revenues 17 percent with the same overall marketing budget.
A credit card company conducted a conjoint and segmentation study among general consumers to ascertain what certain consumer segments are seeking most. The results challenged some important assumptions, while also providing context for some earlier marketing research the company had commissioned.
A fast-moving consumer goods company wanted to adapt its global dashboard to reflect different performance effects across mature vs. emerging countries. A hierarchical linear model showed that advertising awareness and brand love are key in the former but consideration and word-of-mouth are key in the latter.
A software company conducted a user experience and satisfaction study among a sample of its customers. Key driver analysis was performed with an advanced variation of structural equation modeling and the results used for new product development as well as revision of user manuals and customer online help.
More buzz
With the advent of big data, quantitative analytics is receiving more attention and buzz than ever. However, it is our impression that most quantitative analytics is descriptive or predictive, and does not attempt to explain to decision-makers what it is describing or predicting. This appears to be true in marketing research as well.
In some cases this is inevitable given the limitations of the data at hand or the obscurity of the causal mechanism(s) generating the data. Sometimes it reflects low-cost factory-style analytics and the fact that many data scientists have only had superficial education in the research methods and statistics – let alone marketing – that can be used to get at the why and link it to the who, what, when, where and how. Collecting, managing data and running random forests will not do the trick.
We also sense many marketing researchers are feeling overwhelmed by the emergence of new data sources and sophisticated technology for gathering and analyzing data. There has been an inordinate amount of hype about big data and data science, which has proven very distracting and confusing for many of us. All this said, we are confident that, with time, our future shock will diminish sufficiently to allow us to extract more value from data than we now do.
[1] Research conducted by Geert Hofstede and his colleagues will be of interest to marketers working cross-culturally.
[2] Human behavior is not mostly irrational. The animal kingdom, in general, is inclined to be pragmatic, thus we have survived for millions of years.
[3] SEMM is really a family of methods that are complex and highly computer-intensive and have only become feasible for most researchers in the past decade. Structural Equation Modeling: A Multidisciplinary Journal (Routledge) is an excellent source for new developments in SEM and mixture modeling.
[4] Another urban legend has it that, if people aren’t talking about something in social media, it isn’t important. However, even professional bloggers will avoid topics they feel are too sensitive or too obvious, even if they believe these topics are important. There is also the question of who uses social media and why, as well has how accurately comments made in public reflect true opinions.