Taking Quantitative Marketing Research to a Higher Level

Quantitative marketing research is just cross tabs and t-tests, right? Nope. Not even close.

 

In marketing research, quantitative is mostly associated with consumer surveys, simple cross tabs and significance tests. We develop hypotheses with qualitative research and perhaps later test them in a quant survey. Actually, however, most of what a typical qualitative report contains can be extended to quantitative, and intricate and important relationships among consumer background characteristics, priorities and behavior explored in greater depth. 

Statistical innovations once impractical due to the limited power of older generations of PCs now make it possible to take multivariate analysis (MVA) beyond what most marketing researchers have been exposed to. Some familiar methods were also clunky to run on the computers we used until quite recently. Partly because of these barriers, advanced analytics has not penetrated very far into mainstream marketing research.

Everyone’s now talking about Big Data, but marketing researchers often are unaware that data mining and predictive analytics have been part of the marketing research landscape for more than twenty years, albeit under various names and without the fanfare. Clients have sometimes done it themselves or subcontracted it to specialized agencies. Nielsen and IRI, of course, have been in the big data business for decades.

Back to school

Let me quickly review some statistics fundamentals and give a few examples of how stats can be applied in marketing decision-making. In the next section, I’ll include some links to articles and other resources for those of you who’d like more detail regarding specific techniques.

Descriptive Statistics: Frequencies, means, medians, modes, standard deviations and interquartile range will ring a bell with nearly any marketing researcher. These tell us the essential facts about our data.

Inferential Statistics: Most often (though not always) we’re working with a sample of data drawn from a much larger population of consumers, and there is sampling error. How sure can we be that 43% of all laundry detergent users, for example, know of our brand? This is where inferential statistics comes into play, by allowing us to put confidence intervals around our sample statistic, say 40% – 46%, with 43% being our best guess. We might also be curious if differences between groups, e.g., younger and older women, is likely to be real or the result of sampling error. Most of you are probably familiar with chi-square, z-tests and t-tests, which can be used in these instances.*

New Call-to-action

Patterns of Association: Now it starts to get more interesting…The Pearson product-moment correlation is just one tool in a big bag of measures of association designed for various kinds of data (e.g., nominal, ordinal, interval, ratio). Multivariate extensions of this idea include principal components, factor analysis and correspondence analysis, and these are often used for brand and user mapping. Canonical correlation analysis is not a household name in marketing research but is one of many other MVA employed to analyze patterns of interrelationships among many variables. Doing so provides us with a richer and more complete perspective than simple cross tabs or pairwise correlations can.

Segmentation: To many marketing researchers, segmentation either means partitioning respondents based on their responses to attitude or importance ratings with K-means cluster analysis, or profiling pre-determined consumer groups with decision tree methods such as CHAID. There are actually a substantial number of other procedures, such as mixture modeling and random forests, we can now use. A common misperception is that segmentation is used for targeting only, whereas it’s also an extremely useful way to dig deep and understand what makes consumers tick.

Predictive Analytics: This used to be considered just one application of statistics but in the “Data Age” has gotten a lot of attention in the media and among decision makers. Put simply, we are trying to predict the future! One example is to use time-series analysis to forecast sales. Another is using statistical models or machine learners to improve the odds that customers will respond to an ad based on what we know about their demographics and/or past consumption patterns.

While logistic regression and other familiar statistical tools are widely-used in predictive analytics, an enormous number of new methods have been developed in the past two decades, many of them by computer scientists or academics in fields outside of statistics. Statisticians have also reinvigorated their efforts in this area. These methods can also be combined (“stacked”) to improve performance.

Causal Analysis: You may have heard that “correlation doesn’t mean causation.” This is true (but does not imply the absence of a causal relationship, either). Though complex, causal analysis is nevertheless very useful in many fields, marketing research included. It enables us to explore potential causal relationships, which is helpful even if we cannot “prove” that a causal relationship among two or more variables exists. It may also reveal that causal relationships we’d hypothesized are weak or perhaps even run in the opposite direction from what we’d believed!

Examples familiar to many marketing researchers are key driver analysis and marketing mix modeling. An early but still popular example of key driver is when brand mapping is combined with regression. This shows us how brands are perceived by consumers and which attributes are most strongly associated with actual purchase or intent to purchase. Marketing mix modeling can help us improve our “mix” of marketing activities and re-allocate our investments to improve our marketing “bang for buck.”

Trade-off methods such as conjoint aren’t usually considered causal modeling but do overlap with it. In trade-off analysis, experimental designs and sophisticated statistical methods are used to unravel consumer priorities. It is often combined with segmentation, since not all consumers are looking for the same things in a product category. 

Resources

If you’d like to learn more about any of these methods, I’ve listed a number of books and academic journals in my company library. There is much more online, too, as well as in your favorite bookseller’s inventory. To get you off to a quick start, here are some short articles I’ve written, and there are also quite a few more listed on my profile as well as articles by various authors in my Box archive.

I hope you’ve found interesting and helpful!

______________________________________________________________________

*What is less well-known in marketing research is that inferential statistics assumes probability samples, e.g., simple random samples, and does not account for coverage, non-response or measurement errors. This means that the 95% confidence interval reported by our computer might be much too narrow. There are also multiple comparisons to account for, but that is a lengthy topic. (If I didn’t mention these issues, I’d have gotten hammered by statisticians reading this. Please indulge me!)

Kevin Gray is President of Cannon Gray, a marketing science and analytics consultancy. He has more than 30 years’ experience in marketing research with Nielsen, Kantar, McCann and TIAA-CREF.

Arrange a Conversation 

Browse

Article by channel:

Read more articles tagged: Data Visualisation, Featured, Marketing Analytics, Statistics