Research Feature . . .
WHY WE BUY: How a Statistical Technique is Gaining Momentum, and Helping Marketers Connect our Minds to our Wallets
By Pam Frost Gorder
Marketing isn’t about making people buy what they don’t really want, explains Greg Allenby. It’s about figuring out what they do want, and selling them that.
The hard part: figuring out what people want. The professor of marketing at Ohio State says he needs more in-depth knowledge about how people think. But the information he gets from marketing surveys is too broad -- like “a puddle a mile wide and an inch thick.”
Meanwhile, Trish Van Zandt has formulated a theory about how people think when they decide to buy a product. In her experiments, the associate professor of psychology can look in-depth at the behavior of one person at a time, but she can’t test out her ideas on a grand scale.
She sums it up this way: “Here in psychology, we set up fake tasks where people sit in booths and press buttons, and over in marketing they survey thousands of people at once.”
To find the answers they are both looking for -- such as how many people acting individually shape broad marketplace demand -- Allenby and Van Zandt have joined with yet another department on campus: statistics.
There professor Angela Dean coordinates what could be the only research group in the country to merge all three disciplines. She and her colleagues Steven MacEachern and Mario Peruggia in statistics have long collaborated with psychology and marketing separately, and wanted to bring their efforts together.
So when the National Science Foundation began an initiative to fund just this kind of interdisciplinary research, they were ready. This summer, they secured one of the first grants offered -- more than $600,000 -- to use statistics to develop new models of human behavior for marketing.
The problem with typical marketing models is that they assume people think in a linear fashion when they decide to buy something -- as if thought A leads directly to thought B which leads to a purchase. But as Van Zandt and other psychologists have found, humans are decidedly nonlinear.
So, taken to the extreme, what the new project is really trying to do is explain how the human mind works -- a task well beyond the scope of any one study, and a notion Dean finds amusing. “Of course, you have to focus,” she says. “You can’t solve the whole thing in one go.”
Understanding the mind enough to gauge how people will react to a new product is a more attainable goal. What will make such predictions possible is a centuries-old statistical method that was once very controversial but is now gaining a foothold in the scientific community.
People have long wanted to understand probability, particularly in games of chance, and 18th Century mathematician Thomas Bayes was no exception. He formulated an equation that expresses uncertainty using the same kind of language that describes the outcome of a coin toss or a roll of the dice. Bayes’ theorem, in and of itself, wasn’t controversial.
But 50 years ago, a few scientists started applying Bayes’ theorem in a new way that definitely was controversial. The method involves taking an educated guess about the likelihood of something happening, and then gradually making that prediction better by factoring in the results from rigorous, controlled scientific experiments.
To classical statisticians, for whom controlled experiments were the only legitimate source of information (no educated guesses allowed), this new kind of Bayesianism came across as scientific heresy.
Now, says Peruggia, scientists have more widely recognized that the Bayesian method is a useful tool for modeling multi-faceted, non-linear phenomena such as the workings of the human mind -- which would be nearly impossible to analyze with classical statistical methods.
Allenby calls it a breakthrough.
“You wouldn’t think there could be a ‘breakthrough’ -- that’s just not how it’s done,” he says. Most scientific advances that are heralded as breakthroughs are really the result of years of incremental progress. “But modern Bayesian statistics really is a breakthrough, because it replaces complex equations with simple ones that computers can perform over and over.”
Not that Thomas Bayes could have envisioned computers more than 200 years ago, but it does explain why statisticians like Peruggia and MacEachern saw attitudes about Bayesian methods start to change when they were in graduate school in the 1980s. Only then did technology bring the large number of calculations required for a Bayesian analysis within the reach of a typical scientist.
Scientists, including some at Ohio State, are now using Bayesian methods to solve problems in the physical and social sciences. In fact, Peruggia says his department has one of the strongest Bayesian groups around. Other notable ones are at Duke University, Carnegie Mellon University, Harvard University, and the University of Chicago.
In this new project on consumer psychology, MacEachern sees enormous potential to unite different models of human cognition that have been developed over the years and incorporate them into other settings. “We can combine different sources of information, step outside of the mindset of doing one experiment at a time,” he says. “Bayes’ theorem provides an excellent way of doing that.”
Many statisticians haven’t joined the Bayesian community. Dean is one of them. But she will work on methods for testing whether the team’s new Bayesian models carry over from the lab to the market place. She’ll face a real challenge, since any one study may examine dozens of variables at once, and small changes in how the study is done could skew the results.
As an example of the kinds of questions marketers would like to answer, Allenby offers the simple analogy of a company that wants to introduce a new brand of ketchup. To be successful, the company needs to know more than which brands of ketchup sold well in the past.
Shoppers, he says, will consider a series of questions, consciously or subconsciously, that will decide the purchase. What’s special about this new ketchup? Will it help get the kids to eat? Will it go with what I’m serving? Is it priced to suit my budget?
“To really understand whether a new product will sell at the grocery store, you have to think not just about the particular product but about broader issues like family and time pressures and financial pressures and the meaning of dinner,” Allenby says.
Van Zandt uses another analogy: buying a new car. How do people choose, for instance, between a Honda and a Toyota?
The analogies may sound very different at first. “Greg talks ketchup, I talk cars,” Van Zandt says with a wave of her hand. “It’s the same thing.”
In essence, she explains, the ketchup-sized decisions are the same as the Toyota-sized ones. As we think about a product, we gather information over time, marking ticks in a kind of mental “yes” or “no” column until we feel we know enough to buy -- or not. It’s just that we require many more ticks before we commit to buying the Toyota.
That’s just one of the models the team will be testing. If Dean and her colleagues are successful, they will get an unprecedented look at why we buy what we do. Marketing firms will use that theory to shape tomorrow’s products.
The Ohio State scientists want to shape some young minds as well. They hope to start a new graduate specialization that will allow students in psychology, marketing, and statistics a chance to develop crossover skills in all three areas.
Given the rise of Bayesian statistics, the need for better data in marketing, and the possibilities of new psychological theories, the students would be poised to fill a growing niche.