The Conversion Story Of A Cognitive Researcher To An Emotional Researcher
By Todd Powers, PhD
If you are like me, you have probably avoided the whole question of emotions in your market research career. I am perfectly willing to admit that, for the most part, I have prayed each morning to the God of Logic as the ultimate arbiter in the battle for market insights. Routinely, I produced and pledged allegiance to statements like, “Males are significantly more likely than their female counterparts to cite horsepower as an important determinant of brand choice in pickup trucks.” And I had proof, right there in the sample of 500 recent buyers of trucks who had answered questions I asked them about that very issue.
Oh, I knew that people often let their emotions get the best of them, but I also assumed that these were largely secondary factors in decisions that they made. Well, no more. I may be a slow learner, but I am now thoroughly convinced of the dominant influence that emotions have in the purchase process, and more importantly, of the mandate that we have to measure and understand that influence.
So how did this happen? I now realize that the main reason that I (dare I say “most of us” researchers) studied cognitive effects — to the exclusion of emotions — in decision-making was because emotions were just so damn difficult to measure. Really. We tried asking people what they were “feeling” when they bought that refrigerator, but the results were always less than satisfactory. People gave answers that were more a function of social desirability than any real sense of introspection. Qualitative techniques like laddering helped us to understand that platinum-version credit cards made us feel important, and superior, and condescending when we slapped that baby down on the countertop at the hotel check-in, but that stuff is difficult and expensive in a survey of n=500. So we didn’t. Instead, we just asked our respondents to rate each of 8 or 10 different competitive brands of say, toothpaste, on things like price and availability and flavor.
Consider this: One of the more common models of choice behavior, maximum likelihood, can be represented by the following (simplified) equation.
Presumably, we make choices by rating all of the options on a set of attributes, and then assigning weights to each attribute according to how important they are. If having good gas mileage, and leather seats, and four doors, and a high resale value are important to me when buying a car, then the brands that rate highly on those features will be the ones I am likely to choose. So … If we want to predict brand choice, we just ask people to rate brands on those attributes, then we ask the same people to tell us how important each feature is to them, and presto, we get accurate predictions, right? The answer is “No.” Those efforts proved to be fruitless, for the most part. And that’s because the importance ratings are driven by our emotions, not our cognitions. And we got pretty much meaningless answers to questions about those importances, or weights.
Instead, we adopted a practice of asking for attribute ratings, then observing choices, and (using various modeling techniques) inferring the weights, the importance scores. Essentially, that’s what conjoint/discrete choice models do for us.
I guess I could have been perfectly happy working under this set of neatly logical rules. Let’s measure what we CAN measure, and do it really well. But it still bothered me.
And then I encountered two very different academic discussions that convinced me we needed to measure our emotional reactions in decision-making more directly.
First, I read about the work of Antoine Bechara, a psychiatrist working with patients trying to recover from traumatic brain injury due to car accidents and the like. These were people who had lost the functioning in their pre-frontal cortex, the portion of the brain largely responsible for processing of emotions. Repeatedly, he observed that his patients could perform all of the normal activities of day-to-day life, and could manage the cognitive tasks that he asked them to address. What they could not do was “make a decision.” Bechara cited the example of patients who, when asked at the end of the counseling session just when they might be available to come back next week for follow-up counseling, would start listing all of the reasons that Monday was good (or bad), or that Wednesday had potential, or Thursday might or might not work out. But they were unable or unwilling to actually choose a day and time for their next appointment. They struggled deeply with pulling that trigger.
Bechara concluded that choice was highly dependent on emotional processing, due to the individual, personalized nature of weighing the options, and it is hard to argue differently.
A second perspective came to me as I was reviewing some of the considerable body of work that Kahneman and Tversky pioneered in behavioral economics, work that won Kahneman a well-deserved Nobel prize. His book, Thinking Fast and Slow, has become legion for researchers of human behavior, and the principles of System1 and System2 thinking have wide-ranging implications for decision theory.
A key idea is that we have two dramatically different modes of thinking. Our System2 thinking handles the complex, cognitive tasks where we need to concentrate and draw conclusions based on our efforts. So, for example, if I ask you to multiply 17 times 36 in your head, you could do this. But it would take some work. And some time. It is slow and laborious, and we tend to avoid it like household chores and homework. System1 thinking, on the other hand, is fast and intuitive, and happens with little effort. In the course of our day, we make something like 20,000 individual decisions, and the vast majority is made by System1, with little or no direct attention. Behavioral economists point out that our emotions are handled by our System1 brain. We don’t have to concentrate to be happy, or embarrassed, or remorseful. It just happens.
Furthermore, Kahneman and his followers have been able to demonstrate that System1 thinking invades all manner of decision-making. Even the highly involved and complex decisions, like buying a new lawnmower, give way to System1 processes whenever our lazy System2 brain can unload the work. [In fact, I have personally come to believe that this is one of the important functions of brands – to move product choice decisions to System1 processing whenever possible – but that discussion is best pursued elsewhere.]
Earlier, I might have easily acknowledged that emotions play a strong role in purchases centered on fashion. But on form? No way. If I’m buying a new sport coat, where style is more important than say, protection from the elements, then sure, my emotional needs loom as important determinants of choice. But for that lawnmower, I’m all about performance, and this comes in the form of System2 considerations of quality and materials and sale price and such. Yes? Well … yes, but only to an extent. Experimental evidence from behavioral economics has shown consistently that System1 thinking is also pervasive.
Examples like those from Bechara and Kahneman, cited above, left me convinced that we needed a more formal and exact means of incorporating emotional influences into our study of decision-making in general, and purchase decisions in particular. But until recently, that need was largely unfulfilled. This glaring gap was nowhere more apparent than in the Mauss and Robinson article published in Cognition & Emotion in 2009 summarizing the various measurement techniques deployed to measure emotion. The table below, drawn from that article, classifies the basic approaches to emotional measurement, and reflects the general inability to capture emotional specificity using the techniques in play at the time.
|Subjective Experience||Self-report (surveys)||Valence; Arousal|
|Physiology||Autonomic nervous system (ANS)||Valence; Arousal|
|Physiology||Startle response||Valence (at high arousal)|
|Physiology||Central (EEG, fMRI, PET)||Approach/Avoidance|
|Behavior||Vocal amplitude, pitch||Arousal|
|Behavior||Facial behavior (observed)||Valence (some specificity)|
|Behavior||Facial behavior (EMG)||Valence|
|Behavior||Whole body (observed)||Some emotion specificity|
As a rule, the emotional measurement systems we had been using could get valence (or sentiment, which groups emotions into positive, negative and neutral components) or various dimensions of emotions. A good example is the PAD theory of emotions from Meharabien and Russel (1974) that used the three dimensions of Pleasure, Arousal and Dominance to describe emotions. But consider the model as depicted in the figure below.
Here you can see that fear is low in pleasure, high in arousal, and low in dominance. When we are fearful, the feeling is unpleasant, it definitely gets our attention, and makes us retreat. You can see from this cube where bliss and triumph would show up on the dimensions. But where would you put guilt? It is, like fear, characterized as low pleasure, high arousal, and low dominance, so it could occupy much the same place in the cube. But we all know that these two emotions – fear and guilt – are quite different experiences.
It was at this point that my colleagues at Converseon, an award-winning company focused on the technologies in social listening, shared with me the work that they had been pursuing using Robert Plutchik’s psycho-evolutionary theories of emotion. Plutchik and his cohorts at the Einstein School of Medicine had classified emotions into eight basic, or fundamental, emotions that developed in humans due to evolutionary (Darwinian) pressures. So fear, for example, gave rise to the “fight or flight” response, and helped humans with this capability to either triumph or escape, and thus be less likely to be meals of certain predators. The eight basic emotions are:
- Sadness, and
Now, Plutchik’s psycho-evolutionary theories are not new, having emerged in the 1980’s. But they provided a sense of organization – a taxonomy – to the entire world of emotion which, ironically, made such logical sense to me. It is instructive to look at the “flower” or “wheel” depiction of the Plutchik structure of emotion shown below.
In this figure, emotions are arranged so that the relationships are meaningful. The eight basic emotions appear as the middle of three concentric circles on the wheel. Start with Joy, at the top, and go around that middle ring to see all eight. Each of these eight is the anchor for a spoke on the wheel, varying in intensity. In this manner, Rage is the more intense version of the basic emotion, Anger. And Annoyance is the less intense emotion. Also, the placement of the spokes is important, with Joy being opposite of Sadness, for instance. And some emotions are combinations of the more basic emotions. Love, as we all know, is a complex and “many-splendored” thing, but you can see how it can be construed as the combination of Joy and Trust.
The Plutchik taxonomy and validation of his theoretical explanations for emotion are much more complex than the simple wheel shown above, of course, but the basic premise is captured well, and this was what changed my concept of how to think about emotional effects. It finally made sense to me.
But I think that the reason that Plutchik’s view of emotion had never come to my attention was the fact that it provided structure alone, without a means of measuring the emotions in the nicely ordered wheel. And without this tool, the mental construct could not be applied effectively to the study of product purchases, and other pursuits.
The Converseon team essentially solved that problem. The time frame was 2012 – 2013, and as Chief Research Officer of the Advertising Research Foundation (ARF), I had initiated a study we called “Digital and Social Media in the Purchase Decision Process.” We had recognized the growing importance of digital technologies, like search engines and social media, on how people went about buying goods and services, and we were looking to get a better grasp on that process.
For the study, we wanted to explore the purchase process across a broad range of product types. So Kraft joined the effort, and we looked at packaged meats and cookies (Oscar Mayer and Oreo brands) on their behalf. At the other end of the spectrum, GM came aboard to study compact cars. And somewhere in between was Motorola, who tested smartphones. Google (our digital experts) and Young & Rubicam (Y&R, our advertising consultants) were also sponsors. Besides Converseon, who conducted social listening as part of the study, we had comScore (online panel research), Firefly/Millward Brown (qualitative research) and Communispace (online communities) as research partners donating professional services to the cause. Duke University’s Fuqua School of Business served as our academic advisor.
During the course of the study, the research team acknowledged that we needed to understand the emotional effects at play in the purchase process, and both Firefly and Communispace jumped on that challenge. But so did Converseon. In fact, they proposed that we could capture emotions, as expressed in online social conversations occurring in the different basic stages of the purchase process. And we used the Plutchik organizational taxonomy to show the emotional journey that buyers experience as they seek to acquire cookies, or smartphones or cars.
It’s important to know the process that Converseon uses in their approach to social listening. First, they have access to the full range of data on the Internet, including social media (Facebook, Twitter, Google+, etc.), blogs, company websites, wikis, review sites and so forth. The volume is huge, and we gathered over 50,000 posts about Oreos in about a week’s time period. It’s 10 to 20 times that for all cookies in general. At that point a combination of machine- and human-based coding is used to make sense of the posts. In our case we classified social commentary into a) the stage of the purchase process that the author was in at the time, and b) the predominant emotion expressed in the post (anger, excitement, etc.).
The figure above depicts one perspective of the purchase process. Another viewpoint that you see frequently is that of a “funnel.” Now, it is shown here as a linear process, but there is ample evidence these days to indicate that people do not go necessarily in a straight line through this process. They bounce around. They may get half-way through the process and just up and decide they are not really in the market. But this model is a nice heuristic, nonetheless. And that’s because, at any given moment in time, a consumer actively on the purchase journey is engaged in the behaviors associated with these distinct phases. So we classified our posts into these five phases.
We also classified the posts into specific emotions. The Converseon method of classification does not use the common rules-based approach via Boolean logic and similar techniques popular with many social monitoring firms. Instead human coders do the initial classifications, and the machine “learns” from their efforts. Only posts with strong inter-rater reliability (all coders agree on the proper designation) are fed to the machine for the learning process, and only when the computer classification outcomes reach acceptable levels of precision are they allowed to proceed. Throughout the process, individual posts that are deemed “uncertain” are given to human coders for final arbitration.
The figure below shows the results for all products in the ARF study combined. Finally, we have an accurate and consistent means of measuring emotions, and we can use this information to better understand things like the purchase process.
It is clear from the figure that consumers experience a range of emotions in their purchase journeys. Anticipation is a common emotion in the problem definition phase, and in the purchase decision phase, as well. Joy predominates in post-purchase commentary. But people express many emotions, both positive and negative.
Interestingly, when we separated the data in the figure above into our sponsor product categories, we discovered that packaged meats, cookies and automobile emotional journeys were quite similar, but that smartphones were unique. Go figure. The smartphone results are depicted below.
For smartphones, we found that negative emotions, like Sadness, were prevalent in the early, problem definition phase. Inspection of the relevant posts revealed that users were reluctant to give up their old phones, as they had become comfortable and quite attached to their devices. In many cases, they were being forced to purchase new phones because the older models were no longer being supported. Once they started active shopping, in the information-search phase, that negativity turned to Interest as potential buyers began to encounter the wonderful array of new phones with many new features.
The real surprise came, however, when we looked at post-purchase comments. Many of the posts expressed Joy, but there were an equal number of negative comments, reflecting Anger, Disgust, Annoyance and Sadness. Again, we dove into the data and learned that many new smartphone buyers were upset and frustrated, since they could not easily figure out how to use their new devices. One comment I felt summed it up: “Just bought me a new smartphone. Turns out the damn thing ain’t so smart after all.”
This was valuable insight to our sponsor, Motorola. And based on the learnings, we gave them two pieces of advice:
- Don’t just throw your phones over the fence to new buyers. Educate them. Train them. Offer courses to all new buyers and show them exactly how to accomplish all of the things they were accustomed to performing on their old phones.
- Connect the groups via social media. Hook the joyful buyers up with the disgruntled ones, and let them exchange information. The joyful will feel proud and quite pleased to share their knowledge, and the unhappy will welcome input from regular users (instead of helpline employees).
Knowing the specific emotions associated with decision processes can be illuminating indeed. As the results above indicate, marketers can leverage this insight to their competitive advantage. And certainly, this can inform the marketing/communications strategy.
Emotional measurement can actually shed important light in many areas. Interestingly, at the conclusion of the ARF study, I was thinking about possible contexts for interpretation of the emotional data. I knew that my longtime colleague, Amy Shea, had just opened a consultancy focused on the fundamental concepts of Story. I approached Amy with my challenge. Can we use the principles of story-telling to help interpret findings in the brand/advertising/purchase process environment. She spent a day looking over the various findings and reported back: “Absolutely!”
If you are interested in how story-telling actually works, you should read the book, aptly named Story, written by the noted screenwriter Robert McKee. He has described 52 separate and distinct genres of stories, and he has shown how mastery of the fundamentals of plot, and character development, and sub-stories, and the like are assembled beautifully in the great films of our times. Amy is a big fan, and she uses the principles of story to understand the dynamics of consumers and marketplaces.
I will not go into detail here (and there is plenty of it) but just share with you two key learnings that came out of the Converseon data and emotional coding. First, we have realized that, despite what the brand pundits often claim, consumers have not gained ownership of brands. Social media has not allowed savvy consumers to wrest control of brands, moving them in directions they see fit. But it is more of a joint venture these days. What consumers do control is the category story. They decide what the story is around running shoes, or frozen pizza. The brands actually play the roles of different characters in those category stories. That’s how it works. And it is only by capturing the emotional journeys of consumers in their buying process that we can see all of that unfold.
Second, our review of the smartphone data revealed that the story in that category is a love story genre, with a disillusionment plot. Think about it. People just love their smartphones. They take them to bed at night. They keep them in their breast-pocket, next to their hearts. Heck, most people would rather part with their wallets than their phones. But they occasionally get jilted by their lovers, told that they no longer support the relationship. So sad. Now these jilted consumers take to the streets, hoping to rekindle with a new partner. And for some, the result is indeed Joy. Their new partner is terrific. For others, however, the new mate is not so great. Anger and Sadness prevail. Heartbreaking.
There is much more to the story-based interpretation, as I’m sure you can imagine. I just wanted to share enough to underscore the point that the new approach to measuring emotions that I have described can generate value in many ways.
We now have a new and informative way to measure emotions. It is by no means perfect. It relies on social media data, for instance, and Lord knows that source is not exactly “representative.” Many people do not have access. And while some people may only post once a month, others post 10 times a day. It’s a problem if you are wanting to generalize your findings to any known population. But if you are trying to determine what people are likely to encounter when they go online for information and advice, then it is the ideal source, of course. And it is an “in the moment” source, as well. In surveys, we often ask people to remember what they were thinking or feeling when they bought that flat-screen TV. Such recall is often biased and/or inaccurate. By looking at social posts that were made at the time of the experience, we get a better picture of true emotions.
There is much more to this pursuit of emotional effects in purchase decisions. For example, I have been delving into the relative importance of emotions vs. cognitions in the various stages of the “path to purchase,” and this work is showing early promise. Also, there are other measurement techniques, like neuro-methods, that are producing considerable insight. But my argument here is simple: emotions are important determinants of choice behavior, and we now have systematic means of measuring those very emotions during the process, so we have a great opportunity to combine our right- and left-brain influences to better understand these marketplace dynamics.