The Coke and Pepsi brain study: what McClure found in 2004, and what the result can and cannot carry

Origin

The blind taste test was already old. Pepsi had run the Pepsi Challenge in the 1970s and 1980s, and the recurring result, that people preferred Pepsi without labels and Coke with them, was a marketing puzzle. McClure’s group set out to look at what the brain was doing in each condition.

The paper appeared in Neuron in October 2004 under the title Neural Correlates of Behavioral Preference for Culturally Familiar Drinks [1]. Sixty-seven participants took part across the study’s conditions. In the blind condition, people received squirts of Coke or Pepsi without knowing which, and their behavioural preference was roughly even. Activity in the ventromedial prefrontal cortex, a region associated with reward value, correlated with the preference they expressed.

In the labelled condition, participants were told when the drink was Coke. Knowing it was Coke shifted preference toward Coke and recruited the hippocampus, the dorsolateral prefrontal cortex and the midbrain, regions the authors associated with memory and with modifying behaviour on the basis of emotion and affect. Being told the drink was Pepsi produced no comparable shift.

The authors’ own summary was restrained. Visual images and marketing messages, they wrote, had worked their way into the nervous systems of the people who drink the colas [1]. The problem they were addressing was a genuine one: how cultural information about a product changes the brain’s response to it.

What the evidence shows

The behavioural result is solid and not new. Brand knowledge changes stated preference. This was known from the taste tests, from Allison and Uhl’s 1964 beer study [2], and from a long line of work on expectation and perception. McClure replicated it.

The imaging result is a correlation. Different brain regions were active in the labelled condition than in the blind one. That is what the scanner showed. It is consistent with the story that brand memory was recruited. It is also consistent with other stories, which is the problem with inferring a mental state from a region, discussed below.

The sample is small and the design is specific. Sixty-seven people, two colas, one laboratory. The study was not designed to tell marketers anything about advertising, pricing, packaging or any brand other than the two in the cups. Any claim that extends beyond “a familiar brand name changed the measured response to a drink” is an extension the data do not make.

The asymmetry is the most interesting finding and the least discussed. Coke’s label moved preference. Pepsi’s did not. The authors noted it. Marketing writing about the study almost never does, because it complicates the tidy lesson. One brand had built something in memory that the other had not, and the study cannot say what it was or how.

The challenge

The challenge is methodological and it applies to a whole genre, not to this paper alone.

Reverse inference. Russell Poldrack set it out in 2006 [3]. Forward inference says: when people do task X, region Y activates. Reverse inference says: region Y activated, therefore people were doing X. The second does not follow from the first unless region Y is active only during X, and almost no region is. The hippocampus being active does not establish that participants were retrieving brand memories. It is consistent with that. It is consistent with other things. McClure’s paper is reasonably careful about this. The marketing literature built on it is not.

Replication. The study has not been replicated at scale in the form that marketing cites. Imaging studies from that period were typically small and the field has since reckoned with the consequences [4]. This does not make the result wrong. It means the result is one data point carrying a weight it was not built for.

The leap from a region to a strategy. Even taking the imaging at face value, the study says nothing about what Coke did to earn the response. Decades of advertising, distribution and availability all happened before the scanner was switched on. The paper measured the end state. Marketing has used it as if it measured the method.

The literature on brand effects on perception is strong without the scan. The literature on what the scan adds is thin. That is the honest division.

Where I see this go wrong

The study becomes “neuroscience proves brands work”. It does not prove that. It shows that one familiar brand name changed the measured response to one drink in one laboratory. That brands change perception was already established behaviourally. The scan added a picture, and the picture did the persuading.

The small sample disappears. Sixty-seven is rarely mentioned. The image of a lit brain is presented as if it were a population finding. In the advisory work, the question “how many people were in that study” has ended more than one argument.

The asymmetry is dropped. The lesson becomes “brands change the brain” rather than “one brand did and the other did not, and we do not know why”. The second lesson is the useful one, because it points at the work. Something Coke had done over decades had an effect that Pepsi’s equivalent efforts had not. That is a question about brand building. The scan cannot answer it, but the study at least poses it.

Neuromarketing is sold on the back of it. This paper is the founding citation for an industry that offers to scan your customers. Some of that work is careful. Much of it rests on reverse inference and small samples, and the Coke and Pepsi study is cited as the proof of concept. One study, with the limitations above, is a thin foundation for a procurement decision.

Having come from television, where audience research has been sold on similar scientific dressing for decades, I have learned to ask for the sample size before the finding. It is a reliable test.

What to do with it

1. Cite the study for what it shows: a brand name changed the measured response to a product. Do not cite it for how brands should be built or for any claim about advertising.
2. When it appears in a deck, ask for the sample size and the year. Sixty-seven and 2004. Then ask what has replicated it. The answer will be short.
3. Read Poldrack 2006 before quoting any imaging result in a marketing context. Reverse inference is the single most common error in the genre and it takes one paper to inoculate against.
4. Take the asymmetry seriously. If one brand’s name moves response and another’s does not, the difference lies in decades of brand building, not in the scan. Ask what the stronger brand did. That is the useful research question.
5. Rely on behavioural evidence for brand effects on perception. Blind versus labelled tests are cheap, repeatable and tell you what you need. The scanner adds a picture, not a finding.
6. Treat any proposal to scan your own customers as a research procurement. Ask for the sample size, the analysis plan and the inference method before the price. If the vendor leads with the Coke and Pepsi study, that tells you the pitch is twenty years old.

If you run the company rather than the marketing

Someone will show you a brain scan to prove that brand matters. It will be this study or one built on it.

The point being made is correct. A known name changes how people experience the product, and in a category where buyers cannot easily tell competitors apart, that is where the margin lives. You do not need a scan to accept it. Blind versus labelled tests have shown it for sixty years.

What the scan should not do is move you toward paying for neuroscience of your own. The study it rests on had 67 participants in 2004 and has not been replicated in the way marketing describes. If a vendor proposes to scan your customers, treat it as a research purchase and ask the questions you would ask of any research: how many people, what analysis, how do you get from a brain region to a claim about my buyers. If the answers are thin, the behavioural test will tell you more for a fraction of the cost.