In 1969, Robert Zajonc recruited 72 cockroaches and had the insects run a maze. Surprisingly, they had to do it in front of an audience.
Even more surprisingly, the audience was other cockroaches. They were placed in clear plastic boxes next to the course. (The paper doesn’t say whether they urged the maze-runner to go faster.) Some roaches were made to run a straight path to get away from a bright light; others ran a maze with turns. With the audience watching, Zajonc reported, the roaches got faster on the easy path and slower on the more difficult maze.
This research wasn’t about improving insect athletic performance. The topic Zajonc was studying was social facilitation, i.e., whether being watched changes performance. He needed a subject that could sense the presence of others but was highly unlikely to care what they thought of its performance. A cockroach almost certainly isn’t focused on what the crowd thinks of its ability. If you remove self-consciousness, and if the audience still changes performance, you’ve learned something about what causes the change.
Many interesting studies in the social sciences have failed to replicate. Fifty years after the original experiment, the original finding was questioned. In 2020, German scientists ran a preregistered direct replication with a larger number of roaches and found the audience made roaches run slower on both the straight course and the maze. Social inhibition replicated. The more interesting facilitation effect on the easy task, sadly, could not be reproduced. The Germans also pointed out that the original paper never ran a typical statistical analysis on its widely publicized results.
What’s important about this sequence of studies is the experimental design lesson. Zajonc’s setup was built very simply around a single principle. This enabled scientists to re-run it with a larger sample size half a century later and learn exactly what held up.
Experiments are paintings, not photographs
Dan Ariely published a short piece recently on how he comes up with his ideas for experiments, and his analogy is on point. Designing an experiment is like painting. A painter doesn’t try to photographically reproduce the whole scene. Instead, they do their best to capture the essence of it.
Experiments in behavioral science work the same way. One begins by identifying a real-world problem. For example, people don’t save enough for retirement, or people don’t keep taking their medication, or people abandon projects before completion… Then, reduce the problem to a principle. Next, create a small, self-contained world that includes that principle and almost nothing else. Finally, mess with it and record the results.
What a researcher is really asking, according to Ariely, is simple: what’s the essence of the problem? This explains, at least in part why so much behavioral research seems, well, a little crazy. I’ve spoken with Ariely a couple of times on my Brainfluence Podcast, and he made another important point: never let a single clever experiment be the last word on a topic. Consider the result as encouragement to conduct another study. (But, I’d add, don’t wait a half century for someone else to do it.)
Here are four more studies that probably look a bit goofy to the average person but were all designed to illustrate an important point.
The scary bridge
Donald Dutton and Arthur Aron, 1974. An attractive woman approached men who had just crossed one of two bridges over British Columbia’s Capilano Canyon. One of the bridges was a wobbly suspension design, swaying 230 feet above rocks and rapids. The other bridge crossed a stream just ten feet below and was well-constructed of solid wood.
The woman asked each man to write a short story about an ambiguous picture, then gave him her phone number to call if he had questions about the study.
Nine of 18 men who crossed the scary bridge called her. Only two of 16 from the safe bridge did. The stories written by the men who crossed the scary bridge also included more sexual imagery.
This experiment had nothing to do with bridge psychology. The essence being tested was misattribution of arousal. During a frightening experience, your biometrics change. Your heart races and your palms sweat. Your brain looks for an explanation, and it picks the most interesting one available. The scary crossing was just a way to spike a subject’s heart rate that had nothing to do with romance or attraction.
A note of caution before you share this surprising result with your friends: the bridge study methodology has been criticized for fifty years. Subjects weren’t randomly assigned to bridges (conservative types who pick the safe crossing might be more reluctant to call a woman they don’t know) and the sample sizes were tiny. But, there is some good news: replications using roller coasters and horror movies find the arousal-attraction transfer is real, but those differences are modest and don’t last very long. The effect was shown to exist, but was perhaps a bit less potent than originally thought.
What the bridge experiment got right. Why not simply ask the men how attracted they felt to the woman who greeted them? Because, of course, that answer would be inaccurate. If you are a regular Neuromarketing reader, you already know people can’t or won’t reliably tell you why they did something.
The beer menu
Here’s Ariely’s own favorite illustration, and that marketers can use.
With Jonathan Levav, he ran a study at the Carolina Brewery in Chapel Hill. Diners at each table were offered a free sample of one of four beers. At some tables, people weres asked for their orders out loud, in sequence, the way one normally would. At other tables, everyone was told to write their choice privately on a card.
The tables that ordered out loud chose significantly more variety. Nobody, it seemed, wanted to be the fourth one to say “same here.” Here’s the surprise: those people liked their beer less. The variety-seeking subjects experienced more regret. The happiest drinker at the table was reliably the person who ordered first. That subject was only one who got to choose without being influenced by the choices of others..
The essence in this experiment isn’t beer. It’s what happens when a private preference has to be stated in public. If you change the setting to a boardroom, a Zoom call, or a customer picking automobile options with their spouse watching, the same effect can kick in. (I’m not aware of an authoritative replication of the beer study, so don’t treat it as fully settled science.)
Playing with Legos
Ariely, Emir Kamenica, and Drazen Prelec, 2008. Students were paid to construct 40-piece Lego Bionicle figures. In one condition, the finished models were placed where the builder could watch them pile up. In the other condition, the researcher (sadistically?) disassembled each model in front of the subject while they began bulding the next one.
There was absolutely no difference in the task each subject performed.
The subjects who could see their work pile up produced 10.6 models. The builders who watched their work be destroyed made only 7.2.
The essence of this experiment was to see what happens to human effort when you show that human their work is meaningless. This one hasn’t been directly replicated as far as I know, but it is consistent with a large body of work on meaning and motivation. And we’ve all seen good employees mentally check out after a departmental reorganization made six months of their work irrelevant.
The energy drink
Here’s one more that marketers will appreciate. Baba Shiv, Ziv Carmon, and Ariely sold students SoBe energy drink, either at full price or at a discount. Then, they were tasked with solving a bunch of word puzzles.
The full-price group solved 5.8 puzzles, while the discount group managed only 4.2.
The drinks were identical, the onlly thing that changed was what people paid. Subconsciously, their brains adjusted the product’s potency to match. In this case, the pirce tag, purely a marketing element, is acting like a pharmaceutical.
Of the four, this is the study that has the most replication evidence. The same team later showed that full-price placebo painkillers outperformed discounted ones in a 2008 JAMA study of 82 subjects, and other labs have conceptually replicated the marketing placebo effect with non-price variables like scarcity and packaging. This should come as no surprise to Neuromarketing readers. I’ve written before about how price can act like a placebo, and about the broader risk in discounting a product you want people to value.
Implications for your next test
Marketing experiments often fail not because the methodology is bad, but because nobody asked what the the real question, the essence, was before hitting “run.” Finding out that an orange button outperforms a blue button on a specific landing page may be useful but yields no transferable insights.
Four things the cockroach people do that most growth teams don’t:
- Specify the principle before you pick the variable. Don’t test, “headline A vs. headline B.” Instead, think about “does reducing perceived risk increase signups?” The first may help, but the second gives you something you can apply more broadly.
- Exaggerate the manipulation. Ariely didn’t gently reduce the “meaning” variable in the Bionicle experiment. He tore apart the models in front of the subjects who just built them. If you can’t detect an effect at full strength, don’t waste time and effort A/B testing a diluted version of it. Test the maximum first, then dial back.
- Remove everything that isn’t the essence. The bridge study didn’t need an app, a personality inventory, and/or a follow-up questionnaire. It required only a phone number and a tally of who called.
- Change the essence and see if the behavior changes. Many studies don’t do this part. If your explanation is valid, you should be able to flip the underlying principle and see if the result flips with it. That’s what the 2020 roach team did fifty years later, and the story changed. Bad news for the textbooks, but it’s science working the way it is supposed to.
Neuromarketing Takeaway: The best experiments look silly to outsiders because they aren’t directly about their primary subject matter. Before your team plans its next test, ask them to clearly state what they really believe is happening. If nobody can articulate that, you don’t yet have a solid experiment. And even when your test does produce a nice insight, remember the roaches: re-run it before you build a major strategy on it.
