An explanation of why tricks like priming, nudge, the placebo effect, social contagion, the “emotional inception” model of advertising, most “cognitive biases,” and any field with “behavioral” in its name are not real
by a literal banana
Nothing Works That Way
Back in 2014, Kevin Simler published a provocative essay on the nature of advertising called “Ads Don’t Work That Way.” I imagine most people reading this have read it, and if not they really should read it in its entirety, but I’ll summarize as a reminder. Some advertising works in really boring ways, basically reminding potential customers that the product exists. Simler gives the example of drain cleaner; I think this is also how most fast food advertising operates. But there is also a more psychological theory of advertising, which Simler calls “emotional inception,” in which advertisers create a Pavlovian association between their products and positive emotions or other desirable attributes. In the inception theory, our little brains are extremely malleable and vulnerable to such associations, and we can’t help but associate soda with happiness, sweetened fruit snacks with being a good parent, etc.
Simler does not think this is how advertising works. His theory, which he calls “cultural imprinting,” ascribes more rationality and less pathetic malleability to our interactions with advertisements. In this model, brands create cultural messages with their advertising dollars, so that buyers know what message they can expect to be sending when they purchase products that are consumed socially. One of his major examples is beer branding, and he predicts the Bud Light advertising debacle of 2023 with incredible accuracy with one bit flipped:
If I’m going to bring Corona to a party or backyard barbecue, I need to feel confident that the message I intend to send (based on my own understanding of Corona’s cultural image) is the message that will be received. Maybe I’m comfortable associating myself with a beach-vibes beer. But if I’m worried that everyone else has been watching different ads (“Corona: a beer for Christians”), then I’ll be a lot more skittish about my purchase.
For me, this was an introduction to a new way of thinking: perhaps those phenomena that popular media and official science explain with mysterious psychological effects on weak brains – automaticity – might actually be better explained by rational processes. The rationalist community centered on LessWrong, which was an important influence on my thinking, often focused on cognitive biases, taking the work of Daniel Kahneman and even priming studies seriously as evidence for the structures of human reasoning. To their credit, these associations do not seem to have been edited out of their corpus since the replication crisis in social sciences began to demolish the automaticity literature. An important motivation of the rationalist movement, as I saw it, was that we were all very irrational beings, and had to struggle to become more rational. My argument in this essay is that we are actually very rational, but managed to convince ourselves, for a variety of (perfectly rational) reasons using a variety of tactics, that we were helpless idiots.
Ego Depletion, or Thinking Fast and Slow, The Foundation of Priming
I take the term “automaticity” from the priming researcher John Bargh, famous for “proving” that simply solving word scrambles with elderly-related words like “wrinkle” or “Florida” caused undergraduates to walk as slowly as old people. (Fewer people know that he also “proved” that the same primes cause people to be more forgetful!)
Let’s briefly look at what Daniel Kahneman had to say about priming research in his book Thinking Fast and Slow:
When I describe priming studies to audiences, the reaction is often disbelief. This is not a surprise: System 2 believes that it is in charge and that it knows the reasons for its choices. Questions are probably cropping up in your mind as well: How is it possible for such trivial manipulations of the context to have such large effects? Do these experiments demonstrate that we are completely at the mercy of whatever primes the environment provides at any moment? Of course not. The effects of the primes are robust but not necessarily large. Among a hundred voters, only a few whose initial preferences were uncertain will vote differently about a school issue if their precinct is located in a school rather than in a church—but a few percent could tip an election.
The idea you should focus on, however, is that disbelief is not an option. The results are not made up, nor are they statistical flukes. You have no choice but to accept that the major conclusions of these studies are true. More important, you must accept that they are true about you. If you had been exposed to a screen saver of floating dollar bills, you too would likely have picked up fewer pencils to help a clumsy stranger. You do not believe that these results apply to you because they correspond to nothing in your subjective experience. But your subjective experience consists largely of the story that your System 2 tells itself about what is going on. Priming phenomena arise in System 1, and you have no conscious access to them.
He was also a believer, at the time, of another important phenomenon, “ego depletion,” which, as I will explain, is the foundation for priming-style automaticity and in fact most of the phenomena that I label as examples of automaticity:
A series of surprising experiments by the psychologist Roy Baumeister and his colleagues has shown conclusively that all variants of voluntary effort—cognitive, emotional, or physical—draw at least partly on a shared pool of mental energy. Their experiments involve successive rather than simultaneous tasks.
Baumeister’s group has repeatedly found that an effort of will or self-control is tiring; if you have had to force yourself to do something, you are less willing or less able to exert self-control when the next challenge comes around. The phenomenon has been named ego depletion. In a typical demonstration, participants who are instructed to stifle their emotional reaction to an emotionally charged film will later perform poorly on a test of physical stamina—how long they can maintain a strong grip on a dynamometer in spite of increasing discomfort. The emotional effort in the first phase of the experiment reduces the ability to withstand the pain of sustained muscle contraction, and ego-depleted people therefore succumb more quickly to the urge to quit. In another experiment, people are first depleted by a task in which they eat virtuous foods such as radishes and celery while resisting the temptation to indulge in chocolate and rich cookies. Later, these people will give up earlier than normal when faced with a difficult cognitive task.
Ego depletion is an embarrassing moment for science: hundreds of experiments “replicated” an essentially fake phenomenon, and it took the organized efforts of preregistered ManyLabs-style replication police to put it to rest. (I should mention that one of several large replication attempts led by a pro-ego-depletion researcher managed to find a small effect, but this needs to be weighed against the other large replication efforts, the most recent led by another pro-ego-depletion researcher, Kathleen Vohs, that failed to find any such effect, and against the general silliness of the project.)
But why did they fight so hard for ego depletion? It’s because ego depletion is the foundation upon which priming and other automaticity effects rest, and if it falls, they have nowhere to stand, as outlined by Bargh:
Tice and Baumeister concluded after their series of eight [ego depletion] experiments that because even minor acts of self-control, such as making a simple choice, use up this limited self-regulatory resource, such conscious acts of self-regulation can occur only rarely in the course of one’s day. Even as they were defending the importance of the conscious self for guiding behavior, Baumeister et al. (1998, p. 1252; also Baumeister & Sommer, 1997) concluded it plays a causal role only 5% or so of the time.
Given one’s understandable desire to believe in free will and self-determination, it may be hard to bear that most of daily life is driven by automatic, nonconscious mental processes-but it appears impossible, from these findings, that conscious control could be up to the job. As Sherlock Holmes was fond of telling Dr. Watson, when one eliminates the impossible, whatever remains-however improbable-must be the truth.
For priming to exist as an important phenomenon, we must spend most of our time basically unconscious, or in “System 1” as Kahneman puts it, mere puppets of our environment, until some rare challenge forces us to wake up briefly to deal with it, so that we may go back to sleep.
Bargh and Kahneman are adamant that we can’t trust our experience, because our experience by necessity excludes the 95% of the time we spend as unconscious automatons, and that we must trust science instead. This is similar to the promotion of the “emotional inception” theory of advertising, which remains popular in marketing science, such as it is.
Here’s a bit of such science from Shiv, Carmon, and the legendary Dan Ariely, for flavor, which unites priming, “emotional inception” marketing, and the placebo effect, which will be the the subject of the next section:
In this experiment, participants first consumed SoBe Adrenaline Rush (a drink that claims to help increase mental acuity on its package) and then solved a series of puzzles.
Remember SoBe? The researchers told the subjects that they would be drinking this mind-improving beverage, and then had the gall to give them a form explaining that they were charging the subjects’ university accounts for the privilege of drinking it. Half of the subjects were charged the regular $1.89 price; half were told that they were only charged 89 cents, explaining it had been purchased with an institutional discount. What these experimenters “found,” was that if they had the subjects rate their “expectancies” for how much the SoBe would improve their mental functioning, thereby incepting such expectancies, the subjects who got it at a discount solved only 5.8 word jumbles, compared to 9.9 in the full price group. That is, simply drinking discount energy drink caused these poor subjects placebo brain damage to a fairly significant degree.
The example they give of a word jumble is “TUPPIL, the solution for which is PULPIT.” They allowed subjects 30 minutes to solve as many puzzles as possible; as the best group mean performance was about 10, they were taking about three minutes per word at the fastest. When I’ve looked up six-letter word jumble puzzles in the literature, they seem to allow about 20 seconds for subjects to solve them, but perhaps these were unusually hard puzzles.
That is how priming is supposed to work: we are automatons going around in our sleep, and our performance on a simple puzzle can take a major hit simply by being informed that our drink was bought at a discount. We are infinitely vulnerable to our environment, to suggestion, to parlor tricks, that we can experience a major loss of intellectual ability, walking speed, memory, just by exposure to some infinitely subtle stimulus.
From an evolutionary perspective, it seems like a bad design. You could be out there hunting with your elderly father, and suddenly you look at him and start walking slower and can’t remember what you were doing. You’re at your job and see an acronym BIB and start crawling around on the floor. You’re at the grocery store and you buy ice cream not because it is delicious and enjoyable but because the packaging primed warm emotional feelings and you subconsciously need to cool down. You find out that your prescription Adderall was bought with an insurance discount and suddenly lose 40% of your mental capacity.
It’s not just that the relevant science is fabricated, or p-hacked, or uses meaningless measures or flexible measures, although the prevalence of such things does cast doubt on their evidentiary value. It’s that we should have been more skeptical from the start. The automaticity hypothesis is just as woo as spoon bending and precognition, and we should demand as extraordinary evidence for automaticity as we learned to demand regarding the others. See, e.g., Daryl Bem’s 2011 paper purporting to find evidence of clairvoyance, which may have played a role in instigating the replication crisis, and the history of the scientific investigation of spoon bending.
The spoon benders actually had to make a show of bending a spoon. Regarding automaticity, all we seem to ask is that someone wrote a paper claiming they bent a spoon.
Pre-Post-Erous!
Most people take the placebo effect, as “demonstrated” in the SoBe study, for granted. We “know” that placebo pills heal people; maybe we even “know” that a placebo can still heal if it is openly labeled as such, or that more expensive placebos are more effective. If the placebo effect were not real, why would large medical trials of new drugs have to randomize subjects to a placebo condition? And why would they find big improvements in the placebo condition?
I will argue that we should put healing placebos into the “automaticity = woo” mental bucket. Placebos don’t work that way either.
The first piece of the puzzle is how placebos actually function. Placebos have a perfectly valid job in randomized controlled trials: they are an attempt at mimicking the “noise,” or natural variation, from every aspect of the treatment process other than that believed to be efficacious, including time. Blease et al. explain the distinction:
Before reviewing findings from OLP studies, it is crucial to clearly demarcate between two distinctive uses for the term placebo. First, is the usage of placebos in RCTs. Here the term is often understood to refer to a certain kind of ‘thing’ (eg, saline injections or sugar pills). Strictly speaking, this interpretation is incorrect: instead, placebos in RCTs ought to be conceived as methodological tools since their function is to duplicate the ‘noise’ associated with clinical trials including spontaneous remission, regression to the mean, Hawthorne effects and placebo effects. Properly understood, then, these types of placebos are deployed as controls that are specifically designed to evaluate the difference—if any—between a control group and a particular treatment under scrutiny. Ideally, in RCTs, controls should mimic the appearance and modality of the particular treatment or medical intervention under investigation. In contrast, placebos in clinical contexts are interventions that may be intentionally or unintentionally administered by practitioners either with the goal of placating patients and/or of eliciting placebo effects.
Many conditions are episodic or vary in severity, and starting from a bad time, the problem will often get better after a while on its own. When people talk about “regression to the mean” as an explanation for the placebo effect, this is what they mean. It’s mostly the “pre-post” comparison, as hinted in my title for this section. This effect can be enhanced by something called “eligibility creep” in dermatology: in an effort to include more subjects, researchers may exaggerate potential subjects’ condition at the outset of the study, so that an accurate measurement without exaggeration at the end will show a spurious improvement.
The second piece of the puzzle is that placebo effects compared to no treatment are small. Not just “small” in the meaningless sense of effect size, but too small to be noticeable or to make any clinical difference. For pain, a large meta-analysis found the mean placebo effect to be about 3.2 points on a 100-point scale, too small to matter. Similarly, the most recent meta-analysis on the placebo effect in depression found similar results, an effect size of .22, which is smaller than the effect of antidepressants over placebo, about .3, which translates into about 1.7 points on a 52-point scale and is too small to be clinically relevant. This is the case even though subjects in “no treatment” groups may have an incentive to exaggerate their symptoms in order to receive treatment, whereas those on placebo believe they are receiving treatment and have no such incentive. Still, effects are tiny, so tiny as to be meaningless in real life, and definitely tiny enough to turn out to be nothing at all with better methods.
The third piece of the puzzle is that we can’t trust studies that claim to find large placebo effects, for the same reason that we can’t trust priming studies. Here’s a figure from a paper by Waber, Shiv (from the SoBe paper), Carmon (same), and Dan Ariely, demonstrating a massive effect of more expensive placebos on pain from electric shock (both are placebos, the dark one is the normal price and the light one is discount):
We see effects in the “regular price” condition as high as 30 points on a 100-point scale, and often in the 20s, several times higher than the 3.2 point effect on pain we saw earlier, and certainly clinically significant. This level of pain relief would really matter. I was able to find a more recent study using similar methodology, though not labeling itself a replication, that found placebo effects of zero, one, and three points on a one hundred point scale in similar conditions, respectively. These seem much more realistic, and when combined with the meta-analysis results, indicates that the study that got a huge result is a major outlier.
I do not think, contra Kahneman, that we “have no choice” to accept that such effects are real. Even without knowing about all the irregularities of this particular experiment (such as lack of IRB approval for shocking human subjects and deceiving them about pain medication), we should recognize how extraordinary these claims are, and demand appropriate evidence for them, not just one team’s word that such experiments even took place.
Enter the Nudgelords
Nudge is just priming.
If there are stylized sculptures of waifish humanoids in a room, subjects will eat four fewer blueberries. If you sign at the top, you’re more honest than if you sign at the bottom. It’s priming, as a service.
Nudge studies aren’t real and don’t replicate. When they’re attempted in the real world, the effects are much smaller than in the academic studies, and the effective so-called nudges tend to share a curious feature: they operate on rationality rather than automaticity. For instance, one of the most effective “nudges” is apparently for the government to send clearly-worded letters explaining what they want people to do. That seems more like common sense and respect than a “nudge” to me, but I’m not a Nudge Professional.
“Behavioral” anything (economics, finance, etc.) tends to reduce to automaticity explanations.
I won’t spend much time on Nudge because it’s literally the same thing as priming – the same model of reality, the same experiments, often the same researchers, just using a different word.
Cognitive Bias Parlor Tricks
I can’t go over every cognitive bias individually in this format, but I will give a basic pattern of how I think “cognitive biases” are produced as academic products:
First, an experiment or test is devised that people perform “poorly” on in some specific manner or dimension. It could be a list of choices of lotteries with different payoff structures (see e.g. this investigation of problems with a celebrated Kahneman mathematical model and associated experiments), or the famous demonstration of the “endowment effect” many of us experienced firsthand, in which we are given a unique little gift, like a mug, and then shamed for not being “rational” enough about its value when invited to trade it for somebody else’s gift.
Second, this little experiment, and variants of it, are generalized to the entirety of human behavior. Perhaps a study on “conversation” involves strangers chatting in a laboratory about the answers to trivia questions, and the researchers find that this has no effect on accuracy. This may be generalized to a proposition like “conversations serve no purpose” in general, perhaps a “conversation bias” can be introduced, and people can feel smug for not liking meetings at work.
Thus, we have scientific confirmation that a “bias” exists. This is often confusing, because, for example, if the “endowment effect” obtained under normal circumstances, markets would grind to a halt and not be able to function, because everyone would value what they already own more than anyone who didn’t own it.
One solution to this kind of problem, both in cognitive bias and priming research, is to say that you found a “boundary condition” instead of that you failed to find the effect. “Boundary conditions” are a similar kind of cope to “mediator” or “subgroup” analysis, especially in small studies not powered to detect them.
Overall, I think rationality is a better starting assumption for human behavior, and we should demand a great deal of evidence for an important, widespread departure from rationality.
Which of your favorite biases survived the replication crisis, and do they generalize?
Can You Catch Stupid?
“Social contagion” is the idea that behaviors spread through human groups like infections. Trivially, the spread of technologies like hybrid corn or mobile phones can be modeled like epidemics. The only “automaticity” aspect is the tendency to take the metaphor seriously, as if people were really affected by “social contagions” in the same unconscious, unfree way as by germs.
Some phenomena that have been proposed to be socially contagious are suicide, obesity, quitting smoking, and (as a bit) acne, headaches, and height. The latter three are a bit in the sense that they use similar methodologies as the social contagion literature to demonstrate that homophily, the tendency for similar people to cluster in social groups, accounts for most, if not all, of the supposed “social contagion.” A more advanced method is to use time lags, as if time-lagged obesity weren’t as much a factor of homophily as snapshot-in-time obesity.
However, researchers essentially never try to distinguish germ-type “contagion” from social learning. I think our starting hypothesis should be that behaviors that spread in the population arise from social learning, rather than from a mysterious unconscious process of thoughtless copying. Human copying is anything but thoughtless. Copying is an important form of creativity. Nonhuman animals can copy but a tiny fraction of our behaviors, even if we sometimes like to pretend they are capable of stealing our cool cottagecore aesthetics by dressing them up in sick outfits.
People share and copy, but social contagion doesn’t work that way.
The Clockwork Universe With Clockwork People
An extended excerpt from the phenomenologist Gian-Carlo Rota, better known in mathematics for his work in combinatorics, on the fading myth of our time:
The theory of myths asserts that every civilization is ultimately characterized by the series of myths it believes in. These “working myths” are myths that that civilization is not aware of at all. They are not verbalized. The moment such a myth is out into words, it’s no longer something that people authentically believe in. It can now become a subject of discussion. So, as time goes by, a given myth that was universally believed in suddenly cracks and becomes doubted. At such a moment, the members of that civilization split into opposing camps: those who say “Yes, it’s so” and those who say “No, it’s nonsense.” And then the myth will fade, and finally it’s viewed as untenable except by a small group of people, who keep it as a superstitious belief.
….What I’m working towards is that not too long ago, there was a particular myth, universally believed, that in our time is being verbalized, which is the first stage towards its fading. A lot of phenomenology is a discussion and critique of that particular myth.
What is that myth? It’s the myth of the clockwork, the myth of mechanism. It’s the idea that you can explain every phenomenon causally, by finding an underlying mechanism. It has been strongly believed in for a few hundred years. It had its heyday in the nineteenth century. It’s very simple: understand how the wheels work and you’ll understand everything. This went on and on, and that’s the myth that’s cracking in our time. I’m not in any way implying that mechanisms are bad, or that there are no mechanistic explanations. Certain phenomena can certainly be explained mechanistically. But there are other phenomena that do not have such an explanation. When you say this nowadays, tempers run high. Someone will point a finger at you and say, “You are an irrationalist! Either you believe in mechanism, or else you are irrationalist! Either a marble is red or it’s blue! Nothing in between!” But the phenomenological answer is that it is in no way irrational to deny a mechanistic explanation. We are not denying other modes of explanation which are logical and coherent, but just not mechanistic.
Those who will accuse us of irrationalism have an extremely narrow view of rationalism, a view that scientists have abandoned a long time ago. Ever since quantum mechanics came to be, and quantum mechanics is probably the greatest scientific idea of this century (n.b. the 20th -ed.), you can kiss goodbye to causal explanation. It’s very unfortunate that we still do not really understand quantum mechanics. Feynman used to stress that: quantum mechanics works, but there is something so mysterious about it, so completely different from anything we’ve thought about, that no one has succeeded in explaining it. But that works as our ally now. If someone tries to propose causal explanation as the scientific paradigm, we can say, “Phony baloney, science doesn’t work that way anymore! They gave that up decades ago.” Science is far more sophisticated. The strictly marble-like causal explanation is something that’s been thrown out of the window.
Science doesn’t work that way anymore, but few have gotten the message. Phenomenology proposes a different form of causality from the “marble-like” version, one based on conditions of possibility, a “Fundierung” relationship of things being founded on each other in the sense of allowing each other to come into being and be perceived as such. It is outside my scope to attempt an explanation at length, but this model of the world is not a woo model. It is a richer model, a more realistic model, and a model more in accord with careful observation of the world than the received marble-like model that has allowed us to accept so much silliness.
The automaticity theories named here are holdovers from the fading myth of the clockwork universe, of clockwork people. The myth began to be named by the end of the 19th century (by William James at least, who also named the related Religion of Healthy-MIndedness that is still with us today). I hope by naming a more specific incarnation of the myth here that I can promote its thematization so that it can continue to fade. A science of ourselves cannot be established by dressing woo in lab coats, clipboards, and the mathematical ideas of an antique physics.
I invite anyone to be the Lakatos to my Feyerabend, and present Here’s Why Automaticity Is Real Actually, as mine is an extreme case and does not pretend to be a measured, balanced examination of the subject. I would not recommend that anyone superstitious attempt this project, however, for obvious reasons: if priming were true, such an effort could prove lethal.