SHORT COURSE IN BEHAVIORAL
Edge Master Class 2008
Richard Thaler, Sendhil Mullainathan, Daniel Kahneman
Sonoma, CA, July 25-27, 2008
AN EDGE SPECIAL PROJECT
I have one more minor observation.
Dick Thaler assumed that we already understood and accepted the standard models economic discussion making. When you think about it, it is very flattering—he assumed that we had internalized the standard models to such a degree that we needed to be talked out of them.
One reply to Danny Kahneman's reply about optical illusions and priming... After reading your response, I still wonder if the optical illusion analogy isn't quite apt. I say "wonder" because I don't really know and it would require more thinking—and probably some experiments—to really tell. The mechanism of priming that Danny gives in his reply is that the brain unconsciously retrieves (a computer scientist would say "pre-fetches") a context based on environmental cues. This context then indelibly colors the next thought. Even if you are told that the data (such as tree height) is known to be random. This seems entirely plausible, and I accept it (but I have to ask, is there direct evidence of this, or is it a surmise?). But isn't this very closely analogous to what happens in optical illusions? In that case the brain also automatically processes the scene in light of visual context and it reaches a conclusion. It is entirely subconscious and works even if you are told that you are looking at an optical illusion. This seems to me to be almost exactly analogous to priming.
Here is a classic example:
I can tell you the punch line up front, but it won't change the results. The lines are all the same length—but no matter how much you know that, you still see them as different lengths. The arrowheads on the end of the horizontal lines subconsciously change our estimate of the length of the lines. They are like giving a number when talking about tall redwoods.
Of course any analogy is only that—an analogy—and it will break down somewhere. Priming is not identical to optical illusions, but both seem to be due to unconscious processing by the brain, which is highly context dependent. I've read that there are multiple kinds of priming—conceptual and perceptual—and this may not apply to all of them (but I bet it does). I think what Danny doesn't like about this analogy is not how closely it applies, but rather the inference that priming might only be as important as optical illusions—i.e. as "just" an optical illusion. That is a fair comment, since that was partly what I intended. But not as a condemnation of priming—what I wrote is that "I wonder" how close the analogy is. Now I could be foolish for confessing this, but I will admit it: I wonder how important priming is to everyday life, or, in keeping with our theme, to economic life? Rather than wonder indirectly I'll put this as a question to Danny—how important is priming to everyday life, or economic life? How could that be measured?
The processing that goes on in optical illusions exists because it often gives the right answer—often enough that it is very useful. In particular many of the seeming paradoxes in optical illusions arise because the visual system is trying to solve what we know mathematically to be an ill-posed problem that has no unique solution. One example is inverting a 2D picture into a 3D shape. The simple classic is a Necker cube.
This is a perfectly good 2D picture, but we cannot help trying to force into being a 3D object. The 3D reconstruction problem is ill posed—there are two very different solutions, each of which is feasible. So, when you look at it you alternately see one then the other—you can feel it pop in, or pop out. Without a unique solution your brain flips between the possible solutions. If we had a bit more context it would lock in one interpretation and keep it there. To create this class of optical illusion you must carefully balance two different visual interpretations so closely that they compete.
Similarly, the experiments under which priming shows up are also highly artificial contexts—that is not a criticism because to make a controlled experiment you need this. Part of that is asking people questions about things that they don't know—to which there is no solution.
If you asked a park ranger from Redwood National Park, I don't think he'd be influenced by priming up front—because he gives tours to tourists all the time quoting the world's tallest redwood (by the way, it is a tree named Hyperion that is 379.1 feet tall).
My guess is that the largest priming effect is going to occur when you have somebody who either has no clue about the real answer, or who is grasping for a poorly remembered solution, or sitting on the fence between two compelling solutions. That is the case where context sensitivity would seem to be most important.
A lot of life is spent sitting on the fence trying to decide between multiple solutions or courses of action (purchase decisions, for example), and a lot of life is about things that we don't have a clue about or only dimly grasp. So priming could be quite important even if it is restricted to that case. Or maybe it isn't restricted, and we are all at sea in a storm of contextual metaphors.
Anyway to sum up here are my questions:
Finally, I will admit to wondering about more thing, which I am sure you've all be thinking also : why on earth am I arguing with the guy who won the Nobel prize in this area?
Nathan, lots of good questions here, except the last one. Being with you for 36 hours was enough to not be surprised that you are still pushing the debate! Danny is really the right one to have this debate with but I am not sure your questions have good answers. For example, we have no idea what percentage of the time we are fooled by optical illusions. Presumably the answer is a small number, but how would we know?
Let me respond briefly to two of your points regarding my stuff.
First, I agree that unit pricing was a bust, though interestingly, in an experiment done by Jay Russo, when prices were listed in order of unit prices, this had a big effect on behavior. However, I am not excited about that result because we don't know whether it leads to a good outcome if the products are not of homogeneous quality. For example, I pointed out in an old paper (1985) that the price of dish washing liquid per dish washed was inversely related to the price per ounce. So the "most expensive" brands were actually the best values, at least as evaluated by Consumer Reports.
This is an example of the mapping problem Sendhil and I talked about. If consumers don't know how to map their purchase into utility, then markets can break down. Some firms will make money selling diluted dish washing liquid to consumers who use the wrong heuristic for buying (price per ounce). But they won't make huge amounts of money because it costs more to bottle and distribute diluted product, so competition itself does not solve the problem. Nor does Consumer Reports if most people don't read it. (They re-did the study a few years later and got the same result.) My idea of electronic, machine-readable disclosure will not solve these problems completely either, because not everyone will go on-line, but it would make markets more efficient.
The most interesting question is whether it is necessary for the government to mandate the disclosure. Before I address that let's clarify one point. It may not be necessary that the government be the body that decides what the template is for disclosure.
For example, the travel web sites might determine what the disclosure template would be for hotels and airlines. The only problem with this is if the airlines and hotels capture the travel web sites. However, there can be no debate about whether it is necessary for the government to require that the disclosure be made at all.
Consider the case of mortgages. Let's stipulate that some borrowers took out loans that they did not understand, and that mortgage brokers, especially those marketing to the sub-prime sector, were profiting exactly by not disclosing clearly some of the features of the loan. If I am making money precisely because I do not disclose everything, then I cannot be expected to start disclosing voluntarily. (Suppose the guys who play Three-card Monte on the street had to reveal all their tricks. Who would play?)
I have no problems with Nathan's discussion of illusions and the Necker cube—both appear in early chapters in what I have been writing. Indeed, illusions have something in common with priming effects and with most of mental life: the processes that determine them are not represented in consciousness. As Nathan suspected, my objection to the comparison of priming to illusions related to their importance. Illusions do not play a very important role in the modern study of perception, whereas priming is, I believe, critical to understanding how the mind works.
Anchoring effects work on everybody—including professionals, such as judges in actual courtroom cases, who get demonstrably anchored on numbers that have little more content than random numbers. Anchoring also works quite well in negotiations, where getting your number in first gives an advantage (this is experimental, but confirmed by negotiations experts as a phenomenon observable in significant negotiations).
True, anchoring effects are largest in areas in which there is little experience. If you know how tall the tallest redwood is you won't show an anchoring effect. But not everyone knows everything—although there may be one or two exceptions… Anchoring is not a laboratory curiosity. Nor are framing effects, which are best explained as effects of priming—different frames evoke different associations in memory and different emotions.
I cannot resist quoting a paragraph from a discussion of anchoring:
These effects are large and robust—Nathan, they work on you too!
Well, I am going to sound like I was primed by the recent X files movie, but Danny, I want to believe!
Nevertheless it is not easy to believe it. Despite your having primed me, I'm just not getting it. So while I believe you that it does work on me, in this particular instance it seems to be failing me. So I have a couple more questions.
Perhaps practicing on me will help your book :-), but don't feel compelled to answer if you don't have the time. Indeed one reply you give would be "wait for the book".
My natural inclination is to go in two directions with this:
1. There must be a catch—meaning something which winds up lessening the importance of priming from what you seem to be saying. Here is a list (I will try to be brief) of objections that come to mind.
2. Or, I believe. But now explain how to reconcile our intuitive view of life with the strange reality you propose. The naive reaction to taking you seriously is the storm tossed sea of metaphors—can we really be reality? Could unconscious suggestion send us lurching between extremes all the time without our being aware of it? Perhaps, but it deserves explanation. How could it be so hidden? Why aren't we all switching political parties, wives, cars, and other decisions randomly based on some damn number we hear, or points we plot on graph paper? How can we go thorough life so gullible? Most aspects of life do not seem like a totally random walk—yet that is what you seem to be leading us. The incredible constancy and directionality of some aspects of life is hard to reconcile with the notion that we are almost whimsically influenced by cheap metaphors.
I think it would help the uptake of this idea if you gave people some sense that one can reconcile priming with our life experience. Otherwise it becomes easier to reject the whole thing. Here are some attempted reconciliations I have thought of:
Finally, I have one additional question. Have you considered "Bayesian priming"?
Suppose that I am an Econ, not a human, and I am a very good Bayesian. Before making a judgment I recall my priors—the prior distribution has a very strong sway on what I do. Let's assume my prior distribution is accurate. But although I aspire to Econ-hood, I am only human. So if we assume my prior distribution is faulty—for example if the last thing I heard influenced my prior distribution, then I could use a good procedure (Bayes' theorem) and get a bad answer.
The question here would be this—if you do numerical priming experiments (guess tallest redwood after hearing a random number), then one might be able to calculate what people's prior distribution was, and how it changed on hearing the priming number.
The answer might be that people are non-Bayesian, or that they can't do the prior. But it seems like there would be some merit in figuring out what effect the priming number is. If you said the tree was 1 million miles high, or 1 inch high, I bet it would be less priming effect. I would guess that the maximum effect occurs once you are in a zone of plausibility (i.e. has a probability consistent with your naive prior), and that the priming number tends to cause you to update the prior distribution in a systematic way.
Let me postulate a few things:
So yes, of course there are limits to priming effects and to all forms of influence. My point was not that priming could make a person do anything at all. It was that priming has much more influence than people think it could have. Furthermore, people are generally not aware of having been influenced.
Let me respond to your points in your previous message above.
In response to your final question, the experiments you suggest have been done. Totally ridiculous numbers will not work, but you can be quite extreme and still get large priming effects. Dan McFadden (Econ Nobel Laureate) and I reported on a study in which one group of subjects made free estimates of a set of quantities, or answered hypothetical open-ended questions about contributions to various causes. Other subjects were anchored by a dichotomous question (the redwood example was from that study). The anchors were set either at the 5th or at the 95th percentile of the distribution of free responses. We could measure the anchoring effect as the proportion of the distance between the anchors that was spanned by the responses of the anchored groups. The results of such studies produce a robust estimate: about 50%. I have called this measure the anchoring index.
Nathan, priming and anchoring do not dominate your life—but they affect your choices and judgments more than you can tell by introspection. You can see your face in the mirror but not the brain behind it, and you can see some of what goes on in your mind, but most of it is hidden.
Danny complains that I am objecting to something he never said. I suppose that may be technically true, but what it tells me is that I must have done a poor job of explaining why I was bringing these points up in the first place. My apologies, let me try again.
Priming as Danny presents it is quite a strange phenomenon:
I'm pretty sure Danny said each of these, one way or another. Or maybe I was just primed to draw these conclusions myself, but I think they are accurate.
If find that set of characteristics to be fascinating. However, they are also strange, and perhaps a bit alarming if you really take them seriously. It very naturally begs a set of other questions.
That is really the point—the explanation of priming (and particularly those factors above) begs some important questions. When Danny says that I am responding to claims he didn't make, I think he is misunderstands this point. I think that that the claims Danny did make beg the questions I was responding to.
If somebody told me "the sun is green", there are two natural reactions I would have. The first would be to be skeptical and discount the assertion, thinking it is either false, exaggerated or occurs in very weird conditions. The second is to accept it provisionally and say "ok, if the sun is green, help me understand and accept that by explaining further how this could it be that I've lived my whole life thinking the opposite". Even if I want to believe, if I get no answer to this second approach, then I surely will be driven back to skepticism. But hey, maybe that's just me.
When explaining something this strange, it is very helpful to provide preemptive rebuttals to the first line of skeptical theories. That way, even before the first few objections arise you can swat them down. It is perhaps even more important to provide the perspective on "how could it be that I've spent my whole life thinking the opposite". I didn't get either one from the discussion at the Master Class. Of course that may just be me being really dense. The brevity of the discussion is clearly a factor. Danny's forthcoming book probably handles this all brilliantly and I should just shut up and wait for it. But I'm impatient so I sought to clarify them in this response and counter response format.
My previous response cataloged a bunch of possible reactions to priming in each direction. The first was a list of ways of possible objections to discount the effect. The second was a set of possible ways to reconcile the phenomenon of priming with a naive intuitive notion of perception.
Frankly, approach this hasn't been all that successful, so I'll stop. Some of Danny's responses are spot on in providing the extra insight that I am looking for—they either rebut a skeptical point, or they explain that. Mostly however, I seem to have phrased things badly so Danny's responses seem aimed at trying to debate a recalcitrant heathen.
Incidentally, the sun is indeed green! But not usually—it is only green during a rare and very brief phenomenon called the "green flash". The rarity and brevity are why we don't think of it that way.
The novelty of the recent priming literature is in something that I called "Associative coherence" or "the poetry of priming". The characteristic of our responses to stimuli (I believe I used the word VOMIT as an example) is that they are coherent—the entire associative machinery (including the autonomic and skeletal responses that the machinery controls) seems to be reset for a new context. We are more alert, we are prepared to recognize stimuli that are predictable (in a Bayesian sense) in the new context, we are ready to escape etc. This coherent response makes a great deal of sense in an evolutionary context. The fact that some associations appear bizarrely symbolic (e.g., to notions of distance or reminders of money) makes sense in the context of theories of "embodied cognition", which trace some of the concepts people have to early experiences, e.g., of social and physical distance or of the difference between situations in which money and exchange are or are not relevant).
Nathan, thanks for the skepticism. I learned both from the questions and from the answers. In this context, 'recalcitrant heathen' is not pejorative—it refers to someone who says "you will have to do more to convince me", which is a pretty good description of our exchange.
Well, with this I think we can declare our commentary finished! Danny and I seem to be largely in agreement.