Thursday, May 29, 2008

Cognitive Dissonance and Scientific Progress

I recently read Mistakes Were Made (but not by me). The central theme of this book is the psychological phenomenon of cognitive dissonance, which basically describes the feeling of discomfort we have when confronted with inconsistent thoughts. The mind naturally tries to resolve such dissonance, attempting to bridge the gap with some other explanatory device. The book gives lots of good examples, but the one major area it left out is the role of cognitive dissonance in scientific progress.

There's a strong argument to be made the cognitive dissonance is the major driver of science. When we find evidence that contradicts an existing theory, we find ourselves in a dissonant state. Resolution of that dissonance requires some other explanation, which might be that the experiment which gathered the evidence was wrong. But, if the results are repeated enough times, that argument goes away, and we're left with formulation of a new theory, one which encompasses not only the new evidence, but also all of the old evidence supporting the old theory.

Unfortunately, cognitive dissonance is simultaneously a major impediment to scientific progress. One reason is that it is a lot easier to cling to an old theory by shutting your eyes to new evidence, or explaining it away with ad hoc hypotheses (theories which have no actual support, other than accounting for the gap between the old hypothesis and the new evidence). This is actually a major problem. The effect of cognitive dissonance is that established theories are held with such mental strength that it is automatically assumed that the evidence supporting them also carries great weight; conversely, when confronted with evidence contradicting the established theory, one must then assume that either the new evidence has much less weight than the old, or that the ad hoc hypotheses dreamed up to bridge the evidential gap are themselves supported by weighty evidence. The ad hoc hypotheses themselves become part of the "truth", making the whole mental structure even more difficult to dislodge.

I'll start with an example from physics. Physics is supposedly the "queen of the sciences", not only because it deals with the most fundamental physical phenomena, but also the things studied in physics are often "simple", in the sense of lending themselves to highly controlled experiments and being described by mathematical models with a relatively small number of variables. Contrast with something like human metabolism, where we have to deal with a highly complex system of millions of interacting parts. Yet even so, physicists are just as susceptible to cognitive dissonance as anyone else.

The example is that of the theory of "luminiferous aether". This aether was a substance proposed to permeate everything in the universe, providing a medium through which light waves could travel. In retrospect, it was something of an ad hoc theory to start with. Physicists knew that various physical media could transmit waves: strings, water, air, etc. From this it was extrapolated that light waves must also travel through some medium, and thus was born aether theory. In the nineteenth century, James Clerk Maxwell unified the theories of electricity and magnetism, showing that both phenomena were actually one in the same: electromagnetism. One consequence of Maxwell's theory was that the electromagnetic field could be shown to propagate in waves at the speed of light, which quickly led to the conclusion that light was in fact electromagnetic waves.

Maxwell's equations showed that these waves always propagated with the same speed in vacuum. According to Isaac Newton's view of space and time, this then implied that the aether must define a universal frame of reference, and all other motion (like that of the Earth around the Sun) occurred relative to this frame, moving through the aether. But for this to hold required that the aether have some fairly impressive physical properties. It was simultaneously a fluid (in order to fill space) but also more rigid than steel (to support the high-frequency oscillations of light waves). Aether could have neither mass nor viscosity, otherwise it would affect the motions of the planets. Finally it had to be non-dispersive, transparent, incompressible, and continuous at a very small scale. Great stuff, where can I get some?

It gets better. In 1887, Michaelson and Morley published results of an experiment to show the existence of the aether. The idea was simple: if light travels through the aether with a fixed speed, then the motion of the Earth through aether should affect the observed speed of light. So you measure the speed of light in the direction the Earth is moving, and then perpendicular to that direction (the actual Michaelson-Morley experiment is considerably more clever, but that's the basic result). The problem was that they found zero difference in the speed of light in the two directions. This stimulated a flurry of other ad hoc hypotheses. One of these was the Lorentz-Fitzgerald contraction hypothesis, which posited that objects moving through the ether shrunk in the direction of motion by an amount which precisely counterbalanced the difference in the speed of light, thus "explaining" the Michaelson-Morley result. Efforts to preserve aether theory got progressively more complicated and ridiculous until Einstein developed the Special Theory of Relativity, which explained all of the observed evidence quite nicely as resulting from the geometry of spacetime, doing away with the aether once and for all.

So we can see the strength of cognitive dissonance. Initially physicists experienced dissonance because they could not reconcile the idea of light waves traveling in a vacuum with what they knew about the propagation of matter waves. The aether theory was thus developed to resolve this dissonance, even though there was essentially no evidential support. As contradictory evidence arose, lots of smart people went through a lot of trouble to preserve the existing "cognitive harmony", inventing increasingly unlikely and unsupported hypotheses for the sole purpose of hanging on to the aether concept. It would have been logical to simply note that the idea never had any evidential support in the first place, and kick it to the curb. Special Relativity didn't accomplish this in a single stroke either. It took some time to gather independent evidence supporting Einstein's theory and for aether theory's strongest proponents to die off (though Einstein won two Nobel Prizes, neither were for his theory of Special Relativity). That's pretty much the norm in science, believe it or not.

Before going on, you might think the physicists would have learned from this episode. No dice. Modern cosmology (the study of the origins of the Universe) faces its own cognitive dissonance, essentially a disconnect between how the Universe was in the first few instants and how it is now. It was a pretty major fly-in-the-ointment for the Big Bang theory. The idea of Cosmic Inflation was devised to overcome this gap. Amongst other things, it postulates an essentially undetectable field called the "inflaton", permeating all of space, and having some rather precise and fortuitous properties in order to give us the observed Universe. Sound familiar? And the observational evidence for Cosmic Inflation is basically nil; yet challenges to the Cosmic Inflation hypothesis are typically met with considerable skepticism.

So let's examine some scientific matters of more immediate importance in this light. Gary Taubes' book Good Calories, Bad Calories appears to be doing a good job of stirring up cognitive dissonance in the various areas connected with nutrition and metabolism. But is Taubes himself a victim of cognitive dissonance, or possibly taking a contrarian stance to sell books? To try and answer this question, I dove in to some textbooks on human metabolism and nutrition to get some idea of what the mainstream really thought, and what evidence supported their hypotheses. What I found was many examples of cognitive dissonance, mirroring the story of aether theory (I also found that Taubes' science seems to hit the mark). The situation is rather worse in the case of human nutrition and metabolism. First, the system under study is very complex, with many interacting parts. That makes it hard enough to study cause and effect, even for a single cell. In the case of an actual person, further difficulties arise due to ethical concerns, i.e. you can't take a bunch of people, lock them up in a metabolic ward, and feed them nothing but vegetable oil for a year. So there are many gaps in understanding of metabolic phenomena and associated diseases, creating cognitive dissonance. As with the aether, these gaps are generally filled by hypotheses with poor evidential support, which then become very difficult to remove.

So let's take on the development of coronary heart disease (CHD). Roughly, the idea is that CHD is caused by imbalance of lipids (fats and cholesterol) in the blood, along with the lipoproteins (LDL, HDL) that transport this fat-soluble substances through the blood, which is mainly water. In the 1950s, CHD was perceived as a major health problem, and there was a gap in understanding the cause of this disease. In stepped Ancel Keys, who advanced the theory that CHD was caused by too much cholesterol in the blood getting deposited on arterial walls, and that in turn was caused by eating too much cholesterol and saturated fat. This part connecting diet to blood lipids to CHD is called the "diet-heart hypothesis". The evidence supporting the diet-heart hypothesis was Keys "Seven Countries Study", an epidemiological survey which showed a statistical association between consumption of saturated fat and incidence of CHD. Other studies showed a statistical association between blood cholesterol and CHD.

The problems with this evidence are well-documented elsewhere, particularly in Good Calories, Bad Calories. The short version is this: epidemiological evidence like this is extraordinarily weak. Take consumption of saturated fat. People across the world don't just differ in how much saturated fat they eat, but also in how many carbohydrates, other kinds of fat, protein, how much they smoke, daily stress, exposure to sun, yada yada yada. There are probably thousands of interrelated variables, so it's very difficult to pick one as the key cause of a disease, unless (as in the case of smoking) the association is statistically overwhelming AND supported by other biochemical and cellular evidence. For saturated fat and CHD, it's not even close; indeed, if you take a wider sample than Keys did, the association goes away completely.

The other problem is that statistical associations do not show causality. High cholesterol might cause heart disease. Heart disease might cause high cholesterol. Some third variable might cause both to happen. Simply measuring the association between two variables does not distinguish between any of these options. Instead, we need to delve deeper into the underlying physical processes to identify the cause. Smoking is a good example, where you can dive down to the cellular level and show rather precisely how the chemicals from cigarette smoke cause physiological changes leading to CHD, that smoking increases the probability of arterial spasm, etc. We then have (no pun intended) a smoking gun, other evidence supporting our original hypothesis that was generated from the weaker evidence of statistical association. So, is there a smoking gun connecting dietary lipids to CHD?

Apparently not. I've gone through three advanced textbooks on metabolism. First was
Keith Frayn's Metabolic Regulation: A Human Perspective, which had little to say on the topic beyond repeating observations of statistical association. Indeed, going through the discussion of cholesterol production at the cellular level, one might guess that dietary carbohydrates are potentially the culprit for high cholesterol, because key enzymes in cholesterol synthesis are regulated by insulin and the availability of glucose. This point is not discussed at all by Frayn.

So I then moved on to Gropper and Smith's Advanced Nutrition and Human Metabolism, scientifically considerably more dense than the Frayn book. To be fair, I should note that the version I read was from 2005, and apparently there is a 2008 version, which I'll have to try and obtain as well. At any rate, Gropper and Smith have this to say on the topic: "Despite years of investigation, the mechanism by which hypercholesterolemic fatty acids exert their effects has not been conclusively defined." This is followed by four different hypotheses as to why saturated fats (and trans-fats) might raise cholesterol, but none with any real evidential support.

Surely, somebody had something intelligent to say on this topic, right? So I moved on to Lipid Biochemistry, edited by Gurr, Harwood, and Frayn (yes, the same Frayn). This book was from 1991, but I was hoping it's greater focus on the topic of lipids (covered in mind-spinning technical detail) would provide some insight, at least in to the details of reasonable theoretical mechanisms. Here's what they had to say: "The mechanism for the effects of different dietary fatty acids on serum cholesterol concentration is not clearly understood." There was a bit of hand-waving about apparent shifts between esterified/non-esterified cholesterol pools in liver, but no details, all the more surprising given the intense biochemical detail for related topics in this book.

Sound familiar?

Both books had some other nice examples of cognitive dissonance. Gropper and Smith at one point note

Contrary to widespread belief, changing the amount of cholesterol in the diet has only a minor influence on blood cholesterol concentration in most people. This is because compensatory mechanisms are engaged, such as HDL activity in scavenging excess cholesterol and the down-regulation of cholesterol synthesis by dietary cholesterol.

But later on the same page, we find this:

Consumption of the following lipids shows a positive correlation with the risk of cardiovascular disease (CVD), primarily due to a hypercholesterolemic effect or to unfavorable shifts in LDLC:HDLC ratios:

  • Total fat

  • Saturated fatty acids

  • Cholesterol, and

  • Trans fat.

Now, if you already know that the body compensates for dietary cholesterol, and better yet know how it does this, then clearly the association between dietary cholesterol and high blood cholesterol is NOT causative. Dietary cholesterol can't be "hypercholesterolemic". Talk about dissonance: these two statements are on the same page of the book.

It's worth reflecting on the magnitude of the situation with the lipid hypothesis. Despite over 50 years of intense study, the best anyone can say about the biochemical mechanisms connecting dietary lipids and CHD is that they're "not clearly understood". "Not clearly understood", by the way, is scientist-speak for "clueless", because if they had any clue at all, there would be at least some plausible working hypothesis in place. Instead we've got nothing. Zippo. Nada. Zilch. And the situation is actually far worse, because the original epidemiological evidence has been repeated refuted by other studies such as Framingham, which found an inverse correlation between saturated fat consumption and CHD incidence, not the positive correlation touted by Gropper and Smith. Such selective weighting of evidence again shows the strength of cognitive dissonance. We tend to think of people like scientists and doctors as possessing greater-than-average powers of reasoning, to be immune to effects of cognitive dissonance. But they're just people too, in the end, subject to the same psychological flaws as the rest of the population. It's actually worse for them, because they have the same opinion of themselves, of being supremely rational and even-handed in how they weigh evidence. As such, scientists are, in a way, the most susceptible to the effects of dissonance, because they are the most convinced that they have reasoning powers to avoid such irrational behavior.

Bear that in mind next time your doctor starts a sentence with "We know that . . ." Your doctor may "know" that consumption of saturated fat increases risk of heart disease, but if you ask him/her WHY they're so sure, you're going to witness some cognitive dissonance first hand.