Science Lesson: Anchoring Bias And The Mistake Of Comparing Smoking To Vaping
Anchoring bias is the common tendency for people, including experts, to not adjust their beliefs enough in response to evidence. For example, a medical treatment might be widely believed to be effective, but later when some research is done, the studies find no benefit. The correct response to this is “we were just wrong; it does not work,” but the anchoring tendency is to think “well, I guess it does not work quite as well as we thought.”
This tendency creates a major problem when the risk from vaping is compared to that from smoking.
The proper starting point for estimating the risk from vaping is to believe it is effectively harmless until evidence suggests otherwise. Research on smokeless tobacco has long shown that nicotine causes no measurable health risk. In 2008, near the dawn of e-cigarettes, Murray Laugesen reported that the other exposures from the early e-cigarettes he studied were not merely safer than smoking, but “safe in absolute terms on all measurements.” In 2013, Igor Burstyn reiterated this conclusion based on all available reports of e-cigarette chemistry.
There is sufficient health science about most of the exposures from e-cigarettes to know that they do not cause detectable risk. It is quite possible, of course, that there is some small risk from vaping. A novel aspect of the exposure might turn out to be somewhat harmful, and making common what was once a rare exposure (e.g., inhaling a lot of propylene glycol every day) might reveal some small risk that was previously unnoticed. But any evidence that emerges about such risk should be thought of in terms of moving us away from the sensible initial hypothesis that there is approximately zero risk.
Assessments of the risks from vaping are almost always presented by comparing them not to zero, but to the risk from smoking. As a result, that high risk becomes the anchor and lots of evidence is needed to move the bias away from it. In almost every survey, like the recent one in the UK, most respondents who understand that vaping is less harmful than smoking still incorrectly believe it is only modestly less harmful.
Recall the example of the medical treatment. Imagine that there was an initial study that suggested it worked 30 percent of the time, so that was once the best estimate. (In the language of statistics, the “prior probability.”) But over the next few years, five new studies show no effect at all. The revised assessment (“posterior probability”) should be that the treatment rarely or never works. But because 30 percent was the initial anchor, many will still believe that it works about a quarter of the time.
When comparing vaping to smoking, there is not even such a first study. There simply never was any reason to believe that vaping could cause risks remotely similar to those from smoking. In terms of the science, it makes no more sense than starting with the assumption that the risk from eating toast is similar to that from smoking, and then trying to make the case that it is not. But it does not take much to create an anchoring bias. Psychology lab experiments in which subjects are making a quantitative estimate have found that just showing them a random number, which they know is random, will bias their estimates toward that number.
In the case of vaping, the anchor is much more captivating than a random number. It is created by the obvious similarities (vaping fills the same behavioral niche as smoking), even those tell us nothing about the health risk, and reinforced every time someone makes the comparison, “it is less harmful than smoking.” This creates strong prior beliefs, as if there really were studies that showed the risks were similar. It takes a lot of evidence to move people’s beliefs very far from that anchor.
It is obviously tempting to quantify the risk from vaping in terms of the risk from smoking. That is what I did for the risks from smokeless tobacco. I have lived to regret that, however (more here). Even a favorable comparison to smoking encourages the comparison to smoking, and thus reinforces the anchor. Moreover, my estimate that smokeless tobacco is about 1 percent as harmful as vaping was based on hints of risk for cardiovascular disease from nicotine. Another decade of evidence has failed to find any such risk, and that benefits of nicotine probably outweigh any risk for most people, making smokeless tobacco harmless or even beneficial on net. So effectively I created an alternative anchor for those who reject the “similar to smoking” anchor. My estimate continues to be widely repeated even though it is pretty clearly too pessimistic.
The situation for vaping is far worse, however. The smokeless tobacco comparison was based on epidemiology (actual health outcomes, observed after decades of exposure). It showed what it showed. But there is no epidemiology for vaping. All the quantitative comparisons you read are just made-up numbers, plucked out of the air. It is possible to do real estimates of risk based on what we know and, as Laugesen and Burstyn showed, those estimates would be that there is basically zero risk.
But those numbers do not come from scientific estimates. Instead, they are purely products of anchoring bias. Why do people make up claims like “95 percent less harmful than smoking” or “more than 90 percent less” or whatever? They apparently know the risk is much lower; they may even understand that the science suggests approximately zero risk from vaping. But they are still just throwing out guesses that are anchored by the risk from smoking, and they do not adjust far enough from the misleading anchor.
This creates a worst-of-both-worlds situation.
These claims reinforce the anchor, and in so doing, they come across as extraordinary. That is, they effectively suggest that 100 percent of the risk from smoking is the proper starting point for our thinking, and then argue that we have enough evidence to knock this down by 95 percent. This invites the common response that we just do not know enough to make such a bold conclusion. After all, extraordinary claims require extraordinary evidence.
At the same time, if vaping really were 5 percent as harmful as smoking, it would be the second-most harmful behavior (after smoking) that millions of people – Americans, British, and others – choose to engage in. For those who believe it is acceptable to dictate to people what they can do with their own bodies, this makes vaping a legitimate target. Indeed, it seems quite plausible that the “95 percent less harmful” claim – which was crafted and repeated by tobacco controllers who wish to encourage switching –was designed to be low enough to encourage switching but still high enough to justify anti-vaping efforts.
Getting someone to move from their prior belief by 95 percent is a big ask. Thus most people who realize there is a difference still cannot accept that it is that big. Yet if someone accepts that the difference 95 percent, they still believe the risk is far higher than it really is. Getting most people to move from the anchor to a more realistic “99.9 percent less harmful” seems hopeless.
The only way to win this anchoring game is to refuse to play.
The exposure from vaping is similar to the exposure from living in a city with moderate air pollution, or working in a poorly ventilated building. It seems to be less risky than having a taste for pastries. Rather than vainly trying to claw back from the anchor point of smoking, it would be wise for vaping advocates to offer an alternative anchor: “Smoking is harmful. But vaping is in no way similar to smoking in terms of unhealthy exposures. Rather, it is similar to everyday exposures like moderate air pollution.”
Finally, it is worth adding that the “continuum of risk” concept, which is widely seen as a positive statement about low-risk products, is really a part of this same problem. There is no continuum (and that is not just because of the incorrect word choice; an actual continuum would require an infinite number of options). There are really only two points on the distribution of risks: smoking, and all the other popular options – vaping, using smokeless tobacco, or complete abstinence from tobacco products – which pose approximately the same risk. (Cigars and pipes fall in between, though they do not really matter much, and new heat-not-burn products might not be quite at the zero point.)
The “continuum” claim insidiously plants the false belief that the smoke-free options are spread across a wide range of risks. It was crafted by people whose politics or profits benefit from that misperception. All smoke-free options pose approximately zero risk. Vaping advocates would be wise to resist the myth of a continuum, which serves mainly to reinforce the anchoring bias and the harmful perceptions it causes.