Law and Liberty Forum on Opioids: My Reaction to Caulkins

A few months ago, Jeffrey Miron (who has been something of an e-mail pen pal) asked for my commentary on this essay he wrote for Law and Liberty. It's a response to another essay on the same site by Robert VerBruggen. (In my opinion, the VerBruggen piece is incredibly wrong-headed and his narrative is wrong in some pretty basic ways. I'll respond directly to his piece at another time, but I think my post from last September  still holds up well.)

There are three other responses, and I'll try to get to each of them in time. For this post, I'll focus on this one by Johnathan Caulkins.

Caulkins pushes back against the argument that most people who try drugs, even hard drugs, do so responsibly and aren't harmed by them. He recasts the problem from being "proportion of problem users" to "proportion of total use that is problematic." As in, most who people try cocaine don't get hooked. But if you look at the proportion of incidents of cocaine use, or the proportion of cocaine going to problem users, it's very high. Probably a majority by his estimates.

In 1994, Jim Anthony and colleagues published what is still one of the most widely-cited estimates of what proportions of people who ever try various drugs go on to become dependent.[2] Based on data collected between 1990 and 1992 by the National Comorbidity Survey, their estimates for the three major “hard” drugs varied from 11.2 percent for stimulants (which includes methamphetamine but also weaker amphetamine-type-stimulants) to 23.1 percent for heroin. I’ll focus on the proportion for cocaine (16.7 percent) since cocaine was then the most widely used hard drug.
 The 16.7 percent figure does not mean that at any given time five people are enjoying cocaine for every one that is harmed by its use. People who become dependent often suffer through 10 or 20 years of dependence, whereas most of those who do not become dependent use for much less than a decade, and often only quite briefly. So the proportion of days-of-use that pertains to people struggling with dependence is much greater than 16.7 percent.

He dissects some survey data about how many times cocaine users have used during their lifetime. The result of his back-of-the-envelope calculation appears troubling at first glance:

[T]he odds for the average person who tries cocaine are an expectation of three days of misery per day of harmless fun.

Sounds like a pretty bad deal, huh? The implied lesson is that cocaine is more dangerous than it appears according to "addiction per user" ratios.

Thus a naïve interpretation of Anthony et al.’s “capture ratio” is that trying cocaine is like playing Russian roulette, with just one chance in six of disaster. But after recognizing that happy use is transitory and harmful use is long-lasting, the odds are effectively reversed. It is more akin to playing roulette with bullets in five of the pistol’s chambers, not one.

Pretty damning, right? I had just recently written about this topic based on the SAMHSA drug survey data. I had the first-blush common-sense reaction that most drug users don't get hooked and don't persist in their drug use. Caulkins is inviting us to flip the numbers by using a "problematic use per incident of drug use" ratio rather than a "problematic use per user" ratio. I think his analysis is wrong for some basic reasons, and his re-casting to per incident is a mistake.

Admittedly this is getting philosophical; I'm not accusing Caulkins of making a factual or mathematical error. But, as Daniel Dennett says, "There is no such thing as philosophy free science. There is only science whose philosophical baggage has been taken on board without examination." So let me briefly play the role of the probing, groping TSA agent and see if Caulkins has inadvertently snuck something past us. Let's examine away.

Free Will

First of all, doing something that's potentially habit-forming is not like a game of Russian roulette. There isn't a flipping coin, tumbling die, roulette wheel, or spinning barrel of a six-shooter inside our heads. Human beings are sentient. They consciously decide whether to take risks or avoid them. They consciously (or unconsciously) weigh costs and benefits. The person who gets hooked on cocaine makes a series of decisions. An initial decision: "Hmm. I've heard this thing has a bad reputation. I'll try it anyway." As Caulkins himself concedes, most people make it through this step unscathed. There is a subsequent decision to use again: "That felt really good, I think I'll repeat." Or (again, far more typical): "No thanks." Somewhat paradoxically, a really good first experience can lead to a total swearing off. Drug users often quite rationally recognize that a continued dalliance with the pleasant substance might result in a habit that's hard to control. A sort of "That was good. Too good." reaction. Someone has to really indulge repeatedly and quite deliberately to turn it into a bad habit.

See my post on Unbroken Brain for more of these details on drug addiction. Drug users are mostly rational; they don't get ensnared in the "chemical hooks" of the substances they imbibe. I'll admit that it would be pig-headed of me to try to ignore (in my argument) the temptations imposed by chemical dependence and the fact that some people find these temptations irresistible. I think it's equally pig-headed to ignore the fact that most people with a chemical dependence do in fact overcome their addictions and get their lives in order. They choose to do so. They decide to make the change, despite the temptation. I'm not denying the existence of drug addiction or ducking the point here. I'm just trying to put some proper context around the phenomenon of drug addiction.

Caulkins Dismisses Too Many Causal Users

Caulkins presents a useful table showing a breakdown of how many times "lifetime users" of cocaine have actually used. 29% only once or twice; 16% three to five times; 15% six to ten times. (The survey asked on how many days they used, not how many times they used; an evening-long coke-binge in which you bumped 20 times counts as one day of use.) So fully 60% of "lifetime users" have only used it ten or fewer times. As long as Caulkins would grant that the "not even once" propaganda is overblown nonsense, I'll grant that we might want to ignore people who have only touched the stuff a few times. This population was likely never "at risk" because they never used persistently enough. (Then again, see my caveat above about drug users rationally swearing off something that's "too good" after only one or a few uses. I have heard second-hand stories about people doing this, so it can't be too uncommon. Such persons might even describe themselves as having once been "dependent.")

But Caulkins takes this way too far. He points out that anti-tobacco activists ignore people who have smoked on fewer than 100 occasions. That makes perfectly good sense to me with respect to tobacco. But consider someone who used cocaine 20 times. That could be one coke-fueled outing every weekend for the better part of half a year, or every other weekend for the better part of a year. Someone in that category could be said to have dabbled significantly. And someone who does imbibe with that kind of frequency might develop a mild "dependence" or at least a craving for the habit. That's not really frequent enough to do serious cumulative damage (long-term cocaine use damages the heart, among other things). But they might be represented in the "16.7%" figure that Caulkins cites. I think Caulkins is loading his figures by trying to dismiss the all but 14% who have used on 100 or more days in their life. The 40% who have used 10 or more times are fully in play, in my opinion.

From the paper that the 16.7% figure comes from:

[D]ependence was assessed whenever participants reported at least several occasions of extramedical drug use, under the assumption that even as few as six occasions might be sufficient for development of drug dependence, but that drug dependence would be extremely rare or improbable among persons who had used the drug no more than several times.

There Are Gradations of "Dependence"

Caulkins invites us to imagine worst-case scenarios:

People who become dependent often suffer through 10 or 20 years of dependence, whereas most of those who do not become dependent use for much less than a decade, and often only quite briefly.

"Often?" How often? Half the time? Once in every ten? Dependence is just like any other social problem. There is a distribution of severity, with the most severe instances being the least common. The better part of the 16.7% are probably people who remember using a little too much, or perhaps remember a few genuine problems caused by their drug use which quickly prompted them to stop. Most people who become full-fledged addicts age out of it by their late 20s or early 30s. A decade is a typical tenure for someone who's already become an addict, according to various other sources I've read (Unbroken Brain, High Price, sorry I don't have specific academic references handy for this "stylized fact").

His comment about "three days of misery per day of harmless fun" is more than a little bit hyperbolic. No doubt some addicts are completely miserable. But I'll bet that many of the people who strictly meet the criteria for "dependence" still at least somewhat enjoy their habit, even if they recognize it's bad and wish they would stop.

Selection Bias

It's worth keeping in mind (as Caulkins quite appropriately reminds us about halfway down the page) that this data comes from within a regime of drug prohibition. The sample of individuals who imbibe in a prohibition regime is very different from the sample who would imbibe under full legalization. These are people who are disproportionately likely to be risk-takers. By definition, they are people who choose to break the law. We are constantly inundated with information about how dangerous and addictive these substances are. Pause and think about what kind of person ignores these warnings and imbibes anyway.  People who have impulse control problems are going to be over-represented in this sample of the population. People who don't generally have their lives together (unmarried, marginally employed, no dependents or perhaps neglectful of their existing dependents) will be over-represented here. If you have a normal job and family life, certain patterns of drug use are ruled out of the question. If you look at a population where these things are missing, you're going to see a disproportionate number of addicts and persistent drug users. Of course most people who have been users (even of hard drugs) are not dysfunctional, but any population of illegal users is going to have disproportionate numbers of dysfunctional adults. You can't simply apply numbers from this population to the general population and speculate that it's a reasonable estimate for what would happen under full legalization.

The Substances Themselves Differ Under Prohibition Versus Legalization

Bolivian Indians chew coca leaf all day long. They do not inexorably escalate to powdered cocaine or crack. Presumably this is closer to the model of "legal cocaine use." Or look at another class of stimulants. Compare attention deficit disorder medications to methamphetamine. They are substantially the same substances (in fact government surveys and death statistics count them in the same category!), but school  children with ADD prescriptions spend significant portions of their day (every day) under their influence. They don't inexorably escalate to smoking or injecting methamphetamine.

Under legalization, there would likely be some coca tea drinkers and perhaps leaf chewers (lozenges? nasal sprays? tinctures?). But few would escalate to pure powdered cocaine. We likely would not have many more intense users than we currently have. More likely, we'd fill in the lower-dose-but-more-frequent-use left tail of the distribution, which full-fledged prohibition chops off. It's doubtful that the right-tail of intense frequent use would expand much if at all. You might get the occasional tea drinker who occasionally mixes his brew strong enough to get a mild buzz, much like the caffeine buzz you'd get from a tall cup of Starbucks.

The distribution of "days used in lifetime" would probably expand rightward, putting more people in the categories of more frequent use. At that point, we could talk about dropping people who used on 100 or fewer days. But I think that kind of data-censoring is inappropriate given the regime the data comes from.

Adjusting for Implausible Results

Look at the paper that the 16.7% figure comes from. See Table 2 on the 8th page of the document. So supposedly 4.9% of past psychedelic users developed a dependence? 9.1% of marijuana users? Some kind of "bullshit implausibility adjustment factor" needs to be applied here. Psychedelics and marijuana don't cause physical dependence or withdrawal. Any perceived dependence is psychological, and no more concerning than an "addiction" to video games. Maybe these drugs were the vehicle by which some people chose to harm their lives, but it would be unfair to blame the drugs for the problems of people with poor impulse control or other unrelated problems. I made this point on my "Persistence of Drug Use" post (linked to above).

I think what's going on here is that people are recalling their drug use as a "youthful indiscretion". Perhaps many of them are embarrassed about their former habit and recall it as being more harmful than it really was. Supposedly there were quality control checks in place to get accurate measures of "dependence" according to the DSM III definition (read the paper for details). But the psychedelic and marijuana numbers indicate, to me at least, that some kind of misreporting is creeping in. People who are asked about their drug use, years later when they are older and wiser, likely misreport how bad it was.

Picking the Relevant Base for "Exposure"

I'm an actuary, so I'm keenly aware of the problem of "picking the relevant exposure base." If I have a population of 1000 cars, all else equal it will have twice the accidents as a population of 500 cars. If I have a sample of 1000 cars for 2 years, all else equal there will be twice as many accidents as 1000 cars for 1 year. In fact "car-years" is a standard unit of exposure. Then again, I could pick "households" as my basis for exposure. Some households have an old beater than never gets driven plus two or three cars for regular use. The old beater isn't as exposed to risk as the others. Not all car-years are created equally, but then again neither are all households created equally. Perhaps I could use "miles driven." A car that drives twice as many miles, all else equal, will have twice as many accidents. But highway miles are safer than city miles. So maybe "equivalent highway miles driven", something that re-casts all miles driven to an equivalent number of highway miles. Or maybe I just use "vehicle-years" and adjust each individual exposure for risk factors: the guy who drives 6000 miles and the guy who drives 1000 miles each gets one "car-year" of exposure, but the first guy gets a factor of six adjustment when I calculate his accident risk.

There are different ways of doing this, some equivalent to others. But I think using "incidents of use" or "days of active use" as the exposure for "risk of addiction" stacks the deck in a way that a "per lifetime user" basis does not. Likewise, most casino-goers don't have to worry about developing a gambling addiction. But if you recast your base as "per dollar gambled", you'll find a much larger proportion (maybe a majority?) are being gambled by people with gambling problems. If you're trying to assess a priori risk, you want an exposure base that causes the risk to rise linearly as the exposure rises. It would be silly to use, say, "dollars of insurance claims paid" as my exposure base, because this restricts us to automobiles that have already been involved in accidents. Likewise, the problem with addiction is that if you do it a little too much, you will become "captured" and end up doing it a lot too much. The if the exposure base for the social problem you are trying to measure (be it auto accidents or drug addiction) skyrockets when a problem occurs, it's a bad exposure base.

I'll applaud Caulkins for raising an interesting point about what basis to use, but I don't think it's at all clear that the "per days of use" basis is the relevant one. It depends on the question you're trying to answer. "I'm offered cocaine for the first time. Should I try it?" I think the "per lifetime user" basis is the relevant one for answering this question. "I've tried before, and I have the opportunity to acquire some tonight. Should I?" Maybe the "per use" basis starts to look more relevant for this kind of question, especially for the tenth or 20th offering. I think the "per days of use" basis comes dangerously close to being a tautology. Caulkins cuts off the left tail of the distribution (too much so, I argued above) by claiming that those infrequent users aren't really exposed to addiction. Then, having censored the data to only include the right tail of the distribution, he argues that most of this cocaine use is done by people with addiction issues. Of course he brings in data on what fraction of lifetime users experienced dependence (the 16.7% figure cited above), so it's not literally a tautology. But if frequent, persistent use is part of the definition of dependence, we're still trapped inside a tautology. As in, "Let's define persistent, repeated use as problematic. Oh my goodness! Low and behold, most drug use is problematic!" Breaking the tautology depends on the independence of the "drug dependence" question and the "persistent use" question. If these are strongly linked by definition, as I suspect they are, we're stuck in tautology world.

By the way, this is hard. I struggled with the issue of "what exposure base to use" in a previous post. Suppose I want to know the mortality of cocaine users. Are the "past year users" all at risk? Or just the "past month users" who presumably have a more serious and persistent habit? Let me just reiterate that I am not at all faulting Caulkins for raising this issue.

It's Hard to Deter Self-Harm

Suppose I'm wrong about everything and cocaine use really is the three-to-one game of Russian roulette that Caulkins thinks it is. Does that support the notion of drug prohibition? Of course not.

The problem with drug prohibition is that it eats its own tail. It requires the implausible dueling assumptions that drug users are irrational enough to ignore the risks of drug use, but rational enough to be deterred by legal penalties and the paltry price increase imposed by prohibition. ("Paltry" because the full price includes all those nasty risks in addition to the actual dollar price tag.) If you actually try to model this out by stating your assumptions clearly, you find that it doesn't work. Someone who is willing to play a 3-to-1 game of Russian roulette is someone who is not going to be deterred by a legal slap on the wrist, an increase in the market price (even a severe one), or the search costs required to find a dealer. If the bulk of the "cost" of cocaine use is embodied in the inherent pharmacological cost (risk of addiction and self-harm from continued use), then drug prohibition is unlikely to meaningfully deter these users. He cites the example of marijuana legalization leading to a massive increase in daily usage, but this is a distraction. Marijuana is not harmful or addictive, so it's actually plausible that prohibition causes significant deterrence. The legal penalties, higher market price, etc., are a significant component of the total cost in the case of marijuana. Not so in the case of cocaine, if we're to believe Caulkins' estimates of the risk of addiction. Make whatever assumptions you like about drug users, but keep those assumptions consistent. They're irrational? Cool, I can buy that. Then they won't be rationally deterred by anti-drug laws. They're rational after all? Cool, then their drug use must be a rational decision that you simply fail to understand. They rationally respond to legal sanctions while irrationally responding to the pharmacological risks? No, now you are confused. Specify what the demand curve looks like, but once you've done so stick with it and spell out the implications.

Caulkins wrote a very thoughtful essay, and it has given me quite a lot to think about. I just don't buy his bottom line (about cocaine, anyway). I'll try to respond to other essays in the Law and Liberty forum as I have time.

Addendum

I felt a need to respond to this part of his essay:

Second, I concede that prohibition harms many people, probably more than it helps. However, it harms most of them only modestly, whereas some whom it protects benefit enormously.

This strikes me as a pretty blithe dismissal of the suffering caused by drug prohibition. Are we shifting back to a per-person exposure base? I guess if the median person "harmed by drug prohibition" is the casual user who can't score any, or (to use an even broader base) the tax payer saddled with the bill for a useless and ineffective drug war, this statement is literally true. But just as there is a thick right tail to the distribution of cocaine-related harms, there is a thick right tail to the distribution of drug-war related harms. That is, there are infrequent but severe cases of harm that likely dominate the total harm, by any reasonable accounting. Let's fixate on the innocent people whose homes are unnecessarily raided, the people languishing in prison because they triggered a mandatory minimum over an arbitrary weight limit, the families destroyed by the incarceration of their loved ones, the communities destroyed because incarceration has imposed a lopsided male/female ratio, the people overdosing on heroin or cocaine tainted with fentanyl (yes, that is the fault of drug prohibition, as much as the drug warriors would love to take a pass on this one).

Pardon me for dwelling on this, but what a lopsided comparison. Read the second sentence in that excerpt again. Are we comparing the median person "harmed" (implied by his use of the phrase "most of them") to the very most extreme cases of "drug abuse averted" (implied by his use of the phrase "some of whom")? If we do a proper cost-benefit analysis, weighing all costs against all benefits, Caulkins would have a very hard time justifying cocaine prohibition.