Monday, December 30, 2019

Interesting Study On the Origins of the Opioid Crisis

 The study is called Origins of the Opioid Crisis and Enduring Impacts. Here is a link to the working paper. The paper attempts to put the blame on Purdue's aggressive marketing of the drug. It references internal Purdue documents that describe its marketing strategy. There are five so-called "triplicate" states, in which multiple copies of a form had to be filled out whenever a doctor prescribed oxycodone, the active ingredient of OxyContin. (I believe any Schedule II narcotics required the form, not just Oxy). States with these laws included Illinois, California, New York, Texas, and Idaho. The paper shows that these states did indeed see less adoption of OxyContin and subsequently less overdose deaths as compared to other states. Here is the abstract:
Overdose deaths involving opioids have increased dramatically since the mid-1990s, leading to the worst drug overdose epidemic in U.S. history, but there is limited empirical evidence on the initial causes. In this paper, we examine the role of the 1996 introduction and marketing of OxyContin as a potential leading cause of the opioid crisis. We leverage cross-state variation in exposure to OxyContin’s introduction due to a state policy that substantially limited OxyContin’s early entry and marketing in select states. Recently-unsealed court documents involving Purdue Pharma show that state-based triplicate prescription programs posed a major obstacle to sales of OxyContin and suggest that less marketing was targeted to states with these programs. We find that OxyContin distribution was about 50% lower in “triplicate states” in the years after the launch. While triplicate states had higher rates of overdose deaths prior to 1996, this relationship flipped shortly after the launch and triplicate states saw substantially slower growth in overdose deaths, continuing even twenty years after OxyContin's introduction. Our results show that the introduction and marketing of OxyContin explain a substantial share of overdose deaths over the last two decades.
If you're curious about the details I recommend reading the paper, which is quite readable. Even if you're not familiar with the time series techniques it's using (I'm certainly not), it's easy enough to understand what they're doing and follow their graphs and tables. That being enough of a summary, here's why I think their basic story is wrong.

Opioid Abuse Didn't Increase Over Time

I've written about this before. The data simply don't show any increase in opioid abuse over the period of interest, when the number of prescription and tonnage of opioids prescribed increased dramatically. Here is Figure 5 from the paper:

It looks like the OxyContin abuse rate increases from 0.6 to 0.8% of the population in the non-triplicate states from 2004 to 2010 (before again coming back down to ~0.6). Okay. But the abuse rate for other pain relievers decreased from about 4.4 to 3.9% of the population in the same period. (2010 is the year that abuse-resistant OxyContin came on the market; it was harder to crush and thus override the "time release" nature of the pill. I believe this was also the year that people started noticing the increase in overdose deaths and "do something" policymaking started to push back on opioid prescribing practices.) In other words, the total percent of the population abusing opioids decreased over the period of interest. At best, you could call it a 33% increase in OxyContin abuse specifically for the non-triplicate states (that's 0.2 / 0.6). The triplicate states also increased by about 33%, though, going from 0.3% to 0.4%. Also, abuse rates for non-OxyContin opioid are higher in 2010 to 2012 for the triplicate states, who were supposedly spared the full brunt of the opioid epidemic? Adding the numbers together for 2012, it looks like total opioid abuse is about a 4.5% for triplicate states (I'm crudely eyeballing a 0.4% off the left graph and a 4.1% off the right) and 4.5% for non-triplicate states (0.6% plus 3.9%). It seems like an important causal link in their story is broken. 

This is worth pondering, because later in the paper they attempt to blame total drug overdose deaths on the triplicate/non-triplicate difference. They even acknowledge the 2010 introduction of abuse-resistant OxyContin and the subsequent increase in heroin and fentanyl deaths. Here is Figure 6:

You might think that "opioid overdose deaths" implies prescription opioid overdoses, because the paper is ostensibly about the role Purdue Pharma played in encouraging doctors to overprescribe. But their footnote 22 on page 14 implies that they are counting all opioids, including heroin and synthetic narcotics. (It's counting ICD-10 codes T40.0 - T40.4 and T40.6. T40.1 is heroin, T40.2 is "other opiods", the category that includes OxyContin, T40.3 is methadone, and T40.4 is "other synthetic narcotics". Deaths coded with this T40.4 probably mostly involved synthetic subscription opioids prior to 2010 or so. But these deaths started to spike around 2013 when heroin started to be laced with fentanyl at increasing rates, and these deaths would generally be coded T40.4.) In fairness, they do the analysis separately for T40.1, T40.2, and T40.4 in Figure A6. In my opinion it is inappropriate to lead with the analysis on all drug overdoses or even all opioid overdoses if the claim is that OxyContin specifically is the culprit. They can and should do the analysis using T40.2 deaths excluding T40.1 and T40.4. In other words, what does there analysis yield when looking at deaths involving prescription opioids but not involving heroin or illicit fentanyl? Take a look at Figure A6. They show that heroin and synthetic opioid (mostly fentanyl) deaths are higher in non-triplicate states, but the difference is not statistically significant. (For some reason, the confidence intervals are very wide.) If fentanyl and heroin overdoses can't be rightly blamed on Purdue's marketing, it seems they should exclude deaths involving these drugs from the analysis. And in reporting excess mortality rates in non-triplicate states, they should only be reporting "other opioid" (T40.2) mortality, not total drug-related mortality. From the conclusion:
Our estimates (using Table 3, Column 3) show that nontriplicate states would have experienced 4.49 fewer drug overdose deaths per 100,000 on average from 1996-2017 if they had been triplicate states and 3.04 fewer opioid overdose deaths per 100,000.
This is apparently on the basis of total drug overdoses (the 4.49 figure) and total opioid overdoses (the 3.04 figure). Given Figure A6, it seems inappropriate to use these totals. The fixation should be on T40.2 mortality. If they are going to report total drug or total opioid mortality, they should note that they are speculating beyond what their analysis shows. They are in effect blaming heroin and fentanyl overdose death rates on Purdue's marketing, even though their analysis shows the difference in these deaths (between triplicate and non-triplicate states) to be statistically insignificant. Here is Figure A6:


Back to the abuse rates. I think there is a contradiction here. The standard narrative, which this paper is implicitly endorsing, is that OxyContin prescriptions stoked the appetite for other opioids, created a new population of addicts, eventually leading to the increasing rates of heroin and fentanyl overdoses. But take another look at Figure 5. Why didn't abuse rates for non-OxyContin opioid increase? Why does this general appetite for opioids fail to show up in the abuse rates? I really wish that people who comment on the opioid crisis would take this more seriously, because it is a major flaw in their story. My best literal reading of the data is that opioid abuse flattened out by 2000 or so, even though prescriptions continued to skyrocket and drug poisoning deaths continued to shoot up over the 2000 to present period. (It is not clear what they did prior to 2000. Presumably they rose a little, but that's far from obvious.) Did the population of illicit opioid users saturate by 2000? Did Purdue's marketing, which started in 1996, only take four short years to reach this peak? Was the continuing upward trend in deaths a result of more intense use by this (supposedly new) class of drug users? I'm not picking on the authors of this paper here, but people need to be more specific with their timelines.

Including Suicides In the Analysis

Back to footnote 22. They are looking at deaths with ultimate cause of death codes X40-X44 (accidental drug poisonings), Y60-Y64 (suicides involving drugs), X85 (murder involving drugs), and Y10-Y14 (drug poisonings of undetermined intent). Suicides are a relatively small proportion of total drug deaths. But why include them at all? Are the authors implying that deaths involving suicide by opioid wouldn't have happened but for the increase in opioid prescribing? They should redo their analysis just on the accidental overdoses, X40-X44, because these are the only deaths that can be properly considered part of the opioid crisis in the sense that they wouldn't have happened anyway. This is a little odd, because they even have a section on "deaths of despair", Section 5.4.3. They analyze suicides and alcohol-related liver disease and find that triplicate and non-triplicate states don't have different trends in these mortality rates. I suspect redoing their analysis on accidental drug poisonings would make their triplicate vs. non-triplicate differences larger; this one change might strengthen their conclusion. 

Major Metropolitan Areas

Take another look at the list of "triplicate" states. New York, California, Illinois, and Texas include the four largest American cities: New York, Los Angeles, Chicago, and Houston. Population estimates of the major metropolitan areas for these cities implies an enormous share of the population concentrated around these few cities.  It's possible that the differences between triplicate and non-triplicate states are driven by a few major metropolitan areas. It's not to hard to imagine that four or five major cities might just be idiosyncratically different from the rest of the nation. It might be interesting for the authors to redo their analysis on, say Chicagoland vs. southern and central Illinois, or New York City versus a rural, mountainous part of New York state. (New York is a lot of unpopulated mountains and forests with a few big cities. Beautiful to drive through, by the way. Lots and lots of nothing until you reach a big city.) Notably, Idaho, the only state on the list that doesn't have a mega-metropolis, is something of an outlier in the triplicate group. See Figure A3 from the paper.

(Be careful with the metropolitan area link above; some of these span state lines. The Chicago metro area includes part of Indiana and Wisconsin, for example, and the New York metro area includes parts of New Jersey and Pennsylvania, if I'm reading it right. It might be interesting to see if the parts of the metro areas within the triplicate states are different from the parts outside it.)

The paper does try to control for this problem in a couple of ways. See section 5.4.1. They redo their analysis comparing triplicate states to the non-triplicate states with the largest populations and get similar results. They also do their analysis for urban vs. non-urban counties and get similar results. None of this rules out the possibility that the four or five largest cities are just idiosyncratically different from the rest of the nation in ways that have nothing to do with prescription monitoring laws, and that this difference is driving the results. 

Is Purdue's Marketing To Blame or Are Prescription Monitoring Laws To Blame?

There is something implausible about the story that Purdue's marketing is to blame. Purdue apparently decided not to market as aggressively in the triplicate states because it thought that doctors in those states would be less likely to prescribe OxyContin. If Purdue was right about this, then maybe it's actually the presence of the prescription monitoring law and not Purdue's marketing that caused the difference in OxyContin prescribing and overdose deaths.

The paper's story is that Purdue's marketing is to blame. This conclusion relies heavily on a Purdue internal memo describing its marketing strategy. See Figure A1, an image of the memo suggesting they avoid triplicate states. But they don't actually have any data on what Purdue spent on marketing. From the paper:

The statements made in these internal documents suggest that Purdue Pharma viewed triplicate programs as a substantial barrier to OxyContin prescribing and would initially target less marketing to triplicate states because of the lower expected returns. While we do not have data that breaks down Purdue Pharma’s initial marketing spending by state to confirm this directly, we will show that the triplicate states had among the lowest OxyContin adoption rates in the country.
Emphasis mine. 

The paper explicitly considers the possibility that it's the law itself and not Purdue's marketing strategy that caused the difference between triplicate and non-triplicate states. See Section 5.4.2. It tests to see if other prescription monitoring programs provide some level of protection against the opioid crisis. There were many electronic prescription drug monitoring programs, PDMPs, in various states. These should have affected prescribing behavior in a similar way to what triplicate laws, but the paper found no such effect. Their story is that triplicate laws required the actual filling out and storage of a physical paper form by the doctor, a hassle which made the cost more burdensome and the potential scrutiny more salient in the minds of the doctors. The paper also discusses two former-triplicate states that had repealed their triplicate laws in 1994 (prior to the 1996 introduction of OxyContin): Indiana and Michigan. These states should have had a similar prescribing culture to the other triplicate states given the recency of the law change, but the paper found no apparent effect of this prescribing culture on subsequent mortality. From the paper: 
[W]e compare the five triplicate states to the two former triplicate states that had discontinued their programs prior to 1996. In both tests, we find that the five triplicate states have uniquely low exposure to OxyContin and drug overdose rate growth even when compared to states with more comparable prescribing cultures. This evidence supports the role of Purdue Pharma’s marketing rather than cultural factors and entrenched prescribing habits in explaining OxyContin exposure and mortality patterns.
Emphasis mine. This is really quite stunning. They are making Purdue's marketing uniquely responsible for the observed differences between triplicate and non-triplicate states, independent of the laws that Purdue was actually concerned about. Is the take-away here that Purdue might as well have marketed in Texas, Illinois, California, New York, and Idaho? Their marketing instincts were wrong? Their marketing was so powerful and persuasive that they would have successfully convinced doctors to prescribe in those states, too, despite triplicate laws? If I'm taking their results seriously, it seems that's what I'd have to conclude.

Consider some counterfactuals. What if all states had had triplicate laws in 1996? Would Purdue have just marketed everywhere? It seems implausible that they would have simply declined to market anywhere, or marketed less aggressively everywhere, unless the triplicate laws themselves were driving prescribing behavior. Suppose half of all states had had triplicate laws. Does that mean Purdue would have marketed less aggressively in half of the states? Or would they have researched the impact of their marketing more thoroughly and concluded (as the paper apparently does) that marketing trumps the effects of a prescription monitoring law? It's difficult to come up with a reasonable policy implication, even one that we could hypothetically enact in 1996 to prevent the expansion of OxyContin.

Is the conclusion that pharmaceutical companies shouldn't be allowed to market their drugs at all? Keep in mind that "marketing" is communication between doctors and pharmaceutical companies, usually through pharmaceutical reps who may or may not have a scientific background. It seems like this communication has to be allowed to take place through some channel or another, so I don't see any reasonable way to "ban marketing" by pharmaceutical companies. Doctors are generally more scientifically sophisticated than the reps, and they know these people are trying to sell something. They know that the facts they're being presented with are a biased sample of all facts available, and they are capable of checking their veracity. It's harder than you might think to fool people. At any rate, Purdue was correctly informing doctors that opioids are less addictive than everyone assumed. I'd like to see Purdue's critics be more precise about what exactly their deception was. Did they say the risk of addiction was 1% but really it's 2%? Did Purdue claim a rate of addiction that seemed correct by the evidence available at the time but turned out subsequently to be higher after two decades of expanded opioid access? I'd like to see the critics acknowledge that addiction rates were and are quite low by any standard. 

Market Share

It's not discussed in the paper, but it's notable that Purdue's share of the opioid market was actually pretty small. Here's an image lifted from an FT piece titled "Purdue Pharma's One-Two Punch." 


Interestingly, that article is something of a hit piece on Purdue, "exposing" that Rhodes Pharma is a subsidiary owned by the Sackler family. But it's obvious from the chart that even Rhodes plus Purdue's share wouldn't even make the Sacklers the largest single entity. Eye-balling the chart, it looks like they have 6% or 7% of the market. (It's interesting when an author gives you enough information to conclude that their story is fundamentally wrong.) 

In other words, Purdue's marketing was so successful that it created a market several times larger than their actual market share! This is quite surprising to say the least. A result can be surprising while still being true, but it's worth taking a moment to answer some obvious questions. Why couldn't Purdue capitalize more effectively on the market it created? I realize that generics eventually came on the market, and generics tend to be cheaper and thus more popular. Still, it's shocking that they would end up with only about 7% of the market if their marketing was so influential. Do we really suppose that these other companies are all just copy-cats following Purdue's lead? That nobody else would have hit upon the idea of a time-release opioid pill during an era when doctors' attitudes about opioids and pain management were shifting? Does it make any sense to hold Purdue uniquely responsible? Does it make sense to say that, in a but-for sense, many of these overdose deaths wouldn't have happened if not for Purdue's marketing campaign? A lot of implausible claims are leaning on a few sentences lifted from some internal documents from Purdue.

This was an interesting paper and it presents some interesting new facts, but I don't think its basic story is right. I think the most damning point is the flat opioid abuse rates, and this alone ruins their story. But there were some other (admittedly subtle) problems with their analysis and some things that just didn't make any sense. 

Friday, December 27, 2019

Dear Elected Representative

Dear Elected Representative,

Winning a popularity contest does not entitle you to violate our basic human rights. Nor does it imbue you with the expertise necessary to reshape society, nor the wisdom necessary to pick the ideal society from the infinite variety of possibilities.

Your responsibility as a policymaker is very much more circumscribed than what many of your colleagues presume. Your rightful authority extends only to those matters that require society-wide consensus. All of society gets the benefit from pollution mitigation, so it makes some sense for government to help set guidelines and limitations on what kind of pollution we can emit, be it from our personal automobiles or industrial processes that produce the consumer goods we buy. To the extent that crime control is a public good, it makes sense for you to make decision regarding the allocation of public funds to policing and incarceration. To the extent that it is infeasible for us to individually undertake public works, such as building dams and canals and public roads, it makes sense for you to allocate public funds and establish easements across private property and compensate property owners for the confiscation of their property. If judicial rulings are confusing or conflicting, the legislature is responsible for clarifying the law. If some provision needs to be made for national defense, the amount of public spending on that is within your purview, and so is the appropriate allocation to various branches of military, along with decisions about the kind of personnel and technology needed for adequate defense.

This is a fairly narrow range of responsibilities compared to what government currently does. You and your colleagues have gone far beyond your mandate. All of the legitimate functions of government mentioned above, and some that aren't explicitly mentioned, are in the class of "public goods" or "externalities". They entail situations in which the rational choice for a private individual is irrational from the point of view of society as a whole. We might all shirk on chipping in for national defense, even though we all want the benefit. We might like to reserve the option to rob our neighbors, even though we all benefit from general prohibitions on crime and enforcement mechanisms. We might like to reserve the right to be a curmudgeonly hold-out against an annoying confiscations of our property, even though we enjoy the benefits of canals and roads secured by involuntary easements across private property. We might want to pollute more than what is socially optimal, given that we individually get the benefits of polluting while bearing almost none of the costs, which are diluted across society as a whole.

You have a mandate to clarify the law where judicial decisions are confusing or contradictory. You do not have a mandate to create new law out of whole cloth. Consider a labor dispute. Suppose one judge rules that employers may dock their workers pay for perceived losses in productivity, while another judge in the same district rules that such losses are a "cost of doing business" that the employer must simply absorb. Employers and employees alike will be clamoring for some kind of clarity so the law of the land is predictable and reliable for normal people. You are within your rights to settle such conflicting rulings. You are not within your rights to repeal long-standing common law, nor are you permitted to bend society toward your imagined ideal. Legislation should not be used to ban long-standing commercial practices between willing participants. If borrowers patronize lending services where the interest charged is "too high" (in your estimation), the appropriate response for you is to recognize your own failure to understand consumer behavior. It is not appropriate for you to bend consumer behavior until it fits with your preconceived vision. If workers accept wages that are below what you consider sufficient or dignified, that doesn't mean the workers are foolish or the employers are greedy. It means you've failed to understand why someone else's behavior in the marketplace is rational from their point of view. Any mental impressions you hold about the marketplace are a commentary about you, not about the world. The failure of the marketplace to adhere to your idealized vision is a failure of your imagination, not a failure of reality itself.

Human beings have rights. You have a sacred duty to never abridge those rights, even though you wield power and will often have the opportunity to use it beyond your true mandate. At the most basic level, we have bodily integrity. Personal autonomy. Self-ownership. We have sovereignty over our bodies, meaning others cannot initiate violence against us without an overwhelmingly compelling reason. Sometimes people do things with their bodies that other people don't approve of, but they do not require our permission. Left to their own devices, people will eat or drink to excess. Smoke tobacco. Take psychoactive drugs that most of society doesn't approve of. Engage in prostitution, buyers and sellers. Sell their organs. Jump out of airplanes, ski, ride horses, climb dangerous mountains. You are permitted to object. You are permitted to write polemics condemning the behavior as immoral, even to the point of being an unreasonable prude. But you are not permitted to use the power of government to "fix" the problem, to forcibly straighten the crooked timber you see before you. People in your position are often tempted to "do something." They are embarrassed by public health statistics for their region or for the nation as a whole, so they want to reduce the amount of obesity or alcoholism or some other problem for which we compare unfavorably to some other nation. But it is not the proper role of government to eliminate vices, so long as those vices are freely chosen and harming only the persons engaging in them. It is not our duty as citizens to present a pleasing tableau to the world or to our rulers; I have no obligation to conquer my vices and addictions so that the public health statistics look better to observers. If people own their bodies, the state is not permitted to interfere with their decisions to indulge vices. If the state does interfere, it is literally claiming a controlling ownership stake in its citizens bodies. If this talk of rights and self-ownership is too mushy and abstract for you, you should remember that the practical consequences of vice prohibition (particularly drug prohibition) have been terrible: black market violence, massive amounts of unnecessary incarceration, tainted drugs leading to unnecessary overdoses, communicable diseases among intravenous drug users, which often spread beyond that population. It has been truly awful, and it is entirely the fault of social engineers who wished to "do something." They made an existing problem worse, and they barely had any effect on the overall amount of drug use.

Human beings also have the right to freely transact with each other under any mutually agreeable terms, whether money changes hands or not. As long as they are not significantly harming third parties who aren't involved in the transaction, the transacting parties should be allowed to exchange under any terms they agree to. People have an intrinsic right to engage in verbal, commercial, and sexual intercourse without requiring the consent of society at large. Deviations from this principle have historically included prohibitions on interracial and same-sex marriage, penalties against premarital sex, anti-sodomy laws, the jailing of anti-war activists during World War I, and laws preventing free commerce between whites and blacks during the Jim Crow era, to name just a few. It's an ugly history, even if we constrain ourselves to the United States in the 20th century. I think we've mostly crossed that bride together, and we're not going back. Using the machinery of the state to enforce anti-sodomy laws or premarital sex prohibitions seems beyond prudish. If someone actually proposed assembling a police force to harass such "violators", they would seem downright barbaric. (Bear in mind that anti-sodomy task forces used to exist, and police harassment of gay clubs was somewhat routine.) The era of using the state's monopoly on violence to enforce "traditional" sexual norms is over, thank goodness. (At the same time, the police have finally started taking sex crimes, meaning actual crimes with actual victims, seriously. Be clear about the distinction between what is consensual and what is not. Some crafty polemicists try to blur the line or muddy the waters with borderline cases, but most of us are wise to those tricks.)

Commercial transactions and non-commercial forms of human intercourse are not made of different stuff. Nothing special happens when money changes hands. We're free to have interchanges involving speech, be it a private conversation or a rousing speech delivered in a crowded lecture hall. We're also free to set our living arrangements with other adults who agree to the terms. Sometimes this means traditional marriage, sometimes it means a non-traditional romantic relationship, and other times it may simply mean picking a roommate or carrying on a friendship. Almost everyone acknowledges the right to interact with people on such non-commercial terms. It's not controversial that I am free to invite anyone I wish into my home, serve them food, converse with them, invite them to any room in my home, etc. So long as I am not keeping them against their will, and so long as they aren't coming in uninvited, there is essentially no limit to what people can do in their personal lives on their personal property. We should acknowledge that there is nothing inherently corrupting about the exchange of money. It doesn't facilitate exploitation any more than these other forms of non-commercial intercourse do. One can easily imagine one spouse taking advantage of the other, or one roommate shirking and taking advantage of his roommates generosity. Non-commercial relationships can be far more exploitative and far less pleasant than commercial ones, and yet we are free to engage in these transactions with essentially no government regulation at all (until someone actually commits a traditional crime of violence or a crime against property). What you need to keep in mind is that our rights do not cease to exist the moment one of us pays money to the other. You as a policymaker have no more right to regulate commerce than you do to police sexual behavior or platonic friendships or any other aspects of someone's lifestyle. If anyone can prepare and serve a meal, fix a leaky faucet, or have sex for free, they should be able to do these same acts for money, and should be free to do so without interference from you. Even if this talk of "rights" falls on deaf ears, there are practical reasons for the state to not regulate commerce.

Outside commentators may scold one or the other party for being too stingy. They may idly lament that one party gets a raw deal. They may ruminate from their armchair about the social structures or power differentials that lead to injustice in the marketplace. But it must always be held in mind that these parties are transacting willingly. They are often making the same "unfair" transactions repeatedly, implying that 1) it is a beneficial transaction for the "oppressed" party and 2) the "oppressor" party's scheme is apparently not profitable enough to attract competition, which would certainly make the terms more favorable to the oppressed party. The sad reality is often that the "crappy job" is the best possible work given the resources we have available, and any deviation from the existing arrangement would make both worker and employer worse off. The worker's output is probably not worth much more than what they are actually paid for it, and the employer is probably not reaping a huge profit. If you are imagining that the worker creates a large pile of wealth and the employer arbitrarily captures most of it, your model of reality is mistaken. There simply is not much of a surplus to divide up. Attempts to make the transaction more favorable to the worker will quickly reduce the employer's share of the surplus to zero, rendering the transaction pointless from the employer's point of view.

Some low-skilled workers are willing to accept low wages that you may see as undignified. Other workers are willing to work for employers who don't provide health insurance or other benefits. Such workers would rather take their compensation in the form of take-home pay. Others workers might be willing to work dangerous jobs or work long hours without any "overtime" provision. Some consumers are willing to buy health insurance that only covers rare, catastrophic expenses; they will forego coverage of routine medical expenses (routine check-ups, most medications, birth control, Viagra, etc.). They understand that it's not so hard to pay for these things out-of pocket, and they can be financed by the money saved on premiums by purchasing a less "generous" insurance plan. It is not your right to overrule their preferences as workers or as consumers. You might imagine that you can make the terms of these exchanges more favorable to one party (typically whichever party happens to be a more sympathetic interest group to the voting public), but this is economically implausible. You aren't capable of demanding that employers "give" their employees benefits. Those benefits ultimately come out of the workers' paychecks, so there is no net transfer from capitalist to worker. If you successfully compel employers to give their employees $5,000 worth of benefits, then those employees' take-home pay will fall by $5,000, at least in an in-the-long-run-and-on-average sense. Someone who was perfectly happy with $50,000 a year is now (after the passage of a mandate) getting $45,000 plus benefits that cost $5,000. We can presume they are worse off, because they always had the option of spending $5,000 of their own money on those benefits. Mandates that compel employers to provide benefits make employees worse off. They reduce the options available to employees. If your mental model of such mandates is that these are transfers from the capitalist class to the working class, that model is emphatically wrong. Likewise for mandates that health insurance must cover routine medical expenses. Insurance companies are smart, and they price these provisions into their contracts. Their actuaries estimate the expected future cost of those mandatory benefits, and insurance premiums rise by at least that much. People easily fall prey to the fallacy of composition here, thinking that "If government forced my insurer to pay for this procedure, that would make me better off at the expense of the insurer." That might be true of a one-off that was unlikely to ever happen again, but a policy of mandating predictable "transfers" from insurer to insured nets out to no transfer at all.

Some observers imagine the workers and consumers in conflict with large, powerful businesses, and government standing as a bulwark against the corporations. Once again, this is the wrong model. In the cases mentioned in the above paragraph, attempts to intervene yield no benefit. In fact they often cause tremendous harm. And these examples are quite typical of government intervention more generally. Even if you don't acknowledge our right to transact freely (perhaps because you see society as an ant colony that should be optimized for its production or display value), you should bear in mind these very practical reasons not to interfere.

Let me dispel some other illusions. If you imagine that government is "the helm" from which society takes its marching orders, your model of reality is wrong. If the legislature and executive functions of government were to cease completely, commerce would continue. Food and goods would find their way to communities, production would continue. Obviously we are not a centrally planned economy. Millions of individuals acting independent of any mandate from the government, following price signals in a free market, are what make the world go round. In no meaningful sense is the government in charge of this process. It is no in any sense "running" society.

Another illusion is the "mandate from the will of the people." Anyone who has won an election is likely to claim such a mandate. People like to wax poetic about "democracy", the rule by majority. Politicians often use it as an excuse to do all manner of horrible things. Let me simply point out that nobody takes this argument seriously. When the law is wrong but popular, reformers rightly argue that the law is wrong! They do not usually say that imposing the wrong set of laws is the right thing to do until we change enough minds. Popular opinion can be wrong, and everybody knows it. Moreover, when these same people get their way via undemocratic means, they often cheer. Anti-sodomy laws were ended in 2003 and gay marriage was instituted in 2015, both by the action of the Supreme Court. Brown vs. the Board of Education was a court decision, not a popular referendum. Almost nobody who favors these changes laments that the majority didn't get its way. Democracy is great as a check against tyranny, and it is to some degree necessary for those decisions that are inherently collective, like pollution control or national defense. But it was never meant to be a cudgel by which the majority can oppress unpopular minority groups. It isn't necessary to subject private decisions, such as the contents of someone's labor or insurance contract or the conduct in their bedroom, to a popular referendum.

Let's dispose of this "will of the people" nonsense. If you are like most candidates, you almost certainly do not acquire the consent of your constituents for all of your legislative actions. You probably won an election by being the most plain-vanilla, unobjectionable, milquetoast candidate on the ballot. Perhaps you had name recognition on your side, a powerful weapon when your job is to collect votes from mostly inattentive, apathetic, ignorant voters. Most of your constituents couldn't name you, and the ones who could wouldn't be able to describe your voting record in any detail. Likely you were the incumbent, as most winners of elections are. Perhaps you were the "R" candidate in a predominantly "R" voting district. Perhaps you ran unopposed. Do you spell out your policy positions in glorious detail, such that your constituents can actually evaluate you as a policymaker? Or do you keep them close to the vest so you have enough flexibility to switch when the political winds blow a different direction? (My state representative once gave me a cravenly non-committal answer to a question about her position on marijuana legalization and failed to clarify after a follow up question. I presume this is a routine practice. Clearly many politicians were staking out "I might change my mind" positions on gay marriage and marijuana legalization in recent years rather than committing to a position.) You did not acquire a mandate from the voters, certainly not one that permits you to trample our rights. At best, you won a popularity contest. And that's assuming the election was meaningfully contested.

You are constrained by the various constitutions that enumerate and restrict the powers of government. This includes the Constitution of the United States of America and whatever state or local constitution applies to your office. Clever legal theories invented by politicians or government lawyers have sometimes won over pliant judges and allowed governments to act well beyond their mandate. This is wrong. The plain language of the constitution, as it would be understood by us ignorant plebs not of the legal tribe, is the law of the land, even if the government refuses to follow it. To have a rule of law that requires interpretation by scholars who specialize in constitutional arguments is to have no rule of law at all. A true rule of law only exists if we ignorant commoners can understand what the law actually is. We have to be able to know ahead of time what is or isn't allowed. While the plain language of the Constitution is quite legible, the obscure, tortuous legal arguments that currently govern us are often opaque. The federal government often plays "six degrees of interstate commerce" to grant itself new powers, powers clearly not enumerated in the constitution. The various prohibitions on government action spelled out in the Bill of Rights are also binding. In case it wasn't clear enough that the Constitution only grants government the explicitly enumerated powers, the founders reiterated a list of no-nos. "Congress shall make no law..." is pretty damn clear. You will often have the opportunity to flout the Constitution, and you will likely get away with it. You are still wrong for doing so. If you find that the Constitution is too constraining, there is a process for amending it. Please follow it. If you discover that the process is unwieldy and that you can't muster the popular or political support to pass your amendment, that means the constitution is functioning as it should, as a check against government overreach. That is a feature, not a bug.

To be clear, this is not an anarchist manifesto. Nothing said above rules out a minimalist government. I did explicitly discuss the legitimate roles of government above. There are also topics I haven't touched on here, such as the welfare state and appropriate methods of taxation. As far as I'm concerned, a minimal welfare state for the truly needy can be a legitimate function of government, assuming that the problem isn't made worse (e.g. high implicit marginal tax rates leading to low employment for recipients) and assuming that private charity is not up to the task. Government obviously requires some form of taxation to fund its projects. What is not appropriate is for government to restructure society to conform to someone's vision. These rationales for government action should not be used as a wedge to open the door for outright social engineering. There is a big difference between, on the one hand, using taxation to fund a minimal welfare state and on the other hand using a taxation-and-transfer scheme to flatten society and make the "income distribution" more aesthetically pleasing. I'll allow that the former is a rational function of government, but the latter is not.

I put this all down now, because I see your colleagues at all levels of government acting well beyond their mandate on a daily basis. They sometimes need an explicit reminder of what their actual job is, because they've apparently all forgotten. Here is that reminder.

Thursday, December 26, 2019

FDA Approval Process For Government Programs

I’m trying to imagine what it would be like if government policy had to pass and FDA-like bureaucracy, being tested for “efficacy and safety.” Please demonstrate to the satisfaction of the panel that a tweak to the minimum wage (or *any* minimum wage, or any restriction at all on labor contracts for that matter), or a new regulation or subsidy on health care, or a change to entitlement programs, will be “effective” in the sense of achieving its intended benefits and “safe” in the sense of having acceptable costs and side-effects. This would be a much smaller government, one that is much more constrained to doing *only* those things that are legitimate functions of government.  Some of these policies are *literally* medicine. Health care policy is explicitly aimed at improving health measures in some way. In this upside-down world, this kind of medicine is shoved down your throat against your will *without* any kind of vetting procedure, while you are forbidden from taking medicine (pharmaceuticals and other chemical substances) that you actually want to take unless it passes muster with the FDA. This is backwards. The standard for medicine that we’re compelled to take against our will should be higher, not lower.

In fact, forget for a moment about the actual testing for safety and efficacy. Suppose someone proposing one of these social engineering experiments (a.k.a. “legislation”) had to get his proposal past and ethics panel, just as researchers conducting experiments on animals and humans need to do. “Suspected or likely side effects of your proposal include involuntary job loss, reduction of total productivity…benefits are highly speculative. I’m sorry, it would simply be unconscionable to test this ‘medicine’ on unwilling subjects. Denied.” I could understand someone arguing that we just gotta have pollution control, or we just gotta have military defense. Having no policy at all regarding these problems may be just as unethical as having some slightly misguided policy, so there's no moral trump card with which to simply halt all government activity. But most of what the government does is some kind of sinister social engineering, attempting to steer us toward someone's vision of a great society and away from the paths we would choose for ourselves. There are entire categories of government meddling and regulation that don't need to exist at all, most of which almost certainly wouldn't pass the hypothetical ethics panel. 

When Does Pharmacology "Cause" Behavior?

When is it meaningful to speak of the pharmacology of a drug causing a person's behavior? I think that this concept has been overdone, particularly with respect to the "opioid epidemic" narrative.

Start with the easy cases, where we can definitely attribute the response to a drug to its pharmacology. Some drugs clearly reduce a person's capacity, mental and physical. Alcohol slows your reaction time and can even make you fall asleep at the wheel. Opioids can likewise make someone drowsy or unresponsive. Obviously they are used as general anesthetics for surgery. A dose much lower than that used in surgery can have a milder but still noticeable effect. Someone in such a state of diminished capacity might be prone to accidents, but I don't want to call this a "behavior." A drunk driver and a sober driver are both trying to get to their destination safely, it's just that the drunk is much worse at it. The ability of certain drugs to incapacitate their users is "pure pharmacology". That's not what I want to discuss here. I'm trying to get to behaviors, deliberate actions, that are in some sense caused by drugs.

Sometimes people who drink alcohol are more likely to get into fights or behave violently toward family members. The same is true of people who use certain stimulants. Someone in a very excitable state may be on a hair-trigger and might throw a punch in a situation that their sober self would handle more calmly. I think it makes sense to label this kind of thing "pharmacologically induced violence." It's not mere diminished capacity as described in the above paragraph. It's not that the drunk was reaching for his beer, stumbled, and plowed his knuckles into someone's cheek. Starting a fight is different from stumbling on the sidewalk or crashing your car. It's a behavior. There is an underlying intent. It's a choice, and the drunk is simply worse at making choices than his sober self.

There are certain drugs that can drive people into extremely excitable states. Synthetic cannabis (sometimes called Spice or K2) can sometimes make users extremely irrational. A paramedic friend of mine once described to me an episode in which he had to handle such a person, who was clearly out of his mind. This kid smoked too much Spice and was kicking and flailing at the first responders who were trying to help him. Some psychoactive substances, which are sold in head shops as "bath salts",  have a similar reputation for driving people insane. PCP has a mostly unearned reputation for causing similar outbursts; Jacob Sullum corrects the record on PCP in his book Saying Yes. Some of these stories are overblown or apocryphal. The man who famously ate someone's face off was not high on bath salts, even though it was initially reported that he was. (Why is it that the initial report is always 100 times louder than the retraction?) People occasionally do shocking, crazy things, and for whatever reason "He must have been on drugs" always seems like a plausible explanation. Sometimes first responders or medical personnel make incautious statements to the media before actually double-checking the likely causes of someone's outburst, and media outlets dutifully repeat the message to a receptive public. The public loves a good drug-panic story, whether it happens to be true or not. Exaggerated or not, there are certainly cases where drug-induced psychosis is the best explanation for someone's shocking behavior. It makes sense to talk about the pharmacology of the drugs causing someone to have an irrational outburst (shouting obscenities are random passers-by, starting a fight, etc.). The drug can legitimately be considered the cause of the behavior in a but-for sense.

What about opioid addiction and overdose? More to the point, what about legitimate medical prescriptions leading to an opioid addiction? Here I think it's an incorrect framing to posit the drugs as the cause of the behavior. It is true that if someone is on opioids for an extended period of time, they will probably develop a physical dependence on opioids. They will have developed some degree of tolerance, and they will experience withdrawal symptoms when they stop taking them. The physical symptoms can be quite unpleasant. But physical dependence is quite distinct from addiction. Physical dependence is not a behavior. It will not by itself cause a person to seek opioids on the black market. Someone who takes a long-term course of opioids and is then tapered off by his doctor will not be driven to addiction if he follows the doctor's directions. Addiction is what happens when someone deliberately and repeatedly deviates from "use as directed." It makes little sense to speak of prescriptions "turning patients into addicts". Even if we posit that some people get a mildly pleasant buzz from taking their pills regularly, it's just not the case that this pleasure causes them to escalate their dose or to continue to seek opioids after their prescription runs out. There is nothing inherent to the pharmacology of the opioids that causes these behaviors. Indeed, the vast majority of opioids users don't have any kind of problem going off them when their course ends. Some sloppy writers and commentators have tried to cast this as a cosmic game of dice or Russian Roulette in which patients are subjected to a random risk, as if every ten thousandth pill contained the "addiction virus." It's much more useful to think of addiction as a series of deliberate behaviors that eventually become hard to control.

(I once heard Doctor Jeff Singer, someone who I correspond with occasionally, describe a patient's hesitation about opioids. The patient's concern went something like, "I want to be careful about this, because last time I was prescribed opioids I got addicted and experienced nasty withdrawal symptoms." Jeff explained that the patient was emphatically not an addict. He had experienced physical dependence, but not addiction. In fact the patient was quite deliberately avoiding these substances because he'd had such an adverse reaction to the withdrawal. Addiction, by contrast, is the compulsive, continued use of a substance despite the harm its causing you. I thought this was a nice illustration of the difference between addiction and physical dependence.)

Let's grant that physical dependence can be so extreme and the withdrawal symptoms so severe in some opioid patients that continued use is irresistible. Such a person continues to use after their prescription is gone, purchasing pills or even heroin on the black market. It's a stretch to say that the pharmacology of opioids led to their addiction. As difficult as withdrawal is, continued use is a very deliberately chosen path. But forget that for a moment. Let's grant for the sake of argument that this is an instance of pharmacologically induced behavior. It might make sense to blame their use of black market pills or even  heroin on the body's physiological response to an opioid prescription. Certainly the continued use of opioids leads to nasty withdrawal symptoms, and opioids purchased on the black market relieve those symptoms. But take a closer look at what's actually killing these people. "Drug overdose" is a misnomer, a short-hand term that misstates the true cause of death. The CDC actually refers to these deaths as "drug poisonings." The term "overdose" seems to imply that the decedent took too large a dose of a single substance, but that's the exception rather than the rule. Most drug-related deaths are multi-drug poisonings. See where I've written about this previously, here and here and here for example. Let's dispel this notion of a hapless addict popping pill after pill until he just keels over. This just doesn't match the reality. It isn't the case that it's just "so good" that the addict needs another and another until he's taken too much. A much more likely scenario is that someone is taking opioids with benzodiazepines (~1/3 of prescription opioid related deaths involve this combination) or alcohol (~14%) or cocaine (~12%). Maybe some of the benzo plus opioid poisonings are accidental mixtures of medications, but it appears that many opioid users take benzos because it potentiates the high. Presumably many or most of them know it's dangerous but they do it anyway. Even granting the premise that the pharmacology of prescription opioids causes continued opioid use, it's far from clear that it should cause the patient to mix medications or start using cocaine. It just makes a lot more sense to model this behavior as deliberate thrill-seeking than as "I lost a game of pharmacological roulette and it turned me into an addict."

I think it's worth taking a serious look at the risks of prescribing opioids for pain, acute or chronic. Some people are fast or slow metabolizers, depending on whether they have zero, one, or two copies of a certain gene. Some of these people really are effectively playing a game of pharmacological roulette, as in they might have a fatal reaction to a dose that's benign for an average person. And we certainly want to warn people about withdrawal symptoms so they can make an informed decision. Even granting all that, any talk about the "risk of addiction" is badly missing the mark.

By the way, if there is a "risk of addiction" that roughly corresponds to the ratio of addicts to opioid prescriptions, apparently that risk has been coming down on a per-prescription or per-patient basis. Rates of opioid misuse and addiction have been basically flat or even falling over a time period when prescriptions were skyrocketing. This makes little sense if you think that there is some disembodied probability, the "risk of addiction", faced by all opioid patients. But it makes perfect sense if we think there's a more or less fixed population of extreme risk-takers, and they happen to be using whatever drugs are most available. In this light, I find it just atrocious that state governments are going after Purdue and other opioid manufacturers. We need to purge this notion that addiction is something that is done to you, that passively happens to you as you dutifully take your meds. People don't simply sleep-walk into addiction when taking their medication as directed, and there is nothing inherent in the drugs pharmacology that makes them deviate from "use as directed." We need to debunk this absurd idea that smooth, sophisticated pharmaceutical firms just had their way with us, even getting past the doctors who stood as gate-keepers.

I hope it's clear that this isn't mere hair splitting about what to name something. There are some very important implications, political, moral, and legal. If the pharmaceutical companies aren't culpable (as I think they are not), then they shouldn't be sued for selling opioids. They should be able to continue selling to their very willing buyers, some of whom commit suicide when they are involuntarily tapered. If the addicts are deliberately engaging in risky behaviors, we should not be restricting opioid prescriptions for their sake. Restrictions make it harder for desperate pain patients to acquire opioids. It seems to me that the moral weight of a single pain sufferer who is involuntarily cut off from opioids ought to outweigh the voluntarily undertaken self-harm of many dozens or even hundreds of addicts. Tens of millions of Americans suffer from chronic pain at any given moment (I've seen estimates as high as 1 in 3 that will eventually suffer from it), compared to less than one percent of the population being addicted to pain pills (having a "use disorder within the past year" anyway). It would be hard to make this calculus work out in favor of restricting access. Even supposing some utilitarian calculation gave the nod to further restricting opioid prescriptions, most of us are not amoral utilitarians. People rightly apply a fairness standard to public policy. "You can't have opioids, because, while I'm quite sure you won't have any problems, there is some small chance that Jonny over there will steal them from you and abuse them." This seems terribly unfair. It makes sense to worry about who is harmed or helped and to give consideration to whether that harm is self-imposed or not.

If this narrative of pharmaceutical companies involuntarily converting normal patients into addicts is wrong (and I insist it is), then basically all the policy responses and "public service announcements" by our institutions of public health have been deeply misguided. Sure, some individuals have impulse control problems. But we're not zombies. We're not programmable robots who can simply be re-programmed to serve big corporations' earnings goals. Habit-forming drugs don't sink "chemical hooks" into our brains and control us against our will. There is an element of free choice in all of this that is quite separate from the drugs' pharmacological effects.

___________________

Something I haven't touched on in this post is the fact that most opioid addicts don't even get their start with a prescription, an observation that should relieve opioid manufacturers of any culpability. The idea that I'm arguing against in this post, the notion that drugs take over people's minds and make them do things they don't want to do, is sometimes called "voodoo pharmacology." There is apparently some powerful voodoo going on here. Apparently the pharmaceutical companies are hijacking the brains not of their legitimate patients, but of other people in society who acquire opioids illicitly. Manufacturers, distributors, and patients are by and large acting responsibly, but highly motivated addicts are stealing left-over pills from unfinished prescriptions (a peculiar phenomenon, by the way, for a drug that is supposedly so irresistible). It might make some sense to hold the patient culpable for inadequately securing their leftover pills, but to follow this back up to the manufacturers? This is an absurd concept of legal liability, one  which I don't believe anyone has even tried to defend.

Wednesday, December 18, 2019

Why No New Cities? Why Not Move Tech Hubs?

There are two books I've read fairly recently that complement each other well: Hive Mind: How Your Nation's IQ Matters So Much More Than Your Own by Garett Jones and Shut Out: How a Housing Shortage Caused the Great Recession and Crippled Our Economy by Kevin Erdmann. I think some enterprising academic could do a useful synthesis of the concepts contained in these books.

Hive Mind is, roughly speaking, about how IQ affects productivity, and in particular how this is more important for collectives than for individuals. Your nation's IQ is a stronger predictor of national GDP than your individual IQ is of your particular earnings/productivity. There is a discussion of the "O-ring" theory of economic development, an idea for which one of the 2019 Nobel recipients, Michael Kremer, was famous. The idea here is that you need a lot of really smart people working on a particular project, because any weak link is a possible point of failure. The "O-ring" part is a reference to the Challenger space shuttle explosion, in which a single point of failure, literally an o-ring, caused a massive engineering project to fail and killed several people in the process. New technologies are susceptible to many points of failure, so you need a lot of very highly skilled people working on them. You can't just hire two really good software engineers and then a bunch of B-level engineers to do the rest of the work. Nor can you hire ten mediocre software engineers and simply merge their output to be the same as that of one good engineer. If you're building something new and innovative, you need a bunch of top 1% people on your team. Skimping on talent introduces more points of failure, and these accumulate quickly. The chance of failure grows exponentially with the number of points of potential failure, so reducing the probability that any one point will fail is enormously valuable. This theory probably explains why all the really smart programmers go to Silicon Valley to work with other really smart programmers. It explains why some big cities are world hubs for particular industries rather than having the work be dispersed throughout the nation or the world. It could also explain why there might be a large quality gap between the very best and the second best product on the market. If all the top-tier people are working for the industry leader, the second-tier companies are trying to cobble together the same product out of second-tier talent. They produce something of value, but perhaps it's a distant second. (I'm thinking iPod vs. Zune, or the iPhone vs. its early competitors. Possibly Facebook vs. other social media or Google vs. other search engines, but these could be network effects rather than examples of quality differences.) I can't find it now, but I remember reading about the candidate cities for Amazon's upcoming new campus a few years ago. Chicago was briefly a candidate, but someone commented that the limiting factor was talent. Chicago, it seems, has a lot of "top 10%" people but not quite so many "top 1%" people. My midwestern blood initially boiled at the elitist comment, but I calmed down and recognized the point. Couldn't they have just hired more people at the 90th and 95th percentile and used whatever top 1% talent they could find? Apparently not. They were looking for very high concentrations of extremely skilled workers. Whoever was making that decision was engaging in some kind of "o-ring technology" thinking.

Shut Out is about dysfunctional housing markets in major population centers. Supply restrictions (mostly out-of-control zoning laws) in cities like San Francisco, Los Angeles, New York, and Boston lead to extremely high housing prices. Note that these cities are major centers of industry: tech in San Francisco, finance in New York, biotech in Boston. Erdmann's story is that, contra the dominant narrative, there was never a "housing bubble." Rather there was never enough housing in these major population centers. The price of housing rose so high in these "closed access" cities that they caused demand to increase in nearby regions. These nearby regions weren't as restrictive as closed access cities, but nevertheless failed to keep up with skyrocketing demand. Erdmann dubs these "contagion cities." In other words, supply restrictions in major hubs for highly skilled workers were so bad they caused knock-on effects in the rest of the nation. The standard narrative is that uncontrolled speculation drove up the price of housing until it eventually got completely unmoored from the fundamentals, until housing prices eventually crashed. That's kind of hard to swallow when you look at the growth trajectory of total housing. The "supply restrictions" story is a better explanation of the skyrocketing prices than the "artificially inflated demand" story.

What's the connection between these two books? Shut Out had me wondering, why doesn't someone just start a new city? Or start a tech hub in an open access city, where housing can grow without obnoxious restrictions? I think one would have to lean very strongly on a Hive Mind type of argument to explain this lack of new cities or new industry hubs. It kind of makes sense that a data scientist in Silicon Valley would not simply quit his job and go work for a company in Omaha, Nebraska. This  person's skills might make him useful to just one employer in a city that isn't a tech- or finance-hub, but might present many more options in Silicon Valley. If the job in Omaha doesn't work out, the cost of job-switching is much higher. Maybe this isn't such a big deal to any one person. There may be many such people who are willing to move to a slightly more boring city with fewer colleagues of a similar skill level. But it definitely creates a friction. The city with the fewest frictions will likely attract a hugely disproportionate fraction of the top data scientists, because people aren't just evaluating cities in isolation. There are network effects. The city that first starts to attract lots of talented data scientists will tend to attract more talented data scientists. That first city is a focal point, and it's hard to simply designate some other low-cost city as a new focal point. Something has gone terribly wrong here in the sense that the major centers of productivity also have the worst zoning laws and nimby-ism. Parasitic local governments probably understand that their city's status as a hub is unlikely to change, so they don't have to worry too much about providing good government. They can indulge the locals' nimby-ism and economic populism to the hilt without much changing their tax base. Some kind of concerted effort, given enough political will, could conceivably fix their local housing market. But why bother?
___________________________________

Scott Alexander recently wrote a post about billionaire philanthropy. In a post that is generally positive on philanthropy, he writes disparagingly that "...the George Kaiser Family Foundation has decided the best use of its $4.4 billion is to pay techies $10,000 each to move to Tulsa, Oklahoma." If they actually pulled this off and managed to create a second tech hub in the midwest, the benefits to society could be in the many billions of dollars. If it takes a few billion dollars to incentivize the initial movement to a new hub, that could be well worth it. Admittedly it's hard to understand the value of the George Kaiser Family Foundation's initiative in isolation, but it makes a lot more sense in light of Shut Out and Hive Mind.

Monday, December 9, 2019

My Problem Employee

I wrote this all down over a year ago, and it's been sitting in my list of drafts since then. The situation has resolved itself, and this is no longer my problem. Re-reading it now, parts of it seem whiny and I considered just trashing it. But I thought someone else who has had to manage an incompetent employee might commiserate. This was a deep source of frustration in my life for a year and a half, and I was looking for something written by someone dealing with the same challenge. I'll share in case someone in a similar position finds this useful.

I also tie this to Bryan Caplan's book The Case Against Education. In order to explain why education has such strong signaling value, he has to be able to explain why employers don't just hire indiscriminately and fire the poor performers. I wrote about that here. I have a much finer appreciation for Caplan's discussion of firing aversion. I also can appreciate why "employer learning" is so slow; this is the process by which employers figure out their employees' productivity. HR departments sometimes actively censor or purge this information.

Anyway, I'll start with the gossipy details about my under-performing employee.
____________________________________

I had a problem employee who was not competent at his job. I’d given him two major projects that he was supposed to work on semi-autonomously. The pace of the work was incredibly slow, and it often came back done incorrectly. It wasn't to my specifications, so I'd say so and ask for it again, sometimes pointing to the documentation where I specified what  I wanted, such as a prior e-mail or project plan. I could have easily done both projects and had them done sooner and to my specifications when he wasn't even halfway done. I often had to reached into his project with a heavy hand, going so far as to give him working code that finished his job. He often declined to use what I’d given him, having failed to understand the code, and went back to tinkering with whatever half-completed code he was working on.

A huge part of the problem was lack of communication on his part. I would discuss his projects with him in person, but my critiques and suggestions for different directions did not make it into his work product. I confirmed fairly late in the game that my comments were utterly failing to register with him. He basically came out and admitted this was true. He just would not open up and say “I’m confused about what you want. Explain it to me again.” Or “I don’t understand how this discussion affects my projects. Can you clarify?” We have regular group meetings in which I will discuss topics and code strategies directly relevant to his work. I would point out the relevance to his project, then follow up in an e-mail pointing out the relevance and even pointing him to the relevant lines of already-well-commented-code. Despite basically handing him the answer on a silver platter, this stuff still would not make it into his work product, and his work would come back wrong (also late).

All this came to a head when he “confronted” me about my management style. I had been sending him almost-daily e-mails asking him to correct his work. I would ask why stuff had been missed. Was he just being forgetful? Was he deciding to do something differently from what I specified? I needed to know so I could fix the problem. He was declining to answer such questions, because he obviously didn’t care for them. (Another related long-standing pattern: he declines to volunteer any information, even relevant information, like “I’m confused and need more direction.” He declined to answer a pretty direct question in front of my boss in a meeting until pressed multiple times. It was embarrassing. I pointed it out to my boss after the meeting and said this was a pattern.)

He tried to blame the issues on me, which I thought crossed a line. He said that the sheer volume of "details" being thrown at him was too much for one man. (It was not; there was so much spoon-feeding and repetition that it should have been easy for him.) It was apparently unrealistic of me to expect that all of those details would be remembered or make their way into his work product. One of my recommendations that had set him off was to write down better notes and organize them. He had a habit of not taking any notes at all. I think his philosophy must have been "If it's important it'll come up again" or "If it's important, I'll remember it." If his problem is his own poor memory, better note-taking is the obvious solution. I have a pretty damned good memory, but I still miss things. I take notes when my boss or my boss's boss are commenting on my work product. So does every other productive employee. It's basic office-work hygiene. He’d balked at this and other suggestions, suggesting that my style was too controlling and saying that no other manager had ever exerted this level of control over his work. According to him, he’d never had bosses send back so much of his work for corrections. (I now have private knowledge that this isn’t true, having discussed him with a previous manager who'd had the same frustrations.) He even told me a cute story from when he had, in the past, managed a fast food place. When he tried to crack down on his staff and enforce certain standards, many of them promptly quit on him. I tried very careful just to let him talk without interrupting. Why stop a man when he’s offering to hang himself like that? But sarcastic comments presented themselves. (“This isn’t a Burger King.” Or “I really liked the end of your story, where the bad employees all quit. I wonder if that solved their problem.” Of course I did not say any of these things.) The message was loud and clear. His poor performance was not his fault, it was mine. It was a result of my being too demanding and attentive to detail. He also said repeatedly that he’d never had performance issues with other managers, which I knew wasn’t true. I wasn’t about to tell him that I knew what previous managers had said about his work, but it was useful to know how poorly his self-assessment was calibrated to reality.

After letting him go on for a while, I said that letting up on attention to detail was not an option for me, and that, yes, he would have to fix his problem of dropping instructions. Taking notes would help. Once again, he balked at this suggestion. (Gee, if only there were some way to augment my fallible human memory. Alas, there isn’t!) But more fundamentally he didn’t understand the technical and theoretical details of the tasks he’s doing. So instructions, even when they are pretty clear, failed to register with him. What he perceived as “a bunch of details” were actually crucial things that can make his work product unusable if not implemented correctly. Missing them revealed a basic lack of understanding. He was un-self-critical, apparently incapable of doing a mea culpa. He couldn’t simply admit error and try to fix his problems.

I did some questioning after this meeting, and it turns out the problem employee had persistent performance issues with pretty much every manager he'd worked for. He’d been passed around at this company for something like ten years, from one manager to the next. I don’t know why my boss accepted him on the team, or why it was up to me to say this guy is incompetent and needs to go. Maybe it’s because this is the first time he’s worked for someone who is more technically competent and “down in the weeds” than he is. I can reach into his work, spot his math or coding errors, point out conceptual errors, get his work started for him, etc. I’m about the least confrontational guy around, but I finally had enough. It was taking so much time to correct his work and document his poor performance. Plus it was hugely stressful. I was having trouble sleeping at night, sometimes for days, dreading having to have another unpleasant conversation.

All this had me thinking about Bryan Caplan’s review of the “firing aversion” literature. In The Case Against Education, Caplan has a long discussion of why college degrees affect your pay for such a long time. Couldn't employers simply hire indiscriminately (less discriminantly, anyway), then keep the good performers and fire the bad ones? Couldn't they spot employees whose credentials overstate their productivity and demote or fire them? And shouldn't this make credentialing pretty useless? It turns out there is a literature on this topic, called "employer learning." It actually takes quite a long time for employers to figure out how productive their employees really are. Like, years or decades, not weeks or months. People can "fake it" for a while, or put in extra effort and look more productive temporarily when the boss is getting suspicious. Besides, even when there's no doubt about someone's poor productivity (as is the case with my problem employee), managers are often reluctant to fire out of sympathy. (In my particular case, my sympathy had run out long before the problem was resolved.)

Even when managers are willing to fire, Human Resource departments are often hard blocks on doing the obvious thing. "Do you have documentation? Oh, you do! But, sorry, you didn't document it through our HR system, so it doesn't count. Let's put the problem employee on a 'performance improvement plan.'" From HR's perspective, a termination is a potential lawsuit. They want to do everything they can to make the problem employees quit of their own volition. They also want to make sure it's not just a problem of an incompetent manager making a bad call or failing to manage the problem on their own turf. This is very frustrating to deal with. Putting someone on a performance improvement plan is a pretty direct way of saying, "You aren't qualified for your job." You have to do this, then face that person day in and day out, and this humiliating HR ritual is hanging over both of your heads every time you converse with your employee. In his book, Caplan says something about how even slackers can temporarily look busy and conscientious when under temporary scrutiny. ("Look busy! The boss is watching!") Problematic employees "graduate" from the performance improvement plan, and from HR's perspective the problem is solved and the slate is clean. In reality, the underlying problem is still lurking under the surface and likely to reemerge. And managers are reluctant to go through this emotionally draining process.

Following HR's process is especially tricky for a job like mine. We aren't making widgets here. Employee competence are things like "Does the employee understand the statistical concepts necessary to do the work?" Or "Was the statistical analysis done to the outlined specifications?" Sometimes mistakes creep in to statistics work that are hard to define ahead of time as a "not-do." You're just supposed to know that something looks odd, using your judgement, and locate and fix the problem. It would be easy if I could simply say, "Produce X widgets per week or you're not qualified." In an environment where productivity is hard (impossible really) to measure and even harder to define ahead of time, it's very difficult to implement HR's performance improvement plan. And I don't relish the idea of constantly being questioned, "How am I doing? Do I make the cut?"

_________________________

Some people seem to believe businesses have absolute mastery over their employees and engage in cut-throat employment practices. If this were the case, there wouldn't be so much hesitation over canning poor performers. Caplan's story holds up pretty well. Businesses are often paralyzed by the fear of wrongful termination suits or poor morale that result from performance-based firings. Firings and layoffs still happen, it's just that companies are sometimes too hesitant to do the obvious thing. A combination of human psychology (bad performance evaluations are very unpleasant on both sides of the conversation) and bad HR policy creates a lot of friction. People really do get locked into roles they aren't qualified for, sometimes for a decade or more. 

“Do you think *horrible topic* is funny?!?”

Normal Person: (Tells joke about horrible topic X.)
Obnoxious Scold: Wait! Do you think X is funny?!?
Normal Person: What?! No, of course not. X isn’t funny. X is deadly serious...
Obnoxious Scold: Okay. I guess that’s good.
Normal Person: …which is why jokes about X are hilarious!
Obnoxious Scold: Alright, now. I see what you did. Not cool.
Normal Person: It's not like everyone thinks X is a wonderful and happy subject. That's not why jokes about X are funny. The ensuing laughter doesn’t stem from our enjoyment of X actually happening. Jokes about X work because we all share a mutual understanding that X is deadly serious. The joke violates sacred norms. It reminds us of temptations that we resist or thoughts that we banish from our minds as taboo. So we laugh when Emo Philips pushes a man off a bridge over a minor disagreement over religious doctrine. Or when a parent jokes about eating their child for dinner or flushing him down the toilet. Or when a song from a saucy puppet show reminds us that "Everyone's a little bit racist." The joke is a surprising reminder that the veneer of civilization is thin, that something horrible and taboo is lurking just under the surface, that we’re not totally innocent. The different reactions to jokes about X and X itself are like the difference between being prodded in our sensitive parts by a professional torturer and being tickled in the ribs by a trusted love one. It's the difference between being razzed by your best friend and being insulted in public by a rival about your very real shortcomings. It would be awful if I actually committed infanticide and then cannibalism, or if I actually cuckolded or sexually dominated my best friend. Far worse than if my joke about X had been a serious suggestion. Tickling and teasing are harmless for exactly the same reason that crude jokes are harmless. There is enough trust between the tickler and the tickled, the teaser and the teased, the joke teller and the listener, to know that it's not serious. Actually, you fucking well know all this already, so why am I bothering? I have heard you make "lighthearted" comments about wanting to murder someone. You can drop the whole "outrage as a performance piece" act, because I literally saw you joking about physically assaulting or murdering your political opponents. 
Obnoxious Scold: That really was just a joke. Everyone understands that people are joking about wanting to murder or assault annoying people. On the thing you were joking about, some people actually think that's okay.
Normal Person: Okay, but then I saw you cheer and celebrate on social media when someone actually died. On multiple occasions. Some people are actively making the argument that it's okay to use violence against their political opponents, and others are actively engaging in such violence. But, yeah, let's pretend my thing was "dangerous". Let's say I'm the "offensive" one here.
_______________________
Sorry if I'm literally straw-manning the Obnoxious Scold here. Obviously I'm giving Normal Person a lot more space to state his ideas. I couldn't think of a better way to structure this post. I am getting increasingly annoyed with people who understand how humor works but pretend not to. It is positively perverse that the scolds are going after professional comedians and their stand-up material. Comedians are supposed to point out uncomfortable truths (or "truths", because comedians say a lot of shit that's funny but not really true). Their job is to prod at our sacred taboos. We laugh because we are acknowledging that the thought is taboo and that the joke is inappropriate in polite society. Someone who says "There's nothing funny about X!" is failing to understand how humor works. Most humor isn't about something pleasant and wonderful happening.

Wednesday, November 27, 2019

Cannabis Legalization Doesn't Drive Up Violence

Many states have begun to legalize recreational marijuana. Some people, most notably Alex Berenson, have claimed that this is leading to deteriorating trends in certain vital statistics post legalization. He points to the apparent increases in violent crime in Colorado, Oregon, and Washington. I first became aware of this earlier this year, when Berenson published a piece in the New York Times warning about the supposed dangers of marijuana. Malcolm Gladwell and Tyler Cowen amplified the signal, in my opinion far too credulously. It's come to my attention again. Berenson's recent debate with Jacob Sullum (my write-up of that discussion here) led me to do some digging. To get a good idea of Berenson's position, it's worth listening to this Joe Rogan podcast, in which Joe hosts a sort of discussion/debate between Berenson and Dr. Michael Hart. It's a good source to understand the nature of Berenson's claim. His take is that marijuana use causes psychosis in some people; I think he admits that only a small fraction of people are really at risk and most normal people don't  have anything to worry about. Nevertheless, in his telling, increasing the incidence of cannabis use will cause some small fraction of people to turn psychotic. Psychosis is an enormous risk factor in the commission of violence; he mentions a 20x figure in the Joe Rogan podcast, as in people with psychosis are 20 times as likely to commit violent crime. So more cannabis -> more psychosis -> more violent crime.

So how does one decide whether cannabis legalization increases violent crime? The naive first approach would be to look at rates of violent crime before and after legalization. If violence increases post legalization, that's at least suggestive. But of course that doesn't tell you anything about causation. Violence crime rates increase and decrease for lots of reasons. If violent crime is increasing in states that legalize, we need a control group to compare them to. We need to establish the counterfactual, as in what would have happened if not for cannabis legalization. Also, as with anything we measure, there is random noise in these statistics. That's not to say we don't know how many murders actually happened in Colorado in a given year. Even supposing we count those accurately and to a body (and I think the counts are quite precise), the process that generates those body counts is random. Many of these violent crimes are crimes of passion. Someone gets angry in the moment and assaults or murders someone; the same person on a slightly better day wouldn't have become a murderer. Someone fires a gun with the intent to kill, but misses their victim. A murderer experiences a delay in their plans when their victim does something unexpected, giving the would-be murderer time to rethink things. So the actual realized crime rate, literally the count of crimes divided by the population, is a somewhat noisy measure of the thing we're really interested in, the underlying propensity of the population to commit violent crime. So we need error bars. We need confidence intervals to tell us if an X% increase in crime rates is something that could come about by random chance or something that indicates that the underlying drivers of crime have actually changed.

There have been some very good studies that find appropriate control groups and compute the confidence intervals, and I'll get to these later. There seems to be some contention about whether those controls are done appropriately. Perhaps the statisticians, who are often accused of being themselves biased in favor of legalizing cannabis, are using statistical magic to make a real effect disappear. No doubt this is a very real problem in some academic disciplines. There are generally many different model specifications to choose from. Just a few nobs to turn and flips to switch on your regression analysis leads to a combinatorial explosion of possible models. It's always possible that the modeler has done a large number of these and cherry-picked the one that gave him the desired result. I think the appropriate way to deal with this is for the statisticians and critics to come together and discuss the various choices in model design, decide which specifications are more plausible or appropriate, settle on a constrained space of model specs, and openly report the sensitivity of the results to various specs.

For the moment, I'll start with crude methods.

The Raw Data

Below I have plots of violent crime rates by state for the states that  have legalized cannabis. The data all comes from here. (Simply click on Summary (SRS) Data With Estimates and download the estimated_crimes.csv file. That's what I'm using to build these plots.) There are four kinds of violent crime tracked here: aggravated assault, homicide, rape, and robbery. If Berenson's story is correct, we're probably only interested in aggravated assault and homicide. It's unlikely that paranoid psychosis-induced violence would have an effect on the other categories of violence.

Start with aggravated assault, which is more frequent than murder and thus has a stronger signal, statistically speaking. Notice that I'm color-coding based on the year that cannabis was legalized. Colorado and Washington legalized in 2012. Oregon, Alaska, and DC legalized in 2014. California, Nevada, Massachusettes, and Maine legalized in 2016, and Vermont legalized in 2018. The data is only through 2018, so we only get one post-legalization data point for Vermont. (If the figure is overlapping the stuff on my side-bar, note that you can click directly on the figure to enlarge it for a clearer view.)


It's certainly true that some of these states have seen increases in violence, as measured by this statistic. The post-legalization trends for AK, CO, WA, and OR are up. But DC is down. CA, MA, ME, and NV are down, despite the trend for the nation as a whole (marked "US") being slightly up. California has been declining for two decades, despite having medical marijuana since 1996. Many think that medical legalization in California was de facto legalization, as it was permissible for a doctor to recommend it for almost anything. (I don't fully understand this, but doctors in California could issue "recommendations" that weren't officially "prescriptions", but as long as you secured such a recommendation you could get cannabis from a dispensary.) Other states placed tight constraints on what conditions were acceptable for medical cannabis, so the connection between medical cannabis and rates of recreational use is less clear. Note that Colorado and Washington initially decrease before coming back up. This is not a unambiguous "legalization leads to violence" story. Now let's look at homicide rates.

Again, not an unambiguous story. Homicide certainly increased in Oregon, but then in 2018 it came back down to where it was 15 years ago. California is declining? DC is essentially flat. Certainly, Colorado, Alaska, and Washington look like they're up, but it would take some cherry-picking to tell an anti-legalization story.

Let's take a longer view. In the above, I'm purposefully focusing on the year 2000 and later so we can compare post-legalization to recent trends. What if we go all the way back to 1979, where the data begins? Violence was famously much higher in the 80s and 90s. As Stephen Pinker documents in his book Better Angels of Our Nature, violence has been declining for a very long time. Take a look at the recent trends in light of a historical perspective. Here is assault:


And here is homicide.


Other than assault in Alaska, the historical perspective makes the recent trends look puny. Don't get me wrong, it would be horrible if a state's violence rate went back to what it was in the 80s and 90s. I don't want be accused of making the argument: "The effect gets smaller when I rescale my y-axis." But we should try to appreciate the effect of cannabis on crime, whatever its magnitude, in context with other drivers of crime and other policy considerations. It is widely believed that drug prohibition is a major driver of violent crime because of black market violence, and the so-called "crack epidemic" of the 80s and 90s saw high levels violence between rival drug gangs. This was mostly not pharmacologically induced violence. This paper by Jeff Miron suggests that variation in the level drug enforcement explains something like half the variation in the murder rate. Note that's a regression analysis, and statements such as "explains half the variance" are statements about correlation, not causation. But the "causal" story is reinforced by the work of Paul Goldstien, who actually looked at all the murders in New York city from 1984 to1988. He coded them as non-drug related, pharmacologically induced violence, economic crimes to support a drug habit, and black market violence. Violence related to drug commerce was the dominant category. (See the discussion in the Miron paper, which also lists the Goldstein papers in the References.) If that's even close to correct, we should be far more concerned about violence related to drug trafficking than the pharmacologically induced violence that Berenson is warning us about.

Maybe actual use rates post legalization changed more in some states versus others? We don't care about legalization per se unless it drives up cannabis use and leads to these social problems that the critics are worried about. I looked this up, too. It's not exactly well organized, but SAMHSA tracks this by state and publishes reports. They have "past year" and "past month" marijuana use. I'm using past month, because presumably the higher-frequency use is what's going to track with excessive use and thus violence, assuming that causal connection exists.  (Here is one year's data. I could not find a file where this was all in one place for all years. Also, some years have this in a pdf, while others had it in an Excel spreadsheet. I had to awkwardly key values from a pdf into my own spreadsheet for those years. Government agencies need to get a lot better about how they disseminate their data!) The SAMHSA reports compute by-state use rates using two years at a time, so, for example, 2016 on my charts is really 2016-2017. I couldn't find anything in the 2004-2008 time period, but one report had the data for "2002-2003". So we at least have one data point from the period well before legalization. Here is what I saw.

Note that cannabis use trends upward even prior to legalization, but it definitely looks like there is a post-legalization increase. Unfortunately, there just aren't enough post-legalization data points for the states that legalized in 2014 and 2016 (and none yet for VT, which legalized in 2018). Now, one could seize on this and say, "Ah ha! Cannabis use increased more in Colorado, Oregon, and Washington, and those are the states with the clearest pattern of post-legalization increases in violence!" But it also looks like DC and Nevada saw a stark increase in cannabis use, and these jurisdictions bucked the "increasing violence" trend. California saw increasing rates of cannabis use for a decade and a half, but its rates of violent crime were declining over this period. Crudely speaking, this is a period of declining rates of violence and increasing rates of cannabis use.

For comparison, here are some non-legalizing states. It looks like cannabis use rates are increasing in many states that did not legalize. So the increase in the general trend for the US is not simply being driven by legalizing states.

Maybe "past month use" isn't what we care about? Maybe it's only excessive, daily use that leads to psychosis and violence, and if we had that measure it would track with the violence rates? Maybe, but I would at least expect "past month use" and "daily use" to be highly correlated.

Of course, this is all very crude. Eyeballing graphs and computing raw statistics is dangerous. One is likely to fool oneself and spot patterns that aren't really there. We need a more rigorous analysis that controls for confounders and computes confidence intervals. I'm only showing this because some anti-pot crusaders obsess over the raw data. The statisticians who do a more rigorous analysis are sometimes accused of being biased, using statistical tricks to make the effect disappear. With the above plots, I'm merely trying to show that crude methods, like eyeballing the raw trends, don't tell an unambiguous "more cannabis -> more crime" story.

Studies Using Synthetic Controls

I found two good papers on the cannabis-violence question, both of which use synthetic controls. The idea here is that you find a set of "control" states that behaved like the "treatment" state prior to legalization. Actually, if I'm reading this right, the approach uses a weighted average of all "untreated" states, picking a set of weights that best matches the legalizing state's trends prior to legalization. (One paper uses a subset of non-legalizing states, which it calls "donor states".) Having established these weights, one can compute what the trend for the legalizing state should have been if it hadn't legalized, then compare that to what actually happened. One of these papers discusses legalization of medical cannabis, the other recreational legalization.

The paper on medical cannabis is called Joint Culpability: the Effect of Medical Marijuana Laws on Crime. Here is a link to the full paper. Medical marijuana laws are not the same as laws that fully legalize recreational marijuana, but surely they increase the availability of marijuana. Some states had lax prescribing guidelines, such that you could get a prescription for almost condition. In fact, anti-cannabis fuddy-duddies like to complain that medical cannabis was just a wedge issue designed to make recreational cannabis more acceptable. As described above, many of them complained that California's medical cannabis law was de facto legalization, considering the ease with which one could obtain a doctor's recommendation. Supposing these cannabis critics are correct, legalizing medical marijuana should have led to an increase in recreational use. It seems likely that there would be some leakage from legitimate prescriptions to recreational users. (Based on some personal anecdotes, I know that this happens to some extent.) This paper cites several studies that quantified the effect of medical marijuana on overall use; they suggest the increase is in the 10%-20% range. If our hypothesis is that an increase in recreational marijuana use leads to more violence, the experience of medical cannabis states is a useful test.

Generally the paper finds no statistically significant effect one way or the other on violent crime or property crime. But if one fixates specifically on California, there does appear to be a significant reduction (a ~20% reduction) in both types of crime. I'm happy to call California an outlier and just declare this a null result. But the result is not favorable to the "cannabis use causes violence" story. See the comparison of synthetic controls versus actual experience in California.

The second paper is a student thesis by one Paul Lalonde titled Reeform Madness: Crime in the Changing Landscape of Marijuana Regulation. (I love the puns built into the titles!) It limits its analysis to Colorado and Washington. That makes sense. The crime data it uses is through 2016, so that doesn't give it much time to test the states that legalized in 2014 and 2016. It would be interesting to see this analysis redone in the next couple of years when more states have had a longer experience with full legalization.

By the way, if anyone is tempted to dismiss this as "some kid's term paper", note that Ben Hansen, got a similar result. Hansen is an academic economist with publications in the "synthetic controls" literature, notably this one, a working paper about the effects of cannabis legalization on auto fatalities. As far as I can tell he hasn't published on the violence question.

Reeform Madness generally finds that post-legalization violence rates in Colorado and Washington are slightly lower than the synthetic controls, and property crimes are slightly higher, but the results are not statistically significant. (Read the paper for the exact methodology; they're using some test statistics I'm not familiar with. Anyway, it's clear that they're not hitting the traditional "p < 0.05" threshold.)


In sum, while it's true that Colorado and Washington saw increases in violent crime post legalization, states that look like Colorado and Washington saw comparable increases. I'm quite happy to call this a null result or declare it inconclusive. But certainly this doesn't paint a picture of a post-legalization spike in violence, even for those states that definitely did see increases in murder and assault.

Back-of-the-Envelope Sanity Check

I want to do a quick computation just to get an idea of what magnitude of effect Berenson is suggesting. Past-month cannabis use in Colorado was 10.4% of the population in 2011/2012 and 16.4% in 2016/2017. Let's subtract the two and say that 6% of the population turned from non-smokers to smokers because of legalization. The murder rate increased from 3.0 per 100k in 2011 to 3.4 per 100k in 2016, a 13% increase. For 6% of the population to drive a 13% increase in murder at the population level, this must mean that "past month cannabis use" is a huge risk factor for murder. It apparently increases your chance of becoming a murderer by 216% ( = 13% / 6% )! Even making some more conservative adjustments, this is a huge effect. Suppose that in addition to creating new users, legalization doubled the intensity of use among exiting users. So let's say effectively 16.4% of the population are "new users", adding the actual new users to the existing users who are now doing double-duty. And suppose we only blame pot legalization for half the increase in the murder rate (so call it 6.5% instead of 13%). We're still saying 16.4% of the population caused a 6.5% increase in the murder rate, so this population's murder rate must have increased by 40%. These seem like implausibly large figures to me. The people who are attributing even small population-level increased in violence to cannabis legalization are actually positing a huge effect of marijuana's propensity to cause violence. Of course, most of them will say, "No, no, you won't be 40% more likely to commit murder, or any more likely for that matter. It's all due to a small subset of people who can't handle it, have a bad reaction, and turn psychotic." But the smaller we make the "susceptible" population, the larger the effect on violence we must posit. 

There is nothing special about Colorado; I get similar absurdly large "effect sizes" when I look at Oregon and Washington, depending on the end-points. Of course, as implied by the discussion above, doing this exercise for some states would yield negative effects of legalization on crime.

I think that you can probably find news stories about people who smoked way too much, got overly excited or paranoid, and did violence to someone. I even think it's reasonable to say that cannabis use was the cause, in a but for sense, of some of those violent acts. (I'm reminded of an old post at Slate Star Codex in which Scott Alexander describes a colleague's frustration with marijuana. It seems that cannabis use interfered badly with his treatment of some of his schizophrenia patients. This is part of Berenson's story: cannabis use is particularly bad for people with preexisting mental illness, even if it's benign for most users.) But these are exceedingly rare. Maybe it's reasonable to trumpet these stories as a cautionary tale for people with risk factors, but blaming them for population-level increases in violent crime is just implausible. It's also inconsistent with the crime data from various states. 

I hope that someone with more time and better resources than myself picks this up and expands upon this literature. I think there is a lot more work that can be done. A suggestion to the people who are doing this kind of work, it might be helpful if you showed the full details of your analysis, say in an R Markdown or a Jupiter notebook. That way you can really answer your critics and say, "Which step in the analysis do you disagree with? Run it yourself and see if the result changes." That way, when someone furrows their brow at charts like the ones copied above, there is no question as to how they were generated. Give the critics a chance to respond with specific criticisms, and give yourself a means to "shut them up" if they're just blowing smoke.