At the Hudson Institute, David Murray and former drug czar John Walters have responded to my post on marijuana legalization and roadway fatalities in Colorado. It’s an odd response. It includes some condescending lecturing about statistics, a complete misunderstanding of my point and criticism of me for overlooking things that my post actually addresses.

Before I get to their response, I’ll just reiterate that my argument isn’t that roadway fatalities in Colorado have fallen because of pot legalization. My argument is this: Opponents of legalization predicted a surge in roadway fatalities due to drivers under the influence of pot. After seven months, that hasn’t happened. Roadway fatalities have continued to decline. Opponents of legalization also argue that we’ve seen an increase in arrests for “drugged driving” and that we’ve also seen an increase in accidents (in Colorado and in states that have legalized medical marijuana) in which one or more drivers tested positive for pot. My argument is that these tests don’t measure inebriation. Therefore, we can’t say that those accidents were caused by pot, and we can’t say what percentage of those new drugged-driving arrests involved drivers who were actually a threat to the other people on the road. All we can say is that as pot becomes more widely available, a larger percentage of people on the roads are testing positive for pot. It doesn’t necessarily mean they were high while driving, nor that they were significant impaired while driving. As pot becomes more widely available, you would expect a large percentage of people to test positive for pot from any sample group.

There is also the possibility that some people may be substituting pot for alcohol. Since pot tends to be less inebriating for drivers than alcohol, there’s a possibility that easier access to the drug is making the roads safer, but I added that there isn’t nearly enough research to say so conclusively.

Walters and Murray begin their post with a lesson in what it means when numbers decline in value.

If the number of events per year starts to decline steadily over time, it’s possible to look at any stretch of 14 years and look for meaning. But it’s complicated. Traffic fatalities are declining steadily in the United States in recent decades. For example, in Colorado, there were 606 traffic fatalities in 2005. That number fell to only 548 by 2008, and by 2011, they had fallen further to only 447.

I’m not sure what their use of the word only means here. It isn’t clear if the word is meant to suggest the fatalities should have fallen further, or if they’re expressing appreciation for how far fatalities have fallen. Their point seems to be that traffic fatalities have been falling in Colorado for some time. This is true. I would also point out here that Colorado legalized medical marijuana in 2001. So the drop in traffic fatalities has occurred over a period where, again, pot has become increasingly available to the state’s residents. Murray and Walters are correct that traffic fatalities have also fallen nationwide. But a 2012 study in the Journal of Law and Economics found that roadway fatalities have dropped more in states that have legalized medical marijuana than in those that have not. Again, this suggests that the roads are safer because of wider access to pot, although — again — it falls short of completely making the case. But it certainly deflates the argument that wider access to pot has made the roads more dangerous.

Inevitably, just because of the way numbers work, the most recent years’ traffic fatalities will be lower than those from the beginning of the series. Because the phenomenon is declining over time, the fatalities for 2014 will be fewer than the fatalities for 2005, and even lower than those for 2001. If you took the average number of fatalities over time (the statistical mean), you would add all fatalities together for the years in question and divide by the number of years. It follows that the earliest years will likely have fatalities above the average and the later years will fall below the average; that result is contained in the meaning of the terms. For the three years I’ve provided, the average is 534. Sure enough, the earliest year (2005) is above average, while the most recent year (2011) is below average. In any such declining series, earlier years will likely be higher than subsequent years, and the latest year should be the lowest yet.

I do appreciate this mini-treatise explaining how small numbers are smaller than large numbers. Although I’d like to think that I was pretty clear on the concept before reading it.

Yes, it’s true that traffic fatalities have been falling. They have been falling over a period where both social and legal restrictions on pot have loosened. As I acknowledged in the original post, there are a number of reasons our roads are safer, and it’s possible that were it not for legalization, fatalities would be even lower. But I’ve yet to see a study attempt to back that proposition with data. Until then, we can really only say this: Greater access to pot has not resulted in a surge in highway fatalities.

All of this is pretty elementary, but it’s easy to overlook in the quest for a rhetorical point concerning the possible impact of marijuana legalization. To see the impact of such a new factor, you can’t just add up the number of deaths per year and draw a conclusion. You need to examine the rate, which would be the number of fatalities per number of miles driven, and see whether that is changing in relation to the new factor. Should the number of deaths per 100,000 miles driven suddenly go up, for instance, then you would have grounds for suspecting the impact of the new factor. But what if the rate continues downward, but at a different pace, either a faster decline, or a slower decline? To tease out your new factor, you have to do a more complicated analysis than just compare the years in question to the average number of deaths for all the years.

All very true. And all addressed in my original post, in this paragraph:

I should also add here that these are total fatalities. If we were to calculate these figures as a rate — say, miles driven per fatality — the drop would be starker, both for this year and since Colorado legalized medical marijuana in 2001. While the number of miles Americans drive annually has leveled off nationally since the mid-2000s, the number of total miles traveled continues to go up in Colorado. If we were to measure by rate, then, the state would be at lows unseen in decades.

Now it’s true that we don’t yet have miles driven statistics for 2014, or for the past few years. But for the first 10 years after the state legalized medical pot, roadway fatalities continued to fall in Colorado even as residents drove more miles. Just so I’m not misunderstood, I’ll reiterate: I am not arguing that pot is the reason the roads are safer. I’m arguing that this is strong (though not definitive) evidence that pot hasn’t made the roads more dangerous, as prohibitionists have claimed.

These are the factors that Radley Balko’s claims have overlooked.

I did not overlook them. See the preceding block quote.

What about the second claim, that the results are “closer to historic lows”? Balko’s graphs show a line for 2014 and it falls between what are termed “historic highs” and “historic lows.” But these “historic” years are not actual data from real years. Rather, Balko constructed artificial years, with composite results. That is, the highest January from any year was added to the highest February from any year and so on; likewise for the “lowest years.”

Walters and Murray write as if I tried to pull a fast one. This was all clearly explained in my post.

It’s true that the actual data from 2014 falls between these two composites. But that is just probability in action. The odds are overwhelming that any actual year will not “roll the dice” on each of seven months (January to July) and come up successive highs for each month. Hence, to say that 2014 is “closer to the historic lows” follows from both the probability and the fact that there has been a steady decline in fatalities over time.

Yes! All true! But it is also compelling evidence that people like Walters and Murray were off-base when they predicted that legal pot would make the roads more dangerous. And not only have the month-to-month totals “fallen between” the historic highs and lows, they’ve been much closer to the 10-year-low for each month than the 10-year average. They’re nowhere near the 10-year high. Again, the dire predictions from prohibitionists would not have predicted this.

Overall, we need to step back and assess what can be reliably known about the impact of more readily available marijuana on a host of social indicators, affecting both residents of Colorado and those who transit the state and use the drug when there. One of those impacts is likely to be highway safety. There are parallels to be found in the effect of changes to state legal drinking age laws in the past; when some states lowered the age of access to alcohol to 18, there were measurable effects on highway safety, and they sometimes affected surrounding states whose underage residents transited to the state with a lower age, drank immoderately and then drove home.

Are there parallels? What is the evidence that the two scenarios are similar? How do we know that one of the impacts of legalization is “likely to be highway safety?” It seems to me that this is exactly what we’re trying to figure out. Simply stating that there are parallels, or that highway safety has been impacted, doesn’t make it so. I’m trying to look at the hard evidence. Walters and Murray are citing their own intuition. (And let’s be honest, when it comes to criminal justice, Walters’s intuition and powers of prognostication are spotty at best.)

The effect of marijuana use on driving is more complicated; the level of impairment is difficult to assess, and marijuana, unlike the more transient alcohol, remains in the system for a long time. The THC intoxicating compound is more persistent and cumulative, and its presence might well be affecting a host of judgment and reaction time behaviors.

Note the sleight of hand here. It’s true that the level of impairment caused by marijuana is difficult to assess. It’s also true that every attempt to assess has concluded that the drug is less impairing than alcohol. It true that pot stays in the system longer than alcohol. It can be detectable for days. But that’s all well after the drug’s psychoactive effects have worn off. This is why it’s misleading to suggest that a positive test for pot is proof that a motorist was impaired. Walters and Murray are fine with letting you think pot impairment and alcohol impairment are similar when they’re attempting to ascribe the negative effects of alcohol to pot (state-by-state legalization will have the same effects as varying drinking ages, failing to note that pot causes less driver impairment than alcohol) but want you to distinguish between the two drugs when doing so makes pot look worse (pot stays in your system longer than alcohol).

All that said, the specific measure of in-state traffic fatalities is not a very effective arena to assess the potential impact of more marijuana more widely used. Highway safety is subject to a large number of variables, as are accidents involving driver or pedestrian impairment, some of which produce fatalities.

That last sentence is mostly true. But if you’re going to predict that legalizing pot will make the roads more dangerous, you should be prepared to have an honest way to measure that claim after legalization is implemented. Simply pointing to an increase in drivers who have pot in their systems isn’t sufficient, because 1) we can’t yet reliably measure pot impairment on drivers, and 2) post-legalization, any sample population is likely to show an increase in the presence of pot. (I realize I’m repeating myself. But it’s necessary because they keep making the same argument from a slightly different angle.) Right now, highway fatalities are the best data we have. And seven months after legalization, fewer Coloradans are dying on the state’s roadways than before pot was legalized.

Long-term, the effect of more drug use is likely to be detectable, against the multitude of other background factors, which themselves are not constant in their effect. In general, highways are safer today than they were short years ago. Greater use of drugs, however, has not been a factor in making them more safe, and an increase in such use is unlikely to be beneficial.

Aside from the aforementioned possibility (that still needs more research) that some people may be substituting pot for alcohol — which probably would make the roads safer — I mostly agree. But when the prohibitionists were arguing against legalization, the scourge of drugged driving was a major weapon in their rhetorical arsenal. It seems significant that they’re now reduced to arguing that an increase in pot use is merely “unlikely to be beneficial” to improved road safety.

Beyond that, we simply at this point cannot say. To try and detect a signal regarding the impact of legally available recreational marijuana in a single state’s fatality record for the first seven months of the new regulations is simply not productive, at this point, and is not likely to tell us much about the overall, and potentially lasting, effect on society.

Fair enough. I acknowledged in the post that we’re only seven months in, and that the numbers could change. Still, it’s curious that Walters and Murray take the “it’s way too early to tell” line when the early data cut against their position, while in other forums they write things like this:

Ah, but we are told, the marijuana project is an “experiment.” Based on the results to date, showing undeniable and accelerating damage, it is time to end this sham experiment and acknowledge the mistake. We should do so before even more citizens, especially youth, are hurt . . .

Moreover, the state of Colorado has yet to come to grips with the costs being imposed by expanding marijuana use, costs to public safety and public health, to addiction treatment services, and to school and workplace performance . . . In Colorado, the black market thrives, and the state has even become the source of large-scale interstate smuggling . . .

Some experiment. The outcome is foreseeable, tragic, and avoidable. In genuine medical trials or indeed any experiment involving human subjects, there is an overwhelming ethical obligation entailed by the research. When it becomes undeniable that the experimental intervention is actively harming subjects, the trial must be stopped.

The distribution of marijuana, addictive and harmful, meets these criteria, and the costs vastly outweigh whatever benefit was supposed to accrue to society or the individual. It is time for responsible public officials to stop playing along, and shut this failed experiment down.

So when it comes to traffic fatalities, Walters and Murray say seven months of legalization is way too early to start drawing conclusions. Yet three weeks ago, they were more than willing to declare legalization a catastrophic failure and demand it be ended immediately. (Without citing a single bit of data, incidentally.) And a month ago they wrote this:

The evidence to date is stunning. It is time to stop pretending otherwise.

Again, this conclusion was not based not on any quantifiable figures, but on anecdotes, interviews with people who already agreed with them and their own observations upon visiting Denver. (Incidentally, crime in Denver is down since legalization.)

Seven months of hard data on traffic fatalities? Way too early to say that legalization hasn’t made the highway more dangerous. Personal observations, anecdotes and interviews with prohibition advocates after six months of legalization? The evidence is stunning. Stop pretending. Shut down this experiment.

I don’t doubt that Walters and Murray are sincere in their call to end legalization post haste. Returning to prohibition now would prevent us from collecting more data that could further prove them wrong.