When it comes to drug policy, I think a lot of us underestimate the power of substitution.

We think that marijuana might be a gateway drug, for example (which is the same thing as saying it’s a compliment to further drug use).  But then we forget that marijuana is also to some extent a substitute, and thus an anti-gateway drug: if I want to party tonight, I can chose to smoke pot and drink, but I could also chose to smoke or drink.  I also have that choice in the long run–I can identify as a smoker and a drinker, or I can be either a smoker or a drinker.  Whether marijuana and drinking are net compliments or net substitutes is an empirical question, depending on the strength of each effect.  We tend to forget this, and to assume the two are net complementary.

I think the reason we underestimate substitution effects is that we tend to think more about what we can see with our eyes than about what we can’t–we understand presence better than absence–and the substitution effect is all about absence.  The substitution effect is the counterfactual, “that which is not seen.”

This matters for drug policy, far beyond simple prohibition.  Look at what happened when researchers studied abuse-detterent OxyContin pills, for example (pills that were more difficult to crush or dissolve, and thus harder to abuse):

the selection of OxyContin as a primary drug of abuse decreased from 35.6% of respondents before the release of the abuse-deterrent formulation to just 12.8% 21 months later (P<0.001).

Good news, right?  Making the pills harder to abuse probably meant fewer people abused them.  But wait, there’s still the substitution effect:

66% indicated a switch to another opioid, with “heroin” the most common response.  These changes appear to be causally linked, as typified by one response: “Most people that I know don’t use OxyContin to get high anymore. They have moved on to heroin [because] it is easier to use, much cheaper, and easily available.”  …there was no evidence that OxyContin abusers ceased their drug abuse as a result of the abuse-deterrent formulation.  Rather, it appears that they simply shifted their drug of choice.

Figure 1B shows the before-and-after difference (the vertical dashed line at Q3, 2010 is when they started using abuse-deterrent pills):

Notice that whereas the researchers were definitely going to study the direct effects, they had to make a conscious decision to study possible substitution effects, as a separate, secondary line of study.  The direct effects are obvious research interests–it either reduced OxyContin abuse as intended, or it didn’t–but the substitution effects are more subtle, more numerous, and probably aren’t even all captured by this graph.