Tim Requarth discusses the implications of this innocuous but often-misunderstood statement.
The problem is, by some mixture of ideology, confusion, and too-clever-by-half-ness, people often like to argue that an elasticity will be greater then 1 (or less than 0, depending on how you define it). That is, they argue that a proposed policy will elicit such a strong reaction in the opposite direction to have no effect or even backfire. As Requarth discusses, this sort of claim (for example, the argument that wearing seatbelts makes driving more dangerous) don’t make a lot of sense nor are they typically supported by data. Their main role is to offer just enough surface plausibility to muddy the waters, I guess kind of like the anti-vax arguments we’ve been seeing lately.
This reminds me a bit of middle-school debate, where you win by (a) advancing enough arguments that the other side doesn’t get around to rebutting all of them, while (b) making the effort to offer some rebuttal to each of the other side’s arguments. Even a weak rebuttal not based on solid evidence is fine, in that it then forces the other side to waste its time rebutting the rebuttal, etc.
This claim is stated far too broadly to be true or useful.
Consider firm-level price elasticities of demand.
If the elasticity is between zero and one, then if the firm raised prices by a small amount, it’s total revenue would increase.
And as long as its total costs are increasing in the quantity produced, this implies that its profits would increase.
Thus, whenever economists estimate a firm-level demand equation, if the estimated elasticity is between zero and one (more precisely, between zero and negative one), we strongly suspect that the model was misspecified (absent something like a binding regulatory price ceiling).
Sometimes we even constrain the estimation to give only elastic demand, e.g., through the functional form or by jointly estimating both a demand equation and a profit-maximizing pricing equation.
And there is lots of empirical support for elastic firm-level demand equations from studies where the specification allowed for inelastic demand.
I’d encourage you to think more carefully about the domain where we would expect elasticities to be between zero and one — maybe there is a useful rule of thumb in there somewhere.
But as it stands, I think your claim is basically useless.
As an economist, that was my initial thought as well. But I think Andrew’s post is really different than the economist’s use of elasticity of demand (price elasticity). The point raised is that the indirect effect of a policy or change will be greater than the direct effect. The seat belt example is that requiring seat belts directly improves safety in a crash, but may lead to more unsafe driving, thereby resulting in more crashes (and there is some empirical support for this). Another example is that taxing (anything) will directly decrease use, but will also modify behaviors that could increase the problem (e.g., taxing smoking could decrease cigarette purchases but increase a number of alternatives, including illegal cigarettes, to cause more damage than the initial activity being taxed).
But I’m not sure I buy the idea that our initial reaction should be suspicious of such claims. In fact, economists seem to pride themselves on finding such apparent anomalies. What irks me is that the theoretical possibility of such things is often accepted in lieu of actual evidence of it. I think this is similar to the problem of NHST where rejecting the null is taken a support of whatever your favorite theory of an effect is. So, I’d say the problem is not that indirect effects should be viewed as weaker than direct effects, but that their possibility should not be taken as evidential support.
Replying to myself – after reading the linked Requarth article.
The substantive issue is whether we should approach such claims from the skeptical viewpoint that the default assumption should be that the direct effect will dominate any indirect effects. So, masks and other safety actions should be presumed to have a net beneficial effect, with the burden of proof placed on those that would say that the secondary/indirect effects are strong enough to overwhelm any beneficial direct effect.
I’ll be interested to see what others think. My initial reaction is to resist (at least somewhat). I think it is easy to agree using System 1 thinking: safety requirements work, and while there may be indirect effects that work in the opposite direction, surely the direct effect much be larger. This almost sounds like claims regarding interaction effects in statistical models. However, I don’t see any reason (using System 2 type thinking) to assume that the direct effects will usually dominate. Indirect effects can be strong – unintended consequences can be important. I do realize, however, that there is a selection problem. Finding (or even suggesting) situations where unintended consequences are strong are often publishable while finding that direct effects work as intended is less likely to get attention (certainly true in economics).
Steven:
Fair enough; you’re right that it depends on the context.
I would say though that I tend to defend Andrew’s point more. I think the bigger realization here is that generally speaking the current situation in any sort of context is not optimizing for a particular thing. In the example of a firm’s prices, for instance, it’s not actually likely for a specific given firm to have exactly the price that optimises profit. That’s because there’s usually other concerns, like say, social consciousness, wanting to have the appearance of a premium product, making promises that prices don’t change too often, avoiding the bad press of being the first to raise prices, “because everyone else prices at X”, pure laziness and so on.
Some of those factors might affect profitability – or affect it, eventually – but a lot do not. Even in terms of competition, information flow isn’t free and customers don’t all switch suppliers at the drop of a hat. Thus it is true that there there’s probably some direction in which you can change prices where in terms of profit alone, the offsetting effect does not erase the benefit entirely. A fair number of games developers, for example, have found that when they released games at double the price point that other devs do, the demand does not fall very much at all and they make a bunch more profit.
At least in some of these examples, it’s not the drivers overreacting and nullifying the benefit to themselves, but the increase in arbitrarily-sized negative externality against others, like pedestrians, where it is claimed you see net damage from.
I’ve always come at this a little differently: you should assume direct effects swamp indirect effects.
1st-order/direct effect: increasing seatbelt usage -> reduced deaths.
2nd-order/indirect effect: increasing seatbelt usage -> reduced deaths -> more reckless driving -> more deaths.
Re Steven’s example, I’d argue increasing prices has two direct effects: increasing profit per item and reducing quantity sold. Which one wins is harder to say. But perhaps it’s a bit subjective what I think of as direct or indirect.
One last example, I was skeptical of the claims Obamacare would reduce costs through increased preventative care. Spending more on tests should swamp spending more -> catching problems earlier -> reduced treatment costs.
You are right to be skeptical about cost saving through increased preventive care.
In fact, when formal economic analyses are done, there are almost no preventive interventions that result in net savings of money. Nearly all preventive measures increase costs, often substantially. But that’s not the right way to look at it. After all, most people prefer living in a home to being homeless even though it costs a lot more. The right question is whether gains in health and survival are sufficient to constitute good value for the increased costs.
Complex systems can only exist because there are feedbacks that make this happen (after some delay). Eg, many drugs will trigger a tolerance (upregulated liver enzymes, downregulated receptors, etc). If prices rise, this incentivizes more production, which then lowers the price. There are many examples of this.
So unless the effect is so extreme that things break down, a close to net zero response should kind of be the default assumption. Once again, that is after some delay as the system finds a new equilibrium.
This feels related to the piranha principle. Just like we should have strong priors that any given interaction is very small, we should also have strong priors that second order effects don’t overwhelm first order effects. Of course (like interactions) unintended and indirect consequences are real, and sometimes indirect effects even happen to knock your outcome in the opposite direction as direct effects. But a world in which it is very common for rube goldberg chains of indirect effects to reverse the intended direct effects of your actions wouldn’t look much like our own. It would look more like the universe of Wile E Coyote.
Another relation to the pirhana principle is that the possibility of strong interactions (however unlikely) is often used as a cheap debate tactic as well. Interestingly, in the debate setting, the possibility of interactions is usually used to justify why some intervention might be effective even though we haven’t seen any sign that it is, and thus to justify continued action. The possibility of second order effects, on the other hand, is usually used to prevent or stop actions that on their face are effective. So if you want to argue about something in bad faith and sound sophisticated depending on whether you like or don’t like it you can invoke interactions or indirect effects.
I’d think a world without feedbacks/attractors would look like loony tunes. It would not be stable.
There are usually multiple possible stable states, so we can’t necessarily assume whatever effect will eventually be reversed back to how things were before. But you should always expect the system will compensate somehow.
Agreed, of course there are feedbacks. It’s just rare for them to be counter to and stronger than intended direct effects of interventions. Not rare in the sense that you can’t list a bunch of examples, just rare in the sense of it happening for a low overall proportion of interventions.
I think about it more like: for weak interventions you can use the heuristic that there is eventually a reversal (or even rebound effect into the opposite for a time).
For strong interventions then you can expect to transition to some other stable state that may or may not have net consequences you prefer.
I’d suggest Risk Compensation is a specific case of Zero-sum bias. I mean, maybe it shouldn’t be surprising if risk compensation arguments are generally appealing when we’re all taught at an early age that “every action has an equal and opposite reaction”.
Still, most nudge advocates claim that a small action will have large effects. The evidence is questionable, and my own belief is that many nudges have negligible effects. I guess that’s not quite the same thing as saying the elasticity is less than zero, but it’s also not that far off. The two arguments don’t sit side by side comfortably, in my mind. One appeals to the presence of large secondary impacts and the other to them being small. If we avoid the extreme cases of the secondary impacts being larger than the direct impacts, I wonder how far it makes sense to go regarding priors: limit the size of the indirect effects to be no larger than the direct effects? I’m still not comfortable with that position, although I’ll admit that the selective memory of counterexamples may be clouding my judgement.