Cousin Causality

                                                  Link to the Forbes article above

                                                 Link to the Forbes article above

                                                  Link to the Pediatrics editorial above

                                                 Link to the Pediatrics editorial above

                                                 Link to the Pediatrics article above

                                                Link to the Pediatrics article above

It is often said, rightly, that "Correlation is not causation." But one can say even more: "Correlation between A and B with A coming earlier in time than B is not causation." Let's say A and B are correlated, and someone is interested in knowing if A causes B. The most common reasons for A being correlated with B are

  1. A causes B
  2. B causes A
  3. C causes A and C causes B. 

Given that someone is interested in whether or not A causes B, the second possibility, B causes A is called "reverse causality." Unfortuntately, the third possibility does not have a snappy name that I am aware of. Over my entire teaching career I have ended up calling it "A third thing causes both ___ and ____." Unless someone knows a better name, let me propose calling the third possibility "cousin causality" of A and B.

The analogy is that for any pair of cousins, a grandparent or two helped cause both of those two cousins. The two cousins may never have interacted directly at all, and could have been adopted at birth by different families on opposite coasts, but they will share traits in common.

Vaping. Let me apply the old, but newly rechristened concept of cousin causality to article and associated editorial linked above. The article is "Trajectories of E-Cigarette and Conventional Cigarette Use Among Youth by Krysten W. Bold, Grace Kong, Deepa R. Camenga, Patricia Simon, Dana A. Cavallo, Meghan E. Morean and Suchitra Krishnan-Sarin." The "Conclusions" section of the abstract begins impressively:

E-cigarette use was associated with future cigarette use across 3 longitudinal waves, yet cigarette use was not associated with future e-cigarette use.

This is useful information. In this case it makes it harder to maintain a story of reverse causality. But it does nothing to rule out cousin causality. What if (contrary to my own views) some kids

(a) think things like smoking are a cool symbol of rebellion,

(b) at a young enough age, both vaping and smoking count as "things like smoking," but

(c) at an older age, vaping no longer counts because it is the sort of thing that youngsters do,

(d) except for the growing realization that vaping is babyish just discussed, between vaping and smoking there is some tendency for kids to keep doing the one they started doing. 

In this story, if you take away vaping, the kids who think smoking and the like are a cool symbol of rebellion will still take up smoking. The vaping didn't cause smoking, the desire to be cool caused both vaping and smoking. And we can explain the time sequence by saying that at older ages the desire to be cool tends to cause smoking relatively more compared to vaping.  

The authors interpret their results as if there were no chance of cousin causality. Without much sign of caution, the abstract assumes causality from vaping to smoking:

Future research needs to examine mechanisms through which e-cigarette use leads to cigarette use.

In other words, "Because vaping allows one to predict later smoking, we now know that vaping causes smoking, it is just a matter of how vaping causes smoking." Just in case there is any doubt that the authors are claiming that vaping causes smoking, the next line is:

E-cigarette regulation and prevention programs may help prevent future use of cigarettes among youth.

At least this has the word "may" in it, but in combination with the previous sentence, it is a strong claim that vaping causes smoking.

The editorial by Jonathan Klein in the same issue of Pediatrics also has no qualms about asserting that this evidence points to vaping causing smoking:

"With this study, we add longitudinal, causal evidence that the transition from e-cigarettes to combustible products, at least for adolescents, is a one-way street,” wrote Jonathan Klein, MD, from the University of Illinois Pediatrics Department in Chicago, in an accompanying editorial.

If the word "causal" were dropped, the phrase "one-way street" might be a clever way to imply causality while having deniability that one had ever claimed causality:

With this study, we add longitudinal evidence that the transition from e-cigarettes to combustible products, at least for adolescents, is a one-way street

But with the word "causal" the claim that vaping causes smoking is overt. 

Financial Education. There are many other applications where the possibility of cousin causality needs to be considered. For example, suppose we want to know to what extent a given dose of financial education causes people to make better financial decisions. Just looking at how well financial education predicts good financial decisions won't do the trick, because being smart through processes prior to the financial education can both lead someone to get financial education and make someone better at making financial decisions. Take a way the financial education, and the intelligence that preceded the financial education will still lead to better financial decisions. 

Holding the social environment fixed, I find it very difficult to think of anything that would make one person more likely to get financial education than another in the cross-section that doesn't also have a direct effect in improving financial decisions. For example, think about characteristics of parents. The sort of parent who would encourage a kid to get financial education would also be likely to transmit financial knowledge and savvy at home. Personally, I find it easy to believe that this knowledge and savvy transmitted at home might be much more powerful in its effects on financial decisions than a few semesters of formal financial education. 

To get causality of financial education requires something like a randomized trial or the introduction of an additional financial education requirement in a particular state. And randomized trials are difficult because the dosage of financial education that would cause a detectable different in financial decision-making might be quite expensive. 

Long-Term Bond Prices. I have been focusing on cousin causality, but there are cases where A coming before B in some sense can't even rule out reverse causality: B causing A. The standard theory of the term-structure of interest rates says that if the Fed is now expected to raise future short-term interest rates faster than was expected before, long-term bond prices will be driven down—which is equivalent to long-term interest rates going up. So in some sense the level of future short-term rates is causing the current level of long-term rates.

Of course, this isn't quite right. There is something already out there, now, that is likely to cause future short-term rates to be higher than they otherwise would be and is causing current long-term rates to be higher through people's thoughts now about future short-term rates. Seen that way, current long-term interest rates being able to predict future short-term rates is an example of cousin causality. Whatever is driving the Fed's plans is driving both future short-term rates and—through thoughts about future short-term rates—current long-term rates. In any case, the fact that the movement in long-term rates comes before the movement in short-term rates doesn't mean that the movement in long-term rates now caused the movement in short-term rates later. 

Conclusion. I want to end with a challenge. Now that you know the handy term "cousin causality," look for examples of where this might be going on—especially where it might mean that someone's claim of causality is less of an open-and-shut case than they think. Examples won't be hard to find. For both people and nations, good things tend to cluster with good things and bad things tend to cluster with bad things. But in my view, most of the correlations come from cousin causality. It isn't so easy to single out the underlying causes that are the "grandparents" in the analogy behind the naming of cousin causality.