Correlation is a Clue–But No More than a Clue

(Today’s topic deserves a longer discussion, with specific examples from historiography and case studies. But after reading, on the same day, Alexandra Petri’s hilarious spoof AND Gov. Barbour’s crackpot correlation, I couldn’t resist posting this ASAP.)

If one of the skills of historical thinking is to figure out if A caused B (or, more properly, if a cluster of causal factors caused B), then teaching students the difference between correlation and causation (including how to distinguish one from the other) is a critical component. We don’t want students to find a correlation between A and B–that may be no more than a coincidence–and jump to the conclusion that A caused B (and turn this discovery into a thesis statement).

In a science class students will (or should) learn that "correlation does not imply causation." This needs further explanation–as IMPLY is being used here in a technical sense, not in its everyday sense. Among statisticians IMPLY indicates sufficiency (logically: if p, then q). In ordinary speech IMPLY usually means SUGGEST. That is probably what trips up a lot of students–who then fall into the trap of a logical fallacy (1).

If you’re trying to get students to understand just how ridiculous it may be to confuse correlation with causation, assign Alexandra Petri’s "I Blame the SuperMoon" (2). Petri finds oddball correlations between the occurrence of a SuperMoon, when the moon is at its perigee (closest to the earth), and historical events. She goes on to lampoon Lunesta commercials and those who would attribute natural disasters to questionable "causes" (providing a current events segue to the quake in Japan). Print-out for students or give them the URL to read online. This would make a very good BELL-RINGER activity.

Here’s another example: Gov. Haley Barbour of Mississippi doesn’t seem to have learned to the correlation/causation distinction. This is how he formulated a recent indictment of President Obama’s economic policies: "In the last two years, the federal government spent $7 trillion and our economy lost seven million jobs. I guess we ought to be glad they didn’t spend $12 trillion. We might have lost 12 million jobs" (3). Well, that’s just too neat. What about the much larger number of jobs that would have been lost if the government had done nothing?

Certainly, correlation may co-exist with causation. It certainly is a clue, for correlation does present the POSSIBILITY of a causal relationship. Thus, perceiving a correlation should lead to the next step--formulating a hypothesis. This happens in the hard sciences, in social sciences, and in historical disciplines. Note that another key term is the verb PERCEIVE because, especially in history (or, for that matter, any discipline outside those based on laboratory experiments) it is easy to fall for a false perception. Often it is simply the paucity or unevenness of the historical record that leads a historian astray (after all, we try to do the best we can with what we have). Or there may be plenty of evidence, even readily available evidence, but an individual’s circumstances inhibit engaging with it. Expecting students to work with too few resources is poor pedagogy–only partly mitigated by emphasizing the limitations of those resources. (Believe me, without access to JSTOR or a comparable resource, the Web is still an inadequate tool.)

Sometimes the best way to do historical thinking is to figure out that there are insufficient resources to answer a question–and then stop short of drawing a conclusion. Going ahead anyway (ignoring the insufficiency) may elicit a false perception–a correlation that doesn’t exist–and this will lead to a faulty hypothesis. (Inevitably, this sometimes happens but isn’t it better to be up-front about such an outcome?). Students in high school typically have restricted access to material, relative to their counterparts in college. Let them explore–don’t push them towards  premature conclusions.

In the APWH  [Advanced Placement World History] version of historical skills, "historical causation" is listed as part of chronological reasoning. The other sub-skill under this heading is "patterns of continuity and change over time." Yes, for historical causation, we need linearity but isn’t there more to it than that? Like, correlation, chronology is "a necessary but not sufficient condition for causality." In other words, just because we can show that event A occurred in 1923 and event B occurred in 1924, we cannot assume causality between these two events.

Historians, of course, nearly always bump into the "chicken or the egg" conundrum, especially if the case at hand is complicated. There may indeed be an indirect relationship between A and B, if some other event antecedent to B is somehow related to A. Doing history is messy!

Logically, a correlation between A and B points to five possibilities:

1. relationship between A and B is just a coincidence (no causal relationship)
2. A caused B
3. B caused A
4. C caused both A and B (but C could actually be a set of factors)
5. any combination of #2, #3, and #4

When I was teaching world history to first-year college students, one of my goals was to instill in them a sense of the complexity of historical events and trends. If we examine an event or trend with a very low resolution lens, the many converging factors produced it will be obscured. This is one of the dangers of doing macro-history: it encourages us to see the big picture (that’s OK) by erasing the details, turning off the HD (high definition)–and that’s not OK. Alas, a survey textbook rarely, if ever, provides an evidentiary basis for grasping complexity–though it can convey a sense of interconnectedness and encourage thinking about patterns. (More about this another time, I’m going too far off the track of what I intended to focus on.)

I’d much rather ask students to formulate hypotheses and think about how to test them. I want them to grasp the constructive, reconstructive, and model-building efforts that go into "doing history." And, especially if an analysis is aimed at uncovering causation, to comprehend that only a model with interlocking hypotheses (some more closely linked than others) will accomplish the task. The alternative--concocting "a simple A caused B statement" will only obscure the multiple factors (interconnected) and layers (interlaced) of causal explanations. A simple thesis statement, supported by a few cherry-picked bits of evidence from a textbook (and perhaps a few ancillary sources) just doesn’t convey the effort and time-consuming tasks that any semblance of really doing history requires.

NOTES

1) To brush up on this topic (as I did) consult "Causation Does Not Imply Causation," Wikipedia: http://en.wikipedia.org/wiki/Correlation_does_not_imply_causation.

2) Jeff Zeleny, "Barbour Slams Obama on Economy and Energy," The Caucus: http://thecaucus.blogs.nytimes.com/2011/03/14/barbour-slams-obama-on-economy-and-energy/. The print edition version is shorter: "A Possible ‘12 Candidate Starts Testing Themes," New York Times ((3-15-11).

3) Alexandra Petri, "I Blame the SuperMoon," Washington Post (March 11, 2011): http://www.washingtonpost.com/blogs/compost/post/i-blame-the-supermoon/2011/03/03/AB7rJGR_blog.html