Not-so-recently in the sister blog

The role of covariation versus mechanism information in causal attribution:

Traditional approaches to causal attribution propose that information about covariation of factors is used to identify causes of events. In contrast, we present a series of studies showing that people seek out and prefer information about causal mechanisms rather than information about covariation. . . . The subjects tended to seek out information that would provide evidence for or against hypotheses about underlying mechanisms. When asked to provide causes, the subjects’ descriptions were also based on causal mechanisms. . . . We conclude that people do not treat the task of causal attribution as one of identifying a novel causal relationship between arbitrary factors by relying solely on covariation information. Rather, people attempt to seek out causal mechanisms in developing a causal explanation for a specific event.

Interesting.

This finding is supportive of Judea Pearl’s attitude that causal relationships, rather than statistical relationships, are how we understand the world and how we think about evidence. And it’s also supportive of my attitude that we should think about causation in terms of mechanisms rather using black-box reasoning based on identification strategies.

17 thoughts on “Not-so-recently in the sister blog

  1. > Rather, people attempt to seek out causal mechanisms in developing a causal explanation for a specific event.

    I think of it somewhat differently (of course, unlike the authors I don’t have experimental evidence to support my conjecture). We are motivated reasoning machines. We first find a causal relationship we like and then seek (sometimes) leverage correlative associations as we seek out mechanistic explanations to justify our preferred causal phenomena. Often that means we invent mechanistic explanations that aren’t terribly plausible.

  2. What are “mechanisms”? Are they an encoding of one’s beliefs of “this, then that”? Lombroszo mentions parts, processes, and functions. Is a mechanism deterministic, with possibly hidden logic? How can one talk about mechanism, and how can one qualify one’s belief in this or that hypothesis of mechanism? Is a mechanism a path?

  3. There’s no doubt that people are primed to seek causal explanations. However,there’s also plenty of evidence that people tend to systematically latch onto the *obviously wrong* kinds of causal explanations. Think fundamental attribution error, or causal explanations of regression to the mean (e.g. Dunning-Kruger effect, Kahneman’s fighter pilots). Of course, having a good causal model of a phenomena can be an incredibly powerful & beneficial thing, however, the converse is also (necessarily) true: having a bad causal model leads to theory blindness & all sorts of other dangers. So I think it’s generally the smarter choice to temper our judgement & talk more about “associations” & less about “causes”, precisely *because* of our instinct to seek causal explanations, not in spite of it.

    • psych –

      > So I think it’s generally the smarter choice to temper our judgement & talk more about “associations” & less about “causes”, precisely *because* of our instinct to seek causal explanations, not in spite of it.

      I wonder about that..

      I kinds disagree. Yes, we have an instinct to seek casual explanations but we have an instinct to impose theories of causality to explain the patterns of association we see. I don’t see how spending more time on associations is going to help with that, although certainly being more circumspect about assuming causal explanations would be beneficial. I don’t see how spending more time on associations is going to lessen our tendency to believe in faulty causal explanations. Seems to me the better thing to do is to be more disciplined in how we approach assigning causality.

      IMO, an important issue here is that people have a tendency to accept a connection between associations and causality without having a high standard of requiring a plausible mechanistic explanation.

      • You’re right,it is a cop out to say “I’m just talking about associations” while strongly implying causality, which is what we also do often. But that’s another bridge to cross. The way I see it, our mind are chariots driven by two horses: one seeks to create stories about everything that happens to us, and the other wants to run away from uncertainty. We need to onto the reins & steer, otherwise we end up like Skinner’s pigeons, engaging in blind ritualistic behavior.

      • This argues that mechanistic assumptions should be explicit – to be later confirmed, refuted, or at least understood as a grounding context. The question for me is what is the form for describing “mechanism” in a useful way.

    • Depends who “we” are.

      Are “we” statisticians looking at data from a single non-experimental study in isolation? Then yes, we should focus only on describing observed associations rather than quickly jumping to a causal explanation. Are “we” individuals going about our lives and seeing single instances that strike us as odd? Then sure, we should note that it seems weird but may or may not be indicative of any general causal pattern.

      But are “we” scientists looking at many phenomena observed in different contexts that nonetheless seem to have some common aspects? Or are “we” in a position where we need to decide whether or what type of intervention to perform (e.g., policy makers, doctors, engineers, marketers, educators, farmers, etc.)? Then we cannot avoid positing causal explanations. If we are scientists, it is our job to provide such explanations; this is why experiments are so useful! If we need to make a decision, we need to have some idea of the direction of causality. Knowing that foot size is associated with intelligence doesn’t help unless we know *why*; otherwise, schools would be in the business of making feet bigger.

      Our causal explanations can be wrong, but that doesn’t absolve us of the need for them. It just means we need to build our explanations carefully and be willing to abandon them when they don’t work.

      • Yeah, you’re right, we need causal explanations, I just think that our instinct to come up with them is over-active, based on research across many areas of psychology & plenty of personal anecdotes. Additionally, the real danger is that once we do come up with a causal explanation, it “sticks” & it’s really hard to think of different ones. To use a metaphor, the proponents of causal inference argue: “you’ll never get anywhere if you drive 0mph”, which is patently true, I just think our default proclivity is to drive 80 in a school zone.

  4. “we should think about causation in terms of mechanisms rather using black-box reasoning based on identification strategies.”

    Good science uses both strategies. The issue plaguing social and behavioral sciences isn’t the strategy they use to uncover relationships. The issue is in verifying those relationships through repeated testing. There’s nothing wrong with running a massive regression on a thousand columns of data to find relationships, or even in using p-values to determine which *might* be significant. The problem is accepting this significance at face value given the well-known caveats of the method.

  5. Interesting. I have a visualization paper coming out soon that seems related. We did an online experiment where we described a small set of causal models (DAGs) along with visualizations, both static and interactive, of contingency tables. We asked them to allocate probability to the candidate DAGs. We compared their responses to the posterior probabilities that each DAG generated the data (‘causal support’). We asked them to describe their strategies, and while most of them seemed to have some clue what kinds of covariation they should look for, the results were not so good. Plus visualizing the data didn’t seem to help much over giving them numbers.

  6. Andrew uses the phrase “is supportive of” twice in his response to what is in the sister blog. In each instance, the word “supports” would be preferred by any teacher of English–an active verb and fewer key strokes. On the other hand, is there a hidden and diffident hedging of bets? Paradoxically, just the other day, he recommended Alex Beam’s book, “The Feud,” a marvelous page-turner, full of active, not passive, language jousting by Nabokov, Wilson and Beam.

    • Paul:

      When I write books and articles I go through and clean up the writing. I usually wrote blog posts in just one draft, so sloppy expressions can find their way into the writing and not get removed. I agree that “supports” is better than “is supportive of.”

  7. Um, is the main finding here that people seem to know what ‘causal’ means? I guess one lesson for us data folk is that we may not need to spend quite so much bandwidth explaining how logic works (is XKCD 552 finally obsolete!?!), so we can focus more on demonstrating how probabilistic statements correctly connect to reasonably well-functioning native logic (perhaps along the lines of the research Jessica mentioned).

Leave a Reply

Your email address will not be published. Required fields are marked *