How to think about how to think about causality

In suggesting “a socially responsible method of announcing associations,” AT points out that, as much as we try to be rigorous about causal inference, assumptions slip in through our language:

The trouble is, causal claims have an order to them (like “aliens cause cancer”), and so do most if not all human sentences (“I like ice cream”). It’s all too tempting to read a non-directional association claim as if it were so — my (least) favourite was a radio blowhard who said that in teens, cellphone use was linked with sexual activity, and without skipping a beat angrily proclaimed that giving kids a cell phone was tantamount to exposing them to STDs. . . . So here’s a modest proposal: when possible, beat back the causal assumption by presenting an associational idea in the order least likely to be given a causal interpretation by a layperson or radio host.

Here’s AT’s example:

A random Google News headline reads: “Prolonged Use of Pacifier Linked to Speech Problems” and strongly implies a cause and effect relationship, despite the (weak) disclaimer from the quoted authors. Reverse that and you’ve got “Speech Problems linked to Prolonged Use of Pacifier” which is less insinuating.

It’s an interesting idea, and it reminds me of something that really bugs me.

In the Gospel According to Rubin (which I pretty much follow religiously, except that I use graphs all the time, and Don almost never does), we never speak of the causes of effects, only the effects of causes. This makes sense, for all the usual reasons. (For example, why did my cat die? Because she ran into the street, because a car was going too fast, because the driver wasn’t paying attention, because a bird distracted the cat, because the rain stopped so the cat went outside, etc. When you look at it this way, the question of “why” is pretty meaningless,

On the other hand, I ask Why all the time. Part of me wants to make my speech more precise, to avoid asking Why which is a meaningless question, while part of me wants a statistical framework in which the Why question makes sense. (And, no, this has nothing to do with the Rubin vs. Pearl issue, as both those frameworks define causation in terms of potential outcomes.

P.S. Keith writes: Pearl’s latest paper did cover causes of effects. From Pearl:

In particular, the paper surveys the development of mathematical tools for inferring (from a combination of data and assumptions) answers to three types of causal queries: (1) queries about the effects of potential interventions, (also called “causal effects” or “policy evaluation”) (2) queries about probabilities of counterfactuals, (including assessment of “regret,” “attribution” or “causes of effects”) and (3) queries about direct and indirect effects (also known as “mediation”).

My reply (without having actually read the cited paper): yes, I’m sure Pearl is doing something relevant here, but I’m guessing that he’s not answering (or trying to answer) the “Why did my cat die?” sort of question. The “attribution” thing (see item 2 immediately above) is important and does seem like an intermediate step between causes of effects and effects of causes.

13 thoughts on “How to think about how to think about causality

  1. "The horse is pulling the cart. The cart is being pulled by the horse."

    I wouldn't put money on Joe Public interpreting the inverse any differently than the original.

    Unfortunately, accuracy may require an emotional gutting of the language: "The cart is moving forward. The horse is moving forward. It has not yet been determined if they are moving in the same absolute direction, or at the same speed, or if they are even attached. At least one contrarian has proposed the horse is in the cart, while another has proposed taxing axles AND hooves."

  2. I am a fan of Benjamin Lee Whorf's weak version of lingusitic relatively (i.e. the sign empty gas cans suggests less danger than the sign full gas cans) – but I don't see much if any difference here

    "Prolonged Use of Pacifier Linked to Speech Problems" versus
    "Speech Problems linked to Prolonged Use of Pacifier"

    "Lung cancer linked to smoking"
    "Smoking linked to lung cancer"

    grammer has – I believe – has a very weak effect on interpretations…

    Keith

  3. LeChatlier's principle tells us that, near a stable equilibrium, effects inhibit their causes. Particularly outside chemistry, stable equilibria aren't always a good assumption, but looking for causality running the "unexpected" direction is my knee-jerk reaction to correlations that come out with the sign opposite to what I expected. Often, it turns out to make some sense. (To establish causality, obviously, you want to do more work.)

  4. Helmer and Rescher wrote a paper over 50 years ago on explanation vs. prediction in what they term as "inexact sciences": On the Epistemology of the Inexact Sciences. You can get it here:
    http://www.rand.org/pubs/papers/2005/P1513.pdf

    I'm not a social scientist, but I have to deal with these issues in my work as a transportation engineer/planner, where we're trying to predict the effects of transportation improvements on congestion, air pollution, GHG emissions, etc. (and we're doing a very poor job of it). Part of the problem is that the transportation field seems unaware of what's going on in the social sciences. I hope you don't mind comments from an outsider.

  5. Mahoney and Goertz in their "Tale of two Cultures" article argue that this is one of the crucial differences between qualitative and quantitative methods.
    http://www.wsu.edu/~tnridout/mahoney_goertz20061….

    Mahoney has been trying to answer the "how far do we go back" question using set theory and formal logic recently – some interesting ideas there (I don't have a good link to that atm). Generally, the how far do we go back question has figured very prominently in historically oriented research.

    I think as social scientists, causes of effects questions are just too important to ignore – and even a lot of literature that formally does effects of causes is really interested in explaining outcomes (think all of the quantitative macro-development literature, for example).

  6. I had the same reaction as Keith, though the example that sprung to my mind was autism linked to vaccination/vaccination linked to autism. I'm sympathetic to the idea that we might interpret differently structured sentences differently, but I wonder if one of the parts of the relationship being an action or behavior (using a pacifier, smoking, getting vaccinated) trumps the impact of sentence structure.

  7. When I'm teaching student journalists how to talk about quantitative information, I tell them to state non-causal associations in both orders — one way in a topic sentence, the other for the details.

    "A study published by the University of Minnesota's Boynton Health Service confirms a link between certain student habits and academic performance. Based on data from surveys in 2007 and 2008 of 18,000 undergraduate students in Minnesota, the research found that lower GPAs often coincide with behaviors such as inadequate sleep and excessive television and computer use."

    This isn't the silver bullet for logical clarity, but it's something to try with mass media audiences.

  8. I had the same reaction as Keith, though the example that sprung to my mind was autism linked to vaccination/vaccination linked to autism. I'm sympathetic to the idea that we might interpret differently structured sentences differently, but I wonder if one of the parts of the relationship being an action or behavior (using a pacifier, smoking, getting vaccinated) trumps the impact of sentence structure.

  9. As I hear English, {problem} linked to {candidate cause} and {candidate cause} linked to {problem} are indistinguishable. If people understood how weak a functional association correlation is, we could simply assert "{A} correlated with {B}."

    What's needed is writers who could write "The risk ratio of {A} and {B} is 1.35. Statisticians regard risk ratios less than 2.0 insignificant."

    Never happen.

    Regards,
    Bill Drissel

  10. Attribution and causes of effects.
    Andrew, you correctly placed the analysis of "attribution" somewhere between "effect of
    cause" and 'causes of effect". In the latter,
    we ask: what caused this death? while in attribution
    we ask: "to what degree is this specific cause
    responsible for this observed death."
    It is really a question about the probability of
    a counterfactual (Y_x), conditioned on the outcome (Y).
    Formally, Find P(Y_x = y| X=x', Y=y')
    with x' and y' different from x and y, respectively.

Comments are closed.