No causal comparison

cn: sexual assault and victim blaming are discussed briefly as an example.

Often we observe some phenomena or trend, and we wish to explain what caused it. Different people can disagree on the cause. Or perhaps they agree on the causes, but disagree on which causes are important. Bold claim: There is no objective way to assess the relative importance of two causes.

I’m making a purely abstract argument, but I’ll offer a few provocative examples:

1. Is a given human trait caused by genetics, or the environment?

2. Is personal success caused by hard work, or by lucky circumstance?

3. Is terrorism caused by politics, or by religion?

4. If a woman is victim of sexual assault, is that caused by the perpetrator, or by risky behaviors on her part?

5. Is our knowledge of physics the result of scientific research, or is it the result of the continuing absence of an earth-destroying supernova?

Among these examples, we’d obviously like to say that some causes are more important than others. We are welcome to say so, but there is necessarily an element of subjectivity in our words.

Causation and ethics

When we talk about the “importance” of a cause, this is often really a way to talk about ethics. In the case of a trend we don’t like, we’re talking about the best way to mitigate it. In the case of a trend we do like, we’re talking about the best way to amplify it. If you think ethics is subjective, then causality is subjective too.

Of course, some ethical questions can be transformed into empirical ones. For example, suppose a problem is caused by X and Y, and we have some amount of money to address the problem. Would it be more effective to spend that money on X or on Y? Or suppose that I am a blogger without money, and I only have my words. Would my time be better spent writing about X or about Y?

But what if money is more effective at addressing X, and blogging are more effective at addressing Y? Bloggers and charities might end up disagreeing on which causes are more important.

Another complication is that different kinds of solutions have different externalities. Sexual assault provides an obvious example, where attributing assault to the victim’s behavior places undue burden on victims, and on at-risk populations.  (It may also be the case that changing behavior will not really help at all.)

Scientific ways to understand causes

No doubt you’ve seen many journalists discuss the role of nature vs nurture based on scientific studies. If there is no objective way to compare the relative importance of nature vs nurture, then how are scientists doing it? Scientists are actually measuring a quantity called “heritability”.

Heritability is intended to capture an intuitive concept: the relative importance of genes in determining a particular trait.  However, the intuitive concept is incoherent.  In place of the intuitive concept, we substitute a technical definition: heritability is the proportion of trait variance that can be attributed to genetic variance.  Heritability is only defined relative to a particular population, and may be different for different populations considered.  For illustration purposes, I show a cartoon plot where both the trait and genes are one-dimensional variables.

HeritabilityEach point represents an individual within the population.  A linear regression with slope m is shown. σ(g) is the standard deviation in genes, and σ(t) is the standard deviation in the trait.  Heritability (H) is equal to m*σ(g)/σ(t).

While heritability might capture some sense of “importance”, it certainly does not capture the ethical sense of importance. For instance, suppose we wanted to know the best way to reduce a problematic trait. If studies show that the trait has high heritability, it’s still possible that the trait could be addressed by changing the environment. A high heritability could merely indicate that the current variation in environment is small within the given population. If you were to effect a larger change in the environment, then heritability says little about what would happen.

Philosophy of causation

Note that this post is about philosophy, and since I am not a philosopher I’m speaking outside of my field. It’s a good reality check to see what philosophers have said.

The Stanford Encyclopedia of Philosophy refers to the present issue as causal selection. How do we identify causes, and what distinguishes them from mere background conditions? J. S. Mill argued that there is no basis for causal selection, and people seem to select causes capriciously. And then there are various rebuttals to Mill which you can peruse at your leisure.

None of the rebuttals say that causal selection is an objective process. Instead, they argue that causal selection is systematic or predictable, given a particular context. This is in agreement with my own thesis that causation is context-dependent.

Many people are inclined to view causation as a basic, or fundamental property of the world. But in fact, cause and effect are very human concepts.


  1. invivoMark says

    I’m not sure I agree with your bolded claim as it is written. It’s an abstract claim, as you say, so maybe I’m not interpreting it in the way you intend it.

    However, cancer biologists spend a great deal of time determining the relative importance of causes of cancer. It turns out for overall cancer rates, genetics contribute ~15-25% of risk, while the majority of risk is due to random chance (Tomasetti 2015). For specific types of cancer, specific environmental factors (smoking, hepatitis, diet) or genetic factors (BRCA1/BRCA2) may play a greater role.

    But you kind of get at this point yourself, so I’m not sure if I’m telling you anything new. It’s true that this doesn’t give us a precise landscape of the relative importance of addressing one factor over another, but I think it’s clear that that landscape objectively exists. Using scientific exploration, we can get a pretty good idea how much good we can do by addressing the various causes of cancer (or alternatively by ignoring the causes and focusing on early diagnosis and effective treatment).

    So I guess I don’t really understand your point.

  2. says

    That sounds a bit like the example of heritability. Just like with heritability, how much cancer risk you attribute to genetic factors depends on your population. If you have a genetically homogenous population, genetics contribute 0% of the risk. If you had a more heterogeneous population, genetics could contribute a greater percentage of the risk.

    Thinking about it more mathematically, we can quantify the strength of a causal connection basically by a partial derivative (i.e. change in cancer risk over change in environment). However, to compare two different causal connections, scientists must multiply by another factor to achieve the correct units (i.e. change in cancer risk over change in environment multiplied by environmental variance). The multiplicative factor here is almost completely arbitrary, and is simply something that is easily measured in a scientific study. If you were actually interested in reducing cancer rates, you would likely multiply by a different factor entirely (i.e. change in cancer risk over change in environment multiplied by the environmental changes you could enact given a million dollars).

  3. invivoMark says

    Okay, I think I follow you. So your claim, if I understand it right, is that the relative importance of causes is dependent on context.

    I agree with that, but that seems a whole lot less provocative than “there is no objective way to assess the relative importance of two causes.” Maybe I’m not using the word “objective” in the same way you are?

  4. says

    It’s a fair criticism if you think the thesis doesn’t live up to the thesis statement. There are several of different ways to define “objective”, and here I am using it to mean “non-arbitrary” or “fundamental”.

  5. invivoMark says

    I think those terms make a lot more sense to me. Thanks for indulging my comments.

Leave a Reply

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