• Hallucinogen
    250
    Could also post this in epistemology but I think I'll get more the reply I want here.

    I'm looking for the 'official' name for an epistemological/empirical error.

    Let's say that a correlates with c.
    b also correlates with c, but to an even greater extent.
    when b is controlled for, a no longer correlates with c.
    when a is controlled for, b still correlates with c.

    The error: someone says "a causes c".

    I would describe this as "misattribution of a correlated spurious variable".

    It seems to be like a statistical equivalent of affirming the consequent, but I'd really like to know the proper name, since it crops up as an error in people's arguments frequently.

    For example: "a man's looks (b) are a strong factor in attracting a woman (c)".
    "personality (a) also attracts women".
    >> turns out what people label as "personality" correlates strongly with a man's looks.
    The error: "personality attracts women".
  • counterpunch
    1.6k
    On the meaningful level - it seems like "the law of the excluded middle."

    Tertium non datur ("a third is not given")
  • TheMadFool
    13.8k
    Let's say that a correlates with c.
    b also correlates with c, but to an even greater extent.
    when b is controlled for, a no longer correlates with c.
    when a is controlled for, b still correlates with c.

    The error: someone says "a causes c".
    Hallucinogen

    For example: "a man's looks (b) are a strong factor in attracting a woman (c)".
    "personality (a) also attracts women".
    >> turns out what people label as "personality" correlates strongly with a man's looks.
    The error: "personality attracts women".
    Hallucinogen

    A causes B IFF
    1. There's a correlation between A and B (check)
    2. B doesn't cause B [to rule out reverse causation]
    3. There's no C that causes both A and B [to rule out third party causation]
    4. The correlation between A and B isn't a coincidence

    In the example you gave of what women are attracted to, a man's looks causes both a "good" personality and women to be attracted i.e. this is a case of third party causation and is a well-recognized pitfall in causal reasoning.

    One often-cited example is the apocryphal study that "proved" ice cream sales "caused" shark attacks. The ice cream sales correlated well with shark attacks of course. However, the real cause was the summer heat which made people buy more ice cream and also increased both the volume and frequency of swimming (in the oceans and seas).

    If you control for the high temperatures of summer (control for a man's looks) the correlation between ice cream sales (personality) and shark attacks (attractive to women) disappear. The summer thermometer readings (a man's looks) correlate with both ice cream sales (personality) and shark attacks (attractive to women). Simply put, to infer that ice cream sales (personality) causes shark attacks (attractive to women) is to fail to identify the real cause viz. the summer heat (a man's looks).

    This fallacy goes by the name the third cause fallacy aka ignoring a common cause fallacy.
  • fdrake
    5.8k
    Let's say that a correlates with c.
    b also correlates with c, but to an even greater extent.
    when b is controlled for, a no longer correlates with c.
    when a is controlled for, b still correlates with c.
    Hallucinogen

    In addition to correlation does not imply causation. And assuming you're reading that correlation causally by specifying a causal model:

    a correlates with c; could go a->c or c->a
    b correlates with c; could go b->c or c-> a
    a is conditionally independent of c given b; that says something like a->b->c or a->b<-c or a<-b->c, all undirected paths from a to c go through b.
    b is conditionally dependent upon c given a; that says there is at least one path in the causal diagram from b to c or from c to b that does not go through a.

    It looks like you're referring to the condition of confusing a collider (a->b<-c has b as a collider of a and c) for a fork (a<-b->c has b as a fork of a and c) or a chain (a->b->c) maybe?

    Looks->attractiveness
    personality->attractiveness
    looks->personality or personality->looks

    if looks-> personality then attractiveness<-looks->personality->attractiveness
    if personality->looks then attractiveness<-personality->looks->attractiveness

    The first has looks as forking personality and attractiveness as well as a chain to attractiveness through personality, the second has personality forking looks and attractiveness with a chain from personality to attractiveness through looks. So you might also be referring to the idea that a system can have more than one pathway from a variable to another through other variables, so blocking off one path through controlling/conditioning doesn't block off all of 'em just in case there's more than one.

    Whether someone's committing an error of causal reasoning depends on the adequacy of the causal model, though. You can't tell colliders from forks or chains through reason alone, or even correlation/data alone!
  • EnPassant
    665

    As is often the case a correlates with b but there is a hidden variable x that is unknown, such that
    x -> a -> b
    or
    x -> b
    or
    x -> a and x -> b

    These hidden variables or factors can be the cause of much confusion.
bold
italic
underline
strike
code
quote
ulist
image
url
mention
reveal
youtube
tweet
Add a Comment

Welcome to The Philosophy Forum!

Get involved in philosophical discussions about knowledge, truth, language, consciousness, science, politics, religion, logic and mathematics, art, history, and lots more. No ads, no clutter, and very little agreement — just fascinating conversations.