Comment by klodolph | Hacker News Reader

Comment by klodolph | Hacker News Reader

Parent Full thread

There are a couple people picking on this research in the thread. This is a preliminary study, just a "bullshit test" to see if this woman is crazy or if she's not. It's obvious that she's not crazy, because all she's doing is sniffing t-shirts and she doesn't know who they belong to. The sample size is large enough, we have a p-value of no more than p=0.0012, and that's under the cynical assumption that she somehow knew that exactly seven of the twelve samples would have Parkinson's... something that the scientists didn't know.

So we know that something is going on here. The next steps are to control potential confounding variables and to determine the sensitivity and specificity of the test, if it's shown that she's not actually detecting some confounding variable. This is where you go double-blind, you use larger sample sizes, et cetera, now that you have money. You have money because the preliminary research was promising.

And we're not just getting a good test out of this. If there's an actual chemical that she's smelling, then there's some chemical process that's going on in people with Parkinson's which isn't happening in people without Parkinson's (or vice versa). Tracing these chemical pathways could give us clues to the etiology of the disease, which would be a REALLY BIG DEAL. Or maybe it's just a rabbit hole.

One of the great things about science is that we can determine whether a phenomenon is likely real without having to have the slightest clue what the mechanism might be. Evidence comes first, and this looks like solid evidence, and mechanism can come later; the evidence alone proves the existence of something interesting, and that's what's worth a follow-up.

Hypothesis generation is important, because it helps us design the next experiment, but this experiment is already very interesting.

> One of the great things about science is that we can determine whether a phenomenon is likely real without having to have the slightest clue what the mechanism might be.

You just have to ignore the people who shout "correlation is not causation" at every opportunity, appropriate or not.

Because "correlation is not causation" is just plain wrong. What is should be is "correlation does not imply causation". This is where science comes in. To answer the question, is this correlation because of causation?

The statement is only important to people doing statistical analysis not experimental science.

I think XKCD (like usual) sums it up the best, this time in the alt-text of [0]:

"Correlation doesn't imply causation, but it does waggle its eyebrows suggestively and gesture furtively while mouthing 'look over there'."

[0] - https://xkcd.com/552/

>correlation does not imply causation

It sure does. It might not prove causation, or it might not necessitate causation, but it very much implies it.

Somehow people forget that "imply" means: "indicate the truth or existence of (something) by suggestion rather than explicit reference".

In this -- the dictionary and everyday sense -- correlation DOES imply (suggest) causation. It just doesn't secure it.

https://en.wikipedia.org/wiki/Material_conditional aka implication or implies

Definitions differ depending on the context, both definitions are valid.

99.26% correlation is observed between the Divorce Rate in Maine, and the Per Capita Consumption of Margarine

http://www.tylervigen.com/spurious-correlations

You say this implies causation... I have my doubts in any sense of the word implies.

For sure, a correlation could lead to something to investigate, but look at enough data and you will find plenty of correlations that mean nothing. A lot depends on how the correlation is discovered (number of variables involved etc.).

>You say this implies causation... I have my doubts in any sense of the word implies.

Couples divorcing people their partner got fat on margarine?

Besides that's not the best way to check correlation charts. You first have to remove bias components influencing both curves, e.g. the mere act that both are rising over time.

When you do that, do they still match each other, e.g. following increases and decreases? I very much doubt so. So this plot doesn't actually show correlation -- just that both "increase" over time in a similar way.

The same kind of "same plot trends" happens or every set of things that e.g. both have an exponential growth curve -- but it's not correlation unless both change consistently as the other changes.

That can't be right. There are so many spurious correlations that obviously imply nothing.

There's got to be some other required factor before correlation can imply causation. Like "if there's reason to believe something is relevant, and there is correlation, then that implies causation. "

But a "reason to believe something is relevant" is a prejudged correlation.

imply, v. ... (transitive, of a person or proposition) to hint; to insinuate; to suggest tacitly and avoid a direct statement

>correlation does not imply causation

Perhaps we should say, “Correlation correlates with causation.”

(Because causation is a subset of correlation.)

Isn't the full version something along the lines of: "correlation does not imply causation, but merely hints at it"?

There should be a name for those people who are quick to shout out some truism (that is actually true) but used out of context to a cliched degree.

See also: "You're not Google's customer, you're the product"

See also also: argument from fallacy, aka the "fallacy fallacy": claiming that a conclusion is wrong because an argument supporting it is invalid.

Not disagreeing, but there is a fine-line between an argument being wrong and that the conclusion being wrong. Are you really right when you use an incorrect argument/process/information to arrive to the right conclusion? There's a whole epistemological debate to be had about the intersection of belief, knowledge and truth.

Reply on news.ycombinator.com