Hacker News Reader

Comment by sandofsky | Hacker News Reader

I can't see the full study methodology without an account. That said…

The challenge with running studies ethically is that you can't split your test groups into two groups and somehow induce cannabis use disorder and see if major depression, bipolar, etc results from it. Rather, you can see if the people who naturally develop the disorder also have those problems.

The flaw in this reasoning is that people who develop depression and similar issues are likely to self-medicate with cannabis. In related news, umbrellas do not cause rain.

View on HN

Indeed. Your reasoning was echoed in this critique from Prof David Curtis, Honorary Professor, UCL Genetics Institute.

> “Around 20% of young people in Denmark use cannabis but in this study fewer than 1% of the sample had been assigned a diagnosis of cannabis use disorder and these may well have been people who had a pre-existing susceptibility to depression and other psychiatric disorders.

> “Other studies have shown that cannabis use itself is not associated with increased risk of depression, so perhaps we are seeing that people with a pre-existing susceptibility to depression are more likely to be diagnosed with cannabis use disorder. This clearly does not mean that we should infer that cannabis somehow causes depression.

> “I don’t think that this study provides us with much information to help decide the extent to which cannabis use is, or is not, harmful.”

[1] https://www.sciencemediacentre.org/expert-reaction-to-study-...

Obviously this is not a real argument against what you said, but the same thing happened with smoking. The tobacco companies said that there might be some set of factors that both caused lung disease, and caused smoking. As opposed to the alternate hypothesis that smoking caused lung disease. As is generally thought now, umbrellas do not cause rain but smoking does cause lung disease. Cannabis is federally illegal in the United States (the same way manufacturing machine guns without a license is for example), so if you live there don't use it. In some other places, I'd highly advise you to consider only the known benefits of using cannabis (and actually known, not something from a crappy study that you never read past the abstract or something a stoner buddy told you), and weigh that against known and potential risks of using the drug. Treat any form of smoking, including smoking cannabis, with high suspicion, as it involves inhaling large concentrations of combustion products. In conclusion, I would suggest you treat cannabis like how a rational person would treat cigarettes in the 1950s.

> Obviously this is not a real argument against what you said, but the same thing happened with smoking.

They showed the link between smoking and cancer with a mountain of empirical evidence, including tests on mice. https://www.mskcc.org/news/how-do-cigarettes-cause-cancer

> Cannabis is federally illegal in the United States

Nobody cares anymore.

> I'd highly advise you to consider only the known benefits of using cannabis

Sugar has no health benefits and causes way more deaths. If you want to really change the world for the better, I suggest a war on ice cream.

> Treat any form of smoking, including smoking cannabis, with high suspicion, as it involves inhaling large concentrations of combustion products.

Finally, we agree: consume edibles.

> They showed the link between smoking and cancer with a mountain of empirical evidence, including tests on mice.

Yes, it does now. From what I've seen, people complain about how long it took for official action in labeling to take place on tobacco smoking, but it's up to them how much effort they want to put into preventing a repeat situation. For a somewhat related example, some people think the FDA puts so much effort into preventing a repeat of thalidomide that it's a net negative for the entire organization to exist at all, and suggest expanding the category of "legal but not (effectively) mandated" to include almost all proposed drugs. "Never again" is much harder and more costly than many people think, because you have to get it right (or err in the direction of preventing the type of outcome in question) every single time, but if a single thing gets through, it's all over.

Make your decisions as rationally as possible.

> I suggest a war on ice cream.

Ice cream has some pretty funny statistics that I don't really buy as making actual sense. I'd limit usage quite heavily.

>Nobody cares anymore.

This really depends on the context. Note the various news articles currently going around that highlight the increased scrutiny Musk faced when he smoked weed publicly on a podcast. It’s still federally illegal and that can still come with very real consequences.

> Cannabis is federally illegal in the United States (the same way manufacturing machine guns without a license is for example), so if you live there don't use it.

That's not even really true anymore. I live in Texas, which has virtually no legal weed (beyond medical marijuana for a very small number of specific conditions like epilepsy and ALS - it's not like other states where you can say "Hey doc, I've got a headache, can I get a marijuana prescription?"), and yet I can walk into a nice, clean, friendly store in a strip mall and pick up some D9 gummies and brownies, as well as some THCA bud that to me is indistinguishable from "normal" marijuana.

The fact is the 2018 farm bill that legalized hemp had enough loopholes that it essentially legalized marijuana as well. While a lot of this is in a gray area (particularly in the case of THCA bud which appears to me to be a pretty blatant misrepresentation of the law, but I ain't complaining), it's clear there is no appetite anywhere to crack down on these loopholes. For example, Florida considered a bill this year that would crack down on D9 products, but there was already enough entrenched business opposition that the bill was killed.

Weed is legal in the US. If you're worried about legal consequences, just pay a little more and go to a nice dispensary and you're fine.

The ATF recently (in the last few years) updated some of their forms to clarify that cannabis is still an illegal drug. In addition, while the DEA may not enforce as much as they used to, they have a clear position on the matter.

Meanwhile I know people who had cbd-infused lotion as they drove through Texas and had some very expensive legal fees to avoid a felony.

> Obviously this is not a real argument against what you said, but the same thing happened with smoking. The tobacco companies said that there might be some set of factors that both caused lung disease, and caused smoking.

Mental illness and substance use both affect/are effected by the Brain. It seems less likely that the same cause would affect both the Brain and the Lungs.

This precisely. Who’s going to utilize weed more? People going through mental anguish. You know what causes major depression anxiety and psychosis? Trauma. Read the ACE study. A cruel traumatizing environment and behaviors are far more to blame than cannabis.

How is correlation vs causation determined in other studies where a control group cannot be used?

It's important to recognize that we've never successfully answered Hume's general skepticism about the existence of causality as something that is real outside of our own minds. So causality can't be positively confirmed to exist at all, let alone be detected and confirmed through statistics.

Before you dismiss that is philosophical non-sense, causality is closely related to the "Arrow of Time" which is considered an unsolved problem in physics [0]. From what we've observed time appears to be the only asymmetric physical process, largely due to entropy (that is you can immediately tell if a video of a jar breaking is being played backwards because we don't expected a jar to "fall together"). There are Quantum processes, for example, that are time symmetric, that is playing a video of these processes would look the same reversed and forward.

That said, as someone who does a lot of statistics, in practice what we do is model the causal process with a directed acyclic graph and see how well our models behave under causal assumptions. These work okay for answering practical questions about does A cause B. By controlling for correlating variables we can see the impact on what we believe to be a causal variable when we condition on these other correlating factors.

Worth mention that nearly all of the work is done by linear models in practice.

0. https://en.wikipedia.org/wiki/Arrow_of_time

The comment above refers to Hume's 'induction problem' [1].

> we've never successfully answered Hume's general skepticism about the existence of causality as something that is real outside of our own minds

The quote seems to imply a strong consensus, but [1] disagrees.

The phrasing "the existence of causality as something that is real outside of our own minds" sounds like philosophical idealism [2]. One can take that view without discussing causation at all, seems to me. Forgive the following blunt questions: how is philosophical idealism either (a) useful or (b) testable? To me, currently, it seems to be neither.

Maybe I'm missing something? For those that get value from it, what are some practical take-aways you get out of thinking about Hume's 'problem of induction'?

I'm very much into philosophy, but I admit I don't see a lot of utility in digging into this outside of historical purposes. It feels to me that the tools from inferential statistics combined with some deeper study on causation are far more useful and enlightening than trying to figure out the historical context around Hume's writing here. I'm glad other people do it, for sure, but what I've skimmed doesn't "grab" me.

WRT causation, I often return to the idea of "no model is perfect, but some are useful". There are many times when we build useful models that -- when combined with proper experimentation -- very strongly indicate causation. The laws of physics, for example. As soon as people get involved, it gets messy.

[1]: https://plato.stanford.edu/entries/induction-problem/

[2]: https://plato.stanford.edu/entries/idealism/

I completely agree that looking at old epistemology isn't that useful; simply because more recent theorists will need to address the same issues, so generally refer to older theory within their work. If you want a deep understanding then you need to look at the source material, but as you mention this is mostly only interesting for gaining a historical perspective.

Unfortunately many of the people who create introductory science courses seem to mostly only be aware of Popper, so most students are aware of the importance of falsifiability but not of its problems. I suspect this is because reading Kuhn, Lakatos and Feyerabend leaves one with an impression that science isn't an objective endeavour and this is an uncomfortable truth when start from a belief that science is some infallible rationalist project. Personally it transformed my view from "trust the science"(TM) to "question the science, always." Science isn't always science unfortunately.

I don't think Hume's work suggests that causality isn't 'real' but that we can't prove causation using inductive methods. This isn't a problem unless our measure of 'truth' is proving some causal effect to be true. The solution is to pretend our models of causation are correct, run them and compare the results to observation. We never 'prove' the causal effect, but we judge its validity by how well it can predict future observations.

In my view the 'real' problem we currently have is that we cannot test hypothesis in isolation (Duhem-Quine thesis). Thus we rely on heuristics like "scientific consensus" to judge whether the hypothesis that make up a field are gaining predictive power or not. The solution to all these problems is to use the Lakatosian Method of Scientific Research Programmes to track a field's progress over time. Anything less than this just devolves into Kuhnian paradigms, which is basically a popularity contest to determine 'valid' scientific methods.

I think this 'real' problem can explain all the failures of modern science: the replication crisis, abuse of statistical methods, funding influencing results.

Try this: full PDF: https://static1.squarespace.com/static/5fa2eabb04edcf45a6512...

Don't bother trying to read the full text. They ask for all your information then you still can't read it.

> people who develop depression and similar issues are likely to self-medicate with cannabis

Note, the study speaks of risk of depression - depresion happens after they have cannabis use disorder. Hence "Subsequent" in the study title.

Isn't it possible to distinguish casuality between A and B from correlation by simply considering set sizes?

    !A & B
    A & !B
    A & B
    !A & !B
    A
    B
    !A
    !B

Could you explain which distributions you think are correlation and which are causation?

I hoped someone who knows statistics will just tell us, instead of me stretching my brain.

The famous phrase "correlation does not mean casuation" is not equal to "correlation is indistingushable from casuation", but that's how people seem to often treat it.

Ideally, the paper abstract should be clear about what can or can not be concluded. In the title they say "subsequent" as if suggesting casuality, but in the abstract they say "associated". Confusing for a layman such as me.

My intuition of how to distinguish correlation from casuation:

If cannabis and depression were symmetrically connected, then incease of chances for cannabis use between depressed was equal to the increase of chances for depression between cannabis users.

If cannabis increases chances for depression more that depression increases chances for cannabis, than we conclude cannabis tends to cause depression.

Formula: percentage of depressed between canabis users / percentage of depressed in general population > persentage of cannabis users between depressed / persentage of cannabis users in general population.

Something like that.

He-he, the sides of inequality in my formula are always equal :)

    (c&d/c)/(d/total) > (c&d/d)/(c/total)

    c&d*total/c*d > c&d*total/d*c

Looking at it further, it's indeed not so easy. If we have a statistics produced by some casual relationship, e.g.

    function takeMeasurement() {
      // c (the cause) itself has probability 0.5
      // d (the dependent) in absence of c has probability 0.5
      //   but the presence of c increases it to 0.7
      const c = Math.random(1.0) < 0.5;    
      const d = Math.random(1.0) < (c ? 0.7 : 0.5);
      return {c, d};
    }

one can calculate an equally plausible hypothesis of causality in opposite direction, which would produce an equal statistics:

    P(c) = 0.5
    P(d|!c) = 0.5
    P(d|c) = 0.7
    
    P(d) = ?
    P(c|!d) = ?
    P(c|d) = ?

    -------------------------------

    P(d) = P(c)*P(d|c) + P(!c)*P(d|!c) =
      = P(c)*P(d|c) + (1 - P(c))*P(d|!c) =
      = 0.5*0.7 + 0.5*0.5 = 0.35 + 0.25 =
      = 0.6

    P(c|!d) = P(c) * P(!d|c) / P(!d) =
      = P(c) * (1 - P(d|c)) / (1 - P(d)) =
      = 0.5 * 0.3 / 0.4 =
      = 0.374(9)
    
    P(c|d) = P(c) * P(d|c) / P(d) =
      = 0.5 * 0.7 / 0.6 = 0.35 / 0.6 =
      = 0.58(3)
    
    -------------------------------
    Bayes formula:
    
    P(a|b) = P(a) * P(b|a) / P(b)

a resulting function:

    function takeMeasurement2() {
      const d = Math.random(1.0) < 0.6;
      const c = Math.random(1.0) < (d ? 0.58333 : 0.374999);    
      return {c, d};
    }

Is it really hopeless? Maybe not? For example, a boundary case when d only happens in case of c? Are the hypotheses "c causes d" and "d causes c" still equally plausable?

no?

No because you can never know for sure that there isn't a latent factor C which is the real causal factor and A is merely correlating with it (if we presume causality to exist at all, which is a pretty standard assumption).

We can compose DAGs that control for the factors we do know about, but it's impossible to exhaust or even know all possible latent processes that are impacting the outcome.

Ok. But can statistics disprove casuality?

E.g. to show that A does not cause B, while B (or a latent factor B correlates with) causes A?

When people say that the study can equally be interpretted as cannabis disorder causes depression and as depression causes cannabis disorder, can't statistics show that these two hypothesys are not equal?

No so easy: https://news.ycombinator.com/item?id=37633222

That "whitey" effect people getting smoking weed and hash for the first few times is from a sudden release of glutathione, a powerful antioxidant stored in the liver, and a supplement Asian people have been known to take to whiten their skin.

Reply on news.ycombinator.com