Pay attention to power law distributions | Hacker News Reader

Pay attention to power law distributions

jackealtman.com

54 pointsjackaltman33 comments

I agree with the point of you are making, but feel the need to question your use of the term 'wonder drug' when describing accutane, also known as isotretinoin. HN readers should know that isotretinoin is a profoundly dangerous chemotherapy drug which was only by coincidence discovered to cure acne. The side effects of isotretinoin can be permanent [1] to the extent that the drug's inventor Hoffman-La Roche stopped selling it in 2009 because it was losing so many lawsuits. Yet the FDA (and all the rest) refuses to withdraw this drug from market. It is a modern pharmaceutical scandal which is still in progress.

If a woman becomes pregnant while taking isotretinoin, her children WILL have birth defects. The drug also causes depression in far more cases than Roche's data suggest. The side effects are as diverse as they are serious – there exist entire forums dedicated to sufferers of this drug's permanent adverse effects.

This drug cures acne like nothing else we know. However, the same poorly understood mechanism that enables it to dry out one's skin also systematically dries one's joints of the hyaluronic acid that is so necessary for their lubrication. Since ceasing my treatment four months ago after researching causes for the cartilage damage in my right kneecap, I have been transformed from a healthy and sport-playing science undergraduate to finding even typing this message uncomfortable because of pains in my fingers. So much that I have taken for granted as a 21 year old has been taken away from me by this drug, and I'm not alone in this. I don't like to overdramatise things, and my situation should improve to some extent given time, but taking this drug is the worst decision I have ever made, and the young, impressionable people who are all-too-readily prescribed this drug must be made more aware of its risks.

Remarkably effective acne cure, yes. Wonderdrug, absolutely not. Calling it one is just disrespectful to those who have killed themselves [2] and whose lives it has damaged [3].

[1] http://en.wikipedia.org/wiki/Isotretinoin#Permanent_side_eff...

[2] http://rense.com/general32/scu.htm

[3] http://jacquelinefox.wordpress.com/2011/04/01/ibs-sufferers-...

Do you know what dosage your treatment consisted of?

Full 5 month treatment at slightly under 1mg/kg/day (60mg).

A second, unfinished, 10 month treatment aborted after around 7 months at 0.5mg/kg/day (40mg – I'd put on 10kg of muscle). The lower dose was to reduce the almost daily nosebleeds I was experiencing on 1mg/kg/day.

1 mg is an insane dosis. No wonder you had to endure that shit.

A study suggests that doses as low as 0.25-0.4 mg/kg are sufficient: http://onlinelibrary.wiley.com/doi/10.1111/j.1365-2133.2010.....

While I have to agree, 1mg/kg seems to be by far the most commonly prescribed dose at present. Perhaps that will eventually change in light of the recent research you've mentioned.

Yeah, no doubt. That's what bothers me; sometimes, your doctor is prescribing treatment that could harm you, even though they may not know it.

I've taken this drug, the potential side effects are definitely scary. I was lucky enough to only suffer from dry skin and increase sun sensitivity (which is still existing to this day).

I know that when I was taking it (that was 12 years ago), the dermatologist would only prescribe it as a last resort, and, in case of a girl/woman, would always proceed with a pregnancy test first.

I personally am happy I took this drug but I can fully understand it's not your case.

I'm genuinely happy for you. I know what it's like. When isotretinoin works — and it usually does — it has an amazingly powerful effect on one's self esteem.

Since you took the drug 12 years ago, Roche has been quite aggressive in its marketing of accutane for treating more mild acne – not necessarily as the last resort that you describe. I think it's reasonably well prescribed here in the UK, but I get the impression that it is handed out somewhat too readily in many nations, including even the USA.

The wikipedia references for IBD/Accutane association seem to be all case studies except for one population study. However, perusing google scholar shows several population studies that show no correlation: http://dermatology.co.za/images/IsotretinoinIsNotAssociatedW...

Why do you think it causes depression more than the Roche data suggests? The studies about depression and suicide are similarly inconclusive: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1463189/

Is the data from the Roche lawsuits available? I'd like to know what evidence is resulting in these multimillion dollar IBD/Crohn's cases.

I took two courses of Accutane after reading several research papers and population studies of side-effects. I came to the conclusion it's safe, and would be interested in seeing any data that shows otherwise.

You're right to mention the conflicting nature of much of the data available. I quite agree that there is noteworthy evidence both for and against the IBS link, but the fact that IBS lawsuits led Roche to pull the drug off the market in 2009 suggests that there must be some compelling evidence of a link.

It's anecdotally accepted that isotretinoin users have a tendency to avoid reporting depression so as to continue being prescribed the drug. It's common sense – of course there will be underreporting in cases where the reporting will prevent someone being prescribed a drug they wish to take.

I am the dermatologist referred to here. I am sorry this poster had such a bad experience with this medication, but I want to correct some things he/she said. First of all Accutane was not found by coincidence - it was very intentionally synthesized in the mid-late 60s. It has been known for years that high dose Vitamin A clears acne, but at doses so high it was harmful to the liver. So Accutane was developed to maintain the good effect on acne (which is to shrink sebaceous/oil glands) and get rid of the hepatotoxity (ill effect on liver). But like high-dose Vitamin A it is a teratogen, which is a medication that has a bad effect on an embryo. Lots of drugs have this same issue - when you pick them up at the pharmacy they come with a sticker or insert that says "don't take if you are or may become pregnant" but the focus is on Accutane because acne is such a common condition and mostly occurs in teens and early twenties, a sexually active child-bearing population. Many of the other drugs with this same issue are used for less common conditions in much older patients. And yes, it can be sun-sensitizing, like many meds. As for the mood effects: lots of studies looking at this - all designed as retrospective studies comparing group of adolescent acne patients treated with Accutane with acne kids treated with antibiotics (NOT compared to kids who have NO acne, and so have one less thing to be unhappy about). Studies consistently show the Accutane treated groups have LESS negative mood issues than their antibiotic-treated counterparts - basically because their acne is gone for junior high/high school/college. AND this prevents scarring - we can always prevent scarring, and can never really undo it. In fact , there are recent reports of patients suing for scarring that resulted from NOT being prescribed Accutane! I am sorry about all the lawyer trolling of the internet for business and am aware of all the online support groups around purported ill effects of various drugs - nothing I can do about it. It is true that Hoffman LaRoche stopped making name brand Accutane in 2009 - but only because they couldn't compete financially with the four other companies who entered the market with branded generics. And in fact a fifth company has just come out with another competitor this past month.

After the patient and parent watch the video, I go back in the room and spend as much time as they need with whatever questions they may have - and then for girls I excuse the mom/parent to the waiting room for a careful and confidential contraception discussion!

Hmmmm... I think the post is a bit fast and loose. I recall reading one study of happiness that concluded small purchases/investments were best because one some gets used to new things. Can't find a link right now, unfortunately.

Power laws have become a kind of popular science meme, but I don't think they are as common or powerful an explanatory tool as some would like I to believe. Or perhaps I've just been reading too much of Cosima Shalizi's blog.

right, that's the study that says you should buy new clothes one at a time because you get diminishing marginal happiness from each successive item.

it's possible that the prevalence of power laws in normal life have been overstated by people trying to drive home a point ("normal life" curves are definitely less slack than something like VC returns) but I still suspect that the majority of people underestimate them.

"Power laws have become a kind of popular science meme..."

The blog post is essentially restating the Pareto (80/20) principle, first espoused by management consultant Joseph Juran in 1941. In the blog post he calls it a power law -- the Pareto distribution is a a power law distribution -- but it's an observation that has been made before and is known by many.

Power law and the pareto principle are definitely very related, but I think one is an instance of the other. I'd argue that power law is one manifestation under the umbrella of pareto principle, as compared to a stepwise function where 80% of actions/people/whatever act one way, and then another class of 20% act dramatically different from the 80% but similar to each other.

VC fund returns aren't driven by 80/20 (although they may fit that description), but by the much smaller number of outliers.

You have it backwards. The Pareto principle is based on a Pareto distribution with specific coefficients. The Pareto distribution is a power law distribution. Ergo the Pareto principle is based on a power law distribution. There is nothing stepwise about a Pareto distribution.

Maybe VC returns follow 80/20 and maybe they don't. It isn't exactly a public dataset that we can readily get our hands on. The point I was trying to make in my original comment was that although power law distributions seem to be a popular topic of conversation for the past few years, they aren't exactly a new discovery. Given that wealth and income follow a Pareto distribution, perhaps it isn't that surprising that VC returns do too.

You're right, we shouldn't be surprised that VC returns looks like this, and it's well known that they do. The point I'm making in the post is that more mundane things our lives may look like this than we expect.

Pareto power laws can't be stepwise, the point I was trying to make is that 80/20 distribution can be, and they are related to but not necessarily the same as power laws.

To: johnconroy [dead] (I can't reply to that comment.)

Hi! I'm not a site administrator, so I don't know why you were hellbanned. I read your comments history list to see if I could found a cause (for example an extremely offensive comment, or spam) but I couldn't find anything annoying. []

I saw that a lot of your comments are oneliners. In this community we generally don't like them and they can get you a downvote. Try to expand them and add more details. This information may be useful for the other members and it's more probable to get a few upvotes.

For example: http://news.ycombinator.com/item?id=4161475

Says: "Bookmarked ..."

A better comment could be for example: (It depends a lot on your background.) "I work as a web designer. We have a giant banner with the html symbols at the office, but it's a pain in the back to find the correct one. We were tried to make short versions with the most used symbols but that wasn't useful. I will bookmark this right now."

Or sometimes it is good enough to upvote to the history (without any comment).

[]There are (probably) other reasons for hellbanning that can't be found reading the comments, for example massive sockpuppets.

I was left wondering:

- What the hell can be so complicated about "explaining an acne medication to a patient"? It can't be hard to use. Is it the risks associated? How to respond to numerous reactions and side effects?

- Why did it take a dermatologist (which would put her in the ~90th percentile of intelligence) so long to realize she was repeating the same thing over and over and over, and that this could therefore be automated?

- If a video -- not interactive tutorial -- fully replaces everything her in-person explanation provides, then she must not have been getting questions from patients, which makes me wonder how much of it they were actually following and was informative to them.

> Why did it take a dermatologist (which would put her in the ~90th percentile of intelligence) so long to realize she was repeating the same thing over and over and over, and that this could therefore be automated?

IANAL & IANAD, but I'd wager small sums of real money that there are tons of legal implications around a doctor explaining how to use a medicine vs handing it off to a machine or even a nurse or other medical practitioner. Think: "my doctor didn't properly explain how this medicine would interact with this other medicine and as a result I was harmed, now I feel that my doctor should be held liable." Were I a doctor, that would probably be the very first thought in my mind immediately following how to automate an otherwise time-consuming process.

There are also considerations about ease of access. To a technologically lay person, the perceived complication and costs around a professionally edited instructional video can seem insurmountable or not cost-effective.

It's complicated to go through all the allegations of risk, which are almost all unsupported by serious studies. The birth defects are absolutely a huge risk, but the allegations of depression etc are unfounded. Most dermatologists agree that those studies were supported by accutane competitors, and that fair studies show the drug actually reduces depression (by removing the stressing factor of acne).

Because most doctors are not focused on tech. There are probably a lot of smart things you could do that have nothing to do with tech that you just wouldn't think about.

The video doesn't replace everything. It covers the bases and then sometimes a small number of questions remain. She used to get a ton of interrupting questions - she knows the common questions and touches on all of them in the video. Sometimes one-off questions remain.

A recent article in nytimes discussed research that estimates the cost of some government regulations on businesses. It's done by first noticing that company size follows a power-law distribution and then measuring deviations. I wonder if you were able to get enough data on investment returns if you would find any interesting deviations?

http://economix.blogs.nytimes.com/2013/01/02/why-49-is-a-mag...

Both Steven Johnson's "Where Good Ideas Come From" and Steven Pinker's "Better Angels of our Nature" have fantastic discussions on many other occurrences of power law distributions.

This book [1] also goes through a quite a few examples of power law distributions. It also goes through a large number of examples of random walks and the associated concept of Lévy flights [2].

[1] Barabási, Albert-László (2010). Bursts : the hidden pattern behind everything we do. New York N.Y.: Dutton.

[2] http://en.wikipedia.org/wiki/L%C3%A9vy_flight

The hard part is figuring out which 1% of our investments will yield 99% of our returns.

I would argue that no one actually figures that out. The obvious implication of the fact that VC returns follow a power law is that VCs have no real expertise in picking winners, but are merely riding along on the underlying structure of the market. They invest in X number of companies and one of them hits so big for reasons no one can consistently explain that it covers the losses on the remainder. The rest is just ritual and hand waving.

that could definitely be right. There do seem to be a number of funds that consistently hit winners more often year after year (say, something like 15% winners vs. 10%)

an alternative explanation to the best funds being good at finding winners is that they are good at avoiding the investments that have a low chance of being homeruns. this is where I was heading with my last paragraph but realized it requires a whole new post.

Citation? What funds are better at picking winners consistently year after year?

I think Sequoia, YC, USV and some others have probably had statistically significant outsized returns. Always hard to know because if you let enough people try someone is going to flip a coin heads 10 times in a row, but I do think some funds have created sustainable advantages (through winner-picking skills or otherwise).

That may be because they are big. (YC may not be big in terms of money invested, but they are in terms of number of startups.)

It may also be because they somehow help those startups succeed. Thus, they don't pick winners, they make winners.

Another possibility is that successful startup founders self select themselver into those founds.

This is a really interesting question. Humans tend to be pretty good at accepting things that are mostly random, or mostly skill-based, but have a surprisingly hard time with things that involve a combination of skill and luck. For a great example of this, watch amateur poker players.

This is just as true in investing-people tend to espouse either "all skill" (e.g. Warren Buffett") or "all luck" (one common interpretation of the Efficient Market Hypothesis.) Even defining skill in investing is difficult. High-risk investments tend to have a high return, so skill is getting an outsized return compared to the amount of risk involved. But the simple measures of risk (e.g. Beta) are not good actual measure of risk. For example, a company whose stock price suddenly plummeted is often a good investment since the market tends to overreact to bad news. Furthermore, the stock is now lower-risk because it has less to lose. But it will appear extremely high risk by most common measures.

I think it's a decent assumption in VC to treat all investments at stage X as equally risky. E.g. angel investments are one category, series A as another. This is obviously not 100% perfect but is a decent approximation.

Then we can ask the following question: how did Sequoia Capital's series A investments in a certain time period (say 2000-2005) perform compared to all other series A investments? It's important that you look at all of a VC's investments and not just a particular fund, because firms will often bury their bad funds.

I have no idea if that kind of data is available anywhere, though.

That's a good start, but even if no firm outperformed the average, there might still be skill involved. The question is, what is the a priori probability of success for a given class of investment? To answer that, you'd have to account for investment opportunities that weren't taken by anybody. But of course there's no way to know whether an unfunded startup would have been successful. So the VC community can rest easy knowing that they are insulated from objective evaluation, unless a VC emerges whose sole criterion for funding is that everyone else turned the investment down.

PG has a name for trying to capture the potential of the large-payoff potential of investments: "Farming black swans" [1].

[1] http://www.paulgraham.com/swan.html

Reply on news.ycombinator.com