Yeah, this investigation seems like utter bullshit. They set up a fake company to investigate chinese products. They found chinese companies willing to offer to sell to this fake company. Is this supposed to be a surprise?
Then, when they purchase from these cheapo companies that do not do any checks for their front company's genuineness, they get substandard product. Again, is this supposed to be a surprise?
No one questions that it is possible to buy bad product from any country, including China. The important question is whether it is possible to purchase, reliably, good genuine product from China if all reasonable measures and safeguards are taken. That's the realistic case, and the only truly relevant one. That a GAO investigation specifically seeking to purchase substandard items succeeded in doing so should not be 'troubling' at all. It should be completely freaking obvious to everyone.
*Can anyone actually confirm that tinyurl has been blocked in Turkey?*
I'm saying this, because the linked article and the submission doesn't mention Tinyurl (or indeed Pastebin) at all, and Slashdot seems to be the only people stating this news. Is this something that's actually happened or are we spreading false rumours here?
'Factoring out your prior belief' in Bayesian statistics is like asking for water without the wetness. The best you can do is a prior sensitivity analysis that tells you what you have to believe in a priori to conclude a certain way, but the risk here is that you don't know if you are disbelieving in something because you are being reasonably skeptical, or just being stubborn.
There's a placebo (strictly speaking, nocebo) effect in that people may report (and to be fair, often genuinely feel) side effects solely because of the fact they've been given medication, instead of any chemical or biological reason.
"Why not just report likelihoods instead and let the reader multiply it with any prior they want? In many cases, the prior won't make much of a difference anyway, I suppose."
Reporting likelihoods (or rather summary statistics of likelihoods, because generally likelihoods are functions, not single spot values) is precisely what frequentist statistics is. The use of significance tests, p values and so on can be simply considered a standardised representation of the likelihood.
In the case of multiplying with any prior they want, in many (even most...) cases, the reader simply does not know what their prior is. And often, they do not even know, without a lot of work and some experience in statistical theory, what a 'sensible prior' is. For example, setting a flat prior for x (all values are equally likely a priori) can be actually *very* informative if later calculations make use of 1/x.
In most interesting cases, I'm afraid that the choice of prior *does* make a great difference. This is even more complicated if the prior is in the form of hyperparameters, in which case your prior choice can have a dramatic and non-linear effect on your result.
"Sure, do away with empirical Bayes. Anyway, I don't think "When using Bayesian methods, you run the risk of using non-Bayesian methods." is an argument for not using Bayesian methods."
But Empirical Bayes represents a huge chunk of how 'bayesian stats' is done in practice. The point here is that speakers don't define or understand clearly what is, or is not bayesian in the way that is proposed.
"Regardless of what practical obstacles there might be for using Bayesian inference, using something else would be wrong, leading to results that make you take the wrong actions!"
It depends on how the results are represented. Frequentist results are not 'wrong'. They represent a real value of the data. They may simply be irrelevant. And on the flip side, merely switching to Bayesian inference *does not prevent you from taking wrong actions*. What Bayesian inference accomplishes is that it shifts and makes explicit the mistake you are making.
For example, the false positive results I stated can be *exactly duplicated* by using a prior that is excessively permissive of finding positives, and that prior can look extremely reasonable. The real solution to these problems is to *understand what you are doing*, and this can be done in both a bayesian and frequentist way.
No, that would be the right calculation for arbitary combinations of a certain length. We're looking for the number of pairs (drug, side effect), of which there are #drugs*#sideeffects total.
GWAS done in the traditional hypothesis testing way have a *massive* multiple testing problem, which is only partially solved by imposing *extremely* stringent requirements on significance. That's why things like the Group Lasso and so on are being devised to work on them. It's definitely a non-trivial problem.
I agree that my calculations are very naive. But in my reading of the paper, I'm not really satisfied that they've done enough to deal with the multiple testing issue. (There is a mention of corrections for multiple testing, but that's in an unrelated section to the headline 'thousands of drug side effects' claim.) In the authors' defense, they don't claim that the effects they found are necessarily actually there. But the/. OP seems misleading.
I both agree and disagree. In principle people really should stop using frequentist methods. In practice, though... There's still substantial disadvantages with Bayesian methodologies. For example, off the top of my head:
1. Relative slowness of computational algorithms for large datasets 2. Difficulty of presenting results to people with different prior beliefs. (Strictly speaking, in Bayesian terms, the answer you give must always be relative to *someone*.) 3. Ease of 'cheating', even unintentionally, by choosing priors to favour a certain result. 4. Proliferation of methods that pretend to be Bayesian but are in fact probably not. (e.g. Empirical bayes methods)
I'm saying this because this always comes up, but people don't realise the bayesian approach is necessarily a magic bullet either.
I've only done a light skim of the paper, but it seems to me that the OP (but not the paper itself) is being way too positive here. Their methodology seems to be very vulnerable to false positives - with a massive database of drugs and potential adverse effects, you'd expect a *lot* of apparent side effects occuring solely by chance. For example:
"We constructed a database of 438,801 off-label side effects for 1332 drugs and 10,097 adverse events."
Supposing you are doing a hypothesis test at the standard 0.05 significance level, for each of the 1332*10097 drug-side effect combinations. Then, with naive assumptions, on a null hypothesis, you'd be picking up an average of 666k+ 'side effects' anyway, purely by chance. With the drug interactions case, this multiple testing problem gets even worse.
Now, there are ways to correct for multiple testing, but for things as large and complicated as this problem, I'm not sure the standard methods are going to cut it. At best, this study should be considered more a *filter* on the set of potential side effects, than really an enumeration of effects that are actually there. This is ignoring other issues like the placebo effect.
I think more importantly, while things could have gone wrong, the difficulties in the lunar landers were not *malicious* in nature. It's easy to make a system 'probably' safe in an environment of random threats, but in an environment that is actually actively hostile, that unlikely event of failure would rapidly become a certainty.
As TFA says, "more than half of the $45.8 million, about $25 million, would be spent to replace the district’s computers — both laptops and desktops.". So that comes down to 1520/student. More importantly, this is for a program of improvements over the next *ten years*, not an immediate replacement job - as the article argues that the >3 years old computers currently in use are obsolete, I assume the money might fund more than one cycle of improvements. At one cycle per 3 years, we're talking ~500 dollars per student, not accounting for inflation, which seems pretty sensible.
Anyway this all seems like a storm in a teacup.
Greenpeace does a crapload of things other than things relating to climate change. This is far from a valid comparison - you might as well compare Greenpeace's budget to 'skeptic organisation' ExxonMobile's revenues of $486.429 billion.
Uh... what?
Your argument makes no sense whatsoever. Even if you know absolutely accurately that you have 0.12311 probability of getting into an accident in the next 10 years, the increasing precision you have on that belief changes nothing about the fact that *you don't know whether or not you will actually get into an accident*. The point of individualisation is so that I, as a safe driver, would not have to pay extra money to subsidise the premiums of the unsafe drivers with a higher individual risk.
Hang on a minute, this article doesn't pass the smell test.
'More republicans than democrats have bailed on SOPA'. The question on your mind must be: couldn't this be because more republicans than democrats were supportive of SOPA to start with? Of the 12 initial sponsors of SOPA, 8 were Republicans, and the proposer himself was Republican. The opposition in the Senate was led by Wyden, who is, again, a Democrat.
Certainly the statements by the various Republican candidates all came a week *after* Obama spoke out against SOPA on 14th January, and notably after the wikipedia blackout. (While Obama's action came before.) It's not to say that republicans didn't take a role in fighting the bill later (especially after the wikipedia blackout), but the claim by republicans at this time to ownership of killing SOPA - a bill *they created* - seems to be pure propaganda.
This is a good point, but then again, you can imagine that in some cases you might need three manned planes to take the place of a drone - for example, in long period monitoring missions where crews would need to take time off to rest.
ARGH PLEASE. The Laffer curve is absolute garbage. GARBAGE. I don't care if I get downmodded for this, but it needs to be said. It's based on a fundamental delusion - that because you know that the two ends of the graph is at 0, the curve in between those two points must be some smooth determinable surface with a single obvious maximum. Reality follows the neo laffer curve. http://en.wikipedia.org/wiki/File:Neo-Laffer_curve.svg
Right, exactly. The US political system is already sufficiently opaque and unreactive. Why do people think giving random people the ability to gum it up even further is going to help things?
So, in revenge for politicians bothering you, you deny legitimate callers from being able to put their concerns to their elected representatives by DDOSing their phone service. Aren't you a hero.
Dark matter predicted the existence of things like the Bullet Cluster ( http://en.wikipedia.org/wiki/Bullet_Cluster ). I've yet to see a dark matter replacement that as elegantly results in phenomena like that.
Software is never going to completely defend your privacy, and definitely isn't going to defend *mass privacy*, that is the privacy of the millions upon millions of ordinary users who have never heard of your super-awesome encryption software. Only the 'legalware' of challenging government (and non-governmental) intrusion in the courts can ultimately defend your rights.
And no, I think it's absurd that writing encryption software entitles you to lead the struggle vs survelliance.
Slashdot Mirror
Citation needed.
Yeah, this investigation seems like utter bullshit. They set up a fake company to investigate chinese products. They found chinese companies willing to offer to sell to this fake company. Is this supposed to be a surprise?
Then, when they purchase from these cheapo companies that do not do any checks for their front company's genuineness, they get substandard product. Again, is this supposed to be a surprise?
No one questions that it is possible to buy bad product from any country, including China. The important question is whether it is possible to purchase, reliably, good genuine product from China if all reasonable measures and safeguards are taken. That's the realistic case, and the only truly relevant one. That a GAO investigation specifically seeking to purchase substandard items succeeded in doing so should not be 'troubling' at all. It should be completely freaking obvious to everyone.
I'm posting this up high so it'll be read.
*Can anyone actually confirm that tinyurl has been blocked in Turkey?*
I'm saying this, because the linked article and the submission doesn't mention Tinyurl (or indeed Pastebin) at all, and Slashdot seems to be the only people stating this news. Is this something that's actually happened or are we spreading false rumours here?
'Factoring out your prior belief' in Bayesian statistics is like asking for water without the wetness. The best you can do is a prior sensitivity analysis that tells you what you have to believe in a priori to conclude a certain way, but the risk here is that you don't know if you are disbelieving in something because you are being reasonably skeptical, or just being stubborn.
There's a placebo (strictly speaking, nocebo) effect in that people may report (and to be fair, often genuinely feel) side effects solely because of the fact they've been given medication, instead of any chemical or biological reason.
"Why not just report likelihoods instead and let the reader multiply it with any prior they want? In many cases, the prior won't make much of a difference anyway, I suppose."
Reporting likelihoods (or rather summary statistics of likelihoods, because generally likelihoods are functions, not single spot values) is precisely what frequentist statistics is. The use of significance tests, p values and so on can be simply considered a standardised representation of the likelihood.
In the case of multiplying with any prior they want, in many (even most...) cases, the reader simply does not know what their prior is. And often, they do not even know, without a lot of work and some experience in statistical theory, what a 'sensible prior' is. For example, setting a flat prior for x (all values are equally likely a priori) can be actually *very* informative if later calculations make use of 1/x.
In most interesting cases, I'm afraid that the choice of prior *does* make a great difference. This is even more complicated if the prior is in the form of hyperparameters, in which case your prior choice can have a dramatic and non-linear effect on your result.
"Sure, do away with empirical Bayes. Anyway, I don't think "When using Bayesian methods, you run the risk of using non-Bayesian methods." is an argument for not using Bayesian methods."
But Empirical Bayes represents a huge chunk of how 'bayesian stats' is done in practice. The point here is that speakers don't define or understand clearly what is, or is not bayesian in the way that is proposed.
"Regardless of what practical obstacles there might be for using Bayesian inference, using something else would be wrong, leading to results that make you take the wrong actions!"
It depends on how the results are represented. Frequentist results are not 'wrong'. They represent a real value of the data. They may simply be irrelevant. And on the flip side, merely switching to Bayesian inference *does not prevent you from taking wrong actions*. What Bayesian inference accomplishes is that it shifts and makes explicit the mistake you are making.
For example, the false positive results I stated can be *exactly duplicated* by using a prior that is excessively permissive of finding positives, and that prior can look extremely reasonable. The real solution to these problems is to *understand what you are doing*, and this can be done in both a bayesian and frequentist way.
No, that would be the right calculation for arbitary combinations of a certain length. We're looking for the number of pairs (drug, side effect), of which there are #drugs*#sideeffects total.
GWAS done in the traditional hypothesis testing way have a *massive* multiple testing problem, which is only partially solved by imposing *extremely* stringent requirements on significance. That's why things like the Group Lasso and so on are being devised to work on them. It's definitely a non-trivial problem.
I agree that my calculations are very naive. But in my reading of the paper, I'm not really satisfied that they've done enough to deal with the multiple testing issue. (There is a mention of corrections for multiple testing, but that's in an unrelated section to the headline 'thousands of drug side effects' claim.) In the authors' defense, they don't claim that the effects they found are necessarily actually there. But the /. OP seems misleading.
I both agree and disagree. In principle people really should stop using frequentist methods. In practice, though... There's still substantial disadvantages with Bayesian methodologies. For example, off the top of my head:
1. Relative slowness of computational algorithms for large datasets
2. Difficulty of presenting results to people with different prior beliefs. (Strictly speaking, in Bayesian terms, the answer you give must always be relative to *someone*.)
3. Ease of 'cheating', even unintentionally, by choosing priors to favour a certain result.
4. Proliferation of methods that pretend to be Bayesian but are in fact probably not. (e.g. Empirical bayes methods)
I'm saying this because this always comes up, but people don't realise the bayesian approach is necessarily a magic bullet either.
I am a statistician.
I've only done a light skim of the paper, but it seems to me that the OP (but not the paper itself) is being way too positive here. Their methodology seems to be very vulnerable to false positives - with a massive database of drugs and potential adverse effects, you'd expect a *lot* of apparent side effects occuring solely by chance. For example:
"We constructed a database of 438,801 off-label side effects for 1332 drugs and 10,097 adverse events."
Supposing you are doing a hypothesis test at the standard 0.05 significance level, for each of the 1332*10097 drug-side effect combinations. Then, with naive assumptions, on a null hypothesis, you'd be picking up an average of 666k+ 'side effects' anyway, purely by chance. With the drug interactions case, this multiple testing problem gets even worse.
Now, there are ways to correct for multiple testing, but for things as large and complicated as this problem, I'm not sure the standard methods are going to cut it. At best, this study should be considered more a *filter* on the set of potential side effects, than really an enumeration of effects that are actually there. This is ignoring other issues like the placebo effect.
I think more importantly, while things could have gone wrong, the difficulties in the lunar landers were not *malicious* in nature. It's easy to make a system 'probably' safe in an environment of random threats, but in an environment that is actually actively hostile, that unlikely event of failure would rapidly become a certainty.
As TFA says, "more than half of the $45.8 million, about $25 million, would be spent to replace the district’s computers — both laptops and desktops.". So that comes down to 1520/student. More importantly, this is for a program of improvements over the next *ten years*, not an immediate replacement job - as the article argues that the >3 years old computers currently in use are obsolete, I assume the money might fund more than one cycle of improvements. At one cycle per 3 years, we're talking ~500 dollars per student, not accounting for inflation, which seems pretty sensible. Anyway this all seems like a storm in a teacup.
Why on earth do you think the heartland institute comprises the entirity of skeptic funding?
Greenpeace does a crapload of things other than things relating to climate change. This is far from a valid comparison - you might as well compare Greenpeace's budget to 'skeptic organisation' ExxonMobile's revenues of $486.429 billion.
I can't wait for you to leave slashdot.
Uh... what? Your argument makes no sense whatsoever. Even if you know absolutely accurately that you have 0.12311 probability of getting into an accident in the next 10 years, the increasing precision you have on that belief changes nothing about the fact that *you don't know whether or not you will actually get into an accident*. The point of individualisation is so that I, as a safe driver, would not have to pay extra money to subsidise the premiums of the unsafe drivers with a higher individual risk.
You do realise the Republicans proposed the SOPA bill, don't you?
'More republicans than democrats have bailed on SOPA'. The question on your mind must be: couldn't this be because more republicans than democrats were supportive of SOPA to start with? Of the 12 initial sponsors of SOPA, 8 were Republicans, and the proposer himself was Republican. The opposition in the Senate was led by Wyden, who is, again, a Democrat.
Certainly the statements by the various Republican candidates all came a week *after* Obama spoke out against SOPA on 14th January, and notably after the wikipedia blackout. (While Obama's action came before.) It's not to say that republicans didn't take a role in fighting the bill later (especially after the wikipedia blackout), but the claim by republicans at this time to ownership of killing SOPA - a bill *they created* - seems to be pure propaganda.
This is a good point, but then again, you can imagine that in some cases you might need three manned planes to take the place of a drone - for example, in long period monitoring missions where crews would need to take time off to rest.
Labour's really far from the main cost in space research.
ARGH PLEASE. The Laffer curve is absolute garbage. GARBAGE. I don't care if I get downmodded for this, but it needs to be said. It's based on a fundamental delusion - that because you know that the two ends of the graph is at 0, the curve in between those two points must be some smooth determinable surface with a single obvious maximum. Reality follows the neo laffer curve. http://en.wikipedia.org/wiki/File:Neo-Laffer_curve.svg
Right, exactly. The US political system is already sufficiently opaque and unreactive. Why do people think giving random people the ability to gum it up even further is going to help things?
So, in revenge for politicians bothering you, you deny legitimate callers from being able to put their concerns to their elected representatives by DDOSing their phone service. Aren't you a hero.
Dark matter predicted the existence of things like the Bullet Cluster ( http://en.wikipedia.org/wiki/Bullet_Cluster ). I've yet to see a dark matter replacement that as elegantly results in phenomena like that.
Software is never going to completely defend your privacy, and definitely isn't going to defend *mass privacy*, that is the privacy of the millions upon millions of ordinary users who have never heard of your super-awesome encryption software. Only the 'legalware' of challenging government (and non-governmental) intrusion in the courts can ultimately defend your rights. And no, I think it's absurd that writing encryption software entitles you to lead the struggle vs survelliance.