This guy has evidence which your model doesn't account for. You're saying that the evidence can't be right because it isn't accounted for by your model?
That's not science, that's politics.
If he's got evidence, either counter with your own evidence or show that his evidence is fabricated.
Try actually being a scientist, instead of pretending to act like one.
I'm saying I am very skeptical of the "evidence" because it makes no fucking sense at all. Anybody can find statistically significant, completely spurious correlations when given a large-enough mass of data. Would you also suggest that I take these guys seriously?
I never said that the Purdue people shouldn't publish their result. Their paper simply notes a correlation. They don't claim to know why there is a correlation, and there could be many explanations. That's science. The most likely explanation is that the effect is a systematic. I say this because I know many other well-verified facts about how the world works, and this purported correlation is in conflict with all of these things. That's also science.
Uncritically accepting one piece of data and therefore throwing out a century of scientific knowledge is not being a scientist. It's being a nutjob.
This has to be either a systematic or a fluke. The only thing that could conceivably have an influence on nuclear decay rates is the neutrino flux, which would not show the diurnal variations that they claim, and which furthermore would be completely uncorrelated with solar flares, since neutrinos propagate at the speed of light from the solar core through the envelope, while thermal effects take millenia to propagate.
The paper on the effect is in a peer-reviewed journal, and the authors do not appear to be crackpots, but I notice that the abstract at least does not quote a confidence level for the result. And using an effect this speculative to base a solar weather prediction technology on, however, is pure idiocy.
So where's the news here? This nut was never a credible climate scientist in the first place, and I don't think any of his previous views were shared by anybody who is a credible climate scientist.
Lovelock makes a living out of making sensational, half-baked pronouncements and selling them as science. Good for him for admitting he was wrong, but that doesn't discredit any of the actual science.
Second that. Modern FORTRAN kicks some serious butt and has a huge user and support base. Language snobs dismiss it as antiquated but they're usually referring to versions of the language that haven't been used since the 1980's.
c Any language that still relies on line formatting conventions left over from punch cards
c should be taken out back and shot.
"Increasing sample size increases your real confidence only to the point where your error ceases to be dominated by statistical fluctuations and becomes dominated by systematics." --> I have no idea what this sentence means.
All measurements have systematic error at some level, and the systematic error provides a fundamental lower limit on how much you can increase the accuracy of your measurement by increasing sample size. Just how big the systematic error is for any measurement can be devilshly hard to estimate: there is no completely objective way to do so. But one thing for sure is that they are always there at some level.
There is no reason whatsoever to assume that error in any individual's performance on an IQ test is an uncorrelated variable, even if the difference in performance of different individuals averaged over a large populate are known to be uncorrelated. In fact, this hypothesis is in practice impossible to objectively verify: you can't give the same person the same exam a hundred times because the results will be certainly correlated (a systematic), and if you give one person a hundred different tests, then variations in the exams or the conditions under which the exams are taken (morning or afternoon?) will dominate the variation -- also a systematic. For this reason, it is probably pretty reasonable to estimate the order of magnitude of the systematic by the variation in any individual's score when taking the test multiple times. (This is the analog of the millimeter-graded ruler: I can't measure distances smaller than about 0.01 mm, no matter how hard I try, or how many times I do the measurement.) Maybe you disagree, but then you need to substantiate why you expect the size of the systematic error when giving one person one test is smaller than a point or so.
That kind of fine-grained parsing of IQ tests is silly. IQ tests do measure something: if somebody is a couple of standard deviations above the mean for the population, that is a meaningful thing. But if two people differ by a 1.3 points, it means absolutely nothing. Because the individual variation in test scores is very likely systematics-dominated, you can take 2 -> N without changing this conclusion. Note that the the actual research paper concludes that the measured correlation, while formally statistically significant, indicates that any genetic component of intelligence must involve multiple genes exactly because the measured correlation with single genes is so small. It's just the press and the Slashdot summary that turn this into a "gene for intelligence".
The point is that by assuming the measurement error has expected value of zero (and is not correlated with true IQ) we're able to average it out with large N. Two different populations (differing by, say, the presence or absence of a gene) can then be tested using a Chow test for the difference of two means, where confidence will be increasing in N. I'm not trolling here, I'm genuienly curious as to what you think is incorrect in the above working.
As soon as you assume that the measurement error is zero-mean and uncorrelated, you are for all intents and purposes assuming a Gaussian distribution, by the Central Limit Theorem. Increasing sample size increases your real confidence only to the point where your error ceases to be dominated by statistical fluctuations and becomes dominated by systematics.
You are confusing the distribution of the errors with the distribution of the data.
I'll try one more time before I give up. Suppose I wish to measure the size of an atom using a ruler graduated in millimeters. The answer I will get is zero plus or minus (really optimistically) 0.01mm. If I do the measurement million times, that doesn't mean my error is plus or minus 10 nanometers.
The errors only need to have the same law and be independant. No need for them to be Gaussians for the standard deviation of the average of all errors to be equal to the standard deviation of the error of a single measure divided by the square root of the number of measures.
But then the standard deviation doesn't measure anything meaningful. You can calculate it, but it doesn't tell you anything.
Suppose I do an experiment which returns a uniform random value between zero and one. If I perform that measurement repeatedly, I will determine that the average of that measurement is 0.5, and my error on that average will scale like one over the square root of the number of measures, but the real uncertainty on the measurement itself will still be plus or minus 0.5: I am measuring nothing at all, despite the apparent precision of the measurement.
IQs are gaussian by definition. The question isn't whether the statistics are valid here. The question is whether they're biologically meaningful.
The distribution of IQs across a population is Gaussian. The error in any individual IQ measurement is highly unlikely to be Gaussian below a certain level of accuracy. Completely different things.
But I don't think I disagree with your conclusion: just because somebody quotes a small P-value doesn't mean that the effect is real.
Repeat after me: only for Gaussian distributions, only for Gaussian distributions, only for Gaussian distributions. Any measurement can be assigned a standard deviation, but that doesn't make it Gaussian.
If the errors are not Gaussian, the situation is completely different. For example, if the variation in individual measurements follows a flat distribution instead of a Gaussian distribution, then averaging large numbers of measurements accomplishes exactly nothing. For any measurement, if your margin of error is small enough, the Gaussian approximation breaks down and your accuracy does not increase even if your nominal precision does. Mistaking precision for accuracy is a ubiquitous statistical fallacy.
You know about hypothesis testing, right? With enough samples any difference in mean can be distinguished to any arbitrary confidence interval, and any sample variance . Note: this can mean LOTS of samples.
Only if the errors can be added in quadrature. Which these almost certainly can't.
Whoops. Typo there.
For a standard deviation of 3 for one measurement, Gaussian statistics will give you a standard deviation of 0.03 for 10,000 measurements, and 0.003 for 1,000,000 measurements.
You are thinking about the accuracy of an individual measurement, when averaging large numbers with and without the gene you can get a much greater level of accuracy.
Precision is not accuracy. The standard deviation on IQ tests is about three points, but that does not mean that by averaging 1,000,000 IQs you can detect effects as small as 0.03 points -- the test is fundamentally incapable of measuring effects that small in the first place.
If your measurement is bad in the first place, averaging large numbers of measurements accomplishes nothing except giving you a false sense of accuracy. A huge pile of shit statistics is still shit.
Oh, for crying out loud. IQ tests must have a bigger measurement error than plus or minus 2, which means that the 1.29-point alteration is smaller than the measurement error. I.e., no effect.
There is one possible exception, the neutrino is a half spin fermion and if it really is zero mass it would be its own anti-particle.
Actually it's the other way around: massless Fermions are Dirac, because of Chiral symmetry: in the Standard Model with massless neutrinos, all neutrinos are Dirac particles, with neutrinos being left-handed and all antineutrinos being right-handed. Mass terms break chiral symmetry, and a massive neutrino could be either Dirac or majorana depending on how the mass term is generated:
A particle that is its own anti-particle? Sounds pretty special! Of course, that would also describe photons, the commonest particle in the universe.
Come on, science reporting.
Photons are bosons. Bosons being their own antiparticle is nothing unusual. A fermion that is its own antiparticle has never been observed in nature before.
This is not like finding the Higgs Boson. The majorana fermion they created was (hard to tell exactly how from TFA) a condensed matter excitation with the properties of a majorana fermion, not a fundamental particle.
Pretty cool though.
What has been proposed, and should be technically feasible, is dividing the array up by frequency band. The plan already calls for three overlapping arrays of different types of telescopes in order to capture three different frequency bands. (Phased array dipole antennas work great at 100 MHz, whereas you need dishes for 10 GHz.) In principle, one could put the low- and mid-frequency arrays on one site and the high-frequency arrays on another.
Note that MEERKAT is going ahead in South Africa regardless of the outcome of the decision on SKA, and will operate from 0.5-14 GHz.
If they do this then the likely outcome is that SKA will never happen. This sort of "compromise" as a way of avoiding having to make an actual decision is almost always the first step in a death spiral for the project. Case in point: the Joint Dark Energy Mission, which crashed and burned due to pointless infighting between erstwhile collaborators on the mission.
Looks like the Elsevier boycott by academics had an effect. Still, this looks like more of a tactical response than a real change in position for Elsevier.
Let me be the first to say: Fuck you. I don't have to meet your criteria for intellectual seriousness to get access to the fucking internet, any more than I need to convince a librarian I am Serious(TM) enough to read a book that somebody doesn't like.
My previous (US) passport was beat completely to shit by the time it was close to expiration. I got chewed out once in Germany by a passport agent who disapproved, and nearly didn't get let into Egypt at all. The martinet checking passports in Cairo was most offended by my treatment of the document. But I remember returning to the U.S. from Cairo, and the agent at JFK turned the passport over in his hand and said, in a thick New York accent, "Whadya, put a cigarette out innit?" and that was it. I knew I was home.
Slashdot Mirror
Okay, wait.
This guy has evidence which your model doesn't account for. You're saying that the evidence can't be right because it isn't accounted for by your model?
That's not science, that's politics.
If he's got evidence, either counter with your own evidence or show that his evidence is fabricated.
Try actually being a scientist, instead of pretending to act like one.
I'm saying I am very skeptical of the "evidence" because it makes no fucking sense at all. Anybody can find statistically significant, completely spurious correlations when given a large-enough mass of data. Would you also suggest that I take these guys seriously?
I never said that the Purdue people shouldn't publish their result. Their paper simply notes a correlation. They don't claim to know why there is a correlation, and there could be many explanations. That's science. The most likely explanation is that the effect is a systematic. I say this because I know many other well-verified facts about how the world works, and this purported correlation is in conflict with all of these things. That's also science. Uncritically accepting one piece of data and therefore throwing out a century of scientific knowledge is not being a scientist. It's being a nutjob.
This has to be either a systematic or a fluke. The only thing that could conceivably have an influence on nuclear decay rates is the neutrino flux, which would not show the diurnal variations that they claim, and which furthermore would be completely uncorrelated with solar flares, since neutrinos propagate at the speed of light from the solar core through the envelope, while thermal effects take millenia to propagate.
The paper on the effect is in a peer-reviewed journal, and the authors do not appear to be crackpots, but I notice that the abstract at least does not quote a confidence level for the result. And using an effect this speculative to base a solar weather prediction technology on, however, is pure idiocy.
So where's the news here? This nut was never a credible climate scientist in the first place, and I don't think any of his previous views were shared by anybody who is a credible climate scientist.
Lovelock makes a living out of making sensational, half-baked pronouncements and selling them as science. Good for him for admitting he was wrong, but that doesn't discredit any of the actual science.
Second that. Modern FORTRAN kicks some serious butt and has a huge user and support base. Language snobs dismiss it as antiquated but they're usually referring to versions of the language that haven't been used since the 1980's.
c Any language that still relies on line formatting conventions left over from punch cards
c should be taken out back and shot.
Meant to include a link: http://en.wikipedia.org/wiki/Systematic_error
"Increasing sample size increases your real confidence only to the point where your error ceases to be dominated by statistical fluctuations and becomes dominated by systematics." --> I have no idea what this sentence means.
All measurements have systematic error at some level, and the systematic error provides a fundamental lower limit on how much you can increase the accuracy of your measurement by increasing sample size. Just how big the systematic error is for any measurement can be devilshly hard to estimate: there is no completely objective way to do so. But one thing for sure is that they are always there at some level.
There is no reason whatsoever to assume that error in any individual's performance on an IQ test is an uncorrelated variable, even if the difference in performance of different individuals averaged over a large populate are known to be uncorrelated. In fact, this hypothesis is in practice impossible to objectively verify: you can't give the same person the same exam a hundred times because the results will be certainly correlated (a systematic), and if you give one person a hundred different tests, then variations in the exams or the conditions under which the exams are taken (morning or afternoon?) will dominate the variation -- also a systematic. For this reason, it is probably pretty reasonable to estimate the order of magnitude of the systematic by the variation in any individual's score when taking the test multiple times. (This is the analog of the millimeter-graded ruler: I can't measure distances smaller than about 0.01 mm, no matter how hard I try, or how many times I do the measurement.) Maybe you disagree, but then you need to substantiate why you expect the size of the systematic error when giving one person one test is smaller than a point or so.
That kind of fine-grained parsing of IQ tests is silly. IQ tests do measure something: if somebody is a couple of standard deviations above the mean for the population, that is a meaningful thing. But if two people differ by a 1.3 points, it means absolutely nothing. Because the individual variation in test scores is very likely systematics-dominated, you can take 2 -> N without changing this conclusion. Note that the the actual research paper concludes that the measured correlation, while formally statistically significant, indicates that any genetic component of intelligence must involve multiple genes exactly because the measured correlation with single genes is so small. It's just the press and the Slashdot summary that turn this into a "gene for intelligence".
The point is that by assuming the measurement error has expected value of zero (and is not correlated with true IQ) we're able to average it out with large N. Two different populations (differing by, say, the presence or absence of a gene) can then be tested using a Chow test for the difference of two means, where confidence will be increasing in N. I'm not trolling here, I'm genuienly curious as to what you think is incorrect in the above working.
As soon as you assume that the measurement error is zero-mean and uncorrelated, you are for all intents and purposes assuming a Gaussian distribution, by the Central Limit Theorem. Increasing sample size increases your real confidence only to the point where your error ceases to be dominated by statistical fluctuations and becomes dominated by systematics.
You are confusing the distribution of the errors with the distribution of the data.
I'll try one more time before I give up. Suppose I wish to measure the size of an atom using a ruler graduated in millimeters. The answer I will get is zero plus or minus (really optimistically) 0.01mm. If I do the measurement million times, that doesn't mean my error is plus or minus 10 nanometers.
The errors only need to have the same law and be independant. No need for them to be Gaussians for the standard deviation of the average of all errors to be equal to the standard deviation of the error of a single measure divided by the square root of the number of measures.
But then the standard deviation doesn't measure anything meaningful. You can calculate it, but it doesn't tell you anything.
Suppose I do an experiment which returns a uniform random value between zero and one. If I perform that measurement repeatedly, I will determine that the average of that measurement is 0.5, and my error on that average will scale like one over the square root of the number of measures, but the real uncertainty on the measurement itself will still be plus or minus 0.5: I am measuring nothing at all, despite the apparent precision of the measurement.
IQs are gaussian by definition. The question isn't whether the statistics are valid here. The question is whether they're biologically meaningful.
The distribution of IQs across a population is Gaussian. The error in any individual IQ measurement is highly unlikely to be Gaussian below a certain level of accuracy. Completely different things.
But I don't think I disagree with your conclusion: just because somebody quotes a small P-value doesn't mean that the effect is real.
No, see Margin of error
Repeat after me: only for Gaussian distributions, only for Gaussian distributions, only for Gaussian distributions. Any measurement can be assigned a standard deviation, but that doesn't make it Gaussian.
If the errors are not Gaussian, the situation is completely different. For example, if the variation in individual measurements follows a flat distribution instead of a Gaussian distribution, then averaging large numbers of measurements accomplishes exactly nothing. For any measurement, if your margin of error is small enough, the Gaussian approximation breaks down and your accuracy does not increase even if your nominal precision does. Mistaking precision for accuracy is a ubiquitous statistical fallacy.
You know about hypothesis testing, right? With enough samples any difference in mean can be distinguished to any arbitrary confidence interval, and any sample variance . Note: this can mean LOTS of samples.
Only if the errors can be added in quadrature. Which these almost certainly can't.
Whoops. Typo there. For a standard deviation of 3 for one measurement, Gaussian statistics will give you a standard deviation of 0.03 for 10,000 measurements, and 0.003 for 1,000,000 measurements.
You are thinking about the accuracy of an individual measurement, when averaging large numbers with and without the gene you can get a much greater level of accuracy.
Precision is not accuracy. The standard deviation on IQ tests is about three points, but that does not mean that by averaging 1,000,000 IQs you can detect effects as small as 0.03 points -- the test is fundamentally incapable of measuring effects that small in the first place.
If your measurement is bad in the first place, averaging large numbers of measurements accomplishes nothing except giving you a false sense of accuracy. A huge pile of shit statistics is still shit.
Oh, for crying out loud. IQ tests must have a bigger measurement error than plus or minus 2, which means that the 1.29-point alteration is smaller than the measurement error. I.e., no effect.
There is one possible exception, the neutrino is a half spin fermion and if it really is zero mass it would be its own anti-particle.
Actually it's the other way around: massless Fermions are Dirac, because of Chiral symmetry: in the Standard Model with massless neutrinos, all neutrinos are Dirac particles, with neutrinos being left-handed and all antineutrinos being right-handed. Mass terms break chiral symmetry, and a massive neutrino could be either Dirac or majorana depending on how the mass term is generated:
https://en.wikipedia.org/wiki/Sterile_neutrino#Majorana_or_Dirac.3F
A particle that is its own anti-particle? Sounds pretty special! Of course, that would also describe photons, the commonest particle in the universe.
Come on, science reporting.
Photons are bosons. Bosons being their own antiparticle is nothing unusual. A fermion that is its own antiparticle has never been observed in nature before.
This is not like finding the Higgs Boson. The majorana fermion they created was (hard to tell exactly how from TFA) a condensed matter excitation with the properties of a majorana fermion, not a fundamental particle. Pretty cool though.
What has been proposed, and should be technically feasible, is dividing the array up by frequency band. The plan already calls for three overlapping arrays of different types of telescopes in order to capture three different frequency bands. (Phased array dipole antennas work great at 100 MHz, whereas you need dishes for 10 GHz.) In principle, one could put the low- and mid-frequency arrays on one site and the high-frequency arrays on another.
Note that MEERKAT is going ahead in South Africa regardless of the outcome of the decision on SKA, and will operate from 0.5-14 GHz.
If they do this then the likely outcome is that SKA will never happen. This sort of "compromise" as a way of avoiding having to make an actual decision is almost always the first step in a death spiral for the project. Case in point: the Joint Dark Energy Mission, which crashed and burned due to pointless infighting between erstwhile collaborators on the mission.
Looks like the Elsevier boycott by academics had an effect. Still, this looks like more of a tactical response than a real change in position for Elsevier.
Let me be the first to say: Fuck you. I don't have to meet your criteria for intellectual seriousness to get access to the fucking internet, any more than I need to convince a librarian I am Serious(TM) enough to read a book that somebody doesn't like.
... with disabling the brain's response to endorphins?
Some proportion of that 'beating' was undoubtedly done by the boarder agents themselves.
No, I think it was mostly from the time I stuck it in my shorts mountain biking in South Africa.
My previous (US) passport was beat completely to shit by the time it was close to expiration. I got chewed out once in Germany by a passport agent who disapproved, and nearly didn't get let into Egypt at all. The martinet checking passports in Cairo was most offended by my treatment of the document. But I remember returning to the U.S. from Cairo, and the agent at JFK turned the passport over in his hand and said, in a thick New York accent, "Whadya, put a cigarette out innit?" and that was it. I knew I was home.