Yeah. I can do over 18.5 miles/hour on my mountain bike, EASY. I can do over 25 miles/hour (40 km/h) on a road bike. Where I ride (Boston, MA), traffic often goes much SLOWER than I do, especially when I'm commuting to work.
The numbers I quote are for an average person who is in their mid thirtys, and not a hardcore rider. Have you seen the Tour de France? Those guys are going WAY in excess of 30km/h, on road bikes. They can average in the low to mid 50's.
When you talk about a coin being flipped, you say an unbiased coin flips equally to either side. If you flipped a coin 9 times out of 10 for tails, you'd say you feel 95% confident that the coin is biased. A nice followup is to flip the coin more times, and see if your measurement of the true mean holds up over a larger sample size, which would afford you more confidence. If your initial hypothesis is correct, then the coin is biased, and the results will repeat. Let's just say a 95% confidence sounds like a large number, but it isn't (and a repeat is trivial, so why not increase your power.)
What I'd say here is that there's now a hypothesis, and a second round of data should be gathered to see if this relationship in the data holds.
I think the fallacy you mention abouve is that the TRUE mean for the coin is 50%, but a local fluctuation in the data makes the gambler think the true mean has somehow changed when it hasn't. Thus, I don't think that applying the results of the study are committing a fallacy - but on the other hand I wouldn't be too amazingly confident unless the original data set bore that out. Generally, I like a confidence of 1 in a million a whole lot better than 1 in 20. Some of work I've done generated confidence of 1 in 10^37. That's confidence!
Why not go gather some data? Please remember to gather it from an unbiased, large selection!
Statistically significant boils down to the difference observed between the two means, and the variability of the means. This is easiest to see by plotting the means and the confidence interval around them (the larger the sample size, the smaller the intervals at a given confidence), and seeing if the two confidence intervals overlap or not. However, if you plot the 95% confidence intervals around two means, you aren't testing a 1/20 chance of the two means being the same, it's much closer to a 1/100 chance (so you can't simply compare overlapping confidence intervals, but it does help.) The statistical significance also depends on what you define as significant - the standard often quoted is.05. This is adjusted by the number of hypothesis you test.
As a practical example, if you were to flip a coin 10 times, and the results were 1 head and 9 tails, when you expected 5 of each, the p-value for a chi-square test would be 0.01. That would be a statistically significant result, with only 10 measurements.
The larger the number of measurements, the less the two classes have to be different to detect that they truly are different.
Note, I'm not a statistician, nor do I play one on TV. I do quite a bit of research involving statistics in the life sciences, but I'd always have someone else check my work for a paper...
I am a habitual NPR listener, but everyone I know finds it slow, uninteresting, easily dismissed radio. I try to expose them to intriguing news material that's delivered spin free and very palatable, but have not yet impressed a single person. It's times like these that I just shake my head and sigh.
"a great All Things Considered story from a few years back about Akin Fernandez's 'Numbers' CD, a CD compilation of some of the most interesting strings of randomly read numbers"
Interesting... random numbers... Ok, so my friends were right.
So, the Number Stations story is completely boring, and your friends were right. And yet, you're reading (and posting) about it when it's on Slashdot? What about/. makes the story inherently more interesting?
NPR covered it years earlier, and their story has much better content than the submitted story here. I remember listening to the story in question, and it was very interesting.
So yes, NPR might just not suck.
(comming to you live from a Cambridge, MA, brie eating liberal)
But you describe the Homesian deduction principle: "How often have I said to you that when you have eliminated the impossible, whatever remains, however improbable, must be the truth?" (Arthur Conan Doyle, The Sign Of Four), where the hypotheses, that remain after a bunch of them are eliminated due to contradictions with observations, are the ones that are considered correct until more evidence shows up. That's quite different from induction, where you collect evidence until you spot a pattern and then use the pattern to form a hypothesis.
Eh, you collect evidence until you spot a pattern, and you form a hypothesis. Fortunately, you aren't close minded, so you realize that there are alternative hypothesis which could also be true given the evidence found. This forces you to gather further information to decide if the original hypothesis or one of the alternate hypothesis is the most believable.
We're dealing with statistical levels of proof, and incomplete information (actually, just to say we're dealing with statistics at all means that we haven't exhaustively measured the data in our population of samples.) You do the best you can with the resources you have to come up with a conclusion. You set thresholds of believability before you start the experiment. Sometimes, later on other people are able to gather more data than you were able to (more/different resources, changes in technology) and hypothesis get altered/discarded. As far as I can tell as a professional scientist, that's the way things work.
And, while I understand that it's an appeal to authority, the process I've been describing seems to have been 'good enough' to get multiple publications in Nature Genetics.
And then, just like in real life, someone holds a gun to their head, and reality asserts it's ugly superiority.
Let me know how much use 'control of the Net' is when someone's curbing your ass.
For all of people's views of superiority, when it comes down to it, any two bit criminal can rub out just about anyone on a whim. (exception: people with their own large security force.)
I wouldn't ever say that an alternative hypothesis is 'better' than the original one you're testing without evidence. However, when you say A->B, and you can disprove a bunch of alternatives, it makes the original more attractive, because you've shown that the alternatives are unlikely.
For backround: I'm in the bioinformatics field. One of the last papers we published was on the effects of selection on conserved non-coding sequences. There were a number of hypothesis for the effect we were observing with the data. We had to design a bunch of alternative hypothesis to 'knock out' the other mechanisms for the effect that would have been possible. Once we'd done that with all the reasonable alternatives, we were more sure of our original result. That doesn't mean someone couldn't come up with more data and an alternative hypothesis that is more believable - which would mean their hypothesis would 'take the lead'.
Sorry if I wasn't more explicit, or if I was entirely unclear...:)
The barrier to entry is still much higher than a patent, where the specification of the product and the process to produce it are specifically written out.
That would give you additional time to take advantage of your R&D. I thought that was the point of this thread.
I'll totally agree with you there. Release it as a podcast. Use some of the already made money to fund bandwidth. And/or release it as a torrent. Perhaps wrap the download with an inbedded torrent-client, so people don't have to set up a client themselves.
Well, the movie would have to draw a larger audience than whatever was going to be shown in those timeslots otherwise. Otherwise, the station showing the movie would have less viewers, and thus lesser add revenue.
So, this sounds like an opportunity cost for TV stations, and it might be steep. Not to mention the fact that the movie will be broken up into 10-12 minute segments (at best) with 2-3 minute commercial breaks. Or, it's shown on TV at 3AM, which you know will just get the largest draw of the population to watch it.
This almost seems like a push to go back to 'trade secrets'. You could copy the molecule we developed in the lab, but since there aren't any patents, we're not going to publish the spec of the new compound, or the lab procedure to produce it.
Wasn't that the situation in a way before patents, and the whole reason patents were supposed to be 'for the public good'? I think that patents should stick around, but should be limited to exclude 'business practices', and should be time limited to a reasonable length of time (5-7 years? Meta data studies using existing data should be done to determine a fair 'break even' point.)
Everything should go into the public domain, and trade secrets prevent that. I'm still mad that I don't know the secret blend of 11 herbs and spices.
Eh, just because you haven't found a material that is a superconductor at room temperature doesn't mean that there aren't any. It's easy to say "X can happen, because we have example Y", but you can't say "X can't happen, because we have example Y". All you can do is state all the places you've seen that it doesn't work. Sometimes, you can generalize those results (water from the atlantic ocean is not made of cheese, and thus we have no reason to believe that water from the pacific is either.)
Negative results are still results - they limit the problem space that you can search to find a positive result.
There's some complications to this. There is a physical structure to DNA. What you wind up having in populations is that segments of chromosomes travel in populations (these are haplotypes.) When you have something particularly interesting on some segment (something under positive selection), that entire segment will be kept (as the segment is slowly broken down by recombination, but that can take a LONG time.)
I'd argue that anything that takes energy to maintain but serves no function would be not be selected for. See: Antibiotic resistance in bacteria where their enviornment does not contain the resistance. Bacteria that are not resistant have more energy for reproduction, thus spread faster.
Selection is all about enviornment, though it's got a lot of interesting wrinkles that prevent us (population geneticists - though I can't call myself a 'real' one, as I do bioinformatics on population genetic data) from fully comprehending how all the inputs/outputs are wired together.
Programmers need to type in dozens of lines of code to express a single idea sometimes, and if you do that every day for a few years it really improves your typing speed (and accuracy).
I've found that while my typing speed isn't bad after 6 years of professional programming, my use of an IDE to do the heavy lifting forces my keystrokes to be pretty short.
Example: If I want a BufferedReader, I type Bu[control-space] and usually get it. If I want a variable name, one or two letters is usually enough (short methods = not too many variables in local scope.) Lots of other functions work about the same way. So, if you counted the number of letters or lines of code, it would look high, but that's just because I'm letting something else do the work.
I'm a decent typed, but those few years I spent on IRC constantly in the early 90's are probably what most honed my typing skills...Of course, that didn't teach me how incredibly important [({})] are...
When seeing a count of 0, and expecting a 1, you can't say that the error rate is 1/n. The true error rate in that case is x (number of individuals who voted for that guy) over n. We don't know how many people voted for him, because AFAIK, they did not poll the entire 80 people to find out what the true number of votes is.
All that we know is that an entire class of votes (for this candidate) are absent. That's FAR more worrying to me than a 1.25% error rate.
A few of my friends and I (are we're all in the biology field in some degree, as researchers) refuse to use doctors who don't know what the internet is. I'm glad you do, but it is sad when you talk to a doctor about some large issue you have, and the doctor doesn't know about/use the internet to make sure they're aware of all the treatment issues.
One of the most interesting cases in our group was a friend who had osteonecrosis in one of his knees. Some of the doctors he went to weren't keeping up with modern practice, and they recommended full knee replacements. He finally found a younger doctor who was up to date, and the surgery he had involved boring small holes into his knee, so that blood would enter those areas and rebuild the bone there.
The surgery was a complete success, my friend didn't need an artificial knee (at age 30!), and now he's perfectly healthy. The recovery time for the new surgery was much lower, and it was an all around good solution.
In either case, the question I'd have to ask is: Why do you lean in the direction that you do? Is it some kind of faith based approach, or do you look at your own anecdotal evidence, or are you drawing data to back up your claims based on something else?
I really feel that someone can *feel* that capital punishment is wrong, but I'm not convinced that global warming (or for that matter, any sort of scientific question where there's data to back up a hypothesis) should be something we 'feel' is right, and say that's the end of our thinking. You might have a gut intuition that something should be resolved one way or the other, but when a scientist has that gut feeling, they gather data and try to prove it.
I always loved the classic Yakov Smirnoff sketch called "Shoot the Jew" that YS used to do. I wonder why he stopped? Also, I find his sketch "My sister is #4 prostitute in all of Soviet Russia" sketch to be hilarious. I don't know why Sasha ripped him off.
Read this paper, Nature Genetics, Feb 2006: "Conserved noncoding sequences are selectively constrained and not mutation cold spots"
You can look at the abstract for free.
From it, you may notice that the regions talked about in the paper are what the article calls "Junk DNA". However, if those regions are under active selection, then they are being conserved for a reason - IE, they aren't junk, but are affecting function, and thus are retained.
The whole concept of Junk DNA is cute, but it's out of touch with current research. It would be far more appropriate to say "We're very sure that coding regions affect gene function. We know there are regulatory regions outside of exons that effect regulation / function. We're discovering new classes of regulators, as well as other mechanisms that effect function at a very rapid pace. To say that all of this is "junk" because we're not sure of what functional elements we haven't discovered yet is overstepping our bounds."
If a large number peer reviewed papers is decent journals aren't enough to convince you that the wikipedia is wrong, then I'm not really sure I can help you. Or, if it makes you feel better, you can merely say that the wikipedia article is out of date.
As a coffee fanatic, I'd mod you up if I had the points (despite your being almost off topic.)
I've been able to move my relatives to thermal servers when they serve coffee over longer periods of time, and it's helped them get their coffee up to palatable.
OTOH, I generally drink espresso from a Livia 90 automatic (not super automatic, but not a level system.) The beans are ground 30 seconds before extraction, and the espresso is consumed 3-5 minutes from extraction time. That way, it's never cold...but who could resist fresh espresso anyway?
As a bioinformatician who recently published a paper in Nature Genetics on Conserved Non Coding regions (non gene regions that are more highly similar than expected - the base pairs are the same), I'd have to call "Bullshit!" on this wikipedia article.
Please don't believe everything you read on wikipedia. It might have been right if I'd read that 5 years ago, but my work, and other people's work says otherwise.
Slashdot Mirror
Yeah. I can do over 18.5 miles/hour on my mountain bike, EASY. I can do over 25 miles/hour (40 km/h) on a road bike. Where I ride (Boston, MA), traffic often goes much SLOWER than I do, especially when I'm commuting to work.
The numbers I quote are for an average person who is in their mid thirtys, and not a hardcore rider. Have you seen the Tour de France? Those guys are going WAY in excess of 30km/h, on road bikes. They can average in the low to mid 50's.
When you talk about a coin being flipped, you say an unbiased coin flips equally to either side. If you flipped a coin 9 times out of 10 for tails, you'd say you feel 95% confident that the coin is biased. A nice followup is to flip the coin more times, and see if your measurement of the true mean holds up over a larger sample size, which would afford you more confidence. If your initial hypothesis is correct, then the coin is biased, and the results will repeat. Let's just say a 95% confidence sounds like a large number, but it isn't (and a repeat is trivial, so why not increase your power.)
What I'd say here is that there's now a hypothesis, and a second round of data should be gathered to see if this relationship in the data holds.
I think the fallacy you mention abouve is that the TRUE mean for the coin is 50%, but a local fluctuation in the data makes the gambler think the true mean has somehow changed when it hasn't. Thus, I don't think that applying the results of the study are committing a fallacy - but on the other hand I wouldn't be too amazingly confident unless the original data set bore that out. Generally, I like a confidence of 1 in a million a whole lot better than 1 in 20. Some of work I've done generated confidence of 1 in 10^37. That's confidence!
Why not go gather some data? Please remember to gather it from an unbiased, large selection!
Statistically significant boils down to the difference observed between the two means, and the variability of the means. This is easiest to see by plotting the means and the confidence interval around them (the larger the sample size, the smaller the intervals at a given confidence), and seeing if the two confidence intervals overlap or not. However, if you plot the 95% confidence intervals around two means, you aren't testing a 1/20 chance of the two means being the same, it's much closer to a 1/100 chance (so you can't simply compare overlapping confidence intervals, but it does help.) The statistical significance also depends on what you define as significant - the standard often quoted is .05. This is adjusted by the number of hypothesis you test.
As a practical example, if you were to flip a coin 10 times, and the results were 1 head and 9 tails, when you expected 5 of each, the p-value for a chi-square test would be 0.01. That would be a statistically significant result, with only 10 measurements.
The larger the number of measurements, the less the two classes have to be different to detect that they truly are different.
Note, I'm not a statistician, nor do I play one on TV. I do quite a bit of research involving statistics in the life sciences, but I'd always have someone else check my work for a paper...
"a great All Things Considered story from a few years back about Akin Fernandez's 'Numbers' CD, a CD compilation of some of the most interesting strings of randomly read numbers"
Interesting... random numbers... Ok, so my friends were right. So, the Number Stations story is completely boring, and your friends were right. And yet, you're reading (and posting) about it when it's on Slashdot? What about
NPR covered it years earlier, and their story has much better content than the submitted story here. I remember listening to the story in question, and it was very interesting.
So yes, NPR might just not suck.
(comming to you live from a Cambridge, MA, brie eating liberal)
I've never had to be an admin on a Unix machine, but I've been a molecular biologist/population geneticist for 10+ years.
Let's just say you and I are equally annoyed for completely different reasons.
Eh, you collect evidence until you spot a pattern, and you form a hypothesis. Fortunately, you aren't close minded, so you realize that there are alternative hypothesis which could also be true given the evidence found. This forces you to gather further information to decide if the original hypothesis or one of the alternate hypothesis is the most believable.
We're dealing with statistical levels of proof, and incomplete information (actually, just to say we're dealing with statistics at all means that we haven't exhaustively measured the data in our population of samples.) You do the best you can with the resources you have to come up with a conclusion. You set thresholds of believability before you start the experiment. Sometimes, later on other people are able to gather more data than you were able to (more/different resources, changes in technology) and hypothesis get altered/discarded. As far as I can tell as a professional scientist, that's the way things work.
And, while I understand that it's an appeal to authority, the process I've been describing seems to have been 'good enough' to get multiple publications in Nature Genetics.
Except that "funny" doesn't count towards karma. Mission (not) accomplished!
And then, just like in real life, someone holds a gun to their head, and reality asserts it's ugly superiority.
Let me know how much use 'control of the Net' is when someone's curbing your ass.
For all of people's views of superiority, when it comes down to it, any two bit criminal can rub out just about anyone on a whim. (exception: people with their own large security force.)
I wouldn't ever say that an alternative hypothesis is 'better' than the original one you're testing without evidence. However, when you say A->B, and you can disprove a bunch of alternatives, it makes the original more attractive, because you've shown that the alternatives are unlikely.
For backround: I'm in the bioinformatics field. One of the last papers we published was on the effects of selection on conserved non-coding sequences. There were a number of hypothesis for the effect we were observing with the data. We had to design a bunch of alternative hypothesis to 'knock out' the other mechanisms for the effect that would have been possible. Once we'd done that with all the reasonable alternatives, we were more sure of our original result. That doesn't mean someone couldn't come up with more data and an alternative hypothesis that is more believable - which would mean their hypothesis would 'take the lead'.
Sorry if I wasn't more explicit, or if I was entirely unclear...:)
The barrier to entry is still much higher than a patent, where the specification of the product and the process to produce it are specifically written out.
That would give you additional time to take advantage of your R&D. I thought that was the point of this thread.
I'll totally agree with you there. Release it as a podcast. Use some of the already made money to fund bandwidth. And/or release it as a torrent. Perhaps wrap the download with an inbedded torrent-client, so people don't have to set up a client themselves.
Well, the movie would have to draw a larger audience than whatever was going to be shown in those timeslots otherwise. Otherwise, the station showing the movie would have less viewers, and thus lesser add revenue.
So, this sounds like an opportunity cost for TV stations, and it might be steep. Not to mention the fact that the movie will be broken up into 10-12 minute segments (at best) with 2-3 minute commercial breaks. Or, it's shown on TV at 3AM, which you know will just get the largest draw of the population to watch it.
Or, you just call Oracle SQL Server "Oracle".
As in: Why is &#@*& Oracle ignoring my indexes and forcing a hash join on two 1M+ row tables AGAIN? GAAAAH!
People will know what you mean.
This almost seems like a push to go back to 'trade secrets'. You could copy the molecule we developed in the lab, but since there aren't any patents, we're not going to publish the spec of the new compound, or the lab procedure to produce it.
Wasn't that the situation in a way before patents, and the whole reason patents were supposed to be 'for the public good'? I think that patents should stick around, but should be limited to exclude 'business practices', and should be time limited to a reasonable length of time (5-7 years? Meta data studies using existing data should be done to determine a fair 'break even' point.)
Everything should go into the public domain, and trade secrets prevent that. I'm still mad that I don't know the secret blend of 11 herbs and spices.
Eh, just because you haven't found a material that is a superconductor at room temperature doesn't mean that there aren't any. It's easy to say "X can happen, because we have example Y", but you can't say "X can't happen, because we have example Y". All you can do is state all the places you've seen that it doesn't work. Sometimes, you can generalize those results (water from the atlantic ocean is not made of cheese, and thus we have no reason to believe that water from the pacific is either.)
Negative results are still results - they limit the problem space that you can search to find a positive result.
There's some complications to this. There is a physical structure to DNA. What you wind up having in populations is that segments of chromosomes travel in populations (these are haplotypes.) When you have something particularly interesting on some segment (something under positive selection), that entire segment will be kept (as the segment is slowly broken down by recombination, but that can take a LONG time.)
I'd argue that anything that takes energy to maintain but serves no function would be not be selected for. See: Antibiotic resistance in bacteria where their enviornment does not contain the resistance. Bacteria that are not resistant have more energy for reproduction, thus spread faster.
Selection is all about enviornment, though it's got a lot of interesting wrinkles that prevent us (population geneticists - though I can't call myself a 'real' one, as I do bioinformatics on population genetic data) from fully comprehending how all the inputs/outputs are wired together.
I've found that while my typing speed isn't bad after 6 years of professional programming, my use of an IDE to do the heavy lifting forces my keystrokes to be pretty short.
Example: If I want a BufferedReader, I type Bu[control-space] and usually get it. If I want a variable name, one or two letters is usually enough (short methods = not too many variables in local scope.) Lots of other functions work about the same way. So, if you counted the number of letters or lines of code, it would look high, but that's just because I'm letting something else do the work.
I'm a decent typed, but those few years I spent on IRC constantly in the early 90's are probably what most honed my typing skills...Of course, that didn't teach me how incredibly important [({})] are...
When seeing a count of 0, and expecting a 1, you can't say that the error rate is 1/n. The true error rate in that case is x (number of individuals who voted for that guy) over n. We don't know how many people voted for him, because AFAIK, they did not poll the entire 80 people to find out what the true number of votes is.
All that we know is that an entire class of votes (for this candidate) are absent. That's FAR more worrying to me than a 1.25% error rate.
A few of my friends and I (are we're all in the biology field in some degree, as researchers) refuse to use doctors who don't know what the internet is. I'm glad you do, but it is sad when you talk to a doctor about some large issue you have, and the doctor doesn't know about/use the internet to make sure they're aware of all the treatment issues.
One of the most interesting cases in our group was a friend who had osteonecrosis in one of his knees. Some of the doctors he went to weren't keeping up with modern practice, and they recommended full knee replacements. He finally found a younger doctor who was up to date, and the surgery he had involved boring small holes into his knee, so that blood would enter those areas and rebuild the bone there.
The surgery was a complete success, my friend didn't need an artificial knee (at age 30!), and now he's perfectly healthy. The recovery time for the new surgery was much lower, and it was an all around good solution.
In either case, the question I'd have to ask is: Why do you lean in the direction that you do? Is it some kind of faith based approach, or do you look at your own anecdotal evidence, or are you drawing data to back up your claims based on something else?
I really feel that someone can *feel* that capital punishment is wrong, but I'm not convinced that global warming (or for that matter, any sort of scientific question where there's data to back up a hypothesis) should be something we 'feel' is right, and say that's the end of our thinking. You might have a gut intuition that something should be resolved one way or the other, but when a scientist has that gut feeling, they gather data and try to prove it.
I always loved the classic Yakov Smirnoff sketch called "Shoot the Jew" that YS used to do. I wonder why he stopped? Also, I find his sketch "My sister is #4 prostitute in all of Soviet Russia" sketch to be hilarious. I don't know why Sasha ripped him off.
They are EXACTLY the same.
Read this paper, Nature Genetics, Feb 2006: "Conserved noncoding sequences are selectively constrained and not mutation cold spots"
You can look at the abstract for free.
From it, you may notice that the regions talked about in the paper are what the article calls "Junk DNA". However, if those regions are under active selection, then they are being conserved for a reason - IE, they aren't junk, but are affecting function, and thus are retained.
The whole concept of Junk DNA is cute, but it's out of touch with current research. It would be far more appropriate to say "We're very sure that coding regions affect gene function. We know there are regulatory regions outside of exons that effect regulation / function. We're discovering new classes of regulators, as well as other mechanisms that effect function at a very rapid pace. To say that all of this is "junk" because we're not sure of what functional elements we haven't discovered yet is overstepping our bounds."
If a large number peer reviewed papers is decent journals aren't enough to convince you that the wikipedia is wrong, then I'm not really sure I can help you. Or, if it makes you feel better, you can merely say that the wikipedia article is out of date.
As a coffee fanatic, I'd mod you up if I had the points (despite your being almost off topic.)
I've been able to move my relatives to thermal servers when they serve coffee over longer periods of time, and it's helped them get their coffee up to palatable.
OTOH, I generally drink espresso from a Livia 90 automatic (not super automatic, but not a level system.) The beans are ground 30 seconds before extraction, and the espresso is consumed 3-5 minutes from extraction time. That way, it's never cold...but who could resist fresh espresso anyway?
As a bioinformatician who recently published a paper in Nature Genetics on Conserved Non Coding regions (non gene regions that are more highly similar than expected - the base pairs are the same), I'd have to call "Bullshit!" on this wikipedia article.
Please don't believe everything you read on wikipedia. It might have been right if I'd read that 5 years ago, but my work, and other people's work says otherwise.