|
Read the rules in the OP before posting, please.In order to ensure that this thread continues to meet TL standards and follows the proper guidelines, we will be enforcing the rules in the OP more strictly. Be sure to give them a re-read to refresh your memory! The vast majority of you are contributing in a healthy way, keep it up! NOTE: When providing a source, explain why you feel it is relevant and what purpose it adds to the discussion if it's not obvious. Also take note that unsubstantiated tweets/posts meant only to rekindle old arguments can result in a mod action. |
On November 20 2017 10:35 KwarK wrote:Show nested quote +On November 20 2017 10:30 Aquanim wrote:On November 20 2017 10:09 KwarK wrote:...
You can't get from "the 30% thing happened in a 30:70 odds situation" to "science is wrong". You just can't. Nate Silver never said Trump couldn't win. He said the opposite, that Trump would win 3 times in 10 according to his model. You just apparently don't know how models work. I'm pretty sure he is talking about other modellers who predicted considerably worse odds and margins for Trump, not Silver. That being said I don't agree with the underlying argument that the performance of pollsters and pundits says anything meaningful about the state of scientific endeavour as a whole, even if we stipulate that it counts as "scientific" in the first place. You still can't get from "pollster said it was unlikely" to "science is wrong", when looking at a single event. If there is one thing playing poker taught me it's that unlikely events happen all the fucking time and that in a single hand literally anything can happen. The only way to adequately test the accuracy of a model for predicting an event is to run the same event over and over and compare the outcomes to the model.
The model relies on assumptions. Models that said Trump had a 1% chance of winning relied mostly on the idiotic assumption that polling errors across states are uncorrelated.
You're technically correct in that you can't judge the calibration or probability prediction based on one event, but you're wrong in practice because modeling an election is a separate problem from simple polling, and the underlying model assumptions needed to combine polling results to estimate the probability of an outcome can be evaluated on its own logical and empirical merits--and those models and their assumptions failed badly on both.
|
On November 20 2017 10:56 LegalLord wrote:
If anything, the Nate Silver versus the world situation shows either a widespread statistical illiteracy, a tendency for doctoring data toward a desired result for political ends, or perhaps both. The fact that Nate was criticized for not being strongly enough in favor of saying that Hillary will win it suggests at least one of those options.
Nate Silver got the election 'more right' than his peers because he doctored the data, you yourself wrote a long post about the fact that he took a bayesian approach to the election. Others missed more strongly because the data was just completely unreliable at the state level. (and they went with un-opinionated models)
|
United States44248 Posts
On November 20 2017 11:17 Aquanim wrote:Show nested quote +On November 20 2017 11:14 KwarK wrote:On November 20 2017 10:56 LegalLord wrote:
If anything, the Nate Silver versus the world situation shows either a widespread statistical illiteracy, a tendency for doctoring data toward a desired result for political ends, or perhaps both. The fact that Nate was criticized for not being strongly enough in favor of saying that Hillary will win it suggests at least one of those options. No, it shows nothing. You cannot draw conclusions about the validity of a statistical model from a single datapoint. If Nate says it's 70:30 Hillary favoured and some other guy says it's 999:1 Hillary favoured then you still can't tell which of them was right from a single datapoint. It's entirely possible that Nate was severely overrating Trump's chances AND that Trump won. I think you're overstating this a bit. It's not as though it's impossible to critically evaluate and improve statistical models.
Sure, you can improve models. But when it's modelling single run outcomes you can't test your model against the real world in any meaningful way.
|
(people are saying this is from 2003)
|
|
|
Ah, took a bit of reading but I get the idea. Basically the sketch Tweeden was talking about had been in performance for YEARS before, and there is something in the script about "written just so I could kiss you" which lends a considerable degree of doubt on her claim he wrote the skit to force himself on her in rehearsal. So that video is it being performed with a totally different woman.
KO is awful, but those people are misreading the tweet.
|
On November 20 2017 11:18 Nyxisto wrote:Show nested quote +On November 20 2017 10:56 LegalLord wrote:
If anything, the Nate Silver versus the world situation shows either a widespread statistical illiteracy, a tendency for doctoring data toward a desired result for political ends, or perhaps both. The fact that Nate was criticized for not being strongly enough in favor of saying that Hillary will win it suggests at least one of those options. Nate Silver got the election 'more right' than his peers because he doctored the data, you yourself wrote a long post about the fact that he took a bayesian approach to the election. Others missed more strongly because the data was just completely unreliable at the state level. (and they went with un-opinionated models)
If you don't have enough data to reliably predict things, a frequentist approach is simply inappropriate and the numbers you're coming up with are basically garbage. Similar what I said above, you can't defend presenting such work as a realistic prediction by saying it's "unbiased." It's just bad work and you're misrepresenting the output by confidently claiming a HRC victory (99% or whatever) when you don't have sufficient data to make such a strong claim.
A frequentist approach is never "unbiased" anyway as the choice of the likelihood function (i.e. the model) is always arbitrary in the same way that priors are.
|
On November 20 2017 09:24 mozoku wrote:Show nested quote +On November 20 2017 07:35 Ciaus_Dronu wrote:On November 20 2017 07:09 Danglars wrote:On November 20 2017 06:34 Adreme wrote:On November 20 2017 06:21 Danglars wrote:On November 20 2017 06:09 Gorsameth wrote:On November 20 2017 05:57 Danglars wrote:On November 20 2017 05:53 Gorsameth wrote:On November 20 2017 05:50 Danglars wrote:On November 20 2017 04:40 DarkPlasmaBall wrote:
[quote]
Trump's current* philosophies toward science, civil rights for women and minorities, and social policies are much, much more conservative than Hillary's current philosophies. In fact, the only shared liberal stance I can think of is that they're both fine with gay marriage, even though Trump then went ahead and took the ultra anti-gay Mike Pence as VP runningmate... And with that plus Trump's phobia of anything trans or non-cisgendered, he and his administration are quite clearly not LGBT supporters.
*I say "current" because many of his positions (e.g., abortion) change year to year or even month to month. For example, I don't even know where Trump stands on abortion anymore, and he's accustomed to pandering to everyone so it's hard to pin him down to what he really believes anyway. Half the time, he'll say he believes whatever you want him to believe, if it earns him a little money or power. Lol at your descriptions. Yes, Democratic talking points crafted by Democrats to smear the opposition portray the opposition in a bad light. Holy cow, stop the presses. Science. Now a Democratic talking made to smear Republicans with. Umm Dems have been calling Reps anti-science for a while now. It’s probably just behind the women-and-minorities tripe in frequency. Maybe they wouldn't do so if Republicans didn't keep denying scientific evidence on a regular basis. Climate Change? Clean coal? When was the last time trickle down economics worked? I think there is a tax plan in congress right now where private jets help lower and middle classes. As with so many things Republicans don't like being called. Have you considered not being the thing you don't like being called? The purportedly party of science always tends to shout down the opposition and demand government control and death to industry. Sorry, but America’s energy needs don’t comport with utopian desires for clean energy. I’m not holding my breath for idiotic leftists getting on board with nuclear or other high-power clean energy technology. Dems like their narratives. Science has little tolerance for them. I’ll wait until your policy prescriptions are more science-based and less ideology-based I get that you like to make statements that are nothing more then empty talking points with no basis in reality but you do realize that this administration is the one hiring non-scientists for jobs that used to only go to scientists and trying to purge scientists from various panels. That isnt some debatable point by the way its a thing they are actively doing because there solution when science disagrees with there position is to silence the scientists. I respond to partisan talking points with partisan talking points sometimes. Particularly when the author pretends to be objective. One can very objectively point out that one party (to which they may have a partisan bias independent of the commentary itself) has notably worse policies regarding say, scientific publication, education and application, than the other. That isn't a partisan talking point, that's having a basic grasp of reality. I'm in general agreement that the Democratic party is probably on average more "pro-science", but it's certainly more complicated than an "objective fact." The Democrats are definitely more willing to fund science and like to market themselves as "pro-science", sure. In practice though, the "scientific" left-wing institutions are in a definite rough patch. Academia is in the midst of a replication crisis largely funded by taxpayer dollars (yeah, what a great use of funds...), and these "scientific" election forecasters that all predicted Hillary winning in a landslide obviously aren't really all that "scientific" in practice (there are exceptions of course; I have a lot of respect for Nate Silver's work). Meanwhile, the most economically impactful technology advances atm are coming mostly from barely regulated and not government funded tech firms, whose corporatist mantra is closer to Republican philosophy than Democrat (though ironically, their perceived political allies and enemies are the reverse). There's a large difference between enthusiasm for science and enabling successful science. I'd say the Democrats do well on the first part, but doing good science requires not only enthusiasm and creativity, but discipline and skepticism. It doesn't neatly fit neatly on a left-right scale, and I think it's interesting to consider that that might be a factor in why, historically, scientific success has usually occurred in places where right-left are aligned and working together (e.g. WW2, Cold War, Silicon Valley).
The tech firm model is often to cluster around universities. Universities feed the tech firms talent and technology which they can then refine and bring to market. It's symbiotic with universities increasingly holding onto patents and taking equity stakes in startups.
Knocking part of that relationship (academia), is neither a pro-science stance nor good economic policy.
|
I think we're getting away from the central point here:
On November 20 2017 10:37 Aquanim wrote:
even if you could get to "this pollster was wrong" you can't get from there to "science is wrong".
|
On November 20 2017 11:18 Nyxisto wrote:Show nested quote +On November 20 2017 10:56 LegalLord wrote:
If anything, the Nate Silver versus the world situation shows either a widespread statistical illiteracy, a tendency for doctoring data toward a desired result for political ends, or perhaps both. The fact that Nate was criticized for not being strongly enough in favor of saying that Hillary will win it suggests at least one of those options. Nate Silver got the election 'more right' than his peers because he doctored the data, you yourself wrote a long post about the fact that he took a bayesian approach to the election. Others missed more strongly because the data was just completely unreliable at the state level. (and they went with un-opinionated models)
You're misusing the hell out of the term "doctored." Nate Silver made a model based on reliability of pollsters (in a fairly transparent fashion - his site posts their rankings) and let the results speak for themselves. That has nothing to do with taking a Bayesian approach or not and he definitely wasn't the only one to do so.
Others took a different approach that evidently wasn't as reliable. That's not the issue. It should be fairly clear, from even a layman's perspective, that it wasn't a 99.9% chance of Hillary winning. Anyone who claimed it to be so was largely using ineffective or agenda-driven prediction that was about as accurate as "Trump can't win, he needs to get 7 out of 6 swing states!!!" which evidently didn't reflect reality. Especially if they say "y u no 99.9% like us mang" to Nate Silver's models.
He got it "more right" because he used better statistical analysis and more useful underlying assumptions. Let's not make shit up to try to justify bad predictions made along party leanings.
|
On November 20 2017 11:30 mozoku wrote:Show nested quote +On November 20 2017 11:18 Nyxisto wrote:On November 20 2017 10:56 LegalLord wrote:
If anything, the Nate Silver versus the world situation shows either a widespread statistical illiteracy, a tendency for doctoring data toward a desired result for political ends, or perhaps both. The fact that Nate was criticized for not being strongly enough in favor of saying that Hillary will win it suggests at least one of those options. Nate Silver got the election 'more right' than his peers because he doctored the data, you yourself wrote a long post about the fact that he took a bayesian approach to the election. Others missed more strongly because the data was just completely unreliable at the state level. (and they went with un-opinionated models) If you don't have enough data to reliably predict things, a frequentist approach is simply inappropriate and the numbers you're coming up with are basically garbage. Similar what I said above, you can't defend presenting such work as a realistic prediction by saying it's "unbiased." It's just bad work and you're misrepresenting the output by confidently claiming a HRC victory (99% or whatever) when you don't have sufficient data to make such a strong claim. A frequentist approach is never "unbiased" anyway as the choice of the likelihood function (i.e. the model) is always arbitrary in the same way that priors are.
That's a completely valid statement, it's just not true that anybody had to be 'statistically illiterate' or 'fudged the data' to come to the conclusion that Clinton was, based on polls, very likely to win. They should have showed more scepticism about the data and explained better and explain the uncertainty that is attached to a prediction, but there's no reason to blow it out of proportion.
If the number of 99% would have been replaced with the label 'overwhelmingly likely' we wouldn't really have to discuss anything because everybody believed that regardless whether they were statistically minded or not.
|
On November 20 2017 11:47 Nyxisto wrote:Show nested quote +On November 20 2017 11:30 mozoku wrote:On November 20 2017 11:18 Nyxisto wrote:On November 20 2017 10:56 LegalLord wrote:
If anything, the Nate Silver versus the world situation shows either a widespread statistical illiteracy, a tendency for doctoring data toward a desired result for political ends, or perhaps both. The fact that Nate was criticized for not being strongly enough in favor of saying that Hillary will win it suggests at least one of those options. Nate Silver got the election 'more right' than his peers because he doctored the data, you yourself wrote a long post about the fact that he took a bayesian approach to the election. Others missed more strongly because the data was just completely unreliable at the state level. (and they went with un-opinionated models) If you don't have enough data to reliably predict things, a frequentist approach is simply inappropriate and the numbers you're coming up with are basically garbage. Similar what I said above, you can't defend presenting such work as a realistic prediction by saying it's "unbiased." It's just bad work and you're misrepresenting the output by confidently claiming a HRC victory (99% or whatever) when you don't have sufficient data to make such a strong claim. A frequentist approach is never "unbiased" anyway as the choice of the likelihood function (i.e. the model) is always arbitrary in the same way that priors are. That's a completely valid statement, it's just not true that anybody had to be 'statistically illiterate' or 'fudged the data' to come to the conclusion that Clinton was, based on polls, very likely to win. They should have showed more scepticism about the data and explained better and explain the uncertainty that is attached to a prediction, but there's no reason to blow it out of proportion.
@Aquanim as well.
This is sort of my point. The Left and it's affiliated groups are far more likely to do something like university research or quantitative election forecasting so the perception is that they're much more "pro-science", but that doesn't necessarily mean that they're succeeding at, monitoring it properly, and actually creating useful stuff out of all their science. On the other hand, discipline and focus provided by corporatist tech firms, VCs, the Soviets, whatever it may be, goes a long way towards actually advancing science and tech. And competition is certainly a right-wing ideal.
Keep in mind, when I acknowledged the Left is probably more "pro-science" in the way it's being used. I'm not attempting to excuse climate change denialism (though I haven't personally studied it enough to comment on how much of the effect of manage man-made) or similar nonsense. My point was merely that it wasn't as clear cut as the original poster I quoted made it out to be.
|
On November 20 2017 11:31 JonnyBNoHo wrote:Show nested quote +On November 20 2017 09:24 mozoku wrote:On November 20 2017 07:35 Ciaus_Dronu wrote:On November 20 2017 07:09 Danglars wrote:On November 20 2017 06:34 Adreme wrote:On November 20 2017 06:21 Danglars wrote:On November 20 2017 06:09 Gorsameth wrote:On November 20 2017 05:57 Danglars wrote:On November 20 2017 05:53 Gorsameth wrote:On November 20 2017 05:50 Danglars wrote:
[quote]
Lol at your descriptions. Yes, Democratic talking points crafted by Democrats to smear the opposition portray the opposition in a bad light. Holy cow, stop the presses. Science. Now a Democratic talking made to smear Republicans with. Umm Dems have been calling Reps anti-science for a while now. It’s probably just behind the women-and-minorities tripe in frequency. Maybe they wouldn't do so if Republicans didn't keep denying scientific evidence on a regular basis. Climate Change? Clean coal? When was the last time trickle down economics worked? I think there is a tax plan in congress right now where private jets help lower and middle classes. As with so many things Republicans don't like being called. Have you considered not being the thing you don't like being called? The purportedly party of science always tends to shout down the opposition and demand government control and death to industry. Sorry, but America’s energy needs don’t comport with utopian desires for clean energy. I’m not holding my breath for idiotic leftists getting on board with nuclear or other high-power clean energy technology. Dems like their narratives. Science has little tolerance for them. I’ll wait until your policy prescriptions are more science-based and less ideology-based I get that you like to make statements that are nothing more then empty talking points with no basis in reality but you do realize that this administration is the one hiring non-scientists for jobs that used to only go to scientists and trying to purge scientists from various panels. That isnt some debatable point by the way its a thing they are actively doing because there solution when science disagrees with there position is to silence the scientists. I respond to partisan talking points with partisan talking points sometimes. Particularly when the author pretends to be objective. One can very objectively point out that one party (to which they may have a partisan bias independent of the commentary itself) has notably worse policies regarding say, scientific publication, education and application, than the other. That isn't a partisan talking point, that's having a basic grasp of reality. I'm in general agreement that the Democratic party is probably on average more "pro-science", but it's certainly more complicated than an "objective fact." The Democrats are definitely more willing to fund science and like to market themselves as "pro-science", sure. In practice though, the "scientific" left-wing institutions are in a definite rough patch. Academia is in the midst of a replication crisis largely funded by taxpayer dollars (yeah, what a great use of funds...), and these "scientific" election forecasters that all predicted Hillary winning in a landslide obviously aren't really all that "scientific" in practice (there are exceptions of course; I have a lot of respect for Nate Silver's work). Meanwhile, the most economically impactful technology advances atm are coming mostly from barely regulated and not government funded tech firms, whose corporatist mantra is closer to Republican philosophy than Democrat (though ironically, their perceived political allies and enemies are the reverse). There's a large difference between enthusiasm for science and enabling successful science. I'd say the Democrats do well on the first part, but doing good science requires not only enthusiasm and creativity, but discipline and skepticism. It doesn't neatly fit neatly on a left-right scale, and I think it's interesting to consider that that might be a factor in why, historically, scientific success has usually occurred in places where right-left are aligned and working together (e.g. WW2, Cold War, Silicon Valley). The tech firm model is often to cluster around universities. Universities feed the tech firms talent and technology which they can then refine and bring to market. It's symbiotic with universities increasingly holding onto patents and taking equity stakes in startups. Knocking part of that relationship (academia), is neither a pro-science stance nor good economic policy.
The phenomenon you're describing is completely independent of the quality of research and education actually happening at universities. The tech firms will follow talent. Whether the universities are actually developing talent or merely attracting it is irrelevant from a tech firm's perspective.
|
Dude, you can take the overwhelming majority of scientists at face value and accept climate change to be real.
Just read a IPCC report. And please don't come about with the stupidly misused weather is not climate.
regarding your questioning whether universities creat useful stuff or not:
The anwer to that is to be taken into context of what the respondee considers useful, thus it's higly subjective and a rather tame argument to make. You're also incredibly vague of what you mean so it's hard do understand what exactly your point is. To me it seems like everything non marketable isn't useful by your standards, but that might be just me.
|
It's a very standard practice for companies to cluster around universities. Attributing that to just "tech" is either ignorant or disingenuous.
|
Well you can make a more Conservative case that universities tend to follow economic development rather than the other way around. People have been starting to look a lot at the mismatch between millennials and the labour market, how countries with more corporate schemes distribute economic growth more fairly and so forth.
But yeah that shouldn't devolve into a political debate about "liberal universities" or nonsensical climate change denial.
|
On November 20 2017 12:17 Artisreal wrote:
Dude, you can take the overwhelming majority of scientists at face value and accept climate change to be real.
Just read a IPCC report. And please don't come about with the stupidly misused weather is not climate.
regarding your questioning whether universities creat useful stuff or not:
The anwer to that is to be taken into context of what the respondee considers useful, thus it's higly subjective and a rather tame argument to make. You're also incredibly vague of what you mean so it's hard do understand what exactly your point is. To me it seems like everything non marketable isn't useful by your standards, but that might be just me.
When literally more than half of studies in fields like medicine and psychology can't be replicated, it's not at all a stretch to conclude that some money has been wasted in academia. Esp when most professors are spending minimal attention on teaching and justifying it by emphasizing their focus on research that more often than not can't be replicated.
|
On November 20 2017 10:56 LegalLord wrote:
If anything, the Nate Silver versus the world situation shows either a widespread statistical illiteracy, a tendency for doctoring data toward a desired result for political ends, or perhaps both. The fact that Nate was criticized for not being strongly enough in favor of saying that Hillary will win it suggests at least one of those options.
I agree with the bold part because of how people interpreted Trump's victory as a refutation of the reliability of statistics despite it being a single trial. It's unfortunate that most people don't understand basic statistics and probability, and think Nate Silver is an idiot. Ironically, I'm sure some of these people play the lottery.
|
On November 20 2017 12:37 mozoku wrote:Show nested quote +On November 20 2017 12:17 Artisreal wrote:
Dude, you can take the overwhelming majority of scientists at face value and accept climate change to be real.
Just read a IPCC report. And please don't come about with the stupidly misused weather is not climate.
regarding your questioning whether universities creat useful stuff or not:
The anwer to that is to be taken into context of what the respondee considers useful, thus it's higly subjective and a rather tame argument to make. You're also incredibly vague of what you mean so it's hard do understand what exactly your point is. To me it seems like everything non marketable isn't useful by your standards, but that might be just me. When literally more than half of studies in fields like medicine and psychology can't be replicated, it's not at all a stretch to conclude that some money has been wasted in academia. Esp when most professors are spending minimal attention on teaching and justifying it by emphasizing their focus on research that more often than not can't be replicated.
An extremely uneducated claim.
Replication failure is correlated with how much human factor you involve. In other words, the more you're getting away from the "hard science" and are studying things like human interaction, behavior, or the effects of things on humans, the more you'll see replication error.
However, there is significantly less error when it comes to a lot of the initial trials in medicine, and when you get away from fields like social and developmental psychology and focus more on clinical psychology and cognitive psychology.
Claims about "over half of all studies" just make you lose all credibility when the number is significantly lower.
|
On November 20 2017 12:42 DarkPlasmaBall wrote:Show nested quote +On November 20 2017 10:56 LegalLord wrote:
If anything, the Nate Silver versus the world situation shows either a widespread statistical illiteracy, a tendency for doctoring data toward a desired result for political ends, or perhaps both. The fact that Nate was criticized for not being strongly enough in favor of saying that Hillary will win it suggests at least one of those options. I agree with the bold part because of how people interpreted Trump's victory as a refutation of the reliability of statistics despite it being a single trial. It's unfortunate that most people don't understand basic statistics and probability, and think Nate Silver is an idiot. Ironically, I'm sure some of these people play the lottery.
The problem is that there is a widespread problem in statistical practice, so even if the people you're referring to are wrong in their reasoning for coming to said conclusion (for the reason you stated), they're not altogether wrong. (Educated) People are, in general, far too trusting of research results in general. It's really unfortunate as well because researchers are supposed to be the experts, so if they're misapplying statistics, how is the average layman supposed to know whether a given result or expert is trustworthy or not? Even if you're a statistical expert, you don't have the time to read through every paper you come across and determine whether it's a legitimate result or not. It's a real problem that undoubtedly has huge economic costs.
Pollsters, to be fair, should be the least of their worries though. And Nate Silver is not a malpracticer.
|
|
|
|
|
|
EPT
