Science is HARD

[thriftyandboringinohio]

The popular press loves to insinuate that science can't be trusted.
They often point to studies about diet or health that contradict each other -first the food is bad for you, then it's good for you, then it's bad again.

Here's a good article about why we see that; why we shouldn't lose hope; and why we are prudent to be skeptical of new findings.

Science Isn’t Broken

It’s just a hell of a lot harder than we give it credit for.

http://fivethirtyeight.com/features/...-broken/#part4

[jk]

Re: Science is HARD

the most frequently revised conclusions are biological - nutrition, hormones and such. these subjects are complex and hard to study, and for these reasons the standards for publication are much lower than e.g. in physics. biological studies are just looking for p<0.05, or 2 sigmas, or 5 random occurrences per hundred. physics' standard is 5 sigmas, or 1 occurrence in 3.5 million, a MUCH more rigorous standard.

[lakedaemonian]

Re: Science is HARD

I wonder if hybrid teams of scientists partnered with AI/machine learning would potentially reduce a chunk of those retractions?

Would that potentially reduce the risk of human failings in forcing data around an hypothesis?

[santafe2]

Re: Science is HARD

Science Isn’t Broken

It’s just a hell of a lot harder than we give it credit for.

Politics and entertainment are attributes people enjoy. Science requires thought and a lot of work. Like the "popular press", iTulip has never been a place where science gained support and respect unless there was money to be made. I would not expect that to change. I gave up discussing science here a long time ago.

[EJ]

Re: Science is HARD

Politics and entertainment are attributes people enjoy. Science requires thought and a lot of work. Like the "popular press", iTulip has never been a place where science gained support and respect unless there was money to be made. I would not expect that to change. I gave up discussing science here a long time ago.

Post VirZOOM the new iTulip.com will incorporate new tech and learnings from 16 years of iTulip experience.

[jk]

Re: Science is HARD

Politics and entertainment are attributes people enjoy. Science requires thought and a lot of work. Like the "popular press", iTulip has never been a place where science gained support and respect unless there was money to be made. I would not expect that to change. I gave up discussing science here a long time ago.

one science that has been most "discussed" is climate change. i think it's clear why i put "discussed" in quotes. another scientific issue that's been discussed is peak cheap oil vs the new oil bonanza of fracking vs alternative energy.

i don't see science or tech discussed apart from society wide scientific issues with major social and economic impact. but i don't think those discussions were necessarily about money being made; unfortunately, they were political, not scientific.

there is a shortage of people willing to discuss dispassionately issues of social and political importance, including scientific ones. if it's not dispassionate, it's not science.

i don't think the problem is that these issues require "thought and a lot of work." people will apply those things to economic issues here. but where passions enter, thought becomes a mere tool used for a political end. thought is not used as an instrument to try to find some kind of truth.

[Woodsman]

Re: Science is HARD

Jk, when you have the time I'd appreciate your thoughts on this piece making the rounds:

...This article isn't merely about gambling, but goes to the heart of what makes polls different among one another, and across time. And what should we be cautious of when interpreting the information, while almost never reading (and sometimes not having access to) all the underlying probability details of the poll generation? In particular, we'll delve into the inconspicuous L.A. Times poll here, where for much of the past month they showed Donald leading Hillary. How did they come to that, and what value is there in paying attention to alleged outliers?

Antagonism isn't perpetual

[jk]

Re: Science is HARD

Jk, when you have the time I'd appreciate your thoughts on this piece making the rounds:

have you read the nyt article about the 19yo black guy in chicago who's a trump supporter? and how trump's numbers in the lat poll jump whenever that guy rotates into or out of the polled group? it's linked in the article you posted.

i think the 538 poll analysis is pretty good. for one thing, it tries to correct for "house effects," i.e. longstanding bias in some poll toward one or the other party, by both adjusting for the bias and comparing poll results to earlier results of that same poll, so as to see trends. [note again, in the article you linked, they mentioned "the conservative-advocating rasmussen" - an example of a well-known bias in that poll.]

there are a lot of assumptions made in generating outputs from the raw polling data. what do you assume will be the turnout for various demographic groups? and how do you slice up the population into salient groups in the first place? the times points out that the source of trouble from that 19 year old comes from the lat poll using unusually small demographic slices. this means each demographic bucket has fewer members, thus potentially magnifying the effect of an outlier response.

the best way to avoid or at least minimize these problems is enlarging the sample size to reduce uncertainty, but that's expensive. amalgamating statistically corrected polls, as done at 538 is one way to do that.

the other problem with THIS election is the high number of undecideds. although undecideds and 3rd party supporters are diminishing, they are still more numerous than usual. this means more dispersion in the predicted possible outcomes. the article you linked mentions the low likelihood of a voter switching from one of the major candidates to the other. but the large number of undecideds, and the "leaning towards" groups, still make the voting population less predictable.

to produce its probabilities of victory, 538 runs a monte carlo simulation with 10,000 runs, and then counts the outcomes. the simulation is done at the state level, so this procedure uses the probability distribution for each state, and of course the electoral votes of each.

i'm not sure, however, if they include cross correlative probabilities in their monte carlo runs. just checked and indeed they do. that is, if e.g. trump in one run is assumed to do particularly well in pennsylvania, then the model also assumes he's simultaneously doing better than expected in other states with similar demographic/geographic/economic characteristics.

i am very mindful of the failed poll-based predictions about brexit and the columbia peace deal. and in other posts i've pointed to the phenomenon of black candidates regularly getting 5% fewer votes than predicted by polls, presumably because of disingenuous poll responses. thus, i won't be surprised if trump gets 5% more votes than predicted. of course 5% is very close to the currently predicted clinton margin, so i think this will be a very close election. HOWEVER, there still remain a lot of undecideds and 3rd party people who may decide for one candidate or the other. this means ANY predictions have a lot of uncertainty or dispersion.

the article you linked to proposes that there is systematic bias in most of the polls, and that this bias is towards the liberal candidate. what i'm saying is that even without any assumption of bias, there's a lot of reason to remain uncertain. i think that's the case even in the face of what seem like overwhelming odds. e.g. 538 has hrc's chances of winning at about 85% currently, the ny times model has it at 89%. i suspect the race is not as uneven as that, but i can't prove it.

[vt]

Re: Science is HARD

Yes the LA Times/USC poll is different but here is why. And don't forget they were extremely close with there prediction on the 2012 race:

http://www.latimes.com/politics/la-n...nap-story.html

[Chris Coles]

Re: Science is HARD

Post VirZOOM the new iTulip.com will incorporate new tech and learnings from 16 years of iTulip experience.

"Post VirZOOM" ? Hi there EJ, can we assume that you are past the point where the delights of running a business operation will be placed on hold to enable a return to your acute thinking on the subject of what lies ahead for all of us?

[EasternBelle]

Re: Science is HARD

nowhere close yet, Chris, I think,
Kind regards,
EasternBelle