1. We have added a Gift Upgrades feature that allows you to gift an account upgrade to another member, just in time for the holiday season. You can see the gift option when going to the Account Upgrades screen, or on any user profile screen.
    Dismiss Notice

[RD] US 2016 election: Poll watching thread

Discussion in 'Off-Topic' started by dutchfire, Jul 14, 2016.

  1. tetley

    tetley Head tea leaf

    Joined:
    Nov 8, 2001
    Messages:
    3,211
    Location:
    Igloovik
    Trump may be too incompetent to rig an election, but the GOP sure as heck isn't.


    By definition of a broken democracy, no they aren't.
     
  2. Sommerswerd

    Sommerswerd Rest in Peace Black Panther

    Joined:
    Oct 10, 2007
    Messages:
    18,928
    Location:
    Wakanda Forever
    Again, that is a premise that you accept as true, not me. I maintain that the democracy isn't broken, and the leaders reflect the tastes of the electorate... the problem is that the electorate does not live up to its own lofty idealized version of itself.
     
  3. SMcM

    SMcM Emperor

    Joined:
    Jan 22, 2016
    Messages:
    1,357
    Gender:
    Male
    Location:
    London, England
    You know, a week ago I'd have been confident Clinton had this in the bag. But it seems to me now that in reality ever since the release of that embarrassing recording of Trump bragging about getting his way with women (or at least being able to), many seem to have just looking for an excuse to support Trump again, and this new FBI situation seems to fit. They don't care when people point out nothing new has really come out about Clinton, they've got their hands over their ears and the facts here don't fit their narrative. So I from where I'm sitting, looking at polls and other information, this could be very close.
     
    Camikaze likes this.
  4. tetley

    tetley Head tea leaf

    Joined:
    Nov 8, 2001
    Messages:
    3,211
    Location:
    Igloovik
    I can see both sides to the premise. On the one hand, there has been foreign influence in U.S. elections for a long time. But on the other, the reason the foreign influence is there is because the country has such a zealous religious right, who immediately cast out anyone who even questions Israel's involvement in U.S. affairs or anything.

    That has had a domino effect. Israel's treatment of the U.S. as their client-state--enabled by the electorate's religious fervor--resulted in the CIA getting extensively involved in elections throughout the Middle East (e.g. Egypt, Yemen, Palestine, Syria...). Add only the latest conflict--Syria--to the mix, and pretty soon you have Russian involvement in the U.S. election.

    I won't go so far as to say religion is at the root of it, but the religious misguidings of the electorate absolutely did start it.
     
  5. tetley

    tetley Head tea leaf

    Joined:
    Nov 8, 2001
    Messages:
    3,211
    Location:
    Igloovik
    My own posts last week can be my witness, I was expecting a curve ball all along. And I got it.
     
  6. Bootstoots

    Bootstoots Deity Retired Moderator

    Joined:
    Mar 2, 2003
    Messages:
    9,437
    Location:
    Mid-Illinois
    Wait, there's a baby in that bathwater you're throwing out. An ugly baby, but still a baby.

    I understand and largely agree with the sentiment behind this post, but the substance is wrong. I am not a fan of reducing people to a set of demographics and polling data weighted by demographic, then constructing a model to predict their behavior, but this is a phenomenon that goes far beyond election polling and makes up most of modern marketing. This is of course a reason to feel even worse about it. The reason this approach is used and that people talk obsessively about it is that it works. It doesn't work perfectly, but it does far better than random chance or asking "experts" with no data. That's the whole reason a model is created.

    The function of all models is to take input data and construct a model of a complex world which works most of the time. Take, for instance, the weather forecast. Your weather forecast for 3 days from now may easily be wrong, but on average it will be far closer to correct than just guessing randomly, or even just always guessing whatever the average weather conditions are for this date at your location. Advances in computing have improved weather forecasting quite a bit in the past twenty years, so that now models do better than chance even 10 days out (not well enough to plan by, just better than chance).

    In election polling, models of poll data have done a fairly good job of predicting the outcomes of US elections for at least the last eight years. That doesn't mean there isn't occasionally a miss, or that polls can't be systematically off by several points. This election is harder to deal with because it's fundamentally unlike any other US election, introducing more sources of uncertainty. But it is highly likely that the final results will be closer to the polling data than to a baseline prediction like just predicting that the vote will be exactly the same as in 2012. Still, there could easily be big surprises as well.

    One particular nitpick: the polls haven't shifted 10% in the past few days. Hillary Clinton's chances of winning have shifted by about 10 percentage points (even 15 as of this point, according to 538 polls-only), but this corresponds to a shift from a predicted margin of 7.0 percentage points to a margin of 4.0. Shifts of 3 percentage points affect the chances of winning by much more than that when an election is fairly close. This sort of shift has not been uncommon at any stage of polling; this race has undergone several shifts of greater than this magnitude, and it is likely that this is a "real" effect in that people are actually changing their minds, rather than just some sort of random fluctuation or herding behavior.

    I'd encourage you to learn more about how models are used in the world of "big data". I find "big data" and associated modeling much more dystopian than utopian, but it works so well on average that it is basically how sites on the internet make money while leaving their sites free to view. Also, try to figure out who's just selectively quoting polls and models in support of their preferred candidate, and who's actually really responding to the data divorced of their political views as much as possible. There are people of both descriptions in this thread, and distinguishing the two is crucial to good criticism.
     
  7. tetley

    tetley Head tea leaf

    Joined:
    Nov 8, 2001
    Messages:
    3,211
    Location:
    Igloovik
    But are the polls reflecting early voting numbers? 25% of registered voters voted before the email news even broke out. And that means more than 25% of those who will eventually vote, already did. However, I tend to think undecided voters predominantly did not vote that early.

    I see no indication that even 538 factors in early voting. Undecided votes just constitute uncertainty, the same as always (but if you early voted, you're no longer undecided).
     
  8. innonimatu

    innonimatu Deity

    Joined:
    Dec 4, 2006
    Messages:
    12,672
    It does not, and that was my point: ultimately both the forecasts and political maneuvering based on polling fail. The problems is what happens in the meanwhile.

    There are two different issues here. One is the accuracy of polling. I don't wish to deal with that at length except to say that people can and will change their mind or mis-report. The "science" of pooling cannot compensate for the unknown in the present (errors and deceit) or in the future (the unexpected). Even if polls could build an accurate picture of people's preferences at the point in time when they are done, the very fluctuation we can observe in polls trough time show that they cannot predict elections in the future. It's a guess! You can call it an informed guess, but to pretend that accurate numeric odds can be calculated for n election in the future is just wrong. It's entertainment, content for the "infotainment" industry.

    But my issue is not with this. My issue is with the other problem, the big problem, with polling: candidates use polls to constantly adjust their message. But their real attitudes, opinions, and plans are not that flexible - those do not change. The infotainment around pooling encourages candidates to become professional liars, much more that the political debates would otherwise. Win this demographic over, with that demographic over... Parties and candidates no longer act as if citizens are their equals: individuals with political rights with whom to engage in political debate. No, instead citizens are treated as given data points to be manipulated, accumulated to reach a set total.
    The perfect politician in that environment is the one that ticks enough "issues" to win the necessary majority. This politician must not believe anything, rather he/she must be able to pretend to believe in enough stuff to win. This politician must not have convictions because those are not "flexible" enough. And polling is the tool for the "technicians of political lies" (aka campaign staff) to know what they should lie about to win an election. It provides the real-time feedback to know whether the lies are working or should be adjusted.

    People are not dumb. This may work at first but we all see through it. We all call politicians liars now. Worse, when one politician who comes across as honest tries to run for an election, party officials will do everything they can to block him because he's deemed "unelectable" (yes, you know whom I'm referring to).
    The workings of democracy, and public faith in democracy, is being destroyed by this method of campaigning. I don't think that the recent failures of pollsters in elections around the world are accidents. I think they are a result of a diffuse resistance by populations to cooperate with this way do doing politics, and with the "technicians" who enable it. And that is what will save democracy from this very serious threat: that people are individualistic and rebellious enough to deliberately make this "science" fail.

    It is not just like marketing, it is marketing. But modern marketing is very much hit or miss, and that the largest companies in the game are deliberately opaque with their data to prevent their customers from evaluating the results of the advertisement they buy.
    Fortunately for us all, it doesn't work very well.

    I happen to have already had to deal with big data. It doesn't scare me because one this is always true: garbage in, garbage out ;) Most companies that invested in big data to mine their piles of random facts ended up canceling their projects because they could not get anything useful out of it. They do like to pretend they have lots of valuable information. The quality of the data being fed into models is the really intractable obstacle, not the cost of crunching data. It turned out that so many correlations were devoid of meaning that it took to much human effort to pick any useful results.
     
    Last edited: Nov 1, 2016
  9. metalhead

    metalhead Angry Bartender

    Joined:
    Apr 15, 2002
    Messages:
    8,031
    Early voting is baked into the poll numbers, in the sense that pollsters should be asking respondents if they voted, and if so, for whom. And then it simply goes into the poll number as part of the sample response. So the polling numbers and, by extension, the forecast ought to include the early vote.

    But I think that exposes a weakness of an exclusively poll-based statistical model. They currently have Nevada - whose EVs are of vital importance to Trump given how few opportunities he has - as little better than a toss-up. But Jon Ralston, who knows Nevada voting perhaps better than anyone, says that early voting patterns are a close match for 2012, where Obama won by nearly 7 points. He says barring some huge shakeup in the voting patterns of party registrants, i.e. either lots of registered Ds or Is voting Trump - which polling says is not going to happen - Clinton is going to win the state.

    Early voting in most other places seems to be showing far less clear results - except for Texas, where most commenters still seem to give a wink and a nudge but I think Clinton has a real chance at winning.
     
  10. Archbob

    Archbob Ancient CFC Guardian

    Joined:
    Oct 25, 2000
    Messages:
    11,774
    Location:
    Corporate USA
    Well, Trump has to have a real good showing since Early voting patterns suggest Clinton up by 16 points or so thus far, he better have huge turnout on the 8th(or the 28th as he thinks).
     
  11. Kyriakos

    Kyriakos Alien spiral maker

    Joined:
    Oct 15, 2003
    Messages:
    59,013
    Location:
    Thessalonike, The Byzantine Empire
    Are those votes actually known? I suspect it would be illegal to announce even if known, so what exactly are those numbers about?
    Or do you wish to have a "problematic democracy Egypt" 4-day running election in voting stations :D
     
  12. Archbob

    Archbob Ancient CFC Guardian

    Joined:
    Oct 25, 2000
    Messages:
    11,774
    Location:
    Corporate USA
    Probably either from exit polls or the number voting from Democrats and Republicans.
     
  13. tetley

    tetley Head tea leaf

    Joined:
    Nov 8, 2001
    Messages:
    3,211
    Location:
    Igloovik
    The pollsters don't count the actual early voting ballots, but they gather statistics for the early voting turnout by region, by party, by race, etc. and they take an educated guess. Like saying absentee ballots in Texas are primarily going to be Republican votes. a) Texas votes are primarily Republican to begin with, and b) Texas does not give absentee ballots to just anybody, like some states do. You have to be physically not present. That means overseas military.
     
  14. SMcM

    SMcM Emperor

    Joined:
    Jan 22, 2016
    Messages:
    1,357
    Gender:
    Male
    Location:
    London, England
    There could always be another dramatic swing in the other direction though. Someone had been sitting on that recording of trump for the best part of a decade, and it was conveniently released in time for the next presidential debate, just weeks before election day. I can easily imagine more dirt being brought to life about Trump within days of the vote.
     
  15. Camikaze

    Camikaze Administrator Administrator

    Joined:
    Dec 27, 2008
    Messages:
    27,234
    Location:
    Sydney
    One thing I'm finding interesting about the 538 model is that it's currently showing that Clinton has a 71% chance of winning the election, but an 81.3% chance of winning the popular vote. So over a third of Trump's possible winning scenarios involve him losing the popular vote but winning the electoral college. It's often thought that the Dems have an electoral advantage, but in a tightened national race the GOP has more of a shot at winning from 2nd place.
     
    SMcM likes this.
  16. tetley

    tetley Head tea leaf

    Joined:
    Nov 8, 2001
    Messages:
    3,211
    Location:
    Igloovik
    I don't think 538 is putting that many eggs in the basket of winning popular but losing electoral, so much as it is losing by other means. Means that have never happened before. Trump refusing to concede and the GOP trying to settle everything in the 8-judge court that they caused comes to mind.
     
  17. AdrienIer

    AdrienIer Deity

    Joined:
    Mar 4, 2010
    Messages:
    3,379
    Location:
    Paris
    No it's just that the national polls are still very much in Hillary's favor while the state polls aren't necessarily. Her gains in Texas are impressive but they probably won't get her any EVs.
     
  18. Thedrin

    Thedrin Deity

    Joined:
    Jan 24, 2002
    Messages:
    2,650
    Location:
    London
    They've written about this 10% being the possibility of Trump winning the electoral college but losing the popular vote. I'm not sure how they're model could produce a result of Trump refusing to concede.
     
  19. Bootstoots

    Bootstoots Deity Retired Moderator

    Joined:
    Mar 2, 2003
    Messages:
    9,437
    Location:
    Mid-Illinois
    What I mean is that there is still value in imperfect models and imperfect data as compared to any other method of making a decision. Advertisements can never be targeted especially well, but modeling consumer data can improve a company's hit rate by enough to be worth investing in. The result when data mining is that the model rarely tells you much about causality and you never really know what correlations are meaningful and what aren't, beyond what you can obtain by throwing statistics at it. Still, while this is all sloppy, and while companies that sell targeted ads have every incentive to hide the data so that they can pretend the results are better than they really are, the results of a competent use of big data are on average measurably better than just having some "marketing expert" decide what an ad should be and throwing it out to everyone without targeting it.

    I mean, we're at the point now where computers are doing reasonably well at image recognition, can drive cars, and so on. These sorts of skills involve processing data into fuzzy categories in a consistent way, and computers are finally getting pretty good at this.

    I think for instance that this:
    shows a misunderstanding about how sloppy real-life data can produce something with nonzero value. If a meteorologist says that there's a 40% chance it will rain in 3 days, there's nothing absolute about that number. There's no way to know what the probability of rain really is, nor is it possible to measure current meteorological variables to infinite precision. The problem with weather forecasting is that the atmosphere shows chaotic, nonlinear behavior, such that even tiny unmeasurable changes in anything (e.g. the thickness of a cloud layer, wind speed at some altitude, etc) can result in totally different outcomes over time. But it is still possible to take a model, make a whole bunch of very small random adjustments (within measuring error) to your initial state, and then report the percentage of models that predict rain in 3 days. You can take a different model and do the same thing, and report the chance of rain according to it, and finally average all of your models with all of your different starting states and get something like 40%.

    And yes, this is sloppy. But the prediction is checked against what actually happens, with constant updating to make sure that forecasts of 40% rain come up with rain as close to 40% of the time as possible. It's not garbage in, garbage out. It's slightly old but still probably okay food in, older and slightly moldy but edible if you scrape the mold off food out. The world is a mess and data are somewhat unreliable but we can come up with things that are better prediction tools than nothing. This is a really subtle point that most people misunderstand.

    In the case of models like 538, of course there's no way to know that Hillary Clinton has a 71.0% (polls-only) chance of winning the election. But that doesn't mean the poll and model numbers are nonsense and have entertainment value only. They do much better than just throwing darts at the map, or making a prediction that it will be exactly the same as 2012, or having "experts" pull numbers directly from their posteriors, or any other baseline prediction. This has been demonstrated consistently since 2008 in 538's case: it misses about as often as it would be expected to miss based on the odds it gives. Substantial polling misses, such as some 2014 Senate races and the Michigan Democratic primary in 2016, are still called incorrectly, but by and large the model does match the real result quite well.

    I do agree with you, though, that data-driven approaches have a sort of toxic effect where people are aggregated together and treated as nothing more than the sum of their demographic groups, and then ruthlessly pragmatic politicians that don't really believe anything make targeted appeals to each demographic group. It's not clear to me that this is actually even an effective approach; it seems to have a vulnerability in that someone who purposely ignores data and just says whatever he thinks at any given time seems far more genuine than someone whose every public word is a targeted appeal, even if the former is actually an even more prolific liar than the latter.
     
  20. FriendlyFire

    FriendlyFire Codex WMDicanious

    Joined:
    Jan 4, 2002
    Messages:
    19,588
    Location:
    Sydney
    :crazyeye:

     

Share This Page