4 ms·

I don't get the effort to lionize (foxitize?) Nate Silver just before this election. The "other side" says: http://dailycaller.com/2012/11/01/is-nate-silvers-va

by guscost 14y ago

I don't get the effort to lionize (foxitize?) Nate Silver just before this election. The "other side" says: http://dailycaller.com/2012/11/01/is-nate-silvers-value-at-risk/ http://dailycaller.com/2012/11/01/is-nate-silvers-value-at-r... "In 2008, the RealClearPolitics polling average was incorrect in two states — Indiana and North Carolina. Silver botched Indiana but correctly called North Carolina. In 2010, it was much worse. State polling averages were wrong in Alaska (they said Joe Miller would be elected), wrong in Colorado (they said Ken Buck would be elected), and embarrassingly wrong in Nevada (they said Harry Reid would be involuntarily retired). FiveThirtyEight incorrectly forecast the winner in each of those states, perfectly reflecting the inaccurate information contained in the state polls. Thus, of the five major state races in which polls were wrong over the last four years, Silver only got one right." Why don't people wait and see what happens?

kmfrk 14y ago
The certainty of Romney's loss is bad for page views and viewer numbers for one. Pundits seem to have an interest in perpetuating uncertainty and turning the election into a narrative of a dead-heat horse race.
- cafard 14y ago
  Well, it worked that way in the Republican primaries, didn't it? There was never much doubt that Romney would be the nominee, but quite a cast got turns at playing the non-Romney option of the week.
btilly 14y ago
Duh. Base a model on polling data and you are only as good as your polling data. If the election is swung by cheating this time (as evidence suggests has happened plenty of times in the past), polls will be spectacularly wrong. Incorporate everything else you can think of and you get only a bit more accurate. However the question is this. Do you have anything more accurate? I've had the experience on NPR of hearing pundits going on at length about what states matter. And it was less informative than a glance at http://www.electoral-vote.com/ http://www.electoral-vote.com/ is. So no, it is not perfect. Nobody, particularly Nate, says that it is. But we don't have anything better.
- jeffdavis 14y ago
  So what does he add to the conversation? We can all see the electoral map. It's just a question of which polls you believe (or want to believe).
  - mikeryan 14y ago
    The five thirty eight model is actually based on accounting for accuracies in the polling data. It take's a poll's historical accuracy into account when weighing its impact on the model. This is where he gets an additional degree of accuracy as opposed to straight poll aggregation. So it actually does "choose" which polls to believe...
    - mayneack 14y ago
      Yes. Unbiased != more accurate. If, for example, there is a poll you KNOW is doing something wrong, you should down-weight it. A pure polling average there would be wrong. However, once you start weighting polls, people start criticizing the way you do it.
- jtfairbank 14y ago
  Check out http://electionanalytics.cs.illinois.edu/ http://electionanalytics.cs.illinois.edu/ for an alternative to Nate Silver. It uses bayesian analysis and dynamic program to come up with a deterministic snapshot of the election. It is based on the polling data, so if the data is off the snapshot will be off, but otherwise it seems to provide a great snapshot of the Election.
  - mkr-hn 14y ago
    I think that's the same method Nate Silver uses. edit: I just looked at your comment history. Repeating the same comment over and over is going to cause problems.
schwabacher 14y ago
The Daily Caller analysis misses an important point - there is a much larger volume of polling data for presidential elections than for congressional elections. Drawing conclusion from one about the other is largely meaningless.
ajross 14y ago
People don't "wait and see" because they want to know now what the likelihood of the future is. That's hardly unique or surprising. This is an important subject. But I don't follow your evidence at all. The primary data output from 538 is a single number -- a chance of winning the election. How is that refuted by the inability to call every single state race? No one, certainly not Silver, claims to be able to do that. But numerate folks still understand that those probabilities can be combined to give a meaningful answer. So let's flip it around: why don't you believe the (currently) 83% figure? If your answer is because some of those states will come out differently than their median expectation, you're completely missing the point.
dbecker 14y ago
Sports fans are constantly arguing about who they think is going to win such-and-such game. It's not important... but people enjoy it. This isn't much different. For both the games and the election, we will wait and see what actually happens. In the mean time, we enjoy discussing and forecasting.

by guscost 14y ago

kmfrk 14y ago

The certainty of Romney's loss is bad for page views and viewer numbers for one. Pundits seem to have an interest in perpetuating uncertainty and turning the election into a narrative of a dead-heat horse race.

cafard 14y ago
Well, it worked that way in the Republican primaries, didn't it? There was never much doubt that Romney would be the nominee, but quite a cast got turns at playing the non-Romney option of the week.

btilly 14y ago

Duh. Base a model on polling data and you are only as good as your polling data. If the election is swung by cheating this time (as evidence suggests has happened plenty of times in the past), polls will be spectacularly wrong. Incorporate everything else you can think of and you get only a bit more accurate. However the question is this. Do you have anything more accurate? I've had the experience on NPR of hearing pundits going on at length about what states matter. And it was less informative than a glance at http://www.electoral-vote.com/ http://www.electoral-vote.com/ is. So no, it is not perfect. Nobody, particularly Nate, says that it is. But we don't have anything better.

jeffdavis 14y ago
So what does he add to the conversation? We can all see the electoral map. It's just a question of which polls you believe (or want to believe).
- mikeryan 14y ago
  The five thirty eight model is actually based on accounting for accuracies in the polling data. It take's a poll's historical accuracy into account when weighing its impact on the model. This is where he gets an additional degree of accuracy as opposed to straight poll aggregation. So it actually does "choose" which polls to believe...
  - mayneack 14y ago
    Yes. Unbiased != more accurate. If, for example, there is a poll you KNOW is doing something wrong, you should down-weight it. A pure polling average there would be wrong. However, once you start weighting polls, people start criticizing the way you do it.
jtfairbank 14y ago
Check out http://electionanalytics.cs.illinois.edu/ http://electionanalytics.cs.illinois.edu/ for an alternative to Nate Silver. It uses bayesian analysis and dynamic program to come up with a deterministic snapshot of the election. It is based on the polling data, so if the data is off the snapshot will be off, but otherwise it seems to provide a great snapshot of the Election.
- mkr-hn 14y ago
  I think that's the same method Nate Silver uses. edit: I just looked at your comment history. Repeating the same comment over and over is going to cause problems.

schwabacher 14y ago

The Daily Caller analysis misses an important point - there is a much larger volume of polling data for presidential elections than for congressional elections. Drawing conclusion from one about the other is largely meaningless.

ajross 14y ago

People don't "wait and see" because they want to know now what the likelihood of the future is. That's hardly unique or surprising. This is an important subject. But I don't follow your evidence at all. The primary data output from 538 is a single number -- a chance of winning the election. How is that refuted by the inability to call every single state race? No one, certainly not Silver, claims to be able to do that. But numerate folks still understand that those probabilities can be combined to give a meaningful answer. So let's flip it around: why don't you believe the (currently) 83% figure? If your answer is because some of those states will come out differently than their median expectation, you're completely missing the point.

dbecker 14y ago

Sports fans are constantly arguing about who they think is going to win such-and-such game. It's not important... but people enjoy it. This isn't much different. For both the games and the election, we will wait and see what actually happens. In the mean time, we enjoy discussing and forecasting.