Comments on FiveThirtyEight: Politics Done Right: Should we be discounting Obama's lead? The next a...

酒店經紀人, 菲梵酒店經紀, 酒店經紀, 禮服酒店上班, 酒店小姐兼職, 便服酒店經紀, ...

2009-10-08T08:28:52.391-04:00

酒店經紀人,
菲梵酒店經紀,
酒店經紀,
禮服酒店上班,
酒店小姐兼職,
便服酒店經紀,
酒店打工經紀,
制服酒店工作,
專業酒店經紀,
合法酒店經紀,
酒店暑假打工,
酒店寒假打工,
酒店經紀人,
菲梵酒店經紀,
酒店經紀,
禮服酒店上班,
酒店經紀人,
菲梵酒店經紀,
酒店經紀,
禮服酒店上班,
酒店小姐兼職,
便服酒店工作,
酒店打工經紀,
制服酒店經紀,
專業酒店經紀,
合法酒店經紀,
酒店暑假打工,
酒店寒假打工,
酒店經紀人,
菲梵酒店經紀,
酒店經紀,
禮服酒店上班,
酒店小姐兼職,
便服酒店工作,
酒店打工經紀,
制服酒店經紀,
酒店經紀,
菲
梵,

艾葳酒店經紀公司提供專業的酒店經紀, 酒店上班小姐,八大行業,酒店兼職,傳播妹,或者想要到打工兼差、...

2009-09-10T03:01:17.980-04:00

艾葳酒店經紀公司提供專業的酒店經紀, 酒店上班小姐,八大行業,酒店兼職,傳播妹,或者想要到打工兼差、打工,兼差,或者八大行業,酒店兼職,想去酒店上班, 日式酒店,制服酒店,ktv酒店,禮服店,整天穿得水水漂漂的,還是想去制服店當上班小姐,水水們如果想要擁有打工工作、晚上兼差工作、兼差打工、假日兼職、兼職工作、酒店兼差、兼差、打工兼差、日領工作、晚上兼差工作、酒店工作、酒店上班、酒店打工、兼職、兼差、兼差工作、酒店上班等,想了解酒店相關工作和特種行業內容,想兼職工作日領、假日兼職、兼差打工、或晚班兼職想擁有快速賺錢又有保障的工作嗎???又可以現領請找專業又有保障的艾葳酒店經紀公司!

艾葳酒店經紀是合法的公司工作環境高雅時尚，無業績壓力，無脫秀無喝酒壓力，高層次會員制客源，工作輕鬆，可日領、現領。
一般的酒店經紀只會在水水們第一次上班和領薪水時出現而已，對水水們的上班安全一點保障都沒有！艾葳酒店經紀公司的水水們上班時全程媽咪作陪，不需擔心！只提供最優質的酒店上班,酒店上班,酒店打工環境、上班條件給水水們。心動嗎!? 趕快來填寫你的酒店上班履歷表

水水們妳有缺現領、有兼職、缺錢卡奴的煩腦嗎?想到日本留學缺錢嗎?妳是傳播妹??想要擁有高時薪又輕鬆的夜間兼職工作,打工機會和,假日打工,假日兼職賺錢的機會嗎??想實現夢想卻又缺錢沒錢嗎!??
艾葳酒店台北酒店經紀招兵買馬!!徵專業的酒店打工,想要去酒店的水水,想要短期日領,酒店日領,禮服酒店,制服店,酒店經紀,ktv酒店,便服店,酒店工作,禮服店,酒店小姐,酒店經紀人,
等相關服務幫您快速的實現您的夢想~!!

^^ nice blog!! ^@^徵信, 徵信, 徵信, 徵信社, 徵信社, 徵信社...

2009-03-11T05:40:00.001-04:00

^^ nice blog!! thanks a lot! ^^徵信, 徵信社,徵信, 徵信社, 徵信...

2009-03-11T05:40:00.000-04:00

情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,...

2009-02-12T22:04:00.000-05:00

徵信社, 感情挽回, 挽回感情, 徵信, 徵信社, 徵信, 捉姦, 徵信公司, ...

2009-02-08T22:44:00.000-05:00

Whoops!The 1968 election wasn't close in electoral...

2008-07-19T21:45:00.000-04:00

Whoops!

The 1968 election wasn't close in electoral votes!

Much can change between now and November -- a gaff...

2008-07-19T21:31:00.000-04:00

Much can change between now and November -- a gaffe here, a dirty trick there, and someone promising to "deliver" a State by making it easy for the "right" people to vote and difficult for the "wrong" people to vote or to tamper with voting devices. Some indications suggest that this election will be close, and that John McCain has a genuine chance. If he can make a marginal nationwide gain, he might pick off Ohio and maybe Pennsylvania or Colorado as well as Virginia, Missouri, and Indiana which are statistical ties... and the election. It wouldn't take much, but nobody can predict how that is possible. Remember well; Kerry was ahead of Bush by about the same amount as Obama is ahead of McCain at the same time in 2004.

But if things go the other way, it won't be a close election in electoral votes. Obama solidifies his hold of Pennsylvania and Ohio, picks off a few States that rarely vote for Democratic candidates for President (including Virginia and Indiana), a bellwether state like Missouri, and seem to have been trending Republican (North Carolina, Georgia, Florida). A close election then becomes a landslide in the contest for electoral votes. Note well that North Carolina, Georgia, and Florida have together 57 electoral votes -- two more than California alone. That scenario gives Obama 400 or so electoral votes.

It's surprising at first sight that the distribution of electoral votes for winners has a gap between 307 (JFK, 1960) and 370 (Clinton, 1992). Presidential candidates seem to win by 40 or fewer electoral votes in really-close elections (1916, 1948, 1960, 1968, 1976, 2000, and 2004) or by 100 or more.

The pattern holds because a gain in national percentage of votes cast results in blocks of states flipping -- and at times one of those blocks is the difference between a close election and a not-so-close election. It will be a close election should Obama win the States that Kerry won and either Ohio or the combination of Colorado, Iowa, and New Mexico. that's still around 300 electoral votes at most. He could win with two of Virginia, Missouri, and Indiana -- but he's not going to win either of those unless he wins Ohio, and if he picks one of those three states, it's still close (about 310 electoral votes). With those three, it's 330 electoral votes -- undiscovered country.

Strategies likely to win in Indiana, Missouri, and Virginia put North Carolina, Georgia, and Florida in reach; those States probably vote together, and who knows what goes with them. That pushes the count to about 400 electoral votes.

Obama's campaign seems to have learned something that neither Gore nor Kerry got -- don't target one particular State. The other side might prove capable of ensuring that that one State doesn't have a fair election. Would Bush have won Florida in 2000 or Ohio in 2004 without such deeds? Who knows? Will Obama or McCain resort to such techniques this year? That depends on their character.

It is possible this year that one of the candidates can have a better record in history by choosing an honest loss over a dishonest win. It is essential that those who have partisan axes to grind remember that winning isn't everything; the credibility of the game is even more important.

But one thing that I don't predict is an Obama victory with 320-360 electoral votes. He will lose by 10 or fewer, win by 40 or fewer, or win by 100.

Anonymous said, "why would you discount each state...

2008-07-02T18:29:00.000-04:00

Anonymous said,
"why would you discount each state by 2.1 percentage points even though their snapshot margins vary from very small to very large?
i think you might need to rethink this."
The reason you have to re-think this 2.1% backswing is that you have used incredible precision in analyzing the snapshot of each state and you throw away this precision by throwing in an arbitrary number. It is akin to measuring the distance from the earth to the moon to the nearest meter. Then, when you realize that the distance changes in the course of the month, you add 100,000m because it makes you feel better.

Nate,Have you looked into Bayesian regression? Th...

2008-06-30T11:19:00.000-04:00

Nate,

Have you looked into Bayesian regression? The general idea is that you explicitly include your prior expectations in the regression analysis. For example, if your prior expectation is that smaller margins of victory are more likely, the model would be biased towards predicting smaller margins of victory. There are a lot of benefits to this approach:

1) It does exactly what you are trying to do here by favoring more realistic margins of victory.

2) It is built on a solid, rigorous mathematical foundation.

3) It enables you to use a large number of variables in your model without having to worry about the noise inherent in variable selection.

4) It makes convergence to the "best" model a smoother process.

5) It is very easy to implement algebraicly. It's really just a matter of adding a single term to your regression.

Some of the characteristics of your model appear to be attempts to compensate for the fact that you are inherently thinking about this problem in a Bayesian sense, but you have not modeled it as such. The Bayesian approach would be more elegant, informative, and arguably more rigorously defensible.

I have a question. Have you factored in the possib...

2008-06-27T21:27:00.000-04:00

I have a question. Have you factored in the possible point shaving effect of race?

I have run simulations that show a solid 2 point shave and a probable 4 point shave and a possible 6 point shave from Obama's total.

That would make the race much closer (or a loser for Obama) that your forecast shows. I COULD BE WRONG.

I dont have expertise in this area or a broad enough data set to draw any conclusions I'd place money on. And some of the racial aspect of the data was infered.

So I'd like to know your thoughts on this.

Lastly, later you might want to look into neural networks to augment/replace your regression. They can significantly improve your results if the data isn't linear and can sometimes infere effects like race.

My issue is that people may believe Nate's predict...

2008-06-27T17:45:00.000-04:00

My issue is that people may believe Nate's predictions are fact and it therefore could suppress voting, which is bad for everyone. Especcially with the recent publicity, and the none mathematical public masses viewing these predictions.

Oh, please. Nobody reads this sight (or TNR, for that matter) who isn't a political junkie who is going to vote. People need to stop thinking that websites with a few thousand readers are going to have some kind of macro impact on the election.

That revised lead seem paper thin to me. Bradley ...

2008-06-27T16:44:00.000-04:00

That revised lead seem paper thin to me. Bradley effect? Obama wasting votes in the NE and CA piling up big majorities there, leaving him fewer votes to spread over the battleground states he needs to win.

It should be an interesting night on Nov. 5.

Watch Maine!

2 points: 1. I think your model should take into a...

2008-06-27T15:19:00.000-04:00

2 points:

1. I think your model should take into account undecideds. As I have said before, I suspect that a state with higher undecideds is more uncertain, (and thus the last-minute-movement correction should be higher), but that you should correct for the effect of different polling outfits having different undecideds.

2. A much stronger effect than elections closing at the last minute is 3rd-party candidates collapsing at the last minute. I think that your 3rd-party gap should be about half as big. Luckily, this changes nothing about your results, so you can study the issue and do it later. Still, it is really hard to justify fixing one thing without the other.

why would you discount each state by 2.1 percentag...

2008-06-27T14:49:00.000-04:00

why would you discount each state by 2.1 percentage points even though their snapshot margins vary from very small to very large? the proportionality should apply at the state level too, no??

plus, your 2.1 rule actually swings some states from one candidate to another. that seems silly, because your whole point is that margins move toward zero over time...why would it start moving away from zero after it crosses to the other candidate?!?

i think you might need to rethink this.

I follow Nate's site not because he can in any way...

2008-06-27T14:41:00.000-04:00

I follow Nate's site not because he can in any way shape or form predict the future and the election. I follow because it is by far the best sight at analyzing the current trend of likely voters in the populace and tells both parties what they are up against going into the closing months. It is extremely valuable for that lesson.

When I learn that Obama has an innate advantage in a certain state or area because of its demographics and party ID and learn how to balance that against the voter trend and sentiment, I can tell more exactly the swing needed to turn the election.

I just do not believe it is predictive, it is only taking basic trends as reflected in polls and balancing those numbers in numerous weighted ways to fully reflect the complexity of the current situation.

But then again I do not think polls are by there nature predictive of outcomes at this stage, nor should they be. Too much debate on issues needs to take place first.

My issue is that people may believe Nate's predictions are fact and it therefore could suppress voting, which is bad for everyone. Especcially with the recent publicity, and the none mathematical public masses viewing these predictions.

I try not to snark on other commenters too much, b...

2008-06-27T12:58:00.000-04:00

I try not to snark on other commenters too much, but I find it amusing the number of comments on this post taking the position "it's impossible to predict the future."

Might I humbly suggest that if you believe that, a site whose whole purpose is to project the outcome of the election may not be the place for you? Critiquing the method is fine, but if you declare that the results are meaningless because the objective is impossible, you're essentially saying "stop doing this," which Nate is unlikely to do (nor, based on the traffic, do most readers want him to.)

And on a lesser note, for those who believe it is impossible to predict without directly taking into account all the factors that have produced election results in the past, keep in mind that even for past elections, most of the "reasons" for the outcome are speculative at best; only marginally better than the daily "reasons" cited by financial analysts for why the stock market moved the way it did. Statistical modeling does not require knowing the exact cause and effect for everything; indeed, it can be used to discover effects that contradict conventional wisdom.

Higglytown:In looking at you comment about the mar...

2008-06-27T12:15:00.000-04:00

Higglytown:

In looking at you comment about the margin of error it sounds as though you are not taking into consideration that if you have a collection of polls, averaging them means that the margin of error for that combination is signficantly smaller than the indiviual sampling errors of each poll.

If you have three polls each with an n of 400, the samplng error of each (all other things being equal) is 5%, but the sampling error of their average (all other things being equal) is 2.89.

The other really telling number in this posting is...

2008-06-27T12:08:00.000-04:00

The other really telling number in this posting is that in elections where both candidates have led in polling at any particular point, a person with at least a 6 point lead has lost 7 out of 9 times. 6 points does not seem nearly as secure as this website makes it sound. It sounds like Nate is saying at most a 6 point lead at this point can regress to 3.8 points. That is simply not true, numbers historically move much greater and faster than that.

Are numbers already adjusted down for the Time, Ra...

2008-06-27T11:59:00.000-04:00

Are numbers already adjusted down for the Time, Rasmussen (6/27) and Gallup (6/26) polls (O +4, O +4, Tie) respectively? Cause in my math this thing is getting closer and closer to the margin of error on major polling.

I like the modification that distinguishes between...

2008-06-27T11:55:00.000-04:00

I like the modification that distinguishes between a snapshot and a projection. However, I would like to know what assumptions are being made regarding the corresponding uncertainty of the projected result.

My gut tells me that along with the narrowing of the margin of victory, there is a corresponding narrowing of the probability distribution, i.e. victory margins are shrunk, but the uncertainty of the outcome is diminished.

Perhaps there is useable information in the variability of the historical polling data that can improve the precision of the estimates as the election day approaches. (And then again, maybe this has already been taken in account.)

A couple of other thoughts:When there is "no info...

2008-06-27T11:52:00.000-04:00

A couple of other thoughts:

When there is "no information" the hypothesis needs to be 50%.

I have previously posted my preference for Shewart's/Deming's thoughts on changepoints (of which "trends" are one type - when to throw out data that has a special cause being another). The IndX/Mr chart would be useful I think. But I was just reading about Page,Shiryayev and Roberts. (Devlin , Lordon, "The Numbers behind Numb3ers" a birthday gift.) I'm thinking something like considering all your past and current simulation run and making the likelihood of IN going for Obama the canary in the coal mine.

If you are going to use long term trends from prev...

2008-06-27T11:42:00.000-04:00

If you are going to use long term trends from previous elections and want to project all the way to election day, why not also incorporate the long-term standard deviation (i.e. between polling averages 130 days before the election & final outcome)?
It would probably make the election at this stage a complete toss-up, but that would be more realistic.

In essence it would include all the noise of actual events.

Of the two changes made, I think the second is bas...

2008-06-27T11:39:00.000-04:00

Of the two changes made, I think the second is based on a slightly doubtful assumption. On the other hand, the first change more or less corrects the error I think comes with the second (and also came with the previous model).

I would argue that the two problems addressed by the new changes are connected: undecideds making up their mind is a significant part of the reason why races tighten when approaching election day. With their numbers now close to 20 percent in several states they will obviously affect the race.

Some comments have raised the concern that undecideds might not break 50-50, and that they may break differently in different demographics and thus differently in different states.

Discussing the first change, Blame asks if races don't tighten towards party ID rather than 50-50, and Lilnev makes the case that a state should regress towards its own mean, not the national mean.

I think there's much merit to this line of reasoning, and I think the main reason for this is that the undecideds who eventually vote (some probably won't), tend to vote according to their habits from previous presidential elections. If this is factored into the model, the result will be that states regress towards their own mean, and that a national popular-vote advantage will shrink.

Let's look at Utah, where Bush beat Kerry 72-26, but where McCain's lead in the trend-adjusted average is currently only 54-32 with approximately 14 percent undecided. Ignoring the rather improbable scenario that McCain has made significant inroads in Kerry's support but lost even more Bush voters to Obama, this should mean that almost all the undecideds voted Republican in 2004. My guess is that in this group, the majority of those that bother to vote in 2008 will go Republican again (possibly holding their noses while doing it). If this is the case, Utah won't approach 50-50, but rather move some distance towards 72-26.

Since the last two elections have been close, the accumulated nationwide effect of these trends would be that undecideds break in a way that tightens the race. In creating a method to factor in that a national lead in polling tend to shrink, Nate has basically taken this into account.

Some other commentators have argued that it doesn't make sense to calculate a nationwide adjustment towards an even split and then applying it evenly across all states. I agree with this, but it's also fairly obvious from the Utah example above that it wouldn't make sense to assume that all individual states will gravitate from current polling towards 50-50. Sure, many states will do just that, but mainly because the voters in these states have a tendency to divide their votes fairly evenly in presidential elections.

If polling history seems to agree with what I'm saying, it may be worth adjusting the calculation of how undecideds break. Since this will to some extent work in the same direction as the national polling advantage discount, the algorithm used for calculating the discount probably needs readjusting in order to keep it in line with previous election trends.

1. I think this trend-adjustment stuff has too mu...

2008-06-27T11:33:00.000-04:00

1. I think this trend-adjustment stuff has too much voodoo involved with it. The way to do "trends" would require understanding the correlation among neighbor states and to do a epidemiological "viral" simulation (like how disease spreads). Its possible to do but is way too complex I think to do.

2. Post the General Election Model regression models and coefficients on the FAQ. Who cares about Clinton model and primary race regression models.

3. Allocation of undecideds: Obama is change/reform/forward looking candidate. McCain is status quo/"experience" (whatever that actually means)/incumbent-light as 3rd term of Bush/Cheney.

There surely is some research (think Kaneman and Turversky) that shows some insights into why and under what circumstance undecideds will break (unless there are right conditions) for status quo or what seems most like status quo. The primary season gave us a preview: Clinton was incumbent-light. Obama was change/reform. Late deciding undecided tended to break for Clinton because of some of the underlying psychology behind "decision making" unless there are certain other circumstances which overcome risk-averse bias.

Consider the difference between incumbent and incumbent-light

Nixon - incumbent light/status quo
Kennedy - change/reform

Johnson incumbent/status quo
Goldwater - change

Humphrey - incumbent light/status quo except on civil rights
Nixon - change "back"
Wallace - change "back"

Ford - incumbent light/status quo
Carter - change

Carter - incumbent light because of hostages and Kennedy challenge
Reagan - change

Reagan - incumbent/status quo
Mondale - change (back)

Bush - incumbent light
Clinton - change
Perot - change

Clinton - incumbent/status quo
Dole - status quo but not incumbent
Perot - change

Bush - change from Clinton
Gore - incumbent light/ status quo

Bush - incumbent (but maybe incumbent light because of War)
Kerry - he was for change before he changed his mind on being not really change, etc. etc.

Obama - change
McCain - incumbent light