Comments on FiveThirtyEight: Politics Done Right: A Refinement to the Adjustment, Part I

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、...

2009-09-10T01:31:03.167-04:00

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~!!

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,...

2009-02-12T22:50:00.000-05:00

You suggest that "There are three possible answers...

2008-06-16T17:25:00.000-04:00

You suggest that "There are three possible answers to that question.
1. The random walk hypothesis.
2. The bounce hypothesis.
3. The trend hypothesis.

I suggest a 4th answer based upon Shewart/Deming. Hypothesis 1 creates the error of treating "special cause" data ("a putative trend") as a systemic cause. Hypos 2 and 3 create the error of treating "systemic cause" data (random variation) as "special cause" data.

In other words, when does it make sense to call a trend a trend, i.e. a special cause and treat it as such?

Solution:
a. Create a control chart (IndX/Mr is a perfect chart because it accounts for changes over time).
b. determine "trend" by using so-called Western Electric rules.
c. If a trend exists by WE rules, then and only ten apply the trend hypothesis.
d. If there is no basis for suggesting trend/special cause then you must assume "randomn walk" within the parameters establish by calculated Upper and Lower Control Limits.

see
http://en.wikipedia.org/wiki/Control_chart

Nate Silver: "In other words, suppose that we are ...

2008-06-16T16:13:00.000-04:00

Nate Silver: "In other words, suppose that we are holding an election for the President of Hell. The candidates are Gary Condit and Mark Foley."

Laughing my butt off. Thanks, Nate.

I always thought the hell match-up would be Lieberman v. Inhofe or Delay. but Condit v. Foley even beats that.

Of course, there's always the Spitzer v. Vitter whoremonger special...

.

Re McCain and Indiana: I suspect the reason the re...

2008-06-16T13:01:00.000-04:00

Re McCain and Indiana: I suspect the reason the regression favors McCain by so much is that the variables that favor Obama are not characteristics of Indiana (not a Kerry state in 2004, not a big Hispanic population - yet), and the ones that favor McCain do seem to be.

That seems to match my personal observations in that this has been a red state for a long time, despite the enclaves of Democratic support in the corners and occasionally in Marion County (although the votes must come from somewhere if we have a 5-4 Democratic edge in the House). One of the things that brought me to this site (for which I am grateful in multiple ways) was the note in Newsweek that Indiana was in play. What a pleasant change from being a 6:05 state ...

Mike, I suggested that exact same indicator a whil...

2008-06-16T11:28:00.000-04:00

Mike, I suggested that exact same indicator a while back. I would love to see Nate incorporate it. Call it Decisive States.

The media is likely to make concerted efforts to m...

2008-06-16T10:43:00.000-04:00

The media is likely to make concerted efforts to make the "Bounce" hypothesis the operant one. Based on the the last couple of elections it looks like they're getting pretty good at it. I'd guess cable news is a big factor.

Do you have data from 1980? It seems to me that w...

2008-06-16T10:36:00.000-04:00

Do you have data from 1980? It seems to me that was the last "throw the rascals out" presidential election and thus the one closest in mood to the current climate, and most likely to accurately reflect whether we should be looking at a bounce, a trend, or a random walk over time from June to November. Admittedly, it was against a sitting president, so that throws things off. McCain, regardless of his policy views, is not Bush in person, and there do seem to be a lot of voters who choose a president on personality rather than issues. That trend is less proncounced for downstream races, so I expect a sweeping change in Congress, but the Presidential race could be either a nail-biter or a landslide; it's sort of poised on a tipping point.

Isabel Lugo scores again! Those logaritmic formula...

2008-06-16T10:14:00.000-04:00

Isabel Lugo scores again! Those logaritmic formulae are very interesting and useful (not just for Nate, but in general), and I did not know of them.

The point that I wanted to make was that your LOESS regression is still biased in favor of the "trend" hypothesis. Imagine some data with a perfect trend: win percentage is increasing by (say) 3% per week every week (that is, .12 in the log formula :). Your LOESS regression would hit every point, just like the trend hypothesis. Worse, if the last week suddenly showed no change after weeks of steady change, the regression could come in a hair *over* the last week's number.

A "random walk" hypothesis would lag by a very slight amount, proportional to the error on each week's data - essentially assuming that some small part of each increase might be polling error. And a "bounce" hypothesis would lag by more, supposing that those who change their mind one week are more likely to change it back the next.

Ray,(disclaimer: if it ever turns out to matter, a...

2008-06-16T08:56:00.000-04:00

Ray,

(disclaimer: if it ever turns out to matter, all logarithms in this comment are natural.)

I agree that log(A/B) might be a better measure of candidate A's lead over candidate B. If A and B are close together this is very nearly 4(A-B), so it's just a rescaling in close elections. But as A or B approaches 1, this measure of A's lead approaches positive or negative infinity, which is a property you'd want.

I'm not a statistician, but it's my understanding that this particular measure is used in logistic regression.

It's actually analogous to something that occurs in baseball forecasting. If you have a team that wins games against "average" opponents with probability A and a team that tends to wins games against "average" opponents with probability B, then when A plays B, A should expect to win in A(1-B)/(A(1-B) + B(1-A)) of the games it plays. This is Bill James' log5 formula. The formula seems kind of arbitrary but fits the available data quite well. But it turns out that if you use log(A/(1-A)) as your measure of A's ability, and similarly for B, then this becomes

log(p/(1-p)) = log(A/(1-A)) - log(B/(1-B))

where p is the probability of A winning over B.

This measure has some interesting properties. For example, consider a team with .500 winning percentage. They win one-third of their games against a team with .667 winning percentage. If you work out the winning percentage of a team that a .667 team will win one-third of the time against, it's .800. So the gap from .500 to .667 is the same, in some sense, as the apparently smaller gap from .667 to .800. These could of course be polling percentages instead of winning percentages.

I wonder how the 538 regression gives McCain a mar...

2008-06-16T04:48:00.000-04:00

I wonder how the 538 regression gives McCain a margin of 8.8 in Indiana, when none of the polls are that generous?

They should hold a general election every few week...

2008-06-16T04:39:00.000-04:00

They should hold a general election every few weeks to see if your models and methadology are right, Natster.

I think one reason people do not like the random w...

2008-06-16T04:26:00.000-04:00

I think one reason people do not like the random walk hypothesis is that not all 10 point swings in the polls are uniformly likely: Going from a 5 point lead to a 5 point deficit is much easier to do than going from an 90 point lead to a 100 point lead. And going from a 95 point lead to a 105 point lead is downright impossible.

Thus, after a 5 point bump, the candidate is more likely to drop 5 points in the polls than to rise an additional 5 points, but this does not mean that the candidate isn't equally likely to rise or fall in the polls -- just that if the candidate rises, the absolute change in polling is likely to be smaller.

I suspect that a more uniformly distributed metric of a candidate A's lead over candidate B is more like log(A / B) rather than A - B.
(This formula is borrowed from chemistry -- it computes the free energy difference between a reactant and product based on their
equilibrium concentrations.)

Thank you for this post (especially the last parag...

2008-06-16T04:13:00.000-04:00

Thank you for this post (especially the last paragraph). I still think it overemphasizes current trends, but now I better understand the reasoning behind the change. I still love this site though!

I have 1024x768 resolution, yet the page is just a...

2008-06-16T04:09:00.000-04:00

I have 1024x768 resolution, yet the page is just a bit too wide in a maximized browser. There seems to be some margin on the left side of the page from either the images or the menu, and on the right side, the graphics are of varying widths, making it look like there's a margin over there, too.

Obama's never getting zero EV. The lowest value y...

2008-06-16T03:59:00.000-04:00

Obama's never getting zero EV. The lowest value you see is 3 (Washington DC). Then you get some values and 6, 7 and 10 based on his two next-best states, which are Vermont and Hawaii.

But the reason you do see some numbers for him in the 3-10 range is because Illinois has not been polled in forever, and so the model has no choice but to assign a very large error bar to the Illinois forecast. Once Illinois is polled again, our estimate of Obama's potential range there will tighten, and you should see very few scenarios worse than about ~25 EV.

Anon @ 2:39 AM wrote:"This is a somewhat different...

2008-06-16T03:54:00.000-04:00

Anon @ 2:39 AM wrote:

"This is a somewhat different exercise, at least for the general election, since there's just one outcome this year. (On the other hand, in the primaries there were multiple chances, and while Nate missed some projections he did quite well on average.)"

Not that much different. We have here 50 different contests, same number as the primaries. If Nate can predict the outcome of these races with a better accuracy than others then on an "average" he is more likely to get the GE outcome (with the margins) right.

Nate, what's with the increased chance of Obama ge...

2008-06-16T03:47:00.000-04:00

Nate, what's with the increased chance of Obama getting 0-10 EVs, and why don't we see the same for McCain?

When you're in the forecasting business you have a...

2008-06-16T03:39:00.000-04:00

When you're in the forecasting business you have an acute sense of the laws of probability. If you're making a lot of forecasts, as Nate does for baseball players (about 1,600 individual players per year) and teams (30), then you're looking to do a better forecast than everyone else on average but you know that you're going to be wrong a lot of the time, too.

This is a somewhat different exercise, at least for the general election, since there's just one outcome this year. (On the other hand, in the primaries there were multiple chances, and while Nate missed some projections he did quite well on average.)

Fortunately, Nate seems to have a pretty thick skin, and he's also open to constructive advice. It's enjoyable watching him think through this problem and asking us for advice and criticism.

Nate, I think your efforts to impose order on chao...

2008-06-16T03:28:00.000-04:00

Nate, I think your efforts to impose order on chaos are fantastic. I'm grateful that someone reputable (scary, isn't it?) is trying to freshen up the data. I do expect that as time goes on and the numbers flood in you'll get to see how well the method works in real life. Of course we all want you to be exactly right, so it takes some courage to stand by your method. Keep it up; this civilian is fascinated.

Slack: The advantage of my method(as opposed to a...

2008-06-16T03:16:00.000-04:00

Slack: The advantage of my method(as opposed to a demographic one, which to a certain extent breaks down if it isn't fine grained enough, like West Virginia working class whites vs. Oregon working class whites), is that it is purely statistical, the correlation between the delta within a state over time is a relationship which, should, over time become more and more accurate, in seeing how different states respond to the same events.

Useful explanation."This is one thing, by the way,...

2008-06-16T03:15:00.001-04:00

Useful explanation.

"This is one thing, by the way, that I think some of the McCain supporters around here are missing."

Do not underestimate the number of Hillary trolls masquerading as McCain sock-puppets. A good percentage of Hillary supporters (not just the Hillaryis44 crowd) are 'secretly' hoping for an Obama loss -- you know for their "I Told You So" moment -- or at least they don't mind at all as a Democrat if McCain wins the Bush third term.

I also get the feeling, a good number of undecideds and soft supporters (especially among the independents) will back the front runner in the polls (also depending on the media narrative of how close is this race.)

Its akin to people being undecided about rating a movie as good or bad, but will follow the pre-dominant opinions after hearing the talk among their fellow movie goers; friends, movie critics etc.

The point is, if Obama makes a considerably high gain in favorablity ratings -- with the help of the media narrative-- his margin will increase even further.

The role of media is going to be huge in the coming months -- MORE than the previous election years. It can literally make or break the margins of one candidate or the other over a long period of 5 months.

There seems to be some evidence that elections clo...

2008-06-16T03:15:00.000-04:00

There seems to be some evidence that elections close at the end, although in quite a few cases (definitely 1992 and 1996 -- Kerry was also arguably closing in 2004) the closure was not enough to change the outcome of the election. Then you have a case like 2000, when the election seemed to be getting away from Gore after he'd spent all year catching up, but he actually made up about 3-4 points in the last 72 hours (possibly because of the news of Bush's DUI).

...

I thought about running both versions of the analysis and letting the users pick, but:

(i) this is pretty labor intensive. It requires generating twice as many charts and graphs each day, which is time consuming, even if I figured out some JavaScript to display them effectively.

(2) I stand behind the 'trend-adjusted' version and think 80 percent of the users will too once I finish explaining the couple of refinements that I've made to it.

Nate, there's a bit of conventional wisdom that se...

2008-06-16T03:07:00.000-04:00

Nate, there's a bit of conventional wisdom that seems to have been true from my anecdotal observations, which is that the candidate with the momentum at the end of a campaign tends to win. You've got the data before you, and I don't. Do you see any truth to that? If so, would smoothing the curve at the end make sense?

Also, rather than enforcing your best guess, would it be optimal to allow users to select which model to use, with your explanations and advice provided?

This change makes a lot of sense, and the second o...

2008-06-16T03:04:00.000-04:00

This change makes a lot of sense, and the second one sounds good as well. I also don't think anyone who regularly visits think that you are trying to introduce bias.

Your explanation helped as well - the approach you take splits the middle between random walk (very low LOESS parameter) and bounce (no trend compensation).

I'm still concerned about presenting what seems to be hard poll data that changes week by week, but I can't come up with a better way of presenting your trend analysis beyond titling the Super Tracker as being for past poll adjustment.

Alex: I thought about suggesting this in relation to the state groups Nate has on the left hand side - it will probably be better left to a demographic analysis in the future, possibly with a geographic component because of state borders.