Pollster.com

I'm not a fan of the one day survey techniques used by Rasmussen and Insider Advantage for their state polls because I think they can lead to non-response bias.

A two hour survey? In the middle of the day?

I like the phrase 'snap poll'

Quicker is not better.

Sorry for the lengthy detour but I have to address something I saw today on MyDD, a diary regarding Nate Silver's claim that Intrade is possibly being manipulated. Silver is apparently miffed that Obama's trading price has not soared above 60 or 70 due to the recent favorable polling.

These are the opening two paragraphs from semiquaver's diary on MyDD. I posted a reply there, but the diary plunged quickly. I think it needs to be seen here, and not buried among 100+ replies:

"Nate Silver at FiveThirtyEight has an interesting post about why Intrade's futures prices have barely reflected the major upswing in Obama's polling over the last week. After the GOP convention, when McCain passed Obama in the polls for the first time, bettors priced McCain as high as 54% midday on Sep 13, a really scary number. But when the polls began to move back Obama's way last week, prices were sluggish to respond, and still have only got Obama at 52, which given the electoral math, is almost certainly undervalued. Most other betting sites currently price Obama above 60%, while 538 itself puts him above 70%. For markets to be this far out of sync is bizarre:

Silver: "This is the equivalent of the Giants being 3-point favorites at the Bellagio Sportsbook, and 7-point favorites at the Mirage down the block. Those things just don't happen in efficient, sufficiently liquid markets, because they create arbitrage opportunities: you'd lay $10,000 on the Giants at the Bellagio and $10,000 on their opponents at the Mirage. Any time the Giants win by fewer than 3 points or more than 7 points, you lose nothing, since your two bets cancel out. But any time they win by fewer than 7 points but more than 3, you win both bets, and take home $20,000 (less the casino's vigorish) for absolutely no risk. Pretty good deal, right? That's exactly what's happening with these futures contracts."

***

Nate Silver needs to stick to politics and leave out the sports betting references. There were so many basic errors in that one paragraph they would literally make all my friends in the major sportsbooks roll in laughter. I was staring at them in disbelief.

First of all, the money line difference between a 3 point football favorite and 7 point favorite is massive, far beyond the gap between 53 and 60, or anything comparable. A 3 point favorite is roughly -155 in man to man terms. Once you get to 7 the money line is roughly -280 in man to man terms (no vigorish).

If we applied that to Intrade numbers, a -155 would be a trading price of 61, while a -280 would be 74. Anyone who has ever played Intrade realizes the vast, vast difference between a contract at 61 and 74. It's laughable that Silver tries to assert it's equivalent to a gap between 53 and 60, or similar. A 53 is basically -113 and 60 is -150. That's interesting but not particularly remarkable. Again, anyone who has ever played Intrade understands the mid levels near 50 are much more fluid. Silver, of all people, should get it because his game is supposedly baseball. It's hardly atypical for a baseball price to move 20 or 30 cents on the money line in the 5/7 range, particularly when things are in flux like an overnight injury or two teams going in opposite directions. On the other hand, I don't think I've ever seen a line move from -155 to -280, or anything comparable, without an incredibly significant pitching change, from bum to ace or vice versa.

Silver has got to realize the markets are not mesmerized by polling. That's a major part of this. It's also why Bush remained in the same basic level of favoritism throughout 2004, even when Kerry led the polling average. There's a foundational factor that gamblers rely on. In this case, that foundation is a pro-Democratic climate but in the 60/40 range, not beyond that. It's comical that he thinks a handful of favorable polls are supposed to drive the trading price to 70 or beyond. It's similar to a football team with a base power rating and an injury is not going to alter that rating a TD or more. That's why New England was still a tiny favorite over the Jets in week two, and -13 vs. Miami last week. With a major variable like the debates still to come, you can't get carried away with a short term indication like a week of polling.

If you want the truth of the matter, it was preposterous that Silver used polling to switch wildly to McCain the favorite, and now a week or so later has Obama beyond 70.

And one other thing, I have no idea how Silver thinks betting both sides (playing a middle, in sports betting terms) at $10,000 a pop equates to "you lose nothing" and "absolutely no risk." I would use the word novice but it's far below that. It's amazing ignorance. The house vig is 11 to win 10. It takes $11,000 on each side to win $10,000. A bettor who plays that type of middle is risking a $1000 loss unless the result lands in the middle, or sides, which means landing on an outside number, in this case the 3 or 7. I play middles all the time and while it's profitable in the long run, you can have major stretches where the games don't land, and it certainly doesn't equate to no risk.

BTW, Obama is now trading at 55 on Intrade, last I checked. It's like gas prices, slow to move away from the recent trend. If favorable polls continue, Obama will likely move toward his previous level in the 60 range.

Actually what Nate was saying, was that Obama spikes downward coincided with spikes upward with Clinton. Which shows that there is a chance that one person is manipulating the market.

At least it would be fun to re-interview the Republicans who changed their mind once they learned McCain called for a postponement.

Hillarious!!! Now, the McCain campaign wants to delay Biden vs. Palin debate. Don't you smell fear, guys? OMG!!! These are the same people who are asking us to let them lead our nation, in good or bad!!!

I realized my lengthy response left out one key point and was unfair to Nate Silver in some respects. Silver is correct there is some maneuvering for guaranteed profit going on. I mentioned that in a comment here the other day, regarding a friend of mine who has played both sides of the Virginia contract.

But Silver used the wrong sports betting reference point. That's what irked me. He's talking about scalping, not middling. Scalping is when you play both sides on the money line, not the pointspread, and essentially turn the house vigorish in your favor. It requires anticipation, and a willingness to eat a loss if you anticipate poorly.

In Silver's example, a 3 point favorite and 7 point favorite on the same game, a sharp bettor would not only play the pointspread both ways, but the accompanying money line. That would be something like -165 on the Giants and +250 on the Redskins, to pick an opponent. If you bet $10,000 increments on each side it would be $16,500 to win $10,000 on the Giants and $10,000 to win $25,000 on Washington. The gambler would break even if the Giants won or win $8500 if Washington prevailed. Or you could stagger it to profit regardless of who won, but in that case you lose earning potential, particularly the further from even odds. It's not as if you win $4250 either way. Off the top of my head, if you staggered that example it would be closer to + $3200 either way. It's always a dilemma. Seems like when you rely on the dog they fail, but when you split it the dog cruises. Let's be clear: That example is another world, maybe once a year, perhaps a rapidly moving boxing match. Typically it's taking +125 and giving -115, something like that, minor advantage.

There really hasn't been much movement on the presidential race. That's why I'm skeptical it's been high volume scalping. The state prices are a better target, flopping on either side of 50/50 based on over reaction to the poll flavor of the moment. Certainly some speculators who previously had good positions on McCain in the 40 range played Obama when he dipped below 50 for several days. You always maximize advantage by taking plus odds on each side.

I'm more inclined to believe Intrade speculators view this as a close race and aren't eager to get carried away based on a week of polling. Now, if Silver offered his 74/26, or whatever it is, in a trading market, there would be scalping galore until both prices joined.

Mark,

Thank you for your thoughtful guidance.

SurveyUSA will happily re-conduct this survey for you, given the field period that you recommend.

Would that be ...

* Tonight before the President speaks (with results potentially immediately obsoleted by the speech)?

* Tonight during the President's speech?

* Tonight after the president's speech (if so, that precludes Eastern Time and Central Time zone respondents)?

* Tomorrow during the workday, when those who are not home on Wednesdays are also not home?

* Tomorrow night Thursday, when there will almost certainly be a completely different set of facts in place, and by which time another bank may have failed, and the Debate Commission will almost certainly have ruled?

* Or, should SurveyUSA wait and conduct the research during the debate Friday night?

Your comments today are functionally identical, and similarly perplexing, to comments you made from this same podium immediately after Hurricane Katrina, when you suggested the SurveyUSA's hour-by-hour measurement of the government's response to historic and extraordinary events in New Orleans was somehow sub-optimal.

I react today as i reacted then. SurveyUSA's work in response to fast-breaking news is nothing short of remarkable.

As events unfold between now and the debate scheduled to begin in exactly 49 hours, SurveyUSA will refresh our data. In the meantime, if you are presented with data from a competing research firm that you feel does a better job of capturing this extraordinary moment in time, post that data and please delete ours.

Sincerely,

Jay H Leve
CEO
SurveyUSA

Em'

Thanks for the poll Jay, I thought the quick turnaround was amazing...

Political Stunt a Smoke-Screen for Voted Disenfranchisement?:

As I respect most of the opinions on Pollster (barring a few "outliers")...I wonder what your take on this is...

"Media are reporting Wisconsinites have received absentee ballot
applications from the John McCain for President campaign with
incorrect return addresses, which would likely prevent registration
and processing from being completed."

Full story is on http://www.onewisconsinnow.org/page/content/savewisconsinsvote/

My first reaction as a Wisconsinite is to be angry...however, I would like your input

Gary Kilbride
You clearly misread Nate's post and understood almost none of it. The paragraph about sports betting was just a rough example nothing to get upset about. His point was partly that there were inefficiencies between the two markets which is both true and a legitimate chance to make quite a lot of money.

His other point was that there were odd betting patterns in the last couple days which he was theorizing were responsible. Noone really knows but he never suggested that the price should be at 70 rather that it should be in line with other markets around 60 which it looks like its headed to right now.
Perhaps if u ever used intrade rather than just talked about it u might understand this stuff. But anyways thanks for trying!

Actually, I think this poll is MORE reliable if people haven't had a chance to find out that McCain suspended his campaign. It shows just how absurd the whole notion is. If the numbers change after people learn McCain suggested this, then we can identify people who actually think this is a good idea, and people who will say anything to show support for McCain.

>"SurveyUSA will happily re-conduct this survey for you, given the field period that you recommend." - Jay H Leve, CEO, SurveyUSA

Oh, snap! Snark from a pollster! Can you dig it?

Seriously, Jay, you understood the criticism perfectly well. It is not possible to get a representative sample, free of non-response bias, by auto-dialing landlines for two hours in the middle of a workday. The issue is not when you take your measurement, it is how you take it.

I don't see not being able to get non bias response being a big issue as long as the poll is presented in a way that has appropriate disclaimers.

I was happy to get a snapshot however imperfect.

It's better than the WSJ,MSNBC,CNN internet polls in any event.

Nate's post gave the uninitiated a rough feel for the situation. He never said Obama should be up at his 71%, just that he shouldn't be at 51%, especially when he was at 61% on another, bigger site. He was also right to put McCain up for a while, McCain really had bounced into the lead and it showed up across state and national polling.

Nate was also correct in pointing out the market manipulation. I imagine somebody had a system where they would short a contract to drive the price down, then buy it back on the slower rebound. It doesn't take much money to move Intrade quickly when the market sets up certain ways.

As for the SUSA poll today: yes, it's probably horribly unrepresentative. Better to at least poll overnight. However, looking at the crosstabs, here's a skew towards the 18-29 vote at 32%, and a +8 Democratic ID edge. McCain lost the "debate postponement" question 86-10 and the "campaign suspension" question 79-14, if I did my math right. It doesn't seem like any reweighting will come close to making that up.

I say, who cares if it's skewed?

If you think it's skewed, then you have reasons for thinking it's skewed, and you can use those reasons to re-weight, and you arrive at results that you particularly believe in and espouse. If you don't think it's skewed, then you say so, and why, and that's that.

Arguing over whether or not these results are skewed is missing the point! Use the data how you will and continue with the _intelligent_ debate, please. What we need now, more than anything, is intelligent debate.

@Kevin:

It is not possible to "re-weight" to account for the people who cannot be reached in the middle of a workday by auto-dialing landlines, unless you assume that the people who are available have the same opinion as the people who are not available. That is a dangerous and unproven assumption.

@Allen: What leads you to believe the people who you couldn't reach had a different opinion, other than the possibility, and what prevents you from using this belief to account for it? Do you think the data is so unrepresentative that any conclusions anyone made would be inherently unsound? Or do you just think the possibility is there, and without a 100% guarantee, that's what makes them unsound?

@Kelly

Just the possibility makes this unsound. In general, if you are preforming a sample of (for example) registered voters, you either need to take a random sample in which every registered voter is just as likely to be chosen as another OR you need to prove based on prior studies and experience that any skew in your sampling techniques has an insignificant affect on he results. No REPUTABLE pollster would claim that the skew that could result from suto-dialing landlines for two hours during the workday would be insignificant. In fact, any reputable pollster would assume that this would result in a significant skew and they would call their result "a non-scientific survey". In that respect, its not much different than the "polls" you can fill out when you visit a website like CNN. For example, the poll currently on CNN's front page is "Quick Vote: Should Barack Obama and John McCain debate on Friday?". That "vote" IMO is just about as valid as auto-dialing landlines for two hours during the workday. In fact, add a few demographic adjustments, and you've accomplished the same thing. No reputable pollster, researcher or scholar would regard this as a valid scientific survey.