Article

2008-09-08 Home Front: Politix

USA Today Poll: McCain-Palin up by 10 among likely voters

Archived material is restricted to Rantburg regulars and members. If you need access email fred.pruitt=at=gmail.com with your nick to be added to the members list. There is no charge to join Rantburg as a member.

Posted by Mike 2008-09-08 08:37|| || Front Page|| [10326 views ] Top

#1 But do these figures account for the likely military voters whose ballots will be disallowed? And the dead people in Chicago who will vote several times? And the snowbirds who will also vote in Florida?

Posted by Glenmore 2008-09-08 09:24|| 2008-09-08 09:24|| Front Page Top

#2 At any rate, I bet the mood is tense in the Obama camp this morning.

Posted by bigjim-ky 2008-09-08 09:56|| 2008-09-08 09:56|| Front Page Top

#3 I bet the Obama camp is in full panic mode this morning. Look for more Palin smears as that is the typical knee jerk reaction from liberals.

Posted by DarthVader 2008-09-08 10:04|| 2008-09-08 10:04|| Front Page Top

#4 A thousand vote sample out of tens of millions or so isn't much of a sample.

Posted by Mike N. 2008-09-08 10:15|| 2008-09-08 10:15|| Front Page Top

#5 With all due respect, the only "poll" that matters is the one administered on election day.

Posted by Swamp Blondie in the Cornfields 2008-09-08 10:48|| 2008-09-08 10:48|| Front Page Top

#6 Biden's the one who needs to watch out.

Posted by DoDo 2008-09-08 10:53|| 2008-09-08 10:53|| Front Page Top

#7 Yeah, I know, but it's a major milestone to see the pollsters actually publish a poll with McCain winning by more than the confidence points.

Posted by bigjim-ky 2008-09-08 11:05|| 2008-09-08 11:05|| Front Page Top

#8 "Something to keep in mind: left-leaning candidates tend to generically do better when the sample includes non-voters."

They do even better when the sample includes dead voters. And felons.

Posted by Frozen Al 2008-09-08 11:18|| 2008-09-08 11:18|| Front Page Top

#9 Mr Mike N

Take a few math/stat classes before posting. The average deviation of sample respctive to "truth" (ie total population) doesn't depend on size of population (except when sample is equal to total poulation in this case it is zero) but depends on sample size.

The theory is more or less like this: consider the outcome of a randomly selected sample in a population who contains a proption of p, say repulblicans and ( 1-p) republicans. Depending on
"luck" during selection the proportion of republicans in the sample will differ more or less respecvtive to the total population. Repeat the proceess an infinite numlber of times. You will find that the set of samples has same average than total population. You will also find aytht standard deviation for the set of samples varies in an inversely proportional way respective to poulation size ie standard deviation of 2,000 sized samples is half the one for 1,000 sized samples. Finally you will find that teh set of samples follows a Laplace-Gauss law.

OK? Now. In a Laplace-Gauss law there is only an infinitesimal probability to get a sample who deviates more than 3 times the standard deviation. So le's imagine that you have 50% republicans and 50% democrats and that given your sample size the standard deviation is 2. It is highly unlikely (from distant memories it only happens one in a thousand times) that if you select a random sample you will get over 56% republicans or democrats.
(BTW, American poll institutions give amargin of error ie an interval of error that emans that there is 95% or 99% probability depending on institutions tht the real result lies berteeen poll_number - interval and poll_number + interval.

Posted by JFM">JFM 2008-09-08 11:19|| 2008-09-08 11:19|| Front Page Top

#10 JFM, and assuming that people tell the truth.

Posted by Bright Pebbles 2008-09-08 11:50|| 2008-09-08 11:50|| Front Page Top

#11 America awakes to the internal danger?

Posted by JohnQC 2008-09-08 12:12|| 2008-09-08 12:12|| Front Page Top

#12 Thus the Bradley effect which had Tom Bradley 10% higher (I think) in polls than he did in the election. People lie to pollsters and conservatives historically are under-represetened because they hang up, or because their answers and inconvenient to the pre-determined outcome. i don't know which.

Posted by rjschwarz 2008-09-08 12:13|| 2008-09-08 12:13|| Front Page Top

#13 Of course. It is a well known problem in many fields where statitics are used: people are afrid to tell the truth to the interviewer. That is why polling institutions try to correct the "lying bias". But that has nothing to do with the problem of estimating something in a population of 100 millions basing on the outcome of a sample of 1,000 individuals.

Posted by JFM">JFM 2008-09-08 12:15|| 2008-09-08 12:15|| Front Page Top

#14 It also depends on how the sample "self-selects" such as "cell phone only" users and those that won't take polls and/or refuse to answer questions as presented.

I happen to represent 2 of 3 of those factors (I have a landline) and I know many that meet all 3. And these are older folks, not the young demographic that is spoken about.

It will be interesting how these polls will compare with the actual results in November.

Posted by tipover 2008-09-08 12:21|| 2008-09-08 12:21|| Front Page Top

#15 I recommend JFM for the endowed chair in statistics at the Rantburg University College of Math And Other Things That Require A Calculator.

Posted by Mike 2008-09-08 12:23|| 2008-09-08 12:23|| Front Page Top

#16 Oh, no. I was a statistician when I was young nad good-looking. But that was a long time ago.

Posted by JFM">JFM 2008-09-08 12:26|| 2008-09-08 12:26|| Front Page Top

#17 Heh-heh. As the game begins.

Posted by Woozle Elmeter 2700 2008-09-08 12:27|| 2008-09-08 12:27|| Front Page Top

#18 Tipover, one of the most historic cases of self-selection in a political poll (from the dim mists of my statistics courses) was the Dewey - Truman election, one of the first phone polls. They forgot to take into account that the presence of a phone was strongly correlated to income. In those days poor and working class were unlikely to have phones and as a result the richer Repubs were more heavily sampled.

As that famous headline showed......ooops.

Posted by AlanC">AlanC 2008-09-08 12:35|| 2008-09-08 12:35|| Front Page Top

#19 I can assure you by his reactions, when a pollster stopped me mowing the lawn, he recorded me as "not at home". Kept trying to put his words in my mouth. Yeech.

Posted by Bobby 2008-09-08 12:54|| 2008-09-08 12:54|| Front Page Top

#20 I have little faith in polls, regardless of when they are for or against my preferred candidate.

That said, these polls are sponsored by media groups which are generally hostile to my side. If they say my guy is ahead then he probably is.

Anyone else notice that only 959 of the 1022 people sampled were even registered?

FWIW: 54/44 looks like a good bet for the final split on election day.

Posted by Iblis 2008-09-08 12:59|| 2008-09-08 12:59|| Front Page Top

#21 I welcome the calls and lie like a rug.

Posted by Nimble Spemble 2008-09-08 13:00|| 2008-09-08 13:00|| Front Page Top

#22 AlanC, that was the famous Landon-FDR Gallup poll in 1936. Telephones were better-distributed by 1948.

The annoying thing about the "grr, my vote's between me and my God" attitude is it makes it damned hard to accurately poll the conservatively-minded.

I'm campaigning for McCain in Pennsylvania, and I keep running into these idiots. How much weight will your opinions have if you keep hiding them from the general public outside of the election proper, people? Have you ever heard of the "herd effect"?

How about influencing your representatives? If Joe Clockpuncher (Lubbock-R) gets elected from an electorate which is actually anti-illegal-immigration, but shows moderately pro-illegal-immigration in the polls because all of the nativist yahoos refuse to answer survey questions, then by golly, Clockpuncher is gonna listen to the will of the people and moderate his natural tendency to stomp around and make noise about the illegal immigration issue.

Posted by Mitch H.">Mitch H. 2008-09-08 13:11|| http://blogfonte.blogspot.com/]">[http://blogfonte.blogspot.com/] 2008-09-08 13:11|| Front Page Top

#23 JFM, with all due respect, open your eyes, Good Sir.

These polls are wrong far more often than they are right. Fancy math aside, there's no accounting for dumb luck. One thousand out of 50 million is too small. People hide their true feelings on any number of matters. People are especially devious about politics. Sure, they got groups of people on both sides telling then something else, but there's no way they can say for sure that there weren't donks, telling them the were voting trunk than the other way around.

Poll 10 times more people and get better numbers.

If they were taking a sample that related to something that didn't have minds of their own, it would be a different story. The human brain isn't a math problen.

Posted by Mike N. 2008-09-08 13:49|| 2008-09-08 13:49|| Front Page Top

#24 Mike, look up a good article on Statistics and margin of error. For a large random population, for all purposes, the margin of error is dependent only on the size of the sample, not the size of the population. Formulas give a margin of error of 3% (w/ 95% confidence or 2 standard deviations) for a sample size of 1000. For 10,000 the margin of error decreses to 1% with 95% confidence.

The math behind it is proven. When human behaviors are included, then bias can creep in. But that is separate from the mathematics of statistics and random processes.

Posted by ed 2008-09-08 14:10|| 2008-09-08 14:10|| Front Page Top

#25
Re statistical sampling.

Everybody's got good points.

The real issue in using any sample of any size to predict the "real" status of a population is randomization and representativeness. First and foremost, it is critical that the sample is pulled randomly from the total population -- that is the only way to remove hidden variables that would otherwise contaminate the sample (e.g., it's just people with phones, or it's all of my neighbors, or . . .). Secondly, the sample must be representative of the population as a whole (e.g., same percentages in income class, race, religion, etc.). Of course, a truly random selection from the population as a whole should result in a representative sample, but checking for representativeness is a good check on the process.

If the job's been done right to get a proper sample, then the size of the sample is not that important unless the effect being studied is very small (think: does cell phone use increase cancer). To study small effects, the sample size needs to be increased to increase statistical power. However, when the effect is large (think: 55% to 45%) not much statistical power is needed. Increased power will give you a tighter confidence interval, but with a 10 point difference, +/- 3 is just about as good as +/- 2.

Finally, the issue of reliability comes into play as well. Are the questions posed to the sample the same as what will be posed to the voter on election day? The closer the questions are, the more reliable the results from polling the sample.

FWIW.

Posted by cingold 2008-09-08 14:46|| 2008-09-08 14:46|| Front Page Top

#26 Mike N., I once -- for a few years -- worked for a Fortune 500 consumer products company. When we wanted to know whether or not a product we were developing would sell, we used a sample size of 1000. When we wanted to know which flavour would sell the best for that product, we used a sample size of 10,000. Every once in a while the product was a bust anyway, because the right question hadn't been asked, or we hadn't understood the answer properly. For a quickie poll, which is what this is, a sample size of 1,000 is correct.

These polls are indicative at best. In the last several elections they've generally been slanted toward the Democratic candidate, which has a good deal to do with the Stolen Election meme out there. Given that, a 54% to 44% result is certainly interesting, and a very, very nice Convention Bounce for the Republicans. But as Swamp Blondie says, the only poll that matters is the election.

I hate statistics. Back when studying such things seemed like a good idea, I asked my Stats professor how we knew the formulae were true. "Well, it seems to work so far," was his answer. I've worked very hard to forgive all the statisticians I meet, ever since, realizing they just can't help it. I 'm still working on forgiving Mr. Wife, who took the same course the next semester. He'd always known the world worked like that, he just hadn't had those lovely formulas to describe it with.

Posted by trailing wife ">trailing wife 2008-09-08 15:44|| 2008-09-08 15:44|| Front Page Top

#27 Mike N

Forgive me for being harsh, since you touched a point of trhe job I was trained for (even if I never worked in it) I went ballistic.

Now what I have done is explain the mathematical theory about surveys, anykind of surveys be them of opinion or quality control (eg when you evaluate quality of one use flash bulbs by taking a small sample of them).

Now I could tell you a lot of resons to take political polls with a grain of salt from dubious methodology (eg phone calls: some categories of people are not at home so poll is oly valid if you have the budget to insist and insist until the guy answers), to people lying (1) or the fact that people can change vote in the last days or hours
or hours (eg when Bush came close in 2000 due to a DUI decades before). Anyway statisticians tend to think poorly about political pollsters. But I couldn't allow you to remain unchallenged about the assertion "you can't tell how is a population
by looking at a sample one tenth of thousand of its size". That is mathematically wrong. You can.

(1) In France, many Communists voters told they voted socialist because they were ashamed to tell it in front of the pollster. So pollsters had a rule of thumb to get the Communist vote right. But when Communist vote crumbled only hardcore communists remained. These had no qualms telling how they would vote, so raw results were very close from the poll to actual numbers. But, out of habit pollsters applied their old formulas so they grossly overrated the communist vote. Also many communists shifted to fascist Front National but since this was very politically uncorrect they told pollters they voted communist. As you can see political polling is not an easy job, that is why they have a lot of ~~soft scoence clowns~~ political science majors in their staff.

Posted by JFM">JFM 2008-09-08 15:44|| 2008-09-08 15:44|| Front Page Top

#28 All this assumes that you are trying to get the right answer when you poll, but we all know some polls are intentionally skewed toward a particular result.

Posted by Iblis 2008-09-08 16:21|| 2008-09-08 16:21|| Front Page Top

#29 When discussing a possibility, there are only two outcomes: either it WILL happen or it WON'T happen.

Therefore, everything, statistically speaking, has a 50% chance.

QED.

OSLT.

Posted by Adriane 2008-09-08 16:39|| 2008-09-08 16:39|| Front Page Top

#30 Mitch - I hear you when it comes to the herd effect. However, I can't help but wonder why we should help the democrats obtain a good estimate of how many dead they need to turn out to vote.

Posted by Betty Grating2215 2008-09-08 18:35|| 2008-09-08 18:35|| Front Page Top

#31 Adriane, that is not stats, you are pulling a Schroedinger cat on us!

Posted by Spike Uniter 2008-09-08 18:53|| 2008-09-08 18:53|| Front Page Top

#32 Betty, a very good point. Recommendations?

Posted by Spike Uniter 2008-09-08 18:56|| 2008-09-08 18:56|| Front Page Top

#33 Every time a poll call comes in (I don't care if it is toothpaste or what) I say "I don't do polls".

When I go to a left blog or site and they have a poll attempting to rag on a pub or the USA ... I take the online poll in such a way that the outcome will upset them...

Posted by 3dc 2008-09-08 19:08|| 2008-09-08 19:08|| Front Page Top

#34 Keep in mind that all polls are bought and paid for, and pollsters like to keep the customer happy.

You tell them what they want to see an their front page. Liberal MSM likes get left leaning poll results, right up to the exit poll.

For pollsters, this just makes good business sense.

Posted by Skunky Glins 5***">Skunky Glins 5*** 2008-09-08 19:14|| 2008-09-08 19:14|| Front Page Top

#35 This week US Today show a significant upward move for McCain-*-Palin.

But next week, they will probably trumpet the "significant drop" in M-*-P support. Just the RNC bounce, folks, Palin's over, nothing to see here, move along back to your regular Obama programming.

(Some of the internal number show huge changes. When asked "Who would do a better job with the ecomomy?" McCain went from 39% last week to 58% this week. This in a week when McCain never mentioned the economy. Except he did. If most people polled equate the economy directly with the price of gas, then bringing on an Alaskan pro-drill governor as VP had an effect.)

Posted by Skunky Glins 5***">Skunky Glins 5*** 2008-09-08 19:29|| 2008-09-08 19:29|| Front Page Top

#36 #32 - Not usually an issue, as in the corrupt precincts they wait until the results are in to start the vote manufacturing and enhancement process.

Posted by Oldcat 2008-09-08 20:09|| 2008-09-08 20:09|| Front Page Top

#37 #32 - Not usually an issue, as in the corrupt precincts they wait until the results are in to start the vote manufacturing and enhancement process.

Posted by Oldcat 2008-09-08 20:11|| 2008-09-08 20:11|| Front Page Top

#38 Liek I have said before, wait until you have 3 consecutive weekday polls to track. Then look mainoy to likely voters (LV) not registered voters (RV). And look at the crosstabs to see how they "normalize" for party identification, age, etc.

The biggest problem is that there is likely an undersampling of 18-25 year olds due to them being unavailble because of cell-only use, and approximating or normalizing is difficult to do accurately with such a large subsample group. Fortunately this group also tends to turn out at a lower percentage than any other. The flip side is there may be a "youth surge" for Obama because the youth are esaily duped by hype, appearance and the MSM's featherweight fluffing.

Anyway, national polls suck. Find me a good one that's representative of LV, is a large sample, completely neutral and random inn its questioning, and normalized -- and there needs to be one of each of these for each of the states. And I want to be able to see the crosstabs including demographics and so on.

Calculate the electoral vote, and THEN you'll have me talking.

Posted by OldSpook 2008-09-08 20:32|| 2008-09-08 20:32|| Front Page Top

#39 If I wanted to know about how many Americans like cookies or about how many people would tune in to watch CSI: Upper Amazon Basin, a thousand person sample would work just fine.

When it comes to politics, a thousand aint enough. That's why two different polls taken on the same race at the same time can have different results. Often outside one anothers margin of error. Too many variables invloved here (emotions, bias etc.) Forget accurately adjusting for them. I would sa its safe to assume that the vast majority of pollster have never even bother to make an attempt to to mathematically find any sort of correlation between polling results and election results.

Like the often mention theory that conservations keep to themselves during polls. Right there we have what many believe to be the causal portion, but to the best of my knowledge, nobody has even thought about doing an actual r squared to see if there's correlation.

Posted by Mike N. 2008-09-08 20:37|| 2008-09-08 20:37|| Front Page Top

#40 I don't do telephone polls because I don't want to be profiled or push-polled by some unknown political group -- and it's a lot easier to say "no, thanks" and hang up. That suggests to me that those polled may be naive and/or anxious to spout off to total strangers. Not exactly a scientific sampling.

Posted by Darrell 2008-09-08 21:10|| 2008-09-08 21:10|| Front Page Top

#41 Apparently the Las Vegas bookies have done the R-squared calculations, Mike N. You might want to check there for something undoubtedly more accurate than the pollsters. I'm sure you're quite right that the pollsters have never actually tested their results against reality -- otherwise they would have successfully called the last few elections.

Posted by trailing wife">trailing wife 2008-09-08 21:31|| 2008-09-08 21:31|| Front Page Top

#42 I don't do phone polls because I would rather use my minutes talking to people I actually want to talk to....

Posted by IG-88 2008-09-08 22:43|| 2008-09-08 22:43|| Front Page Top

#43 Only on Rantburg does a post about a poll turn into a debate with math about sample sizes, statistics and the new replacement phrase,

"Screw the lawyers, let's shoot the pollsters first!"

Posted by Silentbrick">Silentbrick 2008-09-08 23:38|| 2008-09-08 23:38|| Front Page Top

23:52 CrazyFool
23:46 CrazyFool
23:46 JosephMendiola
23:38 Silentbrick
23:30 Silentbrick
23:28 JosephMendiola
23:20 FOTSGreg
23:12 Procopius2k
23:11 Pappy
23:10 JosephMendiola
22:55 Betty Grating2215
22:49 3dc
22:48 mhw
22:47 Barbara Skolaut
22:45 Barbara Skolaut
22:43 IG-88
22:41 3dc
22:39 3dc
22:39 ed
22:38 mojo
22:37 ed
22:33 MoreScotch4Me
22:31 3dc
22:28 bigjim-ky