I'm trying to understand TruthIsAll's analyses

Is this a fair summary?:

In the final exit polls, Mitofsky said that 43% of respondents had voted for Bush in 2000, 37% for Gore, 3% for Nader, and 17% new voters.

How is that even possible, when Bush got 50 million votes in 2000, and 122 million votes were cast in 2004?

50/122 = 41%.

And THAT'S assuming none of those 50 million people died. Let's estimate that 1% of the population dies each year, meaning that 4% of those voters have died. So, AT MOST, the true percentage of people who voted for Bush in both elections is 48/122, or 39%.

Once again, Mitofsky said that 43% claimed to have voted for Bush in 2000. That's a 5 million vote bias IN FAVOR OF BUSH in the exit polls.

Does this not contradict Mitofsky's own assertion that Democrats are more inclined to participate in exit polls?

I've tried reading several of his/her posts and have utterly no grasp of them - math and stats just do not play to my strengths. I think that may be a reason why the "stolen election" argument never caught on among the masses. Too detailed and technical, much easier to just accept that Bush won and move on.

Mostly

has done.

Since they are so convoluted and full of assumptions, there aren't many people who are willing to put in the time to go through all of them and have a rational discourse. In part, such rational discourse is not even welcome and is perceived as pro-repugs to start with...

Heya valis
"Since they are so convoluted and full of assumptions"

Who is 'they'?

Thanx in Advance.

for Bush then Gore, when Gore got more votes last time? And that was before we had a chance to see this administration in action. Do they think we are all stupid?

they don't exactly have the most reliable source of (FOX)news that they tune in to more and more everynight. our other media networks were caught off guard and really haven't been serving the public's interest in terms of reporting hard hitting news..instead its always the next trial...I just hope this new www.Current.tv thing gore is working on pans out...maybe it can reverse the tide of decline in the media today

they do not prove theft

Folks after an election tend to forget "wrong" vote and believe they voted for "winner".

So a report to an exit poll worker that you voted for last times "winner" is expected to run a few points above "truth".

So it does not prove exit poll "fix to actual" - which is a standard procedure - was done in a biased manner.

TIA has, however, shown that the changes required in the model are so large - so beyond any expectations - and so unlikely - that theft is the only reasonable conclusion.

But MSM - and the exit poll takers - will work the fix the exit model side of the street since there is no money in a "theft" storyline.

The fact that other Hispanic exit polls show that the national Hispanic model - after the fixes required to force to the "actual result" - is a total bull shit model - will not be discussed in our media because again - no money in such a discussion - no reporters that did not avoid math in college - no physical evidence as the GOP destroys all evidence as we prepare for the next election -

Indeed the Washington State absentee ballot laugh that was published today - namely they do not know how many votes were absentee - or if all absentee votes correspond with a ballot sent out after an absentee ballot request - or that such requests were valid - is they key to understanding the math.

The GOP is running an absentee ballot scam across the US these days. And the thousand in one ways election folks screw up the process - by accident with no evil intent or bias - is the greatest/best cover for this crime that the GOP could ask for.

This was the discussion - fixing results via absentee ballot games - that was not infrequent in GOP circles in the 80's under Reagan back when I was allowed to hang around but not speak at such GOP gatherings (and I more often heard as attendees unwound in front of me after such "meet-ups").

"post-election bandwagon effect". In another thread I posted links to a study that shows it. It shows that people, when asked some time after the election whom they voted for lie to the pollsters, causing the result to skew toward the winner of the election about which they are asked. There are also numerous papers on the subject of people lying to pollsters who ask them whether they voted or not. So it is not surprising to get skewed results like that - in fact, it would be surprising if you did NOT get such skewed results.

voted for. I'd dearly love to see a follow-up poll. Interesting how we had polls up the yazoo before the election, but not any validating polls afterwards....

post-election bandwagon effect - it is here:

http://www.amstat.org/sections/srms/Proceedings/papers/1993_194.pdf

As for any follow-up polls, the only one I heard of was done by Zogby for someone and Zogby did not release the results. I really have not ever seen any such polls done, either because all the pollsters know that the skew from the bandwagon effect would make such a poll extremely inaccurate, or because such a poll seems rather pointless and no one orders such polls from them.

Get aboard.
We'll take you straight to see Alice in Wonderland.

NOT.

Has there been follow up using just one survey method rather than two. It sounds as if those who are in a 'subordinant' position in society are most likely influenced by the effect. I would suspect that face to face interactions (exit polls) and autonomous surveying (telephone survey) may stimulate in and of themselves, the variation in behavior. Watching my son's behavior towards telephone calls (drop whatever you are doing to answer it), there is something authoritarian associated with being the recipient of a call.

Mike

http://www.springerlink.com/openurl.asp?genre=article&issn=0190-9320&volume=21&issue=3&spage=197

If you have a better library around you than I have, you may be able to find it there - the journal is "Political Behavior".

In addition to the post-election bandwagon effect (yes, I voted for the winner), there is also a vote overreporting effect (yes, I voted) that is well documented. This is from http://polmeth.wustl.edu/polanalysis/PA84-389-398.pdf

"Campbell et al. (1960) used conservative calculations to estimate that misreporting accounted for about 3 percentage points of the bias in the 1952 and 1956 surveys. Just a few years later, Clausen (1969) used the first and crudest vote validation study and Census data to estimate more accurately that response error in the NES was as high as 8.3 points. Sigelman (1982) later found that reported and validated vote estimates differed by 11.5 percentage points in the 1978 study. The same magnitude appeared in the 1976 study 2 years earlier (Katosh and Traugott 1981) and the 1980 NES 2 years later (Anderson and Silver 1986). Internal NES technical reports also verify that as much as 12 to 14 percentage points of the gap between official and survey turnout estimates is due to misreporting alone."

But in quickly looking over the abstract it appears the actual variable making the false voter report increase is the increasing non-response rate. I will give it a closer going over later.

This varible of interveiwing people who actually did not vote is going to be a function of distance from the polling place, and can in part explain those precinct's higher WPE.

Mike

interviewing people who did not vote in the elections that the exit poll is FOR. It is the people who JUST voted but did NOT vote in the previous elections but, when asked, answer that they did.

I think that the voters in a previous election will form a high percentage of the next election's base.

The question here is whether or not we have a reliable electorate, and I would think that we do. The test would be to take one's own behavior and universalize it. Would your response to a question regarding who you voted for and when, change?

Now if it is your boss or someone in authority that asks this, would you respond truthfully, or provide a response that you anticipate to be the one they desire? I think the measures of band wagon effect by different sample methodologies disguise this very understandable element of human behavior.

I think you have to take the measure in the exit poll as accurate, but was not properly reweighted because there is no interest in how Gore/Bush voters behave in 2008. Additionallly, the behavior is probably subsumed within previous weightings.

With the final national election poll's MOE at 6.5%, you bracket that around the results, and the percentage could be anywhere. 43-6.5 =36.5% (which I think is perfectly acceptable, though low estimate)

I think that there are several explanations for the phenomena, I just don't think the band wagon effect is apt.

Mike

Now you have gone from the sublime to the ridiculous.

There you go again.
Back to the your original RW talking points:
Big MOE.
Unreliable exit polls.

43% is too high by at least 3.5%
I have proved it.
But you ignore it.

It has NOTHING to do with sampling.
It has to do with how many voted for Bush in 2000.
And how many were still alive in 2004.
And how many voted in 2004.

That's what it's all about, mgr.
So try another one.

Oh, and the MOE is 1.0% for 10,000 respondents.
When they are asked the question.
Mitofsky said so himself.

No, mgr, there is no MOE as far as people are concerned.
They are either alive or dead.
No uncertainty there.

Do you believe in reincarnation, mgr?
Your desperation has led you down the path to the Land of Make Believe. Where logic stands on its head.

Where even mathematical truth has no meaning.

How can you explain that some polls during the day have a "bandwagon" and others don't...or that some states and counties have a "bandwagon" and others don't...or that the "bandwagon" doesn't go in the direction expected in some cases? I don't think the effect applies here,or it would be constant.

Message removed by moderator. Click here to review the message board rules.

There have been so many analyses of the differences in accuracy by voting method, swing states, etc., etc...that it is silly to think you haven't seen them...and they demonstrate that something other than bandwagon was the confounding variable. Bandwagon as an effect doesn't make sense here: expecially with subgroups like new voters and independents.

I suspect you have no real experience with survey's and polls...any experienced pollster can spot "bandwagoning", and Edison-M. had a chance to mention that and THEY didn't think it was important and they have the raw data in front of them! It's not the issue in this situation.

The ONLY issue that E-M has been able to use to explain the discrepency is self-selected sampling error - that is a threat to validity in every basic research book since (Campbell & Stanley) 1963! Any pollster knows about it, checks for it, and reports it when it happens. That's why the raw data is needed to demonstrate if TruthInAll's observations are supported, but bandwagon is not the answer.

Message removed by moderator. Click here to review the message board rules.

Look, folks...I'm no analyst, statistician, market research guru, whatever. But I am gathering information for others to use in "reverse engineering" the exit poll / final result anomaly. I know that the E/M polling was to be nonverbal...according to their letters to BoE Directors in late September. Nonverbal suggests a clipboard with response form filled-in by voter, then handed back to pollster. Seems very anonymous, other than the pollster inquiring if voter wishes to respond. So wouldn't this downplay the face-to-face (embarrassing...) issue?

I think the completed forms were taken to a compiling office in each county, then results phoned to E/M offices in NJ. I can't imagine each pollster being able to quantify their forms on the spot. Delivering the forms to some collection point for analysis
seems more likely. Now my questions for y'all. How can a bandwagon effect occur, especially in the early polling, when no vote results have been released? And what do post-election effects have to do with the E/M early OR final results?
Educate me on your varying approaches...

Message removed by moderator. Click here to review the message board rules.

Sorry, I didn't realize a bandwagon effect could have a life of FOUR years. What about six or eight years earlier? Is there a "cutoff" point in the past when the bandwagon no longer applies?

Re: the 2004 NEP polls; there weren't answers to oral questions; according to E/M's letter to BoE's, the polling was to be nonverbal, so why would an anonymous written/check-off answer have any relation to whether the responder voted D or R four years earlier? I don't see the possibility of an "embarrassment" factor or other reason for a
skewed answer to the 2000 question. I probably don't get
the methodology used as a basis for your POV. BTW, these discussions are taking on the appearance of a pissing match...or antler-butting, IMHO. Sheesh...

Message removed by moderator. Click here to review the message board rules.

If the Bandwagon effect resulted in a bias or error in the 2004 polls, then why didn't the E-M people who have all the raw data, created the poll, and have a staff of experts - NOT EVEN MENTION THE BANDWAGON EFFECT in their OWN analysis of their own data?

How about this answer: It wasn't an issue!

Not to mention that;
1.) there's no evidence of such an effect in the RAW data (which we don't have)
2.) there's no reason for E-M to avoid it because it gets them off the hook (until or if the raw data show up and they turn out to be liars).
3.) there's reasonable evidence that a bandwagon effect was inconsequential

They picked the only issue in the RAW data where they couldn't be proven liars: that the sample self-selected as biased...but that is a very weak...(as demonstrated by TruthInAll)...theory - fraud is stronger and with the RAW data an investigation would surely be next.

Message removed by moderator. Click here to review the message board rules.

If there was a "bandwagon" effect - any competent pollster would have known it and reported it....which E-M doesn't do...and E-M is very aware of TruthInAll and other reports of reasons to suspect fraud! The also control for a variety of effects.

The "who did you vote for" question is one you ASSUME has a "bandwagon" effect, but that is NOT apparent as anything but minimal across different types of party registrations, states, etc. Why not report your party loyalty as the winning party? Why not report you are better off than 2000 as supportive of a "fake" vote for Bush who won?

Also, pollsters already have controlled studies which demonstrate which persons may lie for any number of reasons and demographics which show which person's have answers that don't make sense. For example, if I am at a poll and ask a black female democrat who just voted for Betty Castor in Florida and they say, "I voted for BUSH in 2000" - then I'm a damned idot pollster if I don't ask, "why?" or toss the survey as unreliable! If I have 10%, as you suggest, or HUNDREDS of such inconsistent answers, I've got to really, really question what in the hell is going on....because we also know that those who engage in false answers do so on more than one question;..."when did you make up your mind?" and "Are you better off than 2000?", etc.

In fact, there is no RAW data available, and E-M themselves ONLY suggest that the sampling appeared to be an issue...if we take them at their word, then TruthInAll has as much confidence that the bandwagon effect is irrelevent as any other weighted, sorted, and professionally gathered report....on the other hand...if ANY data appears contrived or falsely altered, then TruthInAll is correct that we see evidence of manipulation that is convenient and based on partial real data, but has a pattern of logical interference (a rival hypothesis or alternative conclusion)...not legal proof of fraud...but a good reason to suspect that we need an investigation. If we had the RAW data, what would we find?

then they MUST also accept the following

1) THE FINAL EXIT POLL (BUSH WON 51-48) WHICH WAS MATCHED TO THE VOTE IS BOGUS, SINCE IT REFLECTS INFLATED "BANDWAGON" BUSH SUPPORT (43/37) WHICH WE KNOW IS IMPOSSIBLE ON ITS FACE...

THEREFORE THEY HAVE DONE A 180 DEGREE TURNAROUND AND HAVE NOW CANNED THE RELUCTANT BUSH RESPONDER FOR THE EXUBERANT BUSH RESPONDER.

AND

2) THE PRELIMINARY POLL (KERRY WON 51-48) WAS CORRECT, SINCE THE RESPONDENTS TOLD THE TRUTH AND DID NOT INFLATE THE BUSH NUMBERS (41/39) - NO BANDWAGON EFFECT.

BTW, EVEN 41/39 IS OFF. BUSH 2000 VOTERS COULD NOT HAVE BEEN HIGHER THAN 40%, ASSUMING 100% TURNOUT, DUE TO THOSE WHO DIED.

KERRY EVEN WINS ASSUMING A SMALL BUSH BANDWAGON EFFECT, AS REPRESENTED BY 41% BUSH/39% GORE (PERCENTAGES OF 122 MILLION).

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x354407

Message removed by moderator. Click here to review the message board rules.

Those aren't exit polls...where (in this case) 11% reported they were new voters and 89% reported they were revoters. If we had 18% real new voters, then about 7% said they had voted in 2000 when them didn't. Does that change TruthInAll's posts?

Exit pollsters should be able to spot common pervasive effects:

"Did you vote in the last election?" followed by: "How old are you?", "Where did you vote?", etc....and major effects are visible from multiple questions. Your articles aren't on target for exit polls from experienced pollsters. I don't doubt there are survey effects like bandwagon - just how much do they make a difference here (minimal) and why didn't E-M say so!

Overall, exit polls are predictive and accurate in most years, most states, and most countries. Here, we have SOME inaccurate polls in certain years, swing states, and differing by type of machine and even by precinct! What's the explanation from the data...all E-M has suggested is Reluctant Bush Voters (RBV).....that's not the error seen in the data - and easily demonstrated IF the raw data is available.

How many SPOILED exit surveys were administered and tossed is a good question.

"Since TIA heavily bases his "proof" of fraud on that question, that is the reason why I brought it up".

1 The pre-election state polls favoring Kerry.

2 The pre-election national polls favoring Kerry.

3 The post-election state exit polls favoring Kerry.

4 The post-election national exit polls favoring Kerry - prior to the final exit poll, you know the one with the "bandwagon effect".

5 the 50,000 documented election day "incidents", 99.7% of which favored Bush..

6 The pre and post-election shenanigans by the OH SOS.

7 The Berkley U. study of 200,000 impossible BBV Bush votes in FL

8 The analysis of Freeman, Baiman and other PhD's at uscountvotes.org

9 The fact that the recounts were fixed by the SOS in OH.

10 The fact that Bush stole the election in 2000.

11 The fact that for 42 of 50 states to move from the exit polls in favor of Bush, he had to beat odds of 1 in 1.7 million.

12 The fact that Bush beat 257,348,550,135,457 to ONE odds in exceeding the exit poll MOE in 16 states.


That's just SOME of what I base it on.

Question: D12A

(In the election for president this past November 2nd, did things come up that kept you from voting, or did you happen to vote?) Did you happen to vote for John Kerry, George W. Bush, or someone else?

Questionnaire: November Wave 1
Questionnaire Field Date: 11/19/2004-11/21/2004
Questionnaire Sample Size: 1015

Question Mean: N/A
Question Total N: 855


Scale % N
John Kerry 40.57 347
George W. Bush 51.22 438
Someone else 3.28 28
DON'T KNOW 0.27 2
REFUSED 4.67 40

http://brain.gallup.com/documents/questionnaire.aspx?STUDY=P0411044

I wonder whether they controlled by Party affiliation? Regardless, is this the bandwagon effect? We know that this doesn't mirror the reported 11/2 results.

post-election bandwagon effect, and the skew is even higher than the one in the study that I posted.

It lacks some critical evaluation, but I believe that is your issue that you cannot follow the criticism. If you cannot follow an argument, the last thing you should think is that you cannot comprehend it. Anything that a human being does, it should be comprehensible by another human being. The way to critically assess another's position you do not fully understand is from skepticism, and if you do not understand it, you have the right to assume that it is nonsense. The world is full of charlatans. If it sounds too good to be true, it probably is.

Don't let math daunt you, remember Disraeli's quote: "There are lies, there are damn lies, and then there are statistics."

But, to answer your question--no it does not contradict Mitofsky.

Here is how I can explain it (please don't take it as condescension):

Take 20 coins, have it be 12 California Quarters, and 8 Texas Quarters. Pick eight quarters from a bag, does your sample match the 60 to 40 ratio. What you have just modeled is what happens at a precinct with refusals to be polled, and people who the pollster could not stop to interview. The 12 you did not pick are the ones missing. What should happen is that they should vary a little from the 6:4 ratio, but not out of hand

Now, take your twenty coins, and hang onto one California Quarter, and pull out the other seven. Again compare your sample to the total, now you have modeled the bias that NEP reported.

Now, how do you make your sample consistent with the actual voters, well you have to adjust it to 100%. Simple, right. You have now weighted your sample. Now in the second case, how would you reweigh the sample, well now you make an adjustment for that extra quarter (lets not worry about how you know there is an extra quarter for now), but you would adjust the Texas quarter weight upwards and decrease the California quarter weight downwards. (this is the outcome of the reluctant Bush responder effect)

Now you total up all the other sets of quarters in the state, after making these adjustments to get the state election outcome, or the federal outcome.

Now, lets consider what will influence someone to vote for Bush or Kerry. The biggest factor is probably party membership. I would suggest that party registration may determine how someone is going to vote in a precinct at 90% (in other words, if I consider this alone, it would explain 90% of the outcome). What might be the next factor, gender, ethnicity, age, each of these explains a smaller part of the pattern, and each is shared with a previous category, so I adjust party membership, it will also adjust gender, ethnicity, and age.

Now we come to the Gore voter Bush voter conundrum. It is treated as a distinct category in the polling, but if 90% of the vote for each candidate is related to party registration, then wouldn't Gore voting and Bush voting follow this pattern closely? I think you caught this as a hidden assumption.

Now, I'm done with the election, but I am not happy with how well my poll estimates matched actual results (that quarter I hung on to), so I change the weighting to better analyze future elections. So, I redo all the weightings but don't bother with the Gore Bush adjustment, because I am not going to be interested in who voters from 2000 prefer in 2008. So what happens, I make the adjustment, increasing the value for republicans registered, and lo and behold, I end up with more Bush 2000 supporters in 2004 than there were in 2000.

Mike

I'm not quite sure what statisticians mean by bias. I know what it means to a lawyer.

But it seems to me that the bias Mytofsky is using as justification for descrepancy is bias that is behavorial. A social scientist, or psychologist, I can't remember which, posted on Mystery Pollster some time ago taht he thought mathmaticians, statisticians, and political scientists made horrible experts in the fields of human behavior. I wish I could find his post, but the archives on MP's website don't go back far enough.

To say that a poll is biased because the pollster was too old, too young, too hippie or whatever, to get a non biased response, to me does not seem to be the expertese of stats people.

So when you say the polls show bias and not fraud, what kind of bias are you talking about, and what are your credentials for making this claim if the bias is behavorial?

Bias can either be intentional or incidental. In the case of NEP, I think it is incidental to the experimental design, and can be improved on.

Statistically, bias implies a skew of the data that is not consistent with the phenomena being measured. What has to be recognized is that if the purpose of the exit poll is to explain the behavior of segments of the population, then it will over sample any party that is more heterogeneous in its make up (e.g. the Democrats). The correction to this would be in the weighting and evaluating the consistency of the results against better studied precincts.

What is the desire of any experimental design is to control the bias to where its effect on the measure of central tendency is less than the variance. Each variable in a sampling method has potential to introduce bias or error into your sampling. This is basic. Where bias is greater, then you have a problem, and in the scientific realm, if you cannot explain it, you do not publish, because your results are rubbish.

If a category is quantifiable, it is within the realm of statistics to describe or test. Quite correctly, there may be other factors that have not entered into the analysis, but if a major portion of the bias encountered can be explained by a factor, it requires further evaluation, and should be provisionally accepted as a working hypothesis. It may just be that what you have is a spurious correlation. In science you do not prove a hypothesis is right, just that it has explanatory power for the phenomena measured, but may not be applicable in future outcomes when potentially, a more widely encompassing hypothesis may be formulated.

What NEP appears to have done was run a multiple regression to see what factors explained the bias they ran into, apparently a significant correlation was found between the interviewer and the non response of Bush supporters. One of your experimental controls is how your interviewer comes across, and how well they follow directions. The trend appears to work with higher education reducing non response until you reach the graduate level interviewer. I am almost willing to bet that the pattern is a result of the higher degreed interviewing more closely following the directions that any other category, since they are possibly more in tune with why a stringent methodology must be followed.

It would probably be important to the NEP's methodology to control for the interviewer by limiting what educational attainment and age each interviewer has, or select those consistent with the educational attainment and age within a precinct.

My credentials to evaluate this would be that my education falls within both the behavioral sciences and the natural sciences (I am a geographer with a biological/meteorological/geomorphology emphasis, with a doctoral focus on paleoclimate reconstruction (never completed), and a masters focus on synecology). I have both undergraduate and graduate level training in the application of statistics, and in experimental design. I currently employ statistics in my daily work evaluating whether potential hazardous materials present any human and environmental impacts. I guess that I have legal expertise since I will be called as a witness to explain our agency's methodologies at a trial next month.

Mike

I will be very frank, any one that has claimed to have training in the behavior sciences of psychology, sociology, or political science, and whose statements are consistent with that type of background, I will defer to.

I believe the law recognizes the concept of intentional and unintentional bias in Civil Rights Law, if what I recall regarding a case from New Mexico.

to the 2004 exit poll that showed the 43% for Bush and 37% for Gore?

The Washington Post graphic I have shows 41% and 39%.

Thanks in advance.

CNN Final Exit Poll
2:05pm Nov.3 13660 Respondents
http://www.cnn.com/ELECTION/2004/pages/results/states/US/P/00/epolls.0.html

NEP/WP Preliminary Exit Poll
12:22am Nov. 3, 13047 Respondents
http://www.democraticunderground.com/discuss/duboard.ph...


CNN Preliminary Exit Poll
7:38pm Nov 3, 11027 Respondents
http://www.exitpollz.org/CNN_national2.htm

.