US Senate Exit Polls == Why are they so different than Presidential Poll?

Has anyone been researching the US Senate exit polls? We have the more accurate numbers from the Mitofsky/Edison Media report. What do they tell us?

I could not help but keep going back to the data because several huge discrepancies would not evanesce from short-term memory. Why was the Ohio Senate margin off by over 15%?

Using the "Evaluation of Edison/Mitofsky Election System 2004" data (Exit_EvaluationJan192005.pdf), Best Geo Estimator, Call 3, we find the following:

	
Margin Exit Discrepancies

OH	15.10
CT	14.40
NY	12.80
NH	12.20
VT	10.40
PA	10.30
HI	8.70
AL	8.60
SC	7.80
KS	7.30
LA	6.30
ND	6.00
NV	5.50
NC	5.40
WA	5.20
IL	4.70
MO	4.10
FL	3.90
WI	3.60
UT	3.50
AK	3.20
GA	3.10
IA	2.30
KY	1.60
CA	1.40
MD	1.30
IN	1.30
OK	1.20
AR	1.20
SD	1.00
AZ	0.30
CO	0.00

Who has analyzed this data?

for senate exit polls to be erroneous as well.

Now if the senate exit polls matched the official election results, but the presidential exit polls didn't, you'd have something.

By the way, this data knocks the rug out from under people who claim that the "fraud" was confined to the Presidential election (like, for example, the fact that voting a straight Democratic ticket was not registering a vote for President, so people thought they voted for Kerry when they didn't). If that was the case, and the exit polls were not erroneous in general, you'd see them match in Senate elections and not match in Presidential election. So - either the "fraud" was across the board, in Senate and Presidential elections, or the exit polls had a huge problem in 2004.

people talking having problems with the federal portion of the ballot but not on local issues...

The degree of discrepancy is not exactly the same. Mean Pres. margin discrepancy is 4.98%, Senate is 3.65%. So, we know the Presidential results are further from the exit polls than the Senate results.

However, I'm raising a different question. Why, in particular states, are there such a large differences in the President and Senate discrepancies?

Here are the error margins side-by-side, for Pres. and Senate.

 
 	Kerry		Senate Dem
	Margin Error		Margin Error
AK	-9.40		-4.30
AL	-10.10		-8.66
AR	-4.10		1.14
AZ	-3.50		0.14
CA	-3.00		-1.92
CO	0.80		-0.02
CT	-9.30		-14.38
FL	-3.90		-3.92
GA	-3.60		-3.07
HI	-4.40		8.71
IA	-1.70		2.28
IL	-4.80		-4.69
IN	-1.60		-1.25
KS	0.80		7.38
KY	-1.90		-1.52
LA	-1.40		6.29
MD	-6.70		0.98
MO	-2.80		-4.19
NC	-8.80		-5.40
ND	7.00		-5.83
NH	-13.60		-12.15
NV	-4.00		-5.59
NY	-13.90		-12.72
OH	-8.60		-15.01
OK	0.60		1.28
PA	-11.50		-10.34
SC	-11.10		-7.88
SD	6.80		1.04
UT	-6.20		-3.47
VT	-16.50		-10.74
WA	-3.50		-5.26
WI	-5.30		-3.67

So, why are Florida and Illinois so equal, and Ohio and South Dakota so different? Did 7% of South Dakotans and Ohioans lie about Presidential votes while telling the truth about Senate votes? I don't think so!

In North Dakota, are 7% of the people liars? I don't think so. Maybe one or two voters in the whole state, if that!! And if so, why do they lie that they voted for Bush for Pres. while they also lie that the voted for Byron Dorgan, a Democrat, for Senate?

This is the difference I'm concerned about. If we trust the poll responses, what is wrong, especially in Ohio? The exit polls gave the Ohio's incumbent Senator Voinovich a 12.7% margin, the vote results gave him a 27.7% victory. Go figure!!

flaws, whether in methodology, or in execution, you would expect them to give all-over-the-map results, just like you're showing.

These results are much harder to explain if, as TIA does for example, you claim that the exit polls were accurate.

I'm inclined to believe that's the response to this.

Some of them look (CO, LA, SD etc.) look different from those I've seen in TIA's posts. Are the Mitofsky numbers so different or am I just thoroughly confused?

The numbers represent the percentage differences between the two main contenders in the race. The exit poll data privides the percent of responders stating they voted for a candidate. I used the actual vote numbers (post-election day, 100% of precincts reporting) for the for the vote percents.

TIAs numbers errors instead of margins, are not third call numbers, were rounded to nearest percentage by the networks, and were adjusted to add up to 100 percent for the two main contenders. Compared with the numbers statisticians are using, TIAs numbers are off on average by near 1 percent (error, not margin).

My exit poll numbers I used are those used by Steve Freeman in his studies and are available (along with the report Senate exit poll numbers) online.

Download the exit_poll.xls spreadsheet and others from this spreadsheets page online.

So, I should really have asked, are 3.5% of North Dakotans lying, etc... because a switched vote counts double.

He uses the percentage-point differences in the between the candidates in the exit-poll projection and the actual.

That effectively doubles the errors. In Ohio, for instance, Kerry had about 52% in the exit poll, but only had 48% in the actual. That's a 4 percentage-point difference in the exit poll. But, since Bush and Kerry together had more than 99% of the vote in Ohio and most other states, Kerry's loss was Bush's gain, so he had about a 4 percentage-point gain. Thus, we see Mitofsky's -8.6 percentage-point difference in Ohio in the list.

All of this is somewhat irrelevant because Mitofsky hasn't justified his methodology. I noticed that when I read his report. Now, Dr. Freeman has just produced a detailed rebuttal of Mitofsky's error analysis.

Wittingly or unwittingly, Mitofsky is covering-up for election problems or outright fraud.

which would double TIA's numbers. I was taking that into account when I indicated several states look wrong -they do!

Freeman is one of the 9 authors. The Press release is online today:

"
Prominent Statisticians Urge Investigation of 2004 U.S. Presidential Election Results by Kathy Dopp Monday, January 31, 2005

Prominent Statisticians Refute 'Explanation' of 2004 U.S. Exit Poll Discrepancies in New Edison/Mitofsky Report.

President Bush won November's election by 2.5% yet exit polls showed Kerry leading by 3%. Which was correct?

"There are statistical indications that a systematic, nationwide shift of 5.5% of the vote may have occurred, and that we'll never get to the bottom of this, unless we gather the data we need for mathematical analysis and open, robust scientific debate.", says Bruce O'Dell, USCountVotes' Vice President.
"

http://www.uscountvotes.org/index.php?option=com_content&task=view&id=72&Itemid=41

On average, the Senate exit poll/vote result differences are 1.19% more accurate than the Presidential equivalent in the same state.

Some states cancel differences favoring the other party in other states. Therefore, I examined the absolute value (ABS) of the differences, a more accurate gauge of the discrepancy between the Presidential and the Senate reponses and votes when analyzing the entire group.

In the 32 states with Senate races, the ABS difference (in the discrepancies between the vote and the exit polls) between the Presidential and the Senate races average is 3.423%, the median is 2.243%, and the skew is 1.501. The population standard deviation is 3.303% due to 2 large outliers, North Dakota (13.11%) and Hawaii (12.83%), both states with incumbent Democratic Senators winning by wide margins.

Without these two outliers, the mean difference is 2.787%, the median is 1.695%, and the skew is 0.633.
The population (30 states) standard deviation is 2.271%.
The following states, using the 30 state statistical subset values, have greater than 1.0 z-scores (standard deviation from the mean):

Difference	z-score
HI	13.105	4.544
ND	12.835	4.425
LA	7.690	2.160
KS	6.580	1.671
OH	6.406	1.594
AL	5.797	1.326
VT	5.760	1.309
SD	5.760	1.309
MD	5.716	1.290
CT	5.083	1.011

There are a few differences in the numbers due to corrections in the spreadsheet.

Isn't it obvious?
Either he is a WILLFUL PAWN of the Refucknigans, or he is CRIMINALLY STUPID and is FRAUDULENTLY covering his ASS!
WHAT does it TAKE to PROVE THIS FRAUD?!?!

You can't say Mtofsky is a crook on one hand and then turn around and say his exit poll data is accurate.

I think the exit polls in general are a farce. Thats why ultimately, I am still not 100% convinced of fraud. Like I said, I only _know_ that something about the election STINKS, and we will NEVER know the truth so long as characters like Mitcrapsky do the exit polls!

Although data from such an analysis would be wide open to many interperatations, it might nail down some specifics. For example, it might prove a thing or two about the supposedly selective non-response, like for example, what percent of the nonresponders would have had to have been people who voted in both races versus just one for the theory to hold water.

(EDIT: what I mean to say is, if the "selective nonresponse" was independent on whether the person undervoted in either race, then the correlation of disparity in places where cross-voting was rare should be strong. So I guess I'd have to say add rough cross-voting calculation, though that can be tough.)

I'd be interested to know. If you want to look into this, here are some suggestions of things to check:

In each state, was there a competitive race for senate, or was it expected to be a blowout for either side? What did the pre-election polls say?

What was the comparitive undervote in the final results -- how many more people voted for president than for senate?

I forget if Mitofski provided separate completion rates for the senate, to give an indication of how many people took the exit poll survey but refused to answer one or the other question about who they voted for for President or senate. But if so, that would be good backing info to have.

Please keep on this question. I've been wondering about 2000, and 2002.

Apparently, Georgia was a rotten peach in 2002.

BoredtoDeath posted this on a thread:

<http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=308725#308937>

I'm going to do something with this in class, and I need it from an unbiased source.

http://www.exit-poll.net/election-night/EvaluationJan192005.pdf

page 23, 24

Consider the big picture for few minutes.

You have to BELIEVE IN WHAT THE NETWORKS TELL YOU ABOUT ELCTION IN THE USA.

I suspect that ALL the numbers are altered. I believe the difference between exit polls and the vote count are just the proof of this.

I don't think the votes are counted accurately (if at all) and the exit polls are just showing this VERY CLEARLY.

And the fact that people often vote 1 party for pres, and another for congress helps them fog it up a bit.

affected huge numbers of votes in many states.
Irregularities over straight party vote counts(not what intended)
Too few machines and machines that didn't work in minority precincts in very many states;
Touchscreens defaulting away from Dem candidates
Widespread absentee ballot irregularities
Widespread fraud and malfeasance in count of provisional ballots
Widespread registration malfeasance and dirty tricks

These affected millions of votes; this election was not a serious attempt to determine the will of the voters.

http://www.flcv.com/ussumall.html

big swing that affected Kerry- most Dem races were affected by the widespread fraud, malfeasance, dirty tricks, suppression,etc.

similar for New Mexico, and etc.

http://northnet.org/minstrel/alpage.htm
http://www.flcv.com/fraudpat.html
http://www.freepress.org departments articles

I've always suspected at least Daschle's race. If ever there was someone to target...

Being a software engineer and not a mathemagician, it's beyond me without a bunch of study.

-Hoot

Do the pollters assume that individuals vote straight party tickes all the time. I know several people that will split a ticket, Dem for State and Repug for Pres. Would be hard to accurately measure the difference if you did not make the above assumption.

...would have to be part of anything that gets done with this, yes.

However, there are ways to estimate cross-voting levels.

What I think you may end up with in the end is a proof of something along the lines of "if anyone was selectively responding, it was democrats who voted for Bush". It may not be enough to completely disprove the "selective non-response" theory by itself, but it would be at the very least, interesting to know.

Presidential elections have a unique footprint. In some Senatorial races where the results overlap the Presidential results geographically and demographically, even a 2% differential may be significant. In other cases, the uniqueness of the Senatorial coalition may simply sidestep the key precincts for the exit polls.

They also did exit poll results on all the governor races and about 10 of the key US Rep races. It is also necessary to see these other results.

The exit poll does not guess on Senate results - it asks 'Did you vote for Boxer or Jones?' (or whomever). So to have a huge difference on the error from one race to the next is unexpected, because the President and Senate exit polls use the same sample of respondents.

In Ohio, for example, the Senate race is off more than the Pres. race. Using conventional wisdom/excuses, the reasoning would go: 1) Bush voters either lied or were undercounted; 2) Kerry voters who also voted for Republican Voinovich in the Senate race were even more undercounted (or they told the truth when saying they voted for Kerry but then lied by saying that they voted for Voinovich's Democratic opponent); 3) Kerry voters who also voted Democratic in the Senate race were much, much more likely to respond to the exit pollsters than either other Kerry voters or any average Bush voter.

In Ohio the most under-represented voter in the exit poll is the 12% of the electorate who voted for Kerry and Voinovich (a Republican).
Why would these voters not want to talk to pollsters in Ohio but they were perfectly fine talking to pollsters in Pennsylvania?

Exactly. That's why alarm bells keep going off in my head.

What's your Occam's Razor on this one?

This will not simply explain away.

TIA does such a thorough job of this. Have you pm'd her?

TIA is male.

This EF continues to be shocking.

Help us, T!

http://wid.ap.org/campaign2004/senatenew.html

The other states that are not listed (e.g. Texas) did not have US Senatorial contests in the 2004 Election.

I want to sit back with some popcorn, a beer and the latest Freeman report.

:beer:

The Presidential and Senate poll results derive from exactly the same responders. Therefore the variables, those cited as possibly skewing the exit poll in the Mitofsky/Edison Media report, should not impact the comparison. There is no reason to account for non-responders or missed voters in this comparison, at least in discussing the different accuracy results for the two races.

Therefore, the Presidential to Senate comparison should be very useful in pointing to discrepancies in the voting results.

Does anyone find fault with this argument?

There are some counties in Ohio, for example, that went strongly for Kerry and Voinivich. But we should be able to figure out some other metrics that put a cap on the number of cross-voters, because although we do not have exit polls for stuff lower down on the ticket, my guess is ticket splitting is much more common at the very top of the ticket. So you can likely use the official tally on the bottom of the ticket to get a handle on the number of "dixiecrats" and whatever you'd call the converse.

If we can control ourselves and not run around waving the figures in the air before accounting for cross-voting, though, then just seeing the discrepancies side by side with the WEP and response rate and pre-election tracking polls and anything else that may be of use would be a very intriguing start.

It does not matter that there are cross-voters as long as they do not lie to one question and answer the other truthfully.

What we are assessing is the accuracy of the exit polls and the vote results. Cross-voting should not matter.

Can you explain why it would?

Are we looking at distinct issues?

...that the crockery will extend to saying "it was group X of cross-voters that didn't respond." That's why.

If it is going to be done, it should be done right, and the rhetorical escape route closed beforehand.

between the two contests.

For example in Florida, the GOP Senatorial candidate was Mel Martinez, a Cuban-American and an extreme right-winger. Martinez tended to pick up votes in South Florida, which is heavily Democratic, lose votes in North Florida which is right-wing Republican but anti-Cuban, and lose some votes in central Florida which is moderately Republican but put off by hard over Pubs like Martinez. The Presidential sample precincts may or may not have been optimized to pick up these local variations.

Remember, the NEP uses a very small sample (highly economic) refined over many elections. These local one shots can be wild cards.

I agree wholeheartedly when there is a reasonable match between the two contests.

In Florida, the Senate and the Presidential exit poll discrepancies have the least difference of all 32 states.

What about North Dakota? How do you explain a 12.83% difference there? No Cuban, no Dixiecrats. A rather homogeneous population, actually.

I'm saying that local variations in a specific state contest can introduce a pattern that may not make the sampled precincts generate as random a sample as might be desired. There are "favorite son" effects and others which may impact what is in the end a small sample.

The interesting thing is that it can work either way so that you could argue, I suppose, that the sample in Florida was very accurate DESPITE the variations in the Senate race.

On the other end of the spectrum is Alabama (to name one state) where the Presidential and Senate races are virtual mirror images of each other and yet you get an 8% variation.

I'm not saying anything radical. It is a truism of politics that "key precinct" indicators can be skewed by oddities in a particular contest. That then subjects any analysis to a specific criticism of the peculiar conditions of that race. If, on the other hand, the Presidential and Senatorial races have similar footprints, it is obvious that the exit polls SHOULD have very small discrepancies.

I think that the Presidential exit polls have always been more accurate for the Presidential races than for Senate races (I could be wrong but I think this is true). One approach would be to increase the state sample which is precisely what the polling companies do when they are hired for a specific state race.

I know nothing about North Dakota but Ohio looks like it is worth exploring.

You wrote, "I think that the Presidential exit polls have always been more accurate for the Presidential races than for Senate races..."

I do not know about past elections, but this time the Senate races, on average, were significantly more accurate.

If there is a built-in expectation that the Senate result would be less accurate, and I am unconvinced that there is, the greater accuracy of the Senate exit polls just makes the Presidential results all the more troublesome.

I think we will be hearing a lot more about this issue, as the analysis moves forward. The Senate exit polls indicate problems in the vote results.

Ohio is certainly disturbing at 15% off the mark. Ohio just plain stinks to high heaven!

My memory is a little dated but, I coulda sworn....

... could be Alzheimers, I suppose.

You are right that it stinks, to heaven and beyond.

Since being Cuban-Amer affecting his regional vote more than most candidates. But he has been more moderate in the past and only just recently became radical right wing. He ran to the right of a long time right wing extremist.

You could do an accurate exit poll one of 2 ways. choose statewide precincts for the poll to relect similar demographic/political balance as the state; or choose precincts that have some Repub & Dem strongholds, etc. but not necessarily balanced same as state and develop a regression equation to predict the candidate percent vote, based on past election results.

Either way is viable and there are pros and cons to each way.
Trying to choose precincts for balance may be somewhat easier, but the other might be more accurate unless there are unusual differences between current race and past races that prediction equation built on.

In the case of the Florida Senate race, the swing is explained by
the vote machine, dirty tricks, and suppression documentation for Florida
http://www.flcv.com/fraudpat.html

Depending on the state and the specific race, it can get complicated. Better to start with the obvious cases...

http://www.exit-poll.net/election-night/MethodsStatementNationalFinal.pdf

"The polling places were selected as a stratified probability sample of each state. A subsample of the state samples was selected at the proper proportions for the National exit poll. Within each polling place an interviewer approached every nth voter as he or she exited the polling place. Approximately 100 voters completed a questionnaire at each polling place. The exact number depends on voter turnout and their cooperation.

For the national tabulations used to analyze an election, respondents are weighted based upon two factors. They are: (1) the probability of selection of the precinct and the respondent within the precinct; (2) by the size and distribution of the best estimate of the vote within geographic subregions of the nation. The second step produces consistent estimates at the time of the tabulation whether from the tabulations or an estimating model used to make an estimate of the
national popular vote. At other times the estimated national popular vote may differ somewhat from the national tabulations."

Your point is germaine. Of course, specific issues can skew a state race and the subset that is incorporated in the National exit poll can still be highly accurate.

(I emailed him this post and this is what he wrote back to me.)

...Eight of nine competitive Senate races went Republican by very narrow margins, even in states where the Republican
candidate was a laughing stock (KY, OK, SC). The
other five states were AK, FL, LA, NC, SD. Only
Colorado went Democratic. Now I will look at the list
from Democratic Underground:

> SC 7.80
> LA 6.30
> NC 5.40
> FL 3.90
> AK 3.20
> KY 1.60
> OK 1.20
> SD 1.00
> CO 0.00

I cannot help but notice that all nine states are on
the list, and the only one in which the exit polls
matched the election results was Colorado. I leave it
to others to find out if the margins of victory for
the eight Republican Senators were greater than or
less than the listed discrepancy. I assume that the
shift in the election results was toward the
Republicans or else these numbers would not have been
posted on Democratic Underground. To take control of
the Senate, the Democrats needed to win 6 of 9 seats
if Edwards were elected Vice-President, and 7 of 9
seats if Cheney were elected Vice-President.

You may post this.

Richard Hayes Phillips, Ph.D.

Margin values correspond to Dem's margin (ie. Daschle lost by 1.16%).
Negative Exit Error values are Dem Vote Results less than Exit Results.

We see that in the contested races, the Dems were leading in 5 races, averaging a 0.463% lead in all 8 races. They won only one of the 8, in Colorado. The average Exit Error in the 8 contested races shifts over 4% in favor of the Repugs. Only in SD does the Dem do better in the vote than the exit poll.

		Dem		Dem		Exit
		Final Margin	Exit Margin	Error

PA		-10.844		-0.50		-10.344
SC		-9.682		-1.80		-7.882
NC		-4.597		0.80		-5.397
AK		-4.203		0.10		-4.303
KY		-1.324		0.20		-1.524
SD		-1.160		-2.20		1.040
FL		-1.124		2.80		-3.924
CO		4.285		4.30		-0.015
                 -------------------------------------
Contested races:
count		8		8		8
mean		-3.581		0.463		-4.044
median		-2.763		0.150		-4.114
stdevp		4.613		2.051		3.619
                 -------------------------------------
All races:
count		32		32		32
mean		0.858		4.506		-3.781
median		-1.242		0.500		-3.795
stdevp		29.140		30.386		5.759

Common Dreams

Evidence Mounts That The Vote May Have Been Hacked
by Thom Hartmann

When I spoke with Jeff Fisher this morning (Saturday, November 06, 2004), the Democratic candidate for the U.S. House of Representatives from Florida's 16th District said he was waiting for the FBI to show up. Fisher has evidence, he says, not only that the Florida election was hacked, but of who hacked it and how. And not just this year, he said, but that these same people had previously hacked the Democratic primary race in 2002 so that Jeb Bush would not have to run against Janet Reno, who presented a real threat to Jeb, but instead against Bill McBride, who Jeb beat.

"It was practice for a national effort," Fisher told me.

-snip/more-

<http://www.commondreams.org/views04/1106-30.htm>


Rense has a word or two to say about Reno/Mcbride Primary 2002:

<http://www.rense.com/general29/DADE.HTM>

CNN Exit Polls 2004

Fairly thick site.

<http://www.cnn.com/ELECTION/2004/pages/results/states/OH/P/00/epolls.0.html>