EDA Exit Polls Generally Match 2008 Election Results, But Find "Wide Disparities" in NH Vote Counts

An Election Defense Alliance Investigative Report

Based on Data Obtained from the EDA 2008 Election Verification Exit Poll Project (EVEP)

This report is meant as a warning. It does not provide conclusive proof of election tampering, but what is revealed here is strong enough to suggest that Legislators, Secretaries of State, Attorneys General, AND CITIZENS must pay close attention to what is reported in all future elections. Candidates entering races in 2010 Mid-Term Elections should especially read and understand this report and take notice of the current state of our electoral system.

____________________________

'We know that there are huge disparities between the exit poll data and the official results
at all four polling places in New Hampshire, and that adjusting the raw data
to account for party affiliation does not explain them. . . .
We are forced to conclude that it is very possible that the official results in New Hampshire are not true and correct.'
____________________________

CITIZEN EXIT POLLS ACROSS THE COUNTRY:

AN IN-DEPTH ANALYSIS

Richard Hayes Phillips, Ph.D.

Download the Report [1] (PDF)

Appendices added

ABSTRACT

Data from citizen exit polls conducted at 28 sites in seven states were found to match closely the official results for President, Senate and Congress at nine polling places (five in California, two in Pennsylvania, one in Ohio, and one in New Mexico). At four polling places (two in New Mexico, one in California, and one in Texas), analysis was inconclusive. At two polling places in Michigan, the official results for 2008 were not inconsistent with established voting patterns. At two polling places in Ohio, the exit polls reflected correctly an erosion of support for the Congressional incumbent. At five polling places in California, the presidential preference of the non-responders closely paralleled their party affiliation. But at six polling places (two in Pennsylvania and four in New Hampshire), large disparities remained between the official results and the exit poll data, even after properly adjusting the data to account for party affiliation, gender, age, and race. The large disparities in the vote count were found at three of these, and we conclude that the official results may be wrong at all six.

INTRODUCTION

Citizen exit polls were conducted by trained volunteers on behalf of Election Defense Alliance (EDA) on November 4, 2008 at 37 sites in eight states. The purpose was not only to collect demographic data (gender, age, race, and party affiliation) for election analysis, but also to reach a large enough sample of voters at the polls to verify (or question) the official results. In every state, the presidential election was listed on the questionnaire handed to the voters. For comparative purposes, the Congressional election, the United States Senate contest if any, and some local contests, were included as well. It is the purpose of this paper to compare the exit poll data with the official results and, where large disparities exist, to assess the reasons for those disparities.

There are four possible reasons for a large disparity between exit polls and official results: (1) a basic flaw in the exit poll methodology; (2) many voters lying on the questionnaire; (3) a non-representative sample of voters responding; or (4) the official results being erroneous or fraudulent. The first two possibilities are rendered unlikely by the fact that, at numerous polling places, there was little difference between the exit poll data and the official results. Thus, if the official results are true and correct, any large disparities must be due to exit poll responders being non-representative with respect to gender, age, race, or party affiliation. It is shown in an accompanying paper concerning Propositions 4 and 8 in Los Angeles County that party affiliation is the most important of these parameters.

This underscores the importance of collecting “refusal data,” as was done in this poll. The exit pollsters noted the gender, race, and estimated age of each voter who was approached but declined to respond. These data can be compared to the responses on the questionnaires filled out by the participating voters. In some states, the gender and age of registered voters are specified on the voter rolls. Of utmost importance are the party affiliations of those who voted at the polls, which in some states is a matter of public record, although sometimes difficult to obtain. Based upon this information, the raw data for the exit poll can be adjusted according to gender, race, age, and party affiliation, to better reflect the demographic makeup of the electorate.

OVERVIEW

Not all of the exit polls resulted in worthwhile and useable data. At one polling place in Michigan, only 60 of 835 voters were interviewed; no meaningful conclusions can be drawn from such a minimal data set. In San Francisco, and in three of five polling places in Santa Fe County, New Mexico, questionnaires were either lost or possibly mixed up among precincts, leaving us with incomplete and unreliable data sets. At the four polling places in Colorado, election officials have refused to provide a separate vote count for voters at the polls. In Douglas County, Colorado, for example, early voting and absentee ballots accounted for 86.89% of the votes countywide, and 89.51% of the votes in the three precincts where our exit poll was conducted. We are left with no way to make a meaningful comparison between the exit poll data and the official results. But this still leaves us with 28 polling places in seven states. The raw data are shown below.

TABLE 1: COMPARISON OF UNADJUSTED EXIT POLL DATA

AND OFFICIAL RESULTS FOR PRESIDENT

	Official Results		Exit Poll
	McCain	Obama	McCain	Obama

CA LA Taft 9001037A	201 36.0%	346 61.9%	88 28.7%	214 69.7%
CA LA Long Beach 3850101A	120 20.3%	459 77.8%	55 15.8%	285 81.9%
CA LA Berendo 9005399A	127 18.9%	527 78.4%	72 17.5%	331 80.5%
CA LA Santa Monica 6250005A	128 17.0%	614 81.5%	56 10.6%	465 87.7%
CA LA Topanga 710003A 6A	129 12.0%	918 85.3%	26 4.5%	535 93.5%
CA LA Lockhurst 9006489A 90A	405 34.8%	743 63.8%	151 25.8%	421 72.0%
CA LA Glendale 2550120A 120B 122A	393 23.9%	1218 74.0%	106 16.7%	511 80.7%
CA LA Locke 9001145A 9002566A	47 4.2%	1050 94.4%	24 3.3%	698 96.1%
CA LA Eagle Rock 9006334A 6335A	347 28.9%	827 68.9%	157 21.0%	575 76.8%
CA LA Lynwood 3990015A 16A 18A 19A	253 9.3%	2421 89.4%	106 7.7%	1263 91.7%
CA Alameda 280300 280700	95 17.6%	428 79.1%	43 13.0%	286 86.1%
MI Oakland West Bloomfield 2	433 43.5%	545 54.7%	134 33.7%	257 64.6%
MI Washtenaw Chelsea 1 2	752 39.6%	1115 58.7%	195 27.0%	509 70.5%
NH Hillsborough Manchester 3	1013 33.2%	2003 65.7%	401 26.1%	1106 72.0%
NH Hillsborough Manchester 5	992 35.6%	1761 63.2%	286 25.0%	832 72.8%
NH Hillsborough Nashua 5	2499 47.2%	2741 51.8%	746 41.4%	1022 56.7%
NH Hillsborough Wilton	1026 44.6%	1248 54.3%	416 36.7%	692 61.1%
NM Santa Fe 7 Rio En Medio	27 14.1%	161 83.9%	19 13.3%	120 83.9%
NM Santa Fe 25 33 Alameda	91 21.0%	332 76.7%	51 15.6%	267 81.7%
NM Taos Pueblo 13	13 4.3%	291 95.7%	8 3.4%	224 96.6%
OH Cuyahoga Beachwood A B N	542 40.6%	784 58.7%	229 40.5%	331 58.5%
OH Cuyahoga Independence A C F	1013 57.7%	717 40.8%	307 53.5%	254 44.3%
OH Cuyahoga Rocky River 1D 1E 1G	676 53.4%	571 45.1%	223 46.8%	246 51.7%
PA Allegheny Pittsburgh W 14 D 8	107 21.1%	394 78.0%	45 14.4%	264 84.3%
PA Cambria Munster	239 71.6%	89 26.6%	189 71.3%	67 25.3%
PA Centre Harris Twp 56 57	1379 46.7%	1540 52.1%	490 37.1%	816 61.8%
PA Philadelphia D 30 W 5 10	21 3.8%	535 95.9%	8 1.8%	440 97.8%
TX Harris Houston 34	197 27.9%	483 68.3%	95 20.3%	355 75.7%

NOTE: The official results shown above do not include absentee ballots for California, Michigan and Ohio, and do include absentee ballots for New Hampshire, New Mexico, Pennsylvania and Texas.

As shown in Table 1 above, the raw exit poll data, unadjusted for any sample bias with respect to gender, age, race, or party affiliation, are within 2% of the official results in 8 of 28 polling places (three in California, two in New Mexico, one in Ohio, and two in Pennsylvania).

At 20 polling places, the difference between the raw exit poll data and the official results exceeded 2%, thus equating to a disparity of more than 4% in the “point spread” – that is, the margin of victory or defeat. The simplest analysis is to determine, by subtraction, what the vote count must have been among those who declined to participate in the exit poll (the non-responders, or refusals), assuming that the official results are true and correct. The comparison is shown below.

TABLE 2: VOTE COUNT AMONG NON-RESPONDERS,

ASSUMING OFFICIAL RESULTS FOR PRESIDENT ARE TRUE AND CORRECT

	Exit Poll Responders		Non-Responders
	McCain	Obama	McCain	Obama

CA LA Taft 9001037A	88 28.7%	214 69.7%	113 44.8%	132 52.4%
CA LA Long Beach 3850101A	55 15.8%	285 81.9%	65 26.9%	174 71.9%
CA LA Santa Monica 6250005A	56 10.6%	465 87.7%	72 32.3%	149 66.8%
CA LA Topanga 710003A 6A	26 4.5%	535 93.5%	103 20.4%	383 76.0%
CA LA Lockhurst 9006489A 90A	151 25.8%	421 72.0%	254 43.9%	322 55.6%
CA LA Glendale 2550120A 120B 122A	106 16.7%	511 80.7%	287 28.3%	707 69.7%
CA LA Eagle Rock 9006334A 6335A	157 21.0%	575 76.8%	190 42.0%	252 55.8%
CA Alameda 280300 280700	43 12.9%	286 85.6%	52 25.1%	142 68.6%
MI Oakland West Bloomfield 2	134 33.7%	257 64.6%	299 50.1%	288 48.2%
MI Washtenaw Chelsea 1 2	195 27.0%	509 70.5%	557 47.4%	606 51.5%
NH Hillsborough Manchester 3	401 26.1%	1106 72.0%	612 40.4%	897 59.3%
NH Hillsborough Manchester 5	286 25.0%	832 72.8%	706 43.0%	929 56.6%
NH Hillsborough Nashua 5	746 41.4%	1022 56.7%	1753 50.3%	1719 49.3%
NH Hillsborough Wilton	416 36.7%	692 61.1%	610 52.3%	556 47.7%
NM Santa Fe 25 33 Alameda	51 15.6%	267 81.7%	40 37.7%	65 61.3%
OH Cuyahoga Independence A C F	307 53.5%	254 44.3%	706 59.7%	463 39.1%
OH Cuyahoga Rocky River 1D 1E 1G	223 46.8%	246 51.7%	453 57.4%	325 41.2%
PA Allegheny Pittsburgh W 14 D 8	45 14.4%	264 84.3%	62 32.3%	130 67.7%
PA Centre Harris Twp 56 57	490 37.1%	816 61.8%	889 54.4%	724 44.3%
TX Harris Houston 34	95 20.3%	355 75.7%	102 42.5%	128 53.3%

Note that in all 20 cases shown in Table 2 above, the disparities are in the same direction. Obama runs more strongly in the exit polls than in the official results. The disparities between Obama’s percentage among exit poll responders and his presumed percentage among non-responders averages 15.4%, ranging from 5.2% (at Independence, Ohio) to 22.4% (at Houston, Texas). But these disparities are not necessarily due to a corrupted official vote count. They could just as easily be due to an undersampling of Republican voters in the exit polls.

PARTY AFFILIATION

Adjusting exit poll data to account for party affiliation is not always possible. In Michigan, the questionnaire handed to the voters did not ask for their party affiliation, so we lack the relevant exit poll data with which to compare the voter rolls.

In Texas, voters do not register by party, so the voter rolls lack the relevant information with which to compare the exit poll data. In Ohio and New Mexico, election officials compile data at the precinct level for the party affiliation of all registered voters, but this does not reveal how many persons from each party actually voted.

In Ohio, available spreadsheets do show at the precinct level which persons from each party actually voted in November 2008, but the precinct boundaries have been changed since then, and thus the data are comparable only at the village level. But this still leaves us with 18 polling places in three states where the exit poll data can be properly adjusted to account for party affiliation.

In New Hampshire, the voter signature books and absentee voter lists are readily available for public inspection in the clerk’s offices of the individual cities and towns; the party affiliations can be gleaned and tallied from these, as we have done.

In Pennsylvania, the Secretary of State provided a massive database for each and every registered voter in the state, from which we have gleaned, quite tediously, all the desired information -- gender, age, party affiliation, whether or not the person voted, and whether at the polls or by absentee ballot.

For Los Angeles County, a private company provided us with the data for the party affiliations of voters at the polls at all 10 polling places where exit polls were conducted.

In New Hampshire, the voter signature books do not differentiate between third-party and independent voters, and in Pennsylvania our own databases do not make this distinction, so for these states the exit poll responders and the voters at the polls are divided into three categories only – Republicans, Democrats, and all others.

In California, where there are six official political parties, our questionnaires and databases listed them all, together with a category for no party affiliation, and so we are able to present the data in four categories – Republicans, Democrats, third parties, and unaffiliated voters. The comparison of exit poll responders and voters at the polls is shown below.

TABLE 3: PARTY AFFILIATION OF EXIT POLL RESPONDERS AND VOTERS AT POLLS

	Exit Poll Responders				Voters at Polls
	Rep.	Dem.	Other	None	Rep.	Dem.	Other	None

CA LA Taft	25.5%	54.8%	3.9%	15.8%	29.6%	48.6%	3.2%	18.6%
CA LA Long Beach	11.0%	64.9%	5.4%	18.7%	16.7%	60.8%	4.5%	18.1%
CA LA Berendo	10.2%	66.0%	2.8%	21.0%	13.4%	65.7%	3.3%	17.6%
CA LA Santa Monica	10.5%	66.7%	5.1%	17.8%	14.8%	59.2%	4.5%	21.5%
CA LA Topanga	3.6%	66.4%	10.4%	19.5%	9.0%	61.5%	7.1%	22.3%
CA LA Lockhurst	22.4%	53.1%	8.5%	15.9%	29.8%	48.1%	5.2%	17.0%
CA LA Glendale	13.3%	55.1%	6.6%	25.0%	20.5%	48.1%	4.5%	26.9%
CA LA Locke	2.2%	80.1%	2.7%	15.0%	4.6%	77.8%	3.2%	14.5%
CA LA Eagle Rock	14.3%	63.7%	5.8%	16.2%	22.3%	55.7%	3.8%	18.1%
CA LA Lynwood	4.7%	72.6%	3.1%	19.6%	8.1%	74.1%	3.2%	14.6%
NH Manchester 3	18.4%	42.9%	38.7%		21.9%	41.6%	36.5%
NH Manchester 5	17.5%	48.6%	33.9%		20.7%	46.0%	33.3%
NH Nashua 5	25.4%	35.6%	39.0%		27.2%	30.4%	42.4%
NH Wilton	25.0%	36.6%	38.4%		29.9%	31.2%	38.9%
PA Pittsburgh	10.1%	75.3%	14.6%		12.8%	73.5%	13.7%
PA Cambria Munster	42.8%	47.2%	10.0%		40.4%	52.7%	6.9%
PA Centre Harris Twp	39.5%	49.9%	10.6%		47.7%	40.7%	11.6%
PA Philadelphia	3.5%	80.2%	16.4%		6.7%	82.4%	10.9%

As shown in Table 3 above, Republican voters were undersampled in 17 of the 18 polling places where direct comparisons can be made. The differentials were as much as 8.2% in Centre County, Pennsylvania, and 8.0% in Eagle Rock, 7.4% at Lockhurst, 7.2% in Glendale, all in Los Angeles County, California.

It is obvious that some of the disparity between the exit poll data and the official results is due to undersampling of Republicans, and the exit poll data must be adjusted accordingly, as we have done.

The raw data are broken down into groups according to party affiliation, the voting patterns for each group are left unchanged, but the relative weight of each group is adjusted proportionately so that it matches their strength at the polls. (For example, in Wilton, New Hampshire, where 25.0% of the exit poll responders and 29.9% of voters at the polls were Republicans, the exit poll data for Republican voters are multiplied by almost 1.2, and the data for Democrats and Independents are adjusted by the appropriate ratios).

The calculations are set forth in full detail in the appendix; the methodology is explained on pages 8 and 9 of the accompanying paper concerning Propositions 4 and 8 in Los Angeles County [2]; and the results are summarized below.

TABLE 4: COMPARISON OF OFFICIAL RESULTS AND EXIT POLL DATA

ADJUSTED FOR PARTY AFFILIATION, PRESIDENTIAL ELECTION

	Official Results		Adjusted Exit Poll Data
	McCain	Obama	McCain	Obama

CA LA Taft 9001037A	201 36.0%	346 61.9%	97.6 31.8%	204.3 66.5%
CA LA Long Beach 3850101A	120 20.3%	459 77.8%	71.2 20.4%	269.6 77.4%
CA LA Berendo 9005399A	127 18.9%	527 78.4%	78.5 19.1%	325.3 79.0%
CA LA Santa Monica 6250005A	128 17.0%	614 81.5%	72.6 13.7%	445.9 84.3%
CA LA Topanga 710003A 6A	129 12.0%	918 85.3%	42.0 7.3%	521.3 91.1%
CA LA Lockhurst 9006489A 90A	405 34.8%	743 63.8%	175.2 30.0%	395.7 67.7%
CA LA Glendale 2550120A 120B 122A	393 23.9%	1218 74.0%	136.7 21.6%	478.2 75.7%
CA LA Locke 9001145A 9002566A	47 4.2%	1050 94.4%	31.3 4.3%	690.1 95.1%
CA LA Eagle Rock 9006334A 6335A	347 28.9%	827 68.9%	204.6 27.4%	524.4 70.2%
CA LA Lynwood 3990015A 16A 18A 19A	253 9.3%	2421 89.4%	138.3 10.0%	1229.6 89.2%
NH Hillsborough Manchester 3	1013 33.2%	2003 65.7%	437.1 28.4%	1070.6 69.7%
NH Hillsborough Manchester 5	992 35.6%	1761 63.2%	312.1 27.3%	805.1 70.5%
NH Hillsborough Nashua 5	2499 47.2%	2741 51.8%	790.5 43.7%	982.4 54.3%
NH Hillsborough Wilton	1026 44.6%	1248 54.3%	461.3 40.7%	647.1 57.1%
PA Allegheny Pittsburgh W 14 D 8	107 21.1%	394 78.0%	51.1 16.3%	258.1 82.4%
PA Cambria Munster	239 71.6%	89 26.6%	185.6 69.9%	70.9 26.7%
PA Centre Harris Twp 56 57	1379 46.7%	1540 52.1%	570.0 43.2%	732.7 55.6%
PA Philadelphia D 30 W 5 10	21 3.8%	535 95.9%	13.2 2.9%	434.4 96.5%

As shown in Table 4 above, the exit poll data when properly adjusted to account for party affiliation are within 2% of the official results (and thus the margins between the candidates are within 4%) in 6 of 10 polling places in California, and in 2 of 4 polling places in Pennsylvania. Three of these polling places were listed in Table 2 as having large disparities between the unadjusted exit poll data and the official results, the differentials in the margins being 8.6% in Long Beach, 13.9% in Glendale, and 15.8% in Eagle Rock.

But, as shown in Table 4, most of the disparity in all three cases is explained by undersampling of Republican voters, the disparities between the adjusted exit poll data and the official results being 0.5% in Long Beach, 4.0% in Glendale, and 2.8% in Eagle Rock).

But this still leaves us with 10 polling places (four in Los Angeles County, two in Pennsylvania, and four in New Hampshire) where the exit poll data, even when adjusted to account for party affiliation, differ from the official results by 3% or more, which amounts to a disparity of 6% or more in the margin of victory or defeat.

The highest of these disparities is found at Ward 5 in Manchester, New Hampshire, where Obama defeated McCain by 43.2% in the adjusted exit poll and by 27.6% in the official results. This 15.6% disparity is the amount over and above that which can be explained by undersampling of Republican voters.

For Ohio, for the precincts existent at the time the exit polls were conducted, we do not know how many from each party actually voted, but we do know the number of total registered voters from each party in these precincts. The comparison with the party affiliation of the exit poll responders is shown below.

TABLE 5: PARTY AFFILIATION OF REGISTERED VOTERS AND EXIT POLL RESPONDERS

Total Registered Voters	Rep.	Dem.	Ind.

OH Cuyahoga Beachwood A B N	137 7.0%	934 47.8%	882 45.2%
OH Cuyahoga Independence A C F	462 19.0%	1013 41.7%	954 39.3%
OH Cuyahoga Rocky River 1D 1E 1G	488 27.1%	624 34.6%	689 38.3%

Exit Poll Responders	Rep.	Dem.	Ind.

OH Cuyahoga Beachwood A B N	91 15.8%	317 55.0%	168 29.2%
OH Cuyahoga Independence A C F	186 32.1%	261 45.1%	132 22.8%
OH Cuyahoga Rocky River 1D 1E 1G	190 39.4%	196 40.7%	96 19.9%

One reason why these numbers do not come close to matching is that in Ohio, many voters who consider themselves Republican or Democratic actually register as Independent because this allows them to vote in the primary election of either party. Therefore the data would not be sufficiently comparable to allow any exit poll to be adjusted according to party affiliation even if we did know the party registrations of those who actually voted at the polls. But the data shown in Table 5 do indicate that we almost certainly did not undersample Republicans.

For New Mexico also, we know the number of total registered voters from each party in the precincts at which the exit polls were conducted, but do not know how many from each party actually voted. The comparison with the party affiliation of the exit poll responders is shown below. The pattern is similar to Ohio, with a greater percentage of exit poll responders than of total registered voters identifying their political party. The data indicate that we did not undersample Republicans or Democrats. If anything, we undersampled Independents.

TABLE 6: PARTY AFFILIATION OF REGISTERED VOTERS AND EXIT POLL RESPONDERS

Total Registered Voters	Rep.	Dem.	Other / None

NM Santa Fe 7 Rio En Medio	79 13.6%	365 63.0%	135 23.3%
NM Santa Fe 25 33 Alameda	141 7.1%	1384 69.9%	456 23.0%
NM Taos Pueblo 13	43 5.6%	578 74.8%	152 19.7%

Exit Poll Responders	Rep.	Dem.	Other / None

NM Santa Fe 7 Rio En Medio	25 16.9%	101 68.2%	22 14.9%
NM Santa Fe 25 33 Alameda	37 11.0%	253 75.5%	45 13.4%
NM Taos Pueblo 13	13 5.6%	196 84.1%	24 10.3%

CONGRESSIONAL RACES

All the questionnaires handed to the voters at all polling places asked how they voted in their Congressional election. This was done not only to verify (or question) the official results for Congress, but also as a comparison, to provide a check on the accuracy of the official results for President.

In 6 of the 28 polling places included in this paper, all in Los Angeles County, California, the incumbent ran unopposed for reelection to Congress, so those races are not listed here.

The other 22 polling places had contested Congressional elections, generally with Republican and Democratic candidates. At three polling places there was no Republican candidate; Green or Independent candidates were the opposition. These are noted at the bottom of Table 7, which compares the unadjusted exit poll data with the official results for the Congressional races.

As seen in Table 7, the Democratic candidate for Congress ran more strongly in the unadjusted exit poll than in the official results at 27 of 28 polling places. The lone exception was Munster Township in Cambria County, Pennsylvania, where the same pattern was seen in the presidential contest. At seven other polling places (three in Los Angeles County, one in New Mexico, one in Ohio, and two more in Pennsylvania), the difference between the unadjusted exit poll data and the official results was 2.1% or less.

TABLE 7: COMPARISON OF UNADJUSTED EXIT POLL DATA

AND OFFICIAL RESULTS FOR CONGRESSIONAL RACES

	Official Results U.S. Congress		Exit Poll U.S. Congress
	Republican *	Democratic	Republican *	Democratic

CA LA Long Beach 3850101A *	104 19.2%	439 80.8%	55 17.7%	255 82.3%
CA LA Glendale 2550120A 120B 122A	295 20.1%	1079 73.5%	74 13.9%	435 81.5%
CA LA Locke 9001145A 9002566A *	93 10.7%	775 89.3%	59 8.9%	602 91.1%
CA LA Lynwood 3990015A 16A 18A 19A	256 10.2%	2264 89.8%	103 8.1%	1170 91.9%
CA Alameda 280300 280700	128 25.0%	355 69.5%	56 19.2%	225 77.1%
MI Oakland West Bloomfield 2	440 45.4%	466 48.1%	143 40.6%	191 54.3%
MI Washtenaw Chelsea 1 2	745 40.6%	993 54.1%	186 28.3%	437 66.5%
NH Hillsborough Manchester 3	996 34.5%	1809 62.7%	432 29.3%	972 65.9%
NH Hillsborough Manchester 5	925 34.9%	1632 61.6%	294 27.0%	745 68.3%
NH Hillsborough Nashua 5	2239 45.3%	2580 52.2%	667 39.7%	943 56.1%
NH Hillsborough Wilton	931 42.8%	1203 55.3%	371 35.5%	651 62.2%
NM Santa Fe 7 Rio En Medio	20 10.5%	128 67.4%	10 7.5%	99 74.4%
NM Santa Fe 25 33 Alameda	50 11.7%	308 72.1%	33 10.4%	251 78.9%
NM Taos Pueblo 13	8 2.7%	267 88.7%	4 1.9%	195 90.3%
OH Cuyahoga Beachwood A B N	345 30.6%	777 69.0%	123 28.0%	304 69.1%
OH Cuyahoga Independence A C F	1025 60.9%	622 36.9%	300 56.1%	224 41.9%
OH Cuyahoga Rocky River 1D 1E 1G	719 59.0%	477 39.1%	249 54.1%	196 42.6%
PA Allegheny Pittsburgh W 14 D 8 *	54 11.7%	403 87.2%	27 9.5%	253 89.0%
PA Cambria Munster	193 58.3%	138 41.7%	155 59.6%	105 40.4%
PA Centre Harris Twp 56 57	1616 55.6%	1229 42.3%	577 45.0%	684 53.3%
PA Philadelphia D 30 W 5 10	44 8.8%	457 91.2%	31 8.1%	351 91.6%
TX Harris Houston 34	386 21.8%	1345 75.8%	72 15.7%	380 82.8%

* At three of the polling places listed above there was no Republican candidate for Congress. Instead, vote totals for the Independent candidate are listed at Locke and Long Beach, and for the Green Party candidate at Pittsburgh. In Santa Fe and Taos, New Mexico, the Green Party candidate ran second in a three-way race, besting the Republican candidate.

Bear in mind, however, that a difference of 2% in the columns of both candidates affects the margin, or “point spread,” by 4%. Third-party candidates can also affect the margin between the top two candidates.

It is more precise, therefore, to compare the disparities between the margins in the exit poll with the margins in the official results. The difference between them is sometimes referred to as “within precinct disparity,” although the term is not precise because some of our polling places had multiple precincts.

The disparities between the margins, or “point spreads,” in the unadjusted exit poll data and the official results for the presidential and Congressional races, culled from Table 1 and Table 7, are compared in Table 8 below.

TABLE 8: COMPARISON OF MARGINS OF VICTORY FOR PRESIDENT AND CONGRESS

	Official Results		Exit Poll		Disparity
	President	Congress	President	Congress	President	Congress

CA LA Long Beach 101A	57.5%	61.6%	66.1%	64.6%	8.6%	3.0%
CA LA Glendale 120A 120B 122A	50.1%	53.4%	64.0%	67.6%	13.9%	14.2%
CA LA Locke 1145A 2566A	90.2%	78.6%	92.8%	82.2%	2.6%	3.6%
CA LA Lynwood 15A 16A 18A 19A	80.1%	79.6%	84.0%	83.8%	3.9%	4.2%
CA Alameda 280300 280700	61.5%	44.5%	73.1%	57.9%	11.6%	13.4%
MI Oakland West Bloomfield 2	11.2%	2.7%	30.9%	13.7%	19.7%	11.0%
MI Washtenaw Chelsea 1 2	19.1%	13.5%	43.5%	38.2%	24.4%	24.7%
NH Hillsborough Manchester 3	32.5%	28.2%	45.9%	36.6%	13.4%	8.4%
NH Hillsborough Manchester 5	27.6%	26.7%	47.8%	41.3%	20.2%	14.6%
NH Hillsborough Nashua 5	4.6%	6.9%	15.3%	16.4%	10.7%	9.5%
NH Hillsborough Wilton	9.7%	12.5%	24.4%	26.7%	14.7%	14.2%
NM Santa Fe 7 Rio En Medio	69.8%	56.9%	70.6%	66.9%	0.8%	10.0%
NM Santa Fe 25 33 Alameda	55.7%	60.4%	66.1%	68.5%	10.4%	7.9%
NM Taos Pueblo 13	91.4%	86.0%	93.2%	88.4%	1.8%	2.4%
OH Cuyahoga Beachwood A B N	18.1%	38.4%	18.0%	41.1%	-0.1%	2.7%
OH Cuyahoga Independence A C F	-16.9%	-24.0%	-9.2%	-14.2%	7.7%	9.8%
OH Cuyahoga Rocky River 1D 1E 1G	-8.3%	-19.9%	4.9%	-11.5%	13.2%	8.4%
PA Allegheny Pittsburgh W 14 D 8	56.9%	75.5%	69.9%	79.5%	13.0%	4.0%
PA Cambria Munster	-45.0%	-16.6%	-46.0%	-19.2%	-1.0%	-2.6%
PA Centre Harris Twp 56 57	5.4%	-13.3%	24.7%	8.3%	19.3%	21.6%
PA Philadelphia D 30 W 5 10	92.1%	82.4%	96.0%	83.5%	3.9%	1.1%
TX Harris Houston 34	40.4%	54.0%	55.4%	67.1%	15.0%	13.1%

In this case it is perfectly appropriate to use unadjusted exit poll data, because we are comparing the same thing, apples to apples, for both the presidential and Congressional races. These are the same voters, on the same day, at the same polling places. Thus the comparison is valid even for states where it is not possible to adjust the exit poll data to account for party affiliation.

At any single polling place, the unadjusted exit poll data should differ from the official results by about the same amount in all contested partisan races. In 15 of 22 polling places we find that this is true, as shown in Table 8 above. But at 6 of 22 polling places (one in California, one in Michigan, two in New Hampshire, one in Ohio, and one in Pennsylvania) the disparity in the margin for President exceeds the disparity in the margin for Congress by 4.0% or more. The differentials are as high as 9.0% at Pittsburgh, Pennsylvania and 8.7% at West Bloomfield, Michigan. At one polling place the reverse is true. At Rio En Medio in Santa Fe County, New Mexico, the disparity in the margin for Congress exceeds the disparity in the margin for President by 9.2.

As with the presidential election, the simplest analysis is to determine, by subtraction, what the vote count must have been among those who declined to participate in the exit poll, assuming that the official results are true and correct.

The comparison is shown in Table 9 below. At 21 of 22 polling places, the Democratic candidate runs more strongly in the exit polls than in the official results. The disparities between their percentage among exit poll responders and their presumed percentage among non-responders average 10.2%, ranging as high as 26.5% and 26.6% at Rio En Medio and Alameda, Santa Fe County, New Mexico. Again, these disparities could be due to an undersampling of Republican voters.

TABLE 9: VOTE COUNT AMONG NON-RESPONDERS, ASSUMING

OFFICIAL RESULTS FOR CONGRESSIONAL RACES ARE TRUE AND CORRECT

	Exit Poll Responders		Non-Responders
	Republican *	Democratic	Republican *	Democratic

CA LA Long Beach 101A *	55 17.7%	255 82.3%	49 21.0%	184 79.0%
CA LA Glendale 120A 120B 122A	74 13.9%	435 81.5%	221 23.7%	644 68.9%
CA LA Locke 1145A 2566A *	59 8.9%	602 91.1%	34 16.4%	173 83.6%
CA LA Lynwood 15A 16A 18A 19A	103 8.1%	1170 91.9%	153 12.3%	1094 87.7%
CA Alameda 280300 280700	56 19.2%	225 77.1%	72 32.9%	130 59.4%
MI Oakland West Bloomfield 2	143 40.6%	191 54.3%	297 51.3%	275 47.5%
MI Washtenaw Chelsea 1 2	186 28.3%	437 66.5%	559 50.1%	556 49.9%
NH Hillsborough Manchester 3	432 29.3%	972 65.9%	564 40.0%	837 59.4%
NH Hillsborough Manchester 5	294 27.0%	745 68.3%	631 40.4%	887 56.8%
NH Hillsborough Nashua 5	667 39.7%	943 56.1%	1572 48.1%	1637 50.1%
NH Hillsborough Wilton	371 35.5%	651 62.2%	560 49.6%	552 48.9%
NM Santa Fe 7 Rio En Medio	10 7.5%	99 77.4%	10 17.5%	29 50.9%
NM Santa Fe 25 33 Alameda	33 10.4%	251 78.9%	17 15.6%	57 52.3%
NM Taos Pueblo 13	4 1.9%	195 90.3%	4 4.7%	72 84.7%
OH Cuyahoga Beachwood A B N	123 28.0%	304 69.1%	222 31.9%	473 68.1%
OH Cuyahoga Independence A C F	300 56.1%	224 41.9%	725 63.1%	398 34.6%
OH Cuyahoga Rocky River 1D 1E 1G	249 54.1%	196 42.6%	470 61.9%	281 37.0%
PA Allegheny Pittsburgh W 14 D 8 *	27 9.5%	253 89.0%	27 15.2%	150 84.3%
PA Cambria Munster	155 59.6%	105 40.4%	38 53.5%	33 46.5%
PA Centre Harris Twp 56 57	577 45.0%	684 53.3%	1039 64.1%	545 33.6%
PA Philadelphia D 30 W 5 10	31 8.1%	351 91.6%	13 10.9%	106 89.1%
TX Harris Houston 34	72 15.7%	380 82.3%	314 23.9%	965 73.4%

*See footnote for Table 7.

For 12 of the 22 polling places with contested Congressional elections (in Los Angeles County, New Hampshire and Pennsylvania), we have the data necessary to adjust the exit poll data to account for party affiliation. The calculations are set forth in full detail in the appendix. The adjusted data are compared to the official results in Table 10 below.

At 6 of 12 polling places, the disparity between the margins in the adjusted exit poll data and the official results is 3.2% or less. But at the other six polling places (one in California, one in Pennsylvania, and all four in New Hampshire), the disparity is more than 4%, ranging as high as 10.4% at Ward 5 in Manchester, New Hampshire. Again, this 10.4% disparity is the amount over and above that which can be explained by undersampling of Republican voters.

TABLE 10: COMPARISON OF OFFICIAL RESULTS AND EXIT POLL DATA

ADJUSTED FOR PARTY AFFILIATION, CONGRESSIONAL RACES

	Official Results U.S. Congress		Adjusted Exit Poll Data
	Republican *	Democratic	Republican *	Democratic

CA LA Long Beach 3850101A *	104 19.2%	439 80.8%	63.5 20.5%	246.2 79.5%
CA LA Glendale 2550120A 120B 122A	295 20.1%	1079 73.5%	95.2 18.0%	407.5 77.0%
CA LA Locke 9001145A 9002566A *	93 10.7%	775 89.3%	63.5 9.6%	598.6 90.4%
CA LA Lynwood 3990015A 16A 18A 19A	256 10.2%	2264 89.8%	126.7 9.9%	1154.7 90.1%
NH Hillsborough Manchester 3	996 34.5%	1809 62.7%	463.9 31.4%	943.7 63.8%
NH Hillsborough Manchester 5	925 34.9%	1632 61.6%	317.8 29.1%	722.1 66.2%
NH Hillsborough Nashua 5	2239 45.3%	2580 52.2%	704.8 41.9%	906.8 53.9%
NH Hillsborough Wilton	931 42.8%	1203 55.3%	408.7 39.2%	610.7 58.6%
PA Allegheny Pittsburgh W 14 D 8 *	54 11.7%	403 87.2%	27.7 9.8%	250.8 88.5%
PA Cambria Munster	193 58.3%	138 41.7%	153.9 57.2%	107.2 39.9%
PA Centre Harris Twp 56 57	1616 55.6%	1229 42.3%	660.4 51.4%	603.1 47.0%
PA Philadelphia D 30 W 5 10	44 8.8%	457 91.2%	36.4 9.3%	354.6 90.5%

*See footnote for Table 7.

UNITED STATES SENATE RACES

In four of the states where our exit polls were conducted (Michigan, New Hampshire, New Mexico and Texas) there were contests for the United States Senate. A comparison of the official results with the unadjusted exit poll data is shown in Table 11 below. As in the presidential election, the Democratic candidates carried all ten polling places. However, at 9 of 10 polling places the Republican candidate fares better in the official results than in the exit poll. The lone exception is at Taos Pueblo in New Mexico, where Tom Udall received 95.2% in the exit poll and 96.0% in the official results.

TABLE 11: COMPARISON OF UNADJUSTED EXIT POLL DATA

AND OFFICIAL RESULTS FOR UNITED STATES SENATE RACES

	Official Results U.S. Senate		Exit Poll U.S. Senate
	Republican	Democratic	Republican	Democratic

MI Oakland West Bloomfield 2	338 35.0%	573 59.4%	107 28.9%	244 65.9%
MI Washtenaw Chelsea 1 2	629 34.0%	1152 62.2%	152 23.3%	468 71.7%
NH Hillsborough Manchester 3	1078 36.3%	1794 60.4%	484 31.7%	975 63.8%
NH Hillsborough Manchester 5	993 36.1%	1650 60.0%	315 27.7%	762 67.1%
NH Hillsborough Nashua 5	2337 45.2%	2639 51.0%	722 40.4%	983 55.0%
NH Hillsborough Wilton	1010 45.2%	1148 51.4%	401 36.6%	650 59.4%
NM Santa Fe 7 Rio En Medio	28 14.7%	163 85.3%	19 13.1%	125 86.2%
NM Santa Fe 25 33 Alameda	76 17.6%	357 82.4%	44 13.5%	279 85.8%
NM Taos Pueblo 13	12 4.0%	290 96.0%	10 4.4%	217 95.2%
TX Harris Houston 34	426 23.9%	1296 72.8%	94 20.3%	360 77.9%

In 8 of the 10 polling places shown above, the disparity between the margins of victory in the exit poll and the official results is greater for the presidential election (ref. Table 1) than for the Senate race (ref. Table 11). The comparison is shown in Table 12 below.

In New Mexico, at Taos Pueblo and at Rio En Medio, both disparities are small, and at Alameda in Santa Fe, it is possible that the disparities are due to questionnaires being lost or mixed up among precincts, as happened elsewhere in Santa Fe County.

But in Michigan and Texas (as in New Hampshire) both disparities are large, and the difference between them is 7.1% at West Bloomfield in Michigan, suggesting that even if the exit poll data could be properly adjusted to account for party affiliation, a significant disparity might still exist for the presidential election.

TABLE 12: COMPARISON OF MARGINS OF VICTORY FOR PRESIDENT AND SENATE

	Official Results		Exit Poll		Disparity
	President	Senate	President	Senate	President	Senate

MI Oakland West Bloomfield 2	11.2%	24.4%	30.9%	37.0%	19.7%	12.6%
MI Washtenaw Chelsea 1 2	19.1%	28.2%	43.5%	48.4%	24.4%	20.2%
NH Hillsborough Manchester 3	32.5%	24.1%	45.9%	32.1%	13.4%	8.0%
NH Hillsborough Manchester 5	27.6%	23.9%	47.8%	39.4%	20.2%	15.5%
NH Hillsborough Nashua 5	4.6%	5.8%	15.3%	14.6%	10.7%	8.8%
NH Hillsborough Wilton	9.7%	6.2%	24.4%	22.8%	14.7%	16.6%
NM Santa Fe 7 Rio En Medio	69.8%	70.6%	70.6%	73.1%	0.8%	2.5%
NM Santa Fe 25 33 Alameda	55.7%	64.8%	66.1%	72.3%	10.4%	7.7%
NM Taos Pueblo 13	91.4%	92.0%	93.2%	90.8%	1.8%	- 1.2%
TX Harris Houston 34	40.4%	48.9%	55.4%	57.6%	15.0%	12.2%

As with the presidential and Congressional elections, the simplest analysis is to determine, by subtraction, what the vote count must have been among those who declined to participate in the exit poll, assuming that the official results are true and correct. The comparison is shown in Table 13 below. At 9 of 10 polling places, the Democratic candidate runs more strongly in the exit polls than in the official results. The disparities between their percentage among exit poll responders and their presumed percentage among non-responders average 10.4%, ranging as high as 15.7% at Wilton, New Hampshire. Again, these disparities could be due to an undersampling of Republican voters.

TABLE 13: VOTE COUNT AMONG NON-RESPONDERS, ASSUMING

OFFICIAL RESULTS FOR SENATE RACES ARE TRUE AND CORRECT

	Exit Poll Responders		Non-Responders
	Republican	Democratic	Republican	Democratic

MI Oakland West Bloomfield 2	107 28.9%	244 65.9%	231 38.8%	329 55.3%
MI Washtenaw Chelsea 1 2	152 23.3%	468 71.7%	477 39.8%	684 57.1%
NH Hillsborough Manchester 3	484 31.7%	975 63.8%	594 41.3%	819 56.9%
NH Hillsborough Manchester 5	315 27.7%	762 67.1%	678 42.1%	888 55.1%
NH Hillsborough Nashua 5	722 40.4%	983 55.0%	1615 47.7%	1656 48.9%
NH Hillsborough Wilton	401 36.6%	650 59.4%	609 53.4%	498 43.7%
NM Santa Fe 7 Rio En Medio	19 13.1%	125 86.2%	9 19.1%	38 80.9%
NM Santa Fe 25 33 Alameda	44 13.5%	279 85.8%	32 29.1%	78 70.9%
NM Taos Pueblo 13	10 4.4%	217 95.2%	2 2.7%	73 97.3%
TX Harris Houston 34	94 20.3%	360 77.9%	332 25.2%	936 71.0%

Unfortunately, we lack the data for party affiliation of voters at the polls in Michigan, New Mexico and Texas. New Hampshire is the only state with a United States Senate contest where we are able to adjust the exit poll data to account for party affiliation. The calculations are set forth in full detail in the appendix. The adjusted data are compared to the official results in Table 14 below.

At all four polling places in New Hampshire, undersampling of Republican voters in the exit polls accounts for some, but not all, of the disparities shown in Table 12 above. The disparity is reduced from 8.0% to 4.0% at Manchester 3, from 15.5% to 11.0% at Manchester 5, from 8.8% to 4.2% at Nashua 5, and from 16.6% to 8.6% at Wilton. Again, these disparities are the amounts over and above that which can be explained by undersampling of Republican voters.

TABLE 14: COMPARISON OF OFFICIAL RESULTS AND EXIT POLL DATA

ADJUSTED FOR PARTY AFFILIATION, UNITED STATES SENATE RACES

	Official Results U.S. Senate		Exit Poll U.S. Senate
	Republican	Democratic	Republican	Democratic

NH Hillsborough Manchester 3	1078 36.3%	1794 60.4%	516.5 33.7%	946.2 61.8%
NH Hillsborough Manchester 5	993 36.1%	1650 60.0%	340.4 30.0%	737.2 64.9%
NH Hillsborough Nashua 5	2337 45.2%	2639 51.0%	765.4 42.7%	943.6 52.7%
NH Hillsborough Wilton	1010 45.2%	1148 51.4%	444.7 40.6%	606.3 55.4%

SUMMARY and REVIEW

For 28 polling places in 7 states, exit poll data have been compared to official results for the presidential election and for all contested Senate and Congressional races. For 18 of the 28 polling places, exit poll data have been adjusted to account for party affiliation. For the other 10 polling places, we lack the data to make such adjustments possible.

In 9 of the 28 polling places, no significant disparities were found between the official results and the exit poll data, whether adjusted or unadjusted, for any of the 17 races analyzed. This vouches for the accuracy of our exit polls, and there is no need to study these polling places further. In the other 19 polling places, significant disparities exist between the official results and the exit poll data in at least one election contest, and sometimes two or three.

At 5 of 10 polling places in Los Angeles County, no significant disparities were found in any of the 8 races analyzed. All were less than 3%, ranging from 0.4% to 2.8%. At the other 5 polling places in Los Angeles County, adjusted exit poll data showed disparities ranging from 4.0% to 10.5% in the presidential election. There was no Senate contest in California, and in 4 of these polling places there was no contested Congressional race. At Glendale, where there was a contested Congressional race, the disparity was 5.6%. (See Table 15) Data from all of these polling places have been exhaustively analyzed in the accompanying paper concerning Propositions 4 and 8 in Los Angeles County.

At 2 of 4 polling places in Pennsylvania, where there also was no Senate contest, no significant disparities were found in the presidential or Congressional elections. The disparities ranged from 0.7% to 1.8%. At the other 2 polling places in Pennsylvania, adjusted exit poll data showed disparities in both the presidential and Congressional contests. At Harris Township in Centre County, the disparities were 7.0% and 8.9%, respectively. At Pittsburgh, the disparity in the presidential election was 9.2% and the disparity in the Congressional race was 3.2%. Both polling places will be investigated further. (See Table 15)

At all 4 polling places in New Hampshire, where there was a Senate contest and two contested Congressional races, disparities were found in all 12 cases, and most of them were significant. The disparities, based on exit poll data adjusted to account for party affiliation, ranged from 4.2% to 10.4% in the Congressional races, from 4.0% to 11.0% in the Senate race, and from 6.0% to 15.6% in the presidential election (Table 15). All four of these polling places warrant further investigation.

The disparities found among exit poll data adjusted to account for party affiliation are summarized in Table 15 below, with disparities of less than 3% color coded in blue, as are the names of the 7 polling places where no significant disparities were found.

TABLE 15: DISPARITIES BETWEEN MARGINS OF VICTORY

IN OFFICIAL RESULTS AND EXIT POLL DATA, ADJUSTED

	President	Senate	Congress

CA LA Taft 9001037A	8.8%
CA LA Long Beach 3850101A	0.5%		2.6%
CA LA Berendo 9005399A	0.4%
CA LA Santa Monica 6250005A	6.1%
CA LA Topanga 710003A 6A	10.5%
CA LA Lockhurst 9006489A 90A	8.7%
CA LA Glendale 2550120A 120B 122A	4.0%		5.6%
CA LA Locke 9001145A 9002566A	0.6%		2.2%
CA LA Eagle Rock 9006334A 6335A	2.8%
CA LA Lynwood 3990015A 16A 18A 19A	0.9%		0.6%
NH Hillsborough Manchester 3	8.8%	3.6%	4.2%
NH Hillsborough Manchester 5	15.6%	11.0%	10.4%
NH Hillsborough Nashua 5	6.0%	4.2%	5.1%
NH Hillsborough Wilton	6.7%	8.6%	6.9%
PA Allegheny Pittsburgh W 14 D 8	9.2%		3.2%
PA Cambria Munster	1.8%		0.7%
PA Centre Harris Twp 56 57	7.0%		8.9%
PA Philadelphia D 30 W 5 10	1.5%		1.2%

In 4 states we lack the data to adjust for party affiliation. Of these, three (Michigan, New Mexico and Texas) had Senate contests, and one (Ohio) did not. All 9 polling places in these states had contested Congressional races.

We also lack the party affiliation data for Alameda, California, where there was a contested Congressional race but no Senate contest.

At Taos Pueblo, New Mexico and at Beachwood, Ohio, no significant disparities were found in any of the 5 races analyzed. Even with unadjusted exit poll data, the disparities range from 0.1% to 2.7%. At Rio En Medio in Santa Fe County, New Mexico, the disparities in the presidential and Senate races were only 0.8% and 2.5%, but the disparity in the Congressional contest was 10.0%, a significant differential which warrants special analysis. (See Table 16)

At the other 7 polling places, significant disparities were found between the official results and the unadjusted exit poll data in all 18 races analyzed. The disparities range from 7.5% to 24.7%. At West Bloomfield, Michigan, at Rocky River, Ohio, and at Houston, Texas, there were significant differentials among the disparities, always highest in the presidential election (Table 16).

All of these polling places warrant further investigation, a task complicated by the fact that we cannot know for certain to what extent these disparities are due to undersampling of Republican voters. But there are other analytical techniques available even when there are no exit poll data whatsoever, and these will be utilized in the second part of this paper.

The disparities found among the unadjusted exit poll data are summarized in Table 16 below, with disparities of less than 3% color coded in blue, as are the names of the 2 polling places where no significant disparities were found.

TABLE 16: DISPARITIES BETWEEN MARGINS OF VICTORY

IN OFFICIAL RESULTS AND EXIT POLL DATA, UNADJUSTED

	President	Senate	Congress

CA Alameda 280300 280700	11.6%		13.4%
MI Oakland West Bloomfield 2	19.7%	12.6%	11.0%
MI Washtenaw Chelsea 1 2	24.4%	20.2%	24.7%
NM Santa Fe 7 Rio En Medio	0.8%	2.5%	10.0%
NM Santa Fe 25 33 Alameda	10.4%	7.5%	8.1%
NM Taos Pueblo 13	1.8%	1.2%	2.4%
OH Cuyahoga Beachwood A B N	0.1%		2.7%
OH Cuyahoga Independence A C F	7.7%		9.8%
OH Cuyahoga Rocky River 1D 1E 1G	13.2%		8.4%
TX Harris Houston 34	15.0%	8.7%	13.1%

COMPARISON WITH PAST ELECTION RESULTS IN MICHIGAN

In Michigan, where we have no data on the party affiliation of the exit poll responders, the credibility of the official 2008 election results is established by comparison with previous elections, as shown in Table 17 below. Cases where one of the candidates ran for the same office in consecutive elections are shown in blue. In one case there was a rematch between the same two candidates.

At West Bloomfield, where Barack Obama received 64.6% in the exit poll and only 54.7% in the official results (ref. Table 1), this was still 5.0% better than John Kerry, who received only 49.7% of the vote in the 2004 presidential election.

Similarly, in the Senate race, Carl Levin received 65.9% in the exit poll and only 59.4% in the official results (ref. Table 11), but this was very much in line with the 59.1% of the vote that Levin received in the 2002 Senate race.

In the Congressional election, the Democratic candidate received 54.3% in the exit poll and only 48.1% in the official results (ref. Table 7), but again, this was very much in line with the 48.7% of the vote that the Democratic candidate received in the 2006 Congressional election. I must conclude that the raw exit poll data are not reflective of the electorate, and that the disparities shown in Table 16 above are due primarily to an undersampling of Republican voters, which illustrates the importance of asking exit poll responders to indicate their party affiliation.

At Chelsea City, where Barack Obama received 70.5% in the exit poll and only 58.7% in the official results (ref. Table 1), this was still 4.5% better than John Kerry, who received only 54.2% of the vote in the 2004 presidential election.

In the Congressional election, the Democratic candidate received 66.5% in the exit poll and only 54.1% in the official results (ref. Table 7), but this was only 2.1% less than the Democratic candidate received in 2006 in her second try for the office, and the Republican candidate received almost exactly the same percentage as in 2006.

In the Senate race, Carl Levin received 71.7% in the exit poll and only 62.2% in the official results (ref. Table 11). Chelsea was not incorporated as a city until after the 2002 election, so no direct comparison can be made, but in 2002 most of what is now Chelsea City voted in Sylvan Township, where Carl Levin defeated the Republican candidate by 1114 (59.3%) to 719 (38.3%), which is 2.9% less than Levin received in Chelsea City in 2008. As in West Bloomfield, I must conclude that the raw exit poll data are not reflective of the electorate, and that Republican voters were undersampled in the exit poll.

TABLE 17: COMPARISON OF 2008 OFFICIAL RESULTS WITH PAST ELECTIONS, MICHIGAN

	President		U. S. Senate		U. S. Congress
MI West Bloomfield 2	Republican	Democrat	Republican	Democrat	Republican	Democrat

2008	433 43.5%	545 54.7%	338 35.0%	573 59.4%	440 45.4%	466 48.1%
2006					710 48.8%	709 48.7%
2004	654 49.5%	656 49.7%			747 59.6%	483 38.5%
2002			379 38.8%	577 59.1%	550 56.6%	399 41.0%

	President		U. S. Senate		U. S. Congress
MI Chelsea 1 2	Republican	Democrat	Republican	Democrat	Republican	Democrat

2008	752 39.6%	1115 58.7%	629 34.0%	1152 62.2%	745 40.6%	993 54.1%
2006					885 40.5%	1227 56.2%
2004	846 44.4%	1032 54.2%			943 53.1%	757 42.6%
2002			N.A.	N.A.	N.A.	N.A.

COMPARISON WITH PAST ELECTION RESULTS IN OHIO

In Ohio, where we have party affiliation data for exit poll responders and for total registered voters, but not for voters at the polls, the credibility of the official 2008 election results is established by our highly accurate exit poll at Beachwood, and by comparison with previous elections at Independence and Rocky River, as shown in Table 18 below. Absentee ballots are included for 2008 and 2006 only. Prior to this, absentee ballots in Cuyahoga County were counted on an at-large countywide basis, so the results are not available at the precinct level.

At Independence, Barack Obama received 44.3% in the exit poll and only 41.2% in the official results (40.8% excluding absentee ballots) (ref. Table 1). In 2004, John Kerry received in the official results the exact same percentage that Obama received in the exit poll. However, at Rocky River, where Obama received 51.7% in the exit poll and only 45.2% in the official results (45.1% excluding absentee ballots) (ref. Table 1), Kerry in 2004 received only 39.2% of the vote, so the exit poll numbers do seem less credible than the official results.

The Congressional race provides a more interesting comparison. At Independence, Dennis Kucinich received 41.9% in the exit poll and only 37.1% in the official results (36.9% excluding absentee ballots) (ref. Table 7). But this was far below his official percentage in previous elections (69.4% in 2002, 53.5% in 2004, and 59.9% in 2006). At Rocky River, Kucinich received 42.6% in the exit poll and only 39.2% in the official results (39.1% excluding absentee ballots) (ref. Table 7). This was far below his official percentage in two of three previous elections (61.7% in 2002, 40.5% in 2004, and 49.7% in 2006). Our exit polls clearly picked up on this trend, even if they did not capture its full extent.

TABLE 18: COMPARISON OF 2008 OFFICIAL RESULTS WITH PAST ELECTIONS, OHIO

	President		U. S. Congress
OH Independence A C F	Republican	Democrat	Republican	Democrat

2008	1044 57.2%	752 41.2%	1060 60.7%	648 37.1%
2006			556 40.1%	829 59.9%
2004	918 55.6%	732 44.3%	600 39.4%	814 53.5%
2002			343 29.0%	820 69.4%

	President		U. S. Congress
OH Rocky River 1D 1E 1G	Republican		Democrat
			Republican	Democrat
2008	705 53.4%	597 45.2%
2006			746 58.7%	498 39.2%
2004	643 60.8%	415 39.2%	512 50.3%	505 49.7%
2002			583 56.2%	420 40.5%

If our exit polls in Independence and Rocky River undersampled Republican voters, this could account for the disparity between the raw data and the official results (9.8% at Independence, 8.4% at Rocky River) (ref. Tables 8 and 16). Jim Trakas, the Republican candidate, received overwhelming support among exit poll responders who identified themselves as Republicans, winning 84.6% at Independence and 88.1% at Rocky River, as shown in Table 19 below.

But Trakas also had strong support among exit poll responders who identified themselves as Independents (61.1% at Independence, 47.2% at Rocky River), and had substantial support among those who identified themselves as Democrats (33.6% at Independence, 23.7% at Rocky River). We cannot be certain how much of the disparity between the raw data and the official results is due to non-representative samples with respect to party affiliation. But the raw data clearly show the erosion of support for Kucinich.

TABLE 19: BREAKDOWN OF VOTERS BY PARTY AFFILIATION, OHIO

	Republican	Democratic	Independent
Independence A C F
Paul F. Conroy (L)	2 1.1%	5 2.0%	4 3.5%
Jim Trakas (R)	148 84.6%	83 33.6%	69 61.1%
Dennis J. Kucinich (D)	25 14.3%	159 64.4%	40 35.4%
None	11	14	19

Rocky River 1D 1E 1G
Paul F. Conroy (L)	3 1.6%	7 3.8%	5 5.6%
Jim Trakas (R)	163 88.1%	44 23.7%	42 47.2%
Dennis J. Kucinich (D)	19 10.3%	135 72.6%	42 47.2%
None	5	10	7

CONGRESSIONAL RACE IN SANTA FE, NEW MEXICO

In New Mexico, as in Ohio, we have party affiliation data for exit poll responders and for total registered voters, but not for voters at the polls. Thus we have only unadjusted exit poll data to work with. At Rio En Medio (Precinct 7) our exit poll precisely matched Barack Obama’s official percentage of 83.9% (ref. Table 1), and was within 0.9% of the Democratic candidate’s percentage in the Senate race (ref. Table 11). But for the three-way Congressional race, our exit poll differed from the official results by 7.0% for the Democratic candidate, 3.0% for the Republican candidate, and 4.1% for the Green Party candidate (see Table 20). At Alameda (Precincts 25 and 33), our exit poll overstated the margins of victory for Democratic candidates in all three contests – by 10.4% for Obama, 7.5% for Senator Tom Udall, and 8.1% for Congressman Ben Lujan. The comparisons for the three-way Congressional race, including the Green Party candidate Carol Miller, who actually ran second in many precincts, are shown below.

TABLE 20: UNADJUSTED EXIT POLL DATA AND OFFICIAL RESULTS, NEW MEXICO

	Official Results U. S. Congress			Exit Poll U. S. Congress
	Republican	Democratic	Green	Republican	Democratic	Green

NM Santa Fe 7 Rio En Medio	20 10.5%	128 67.4%	42 22.1%	10 7.5%	99 74.4%	24 18.0%
NM Santa Fe 25 33 Alameda	50 11.7%	308 72.1%	69 16.2%	33 10.4%	251 78.9%	34 10.7%

Interesting patterns emerge when considering how the exit poll non-responders must have voted if the official results are true and correct (see Table 21). It would appear that Republicans were undersampled at both polling places, with McCain receiving more than twice as high a percentage of the vote among non-responders as among responders.

At Rio En Medio, the number of Republican votes among the non-responders is almost equal for all three offices, as might be expected. But among the responders, such party discipline did not occur. The 19 McCain voters were evenly divided in the Congressional race. Of the 16 who made a choice, 7 voted Republican, 7 voted Democratic, and 2 voted Green.

At Alameda, the same pattern appears among the non-responders – no more than 42.5% (17 of 40) who voted for McCain voted Republican for Congress, while at least 57.5% (23 of 40) voted Democratic or Green. Our database of McCain voters at Alameda bears this out. Of the 49 who made a choice for Congress, only 29 (59%) voted Republican for Congress; 17 (35%) voted Democratic, and 3 (6%) voted Green. Altogether, then, our database shows that 5 (8%) of 65 McCain supporters voted for the Green Party candidate for Congress.

At that rate, one can only account for the official results at Alameda if, among the non-responders, 20 (50%) of 40 McCain supporters voted Democratic for Congress, and if 32 (49%) of 65 Obama supporters voted Green for Congress. This is where the analysis breaks down. Of the 267 Obama voters in our Alameda database, only 27 (10%) voted Green for Congress – a substantial number, but nowhere near 50%. And yet, if the official results are true and correct, the Green Party candidate must have received almost three times as high a percentage among non-responders (32.1%) as among responders (10.7%). It seems unlikely that Green voters would be reluctant to participate in an exit poll. Alternatively, our data set or the official results could have been corrupted.

TABLE 21: VOTE COUNT AMONG NON-RESPONDERS, ASSUMING

OFFICIAL RESULTS ARE TRUE AND CORRECT, SANTA FE, NEW MEXICO

	Exit Poll Responders		Non-Responders
President	McCain	Obama	McCain	Obama

NM Santa Fe 7 Rio En Medio	19 13.3%	120 83.9%	8 16.3%	41 83.7%
NM Santa Fe 25 33 Alameda	51 15.6%	267 81.7%	40 37.7%	65 61.3%

	Exit Poll Responders		Non-Responders
U. S. Senate	Republican	Democratic	Republican	Democratic

NM Santa Fe 7 Rio En Medio	19 13.1%	125 86.2%	9 19.1%	38 80.9%
NM Santa Fe 25 33 Alameda	44 13.5%	279 85.8%	32 29.1%	78 70.9%

	Exit Poll Responders			Non-Responders
U. S. Congress	Republican	Democratic	Green	Republican	Democratic	Green

NM Santa Fe 7 Rio En Medio	10 7.5%	99 74.4%	24 18.0%	10 17.5%	29 50.9%	18 31.6%
NM Santa Fe 25 33 Alameda	33 10.4%	251 78.9%	34 10.7%	17 15.6%	57 52.3%	35 32.1%

PRESIDENTIAL ELECTION IN LOS ANGELES COUNTY

Citizen exit polls were conducted at ten polling places in Los Angeles County. All ten polling places are exhaustively analyzed in the accompanying paper on Propositions 4 and 8. At five of these ten polling places, there are substantial disparities between the exit polls and the official results for the presidential election, even after the exit poll data are adjusted to account for party affiliation (ref. Tables 4 and 15). These five polling places are analyzed here.

In Los Angeles County the election results, precinct by precinct, are updated numerous times during the weeks following an election. Our data for the party affiliation of voters at the polls are not the final figures, but they are close enough to trust the percentages. To make analysis easier, the raw data for each category (Republican, Democratic, Other, None) have been adjusted so that the sum total matches the final official number of voters at the polls (not including absentees), as listed in Table 22 below, from smallest to largest polling place.

TABLE 22: PARTY AFFILIATION OF VOTERS AT THE POLLS, LOS ANGELES COUNTY

	Republican	Democratic	Other	None

Taft	169 29.6%	276 48.6%	18 3.2%	106 18.6%
Santa Monica	113 14.8%	451 59.2%	34 4.5%	164 21.5%
Topanga	97 9.0%	664 61.5%	76 7.1%	241 22.3%
Lockhurst	353 29.8%	570 48.1%	61 5.2%	202 17.0%
Glendale	345 20.5%	810 48.1%	76 4.5%	453 26.9%

Republican voters were undersampled at all ten polling places in Los Angeles County, including the five analyzed here. The party affiliation of exit poll responders is given in Table 23 below.

TABLE 23: PARTY AFFILIATION OF EXIT POLL RESPONDERS, LOS ANGELES COUNTY

	Republican	Democratic	Other	None

Taft	79 25.5%	170 54.8%	12 3.9%	49 15.8%
Santa Monica	56 10.5%	356 66.7%	27 5.1%	95 17.8%
Topanga	21 3.6%	388 66.4%	61 10.4%	114 19.5%
Lockhurst	134 22.4%	317 53.1%	51 8.5%	95 15.9%
Glendale	85 13.3%	352 55.1%	42 6.6%	160 25.0%

By subtracting the data for party affiliation of exit poll responders (Table 23) from the reliable estimates for voters at the polls (Table 22), we derive the data for party affiliation of non-responders (see Table 24 below). This allows us to analyze the likelihood that the difference between the exit polls and the official results can be reasonably attributed to the pool of non-responding voters.

TABLE 24: PARTY AFFILIATION OF NON-RESPONDERS, LOS ANGELES COUNTY

	Republican	Democratic	Other	None

Taft	90 34.7%	106 40.9%	6 2.3%	57 22.0%
Santa Monica	57 25.0%	95 41.7%	7 3.1%	69 30.3%
Topanga	76 15.4%	276 55.9%	15 3.0%	127 25.7%
Lockhurst	219 37.2%	253 43.0%	10 1.7%	107 18.2%
Glendale	260 24.9%	458 43.8%	34 3.3%	293 28.0%

Similarly, we derive by subtracting the exit poll data from the official results (ref. Table 1) what the vote count must have been among the non-responders, assuming that the official results are true and correct (ref. Table 2). For these five Los Angeles County polling places, the numbers are repeated below.

TABLE 25: VOTE COUNT AMONG NON-RESPONDERS,

ASSUMING OFFICIAL RESULTS FOR PRESIDENT ARE TRUE AND CORRECT

	Exit Poll Responders		Non-Responders
	McCain	Obama	McCain	Obama

Taft	88 28.7%	214 69.7%	113 44.8%	132 52.4%
Santa Monica	56 10.6%	465 87.7%	72 32.3%	149 66.8%
Topanga	26 4.5%	535 93.5%	103 20.4%	383 76.0%
Lockhurst	151 25.8%	421 72.0%	254 43.9%	322 55.6%
Glendale	106 16.7%	511 80.7%	287 28.3%	707 69.7%

When comparing the data for non-responders in Tables 24 and 25, the official results for President seem, at first glance, easily explainable. One indicator would be to compare the ratio of McCain and Obama voters to the ratio of Republicans and Democrats, as shown in Table 26. The two ratios tend to be very close, as one would expect in a partisan contest.

TABLE 26: PARTY AFFILIATION AND PRESIDENTIAL PREFERENCE OF NON-RESPONDERS

	Presidential Preference			Party Affiliation
	McCain	Obama	Ratio	Rep.	Dem.	Ratio

Taft	113	132	1 : 1.17	90	106	1 : 1.18
Santa Monica	72	149	1 : 2.07	57	95	1 : 1.67
Topanga	103	383	1 : 3.72	76	276	1 : 3.63
Lockhurst	254	322	1 : 1.27	219	253	1 : 1.16
Glendale	287	707	1 : 2.46	260	458	1 : 2.22

At Taft and Topanga, the ratios are so close that a straight party-line vote among the non-responders (and a similar breakdown among third-party and independent voters) would explain the official results. At Lockhurst and Glendale, and especially at Santa Monica, Obama’s vote count exceeds what would be expected based on party affiliation alone. This would be explained if the exit poll data showed that a somewhat larger percentage of Republicans voted for Obama than the percentage of Democrats who voted for McCain.

In fact, the difference is much too extreme to correlate closely with the official results. Among Democratic voters, Obama’s percentage was never less than 90.5%, and McCain’s percentage was never more than 8.3%. Among Republican voters, McCain’s percentage was never more than 75.9%, and Obama’s percentage was never less than 21.5% (see Table 27). It is McCain’s official vote count that needs explaining, as follows.

TABLE 27: PRESIDENTIAL VOTE BY PARTY AFFILIATION AMONG EXIT POLL RESPONDERS

	Republicans		Democrats		Other / None
	McCain	Obama	McCain	Obama	McCain	Obama

Taft	60 75.9%	17 21.5%	14 8.3%	152 90.5%	14 23.3%	45 75.0%
Santa Monica	36 65.5%	16 29.1%	4 1.1%	348 98.3%	16 13.2%	101 83.5%
Topanga	11 52.4%	10 47.6%	6 1.6%	372 98.2%	9 5.2%	153 89.0%
Lockhurst	94 73.4%	28 21.9%	18 5.8%	291 93.3%	39 26.9%	102 70.3%
Glendale	59 71.1%	21 25.3%	12 3.4%	337 95.7%	35 17.7%	153 77.3%

At Taft High School, where McCain’s official count exceeded his exit poll total by 113 votes, there were 90 Republican non-responders (ref. Table 26). If McCain got 75% of them, he only needed 45 (26.6%) of 169 other non-responders (106 of whom were Democrats) to reach his official count (ref. Table 24).

At Santa Monica, where McCain’s official count exceeded his exit poll total by 72 votes, there were 57 Republican non-responders (ref. Table 26). If McCain got 75% of them, he only needed 29 (17.0%) of 171 other non-responders (95 of whom were Democrats) to reach his official count (ref. Table 24).

At Topanga, where McCain’s official count exceeded his exit poll total by 103 votes, there were 76 Republican non-responders (ref. Table 26). If McCain got 75% of them, he only needed 46 (11.0%) of 418 other non-responders (276 of whom were Democrats) to reach his official count (ref. Table 24).

At Lockhurst, where McCain’s official count exceeded his exit poll total by 254 votes, there were 219 Republican non-responders (ref. Table 26). If McCain got 75% of them, he only needed 89 (18.9%) of 470 other non-responders (253 of whom were Democrats) to reach his official count (ref. Table 24).

At Glendale, where McCain’s official count exceeded his exit poll total by 287 votes, there were 260 Republican non-responders (ref. Table 26). If McCain got all of them, he only needed 92 (11.7%) of 785 other non-responders (458 of whom were Democrats) to reach his official count (ref. Table 24).

In short, all of these scenarios seem perfectly plausible. Even if McCain got only 75% of the Republican non-responders and only 5% of the Democratic non-responders, he only needed 40 of 63 others at Taft, 24 of 76 at Santa Monica, 32 of 142 at Topanga, 76 of 217 at Lockhurst, and 69 of 327 at Glendale, to reach his official counts. There is little reason to doubt the accuracy of the presidential election results. This stands in stark contrast to the official results for Proposition 8 (the ban on same-sex marriage), which have been exhaustively analyzed in the accompanying paper and found to be inexplicable.

PRESIDENTIAL AND CONGRESSIONAL RACES IN PITTSBURGH, PENNSYLVANIA

For Pennsylvania we have obtained a database containing the names, ages, party affiliation, and voting histories of every registered voter in the state. Thus we are able to adjust the raw data to account for these variables. However, in Pittsburgh, even after the raw data are properly adjusted to account for party affiliation, the most important of these variables, there are remaining disparities of 9.2% in the presidential election and 3.2% in the Congressional race, as shown in Tables 4 and 10 above. The differential between these two figures is reason enough to examine closely the presidential election.

As in Santa Fe, interesting patterns emerge when considering how the exit poll non-responders must have voted if the official results for Ward 14, District 8, are true and correct. As shown in Table 1 above, Obama defeated McCain by 264 (84.3%) to 45 (14.4%) in the exit poll and by 394 (78.0%) to 107 (21.1%) in the official results. Thus, McCain must have gotten the votes of 62 (32.3%) of 192 non-responders, or else the official results are not true and correct.

There was no Republican candidate for Congress in this district. The Democratic incumbent was challenged by a Green Party candidate who received the support of 54 voters, exactly half of whom responded to the exit poll. Of these, 7 voted for McCain and 18 for Obama. There were 51 voters who made no choice for Congress (including one write-in). Of these, 36 responded to the exit poll, and 15 (41.7%) of them voted for McCain.

If these ratios are representative of the non-responders, then McCain also got 7 of 27 non-responders who voted Green for Congress, and 6 of the 15 non-responders who made no choice for Congress, in which case he must have gotten 49 (32.7%) of the 150 non-responders who voted Democratic for Congress. This would be a remarkable achievement given that our database shows that of the 253 exit poll responders who voted Democratic for Congress, only 23 (9.1%) voted for McCain. Similarly, of the 238 exit poll responders who identified themselves as Democrats, only 8 (3.4%) voted for McCain.

When the 2008 presidential and Congressional results are compared to the preceding elections, other interesting patterns emerge. (There is no use in examining the Congressional results for 2004 and 2002, because the incumbent ran unopposed). The official results, including absentee ballots, are shown in Table 28 below.

TABLE 28: COMPARISON OF 2008 OFFICIAL RESULTS WITH PAST ELECTIONS, PITTSBURGH

	President		U. S. Congress
Pittsburgh W 14 D 8	Republican	Democrat	Green	Democrat

2008	108 21.1%	401 78.2%	54 11.5%	410 87.4%
2006			25 7.3%	312 91.2%
2004	109 21.0%	409 78.7%

The 2008 Congressional election was a rematch between Mike Doyle, the incumbent Democrat, and Titus North, the Green Party candidate (for this reason the numbers are shown in blue). The Green Party vote increased from 7.3% to 11.5% in 2008, largely at the expense of the Democrats, whose share of the vote declined from 91.2% to 87.4%. In the presidential election, the percentages were almost unchanged from 2004. Obama received 78.2% of the vote compared to 78.7% for Kerry, and McCain received 21.1% of the vote compared to 21.1% for Bush.

What makes this worth examining more carefully is the fact that, statewide, Obama defeated McCain by 54.49% to 44.17%, a margin of 10.32%, whereas Kerry defeated Bush by 50.92% to 48.42%, a margin of only 2.50%. One might expect Obama to have made some gains in Pittsburgh.

In fact, when Ward 14 in Pittsburgh is considered as a whole, Obama did receive 1,005 more votes than Kerry, and McCain received only 22 more votes than Bush (see Table 29 below). Although this made little difference in terms of percentages, it obviously affected the margin between the candidates.

TABLE 29: COMPARISON OF 2008 AND 2004 ELECTIONS, PITTSBURGH, WARD 14

	President
Pittsburgh Ward 14	Republican	Democrat	Others

2008	4,318 20.13%	16,870 78.64%	263 1.23%
2004	4,296 21.18%	15,865 78.20%	126 0.62%

But there are 41 districts in Ward 14, and Obama made 63.4% of his gains in just three of them, outpolling Kerry by 285 votes in District 2, 192 votes in District 7, and 160 votes in District 25. In the other 38 districts, there was very little overall difference between the election results of 2004 and 2008. In fact, there were 14 districts where Obama actually received fewer votes than Kerry.

So there remains an unexplained 9.2% disparity between the adjusted exit poll data and the official results, Obama winning the adjusted exit poll by 66.1%, and the official results by 56.9%. This is much greater than the 3.2% disparity in the Congressional race, with the Democrat winning the adjusted exit poll by 78.7%, and the official results by 75.5%.

None of this disparity can be attributed to gender bias – that is, to the exit poll responders not being a representative sample with respect to gender. Of the 302 exit poll responders who disclosed their gender, 147 (48.7%) were men and 155 (51.3%) were women. According to the statewide voter data base there were 445 voters at the polls. Of the 426 for whom we know the gender, 219 (51.4%) were men and 207 (48.6%) were women.

Moreover, there was not much of a “gender gap.” Among exit poll responders, Obama defeated McCain among men by 121 (83.4%) to 21 (14.5%), and among women by 131 (85.1%) to 23 (14.9%). When the raw data are adjusted with respect to gender (the calculations are set forth in the appendix), the results are unchanged, and the disparity remains.

Nor can this disparity be attributed to the exit poll responders not being a representative sample with respect to age. Of the 306 exit poll responders who disclosed their age, 91 (29.7%) were aged 18 to 29, 151 (49.3%) were between 30 and 59, and 64 (20.9%) were over 60.

According to the statewide voter database, of the 445 voters at the polls, 110 (24.7%) were aged 18 to 29, 224 (50.3%) were between 30 and 59, and 111 (24.9%) were over 60. Thus, voters aged 18-29 were overrepresented, and voters over 60 were underrepresented, in our exit poll. But our database shows that Obama did well among voters aged 60 and older. Among exit poll responders who made a choice for president, Obama received 77 of 89 (86.5%) aged 18 to 29, 126 of 151 (83.4%) aged 30 to 59, and 55 of 63 (87.3%) aged 60 and older.

Thus, when the raw data are adjusted with respect to age (the calculations are set forth in the appendix), the results are essentially unchanged (Obama actually gains 0.1%), and the disparity remains.

In short, the 9.2% disparity between the official results and the exit poll data adjusted with respect to party affiliation cannot be attributed to sample bias with respect to age or gender.

While the official results for Ward 14, District 8 do not appear anomalous with respect to the rest of Ward 14 or with respect to the 2004 presidential election, it is possible that our exit poll is correct, and that the official results are wrong throughout the ward.

While I have no evidence of this, I have no evidence to the contrary. The alternative explanation, that the official results are true and correct, requires that one-third of the Democratic non-responders voted for McCain, which is sharply at variance with the voting pattern among exit poll responders and is difficult to defend.

PRESIDENTIAL AND CONGRESSIONAL RACES IN CENTRE COUNTY, PENNSYLVANIA

In Harris Township, Centre County, Pennsylvania, our exit poll was greatly at variance with the official results. Obama defeated McCain by 24.7% in the exit poll, and by only 5.4% in the official results, a disparity of 19.3% (ref. Table 1). In the Congressional race, Mark McCracken, the Democratic candidate, defeated Glenn Thompson, the Republican candidate, by 8.3% in the exit poll, but lost by 13.3% in the official results, a disparity of 21.6% (ref. Table 7).

However, as shown in Table 3, Republicans were undersampled, and Democrats oversampled, in our exit poll. Only 39.5% of exit poll responders identified themselves as Republicans, but according to the statewide voter database, 47.7% of all those who voted were registered Republicans. Conversely, 49.9% of exit poll responders identified themselves as Democrats, but only 40.7% of all those who voted were registered Democrats. But even when the exit poll data are properly adjusted to account for party affiliation, Obama still wins the exit poll by 12.4%, and a 7.0% disparity remains (ref. Table 4). McCracken now loses in the exit poll by 4.4%, but an 8.9% disparity remains (ref. Table 10).

None of this disparity can be attributed to the exit poll responders not being a representative sample with respect to age. Of the 1331 exit poll responders who disclosed their age, 248 (18.6%) were aged 18 to 29, 777 (58.4%) were between 30 and 59, and 306 (23.0%) were over 60. According to the official results there were 2962 ballots cast. According to the statewide voter data base there were 2741 voters at the polls and 225 absentee ballots, for a total of 2966 ballots cast (more on this discrepancy later). Of these, 506 (17.1%) were aged 18 to 29, 1667 (56.2%) were between 30 and 59, and 793 (26.7%) were over 60. Thus, voters under 60 were overrepresented, and voters over 60 were underrepresented, in our exit poll. But our database shows that Obama did well among voters aged 60 and older. Among exit poll responders who made a choice for president, Obama received 156 of 247 (63.2%) aged 18 to 29, 460 of 757 (60.8%) aged 30 to 59, and 191 of 300 (63.7%) aged 60 and older. Thus, when the raw data are adjusted with respect to age (the calculations are set forth in the appendix), the results are unchanged, and the disparity remains.

Nor can this disparity be attributed to the exit poll responders not being a representative sample with respect to gender. Of the 1334 exit poll responders who disclosed their gender, 600 (45.0%) were men and 734 (55.0%) were women. According to the statewide voter database, of the 2857 voters for whom we know the gender, 1401 (49.0%) were men and 1456 (51.0%) were women. But there was not much of a “gender gap.” Among exit poll responders, Obama defeated McCain among men by 360 (61.1%) to 221 (37.5%), and among women by 816 (61.8%) to 490 (37.1%). When the raw data are adjusted with respect to gender (the calculations are set forth in the appendix), the results are essentially unchanged (Obama loses 0.1%), and the disparity remains.

In short, the 7.0% disparity in the presidential election (between the official results and the exit poll data adjusted to account for party affiliation) cannot be attributed to sample bias with respect to age or gender. The same would be true of the 8.9% disparity in the Congressional election.

There was a substantial difference in the election results for President and Congress. Officially, Obama received 52.1% of the vote, compared to 42.3% for McCracken. Conversely, McCain received 46.7% of the vote, compared to 55.6% for Thompson (ref. Tables 1 and 7). Simply stated, nearly 10% of the electorate voted Democratic for President and Republican for Congress. This phenomenon, known as “ticket splitting,” was real. Our exit poll data show the same thing.

Among exit poll responders, Obama received 61.8% of the vote, compared to 53.3% for McCracken, and McCain received 37.1% of the vote, compared to 45.0% for Thompson (ref. Tables 1 and 7). When the exit poll data are adjusted to account for party affiliation, Obama receives 55.6% of the vote, compared to 47.0% for McCracken, and McCain receives 43.2% of the vote, compared to 51.4% for Thompson (ref. Tables 4 and 10).

Altogether, among exit poll responders, 14.6% (119 of 816) of Obama supporters voted Republican for Congress, while only 6.1% (30 of 490) of McCain supporters voted Democratic for Congress. This does explain why the Republican candidate for Congress ran more strongly than the Republican candidate for President, but it does not explain the disparities between the exit polls and the official results for the two contests.

The 2008 Congressional election in Centre County was not a rematch. The incumbent was not on the ballot, and neither candidate had run for Congress before. In the 2004 and 2002 Congressional elections, there was no Democratic candidate for Congress in this district. The Republican incumbent was challenged by a Libertarian candidate, and comparisons of the 2008 Congressional election with those races are not very useful. A historic comparison does show that Obama’s percentage of the vote was 6.0% greater than Kerry’s, while McCracken’s percentage was only 1.3% greater than that of the 2006 Democratic candidate (see Table 30).

TABLE 30: COMPARISON OF OFFICIAL RESULTS, HARRIS TOWNSHIP

	President			U. S. Congress
Harris Twp 56 57	Republican	Democrat	Other	Republican	Democrat	Other

2008	1379 46.7%	1540 52.1%	36 1.2%	1616 55.6%	1229 42.3%	60 2.1%
2006				1301 58.8%	908 41.0%	5 0.2%
2004	1553 53.3%	1341 46.1%	17 0.6%

The more interesting analysis is a comparison of exit poll responders and non-responders. In Pennsylvania, we know the party affiliation of every person who voted. And we know the party affiliation of the exit poll responders. Thus, by simple subtraction, we can calculate how many voters from each party did not respond to the exit poll. These numbers are exact (see Table 31).

TABLE 31: PARTY AFFILIATION OF 2008 VOTERS, HARRIS TOWNSHIP

	2008 Voters (Including Absentees)
Harris Twp 56 57	Republican	Democratic	Other / None

Responders	532 39.5%	673 49.9%	143 10.6%
Non-Responders	883 54.6%	533 32.9%	202 12.5%
Total	1415 47.7%	1206 40.7%	345 11.6%

But when the numbers from Tables 2 and 9 are reexamined (see Table 26 below), it becomes nearly impossible to explain the official results in light of the known party affiliations of the voters. The presidential election is the more plausible.

Officially, McCain received the votes of 889 non-responders, and there were 883 Republican non-responders. True, McCain could not have won them all. McCain won the votes of only 76.7% (408 of 532) exit poll responders who identified themselves as Republicans.

At this rate, McCain would have gotten 677 of 883 Republican non-responders. His remaining 212 votes could have come from among the 533 Democrats and 202 Independents who did not respond to the exit poll.

But McCain won the votes of only 6.7% (45 of 673) of Democrats and 25.9% (37 of 143) Independents who responded to the exit poll. Realistically, McCain must have outpaced all three of these percentages among the non-responders, or else the official results are not true and correct.

The official results for the Congressional election are more difficult to explain. Officially, Thompson received the votes of 1039 non-responders, and there were only 883 Republicans. Thompson could not have won them all. He did win 84.0% (447 of 532) of the exit poll responders who identified themselves as Republicans. But that is not all of them.

At that rate, Thompson would have gotten 742 of 883 Republican non-responders. His remaining 297 votes could have come from among the 533 Democrats and 202 Independents who did not respond to the exit poll.

But Thompson won the votes of only 11.6% (78 of 673) of Democrats and 36.4% (52 of 143) Independents who responded to the exit poll. Realistically, Thompson must have far outpaced all three of these percentages among the non-responders, or else the official results are not true and correct.

TABLE 32: VOTE COUNT AMONG NON-RESPONDERS,

ASSUMING OFFICIAL RESULTS ARE TRUE AND CORRECT

	Exit Poll Responders		Non-Responders
President	McCain	Obama	McCain	Obama

PA Centre Harris Twp 56 57	490 37.1%	816 61.8%	889 54.4%	724 44.3%

	Exit Poll Responders		Non-Responders
U. S. Congress	Republican	Democratic	Republican	Democratic

PA Centre Harris Twp 56 57	577 45.0%	684 53.3%	1039 64.1%	545 33.6%

The truth is that the official results for Harris Township are not true and correct, and the statewide voter database proves it. There are two precincts in Harris Township. Both of them vote at the same polling place. According to the official results for 2008, there were 1447 ballots cast in Harris East, and 1515 ballots cast in Harris West (including 2 and 5 undervotes, respectively).

But according to the statewide voter database, there were 1431 ballots cast in Harris East, and 1535 ballots cast in Harris West. Thus, there were 16 extra votes in Harris East, and 20 missing votes in Harris West. If the number of ballots cast is not correct, the number of votes assigned to the candidates cannot be correct.

The same thing happened in 2006. According to the official results, there were 1093 ballots cast in Harris East, and 1208 ballots cast in Harris West. But while there were 1093 votes counted for Congress in Harris East (including 26 undervotes and write-ins), there were only 1166 votes counted for Congress in Harris West (including 24 undervotes and write-ins). It appears that there were 42 missing votes in Harris West. Again, if the number of ballots cast is not correct, the number of votes assigned to the candidates cannot be correct.

On occasion, voters wrote comments on the questionnaires handed to them by our exit pollsters, and they do shed some light on this matter. Three voters noted confusion between the two precincts in the voter rolls and at the polling place:

“East and West confusion”

“East West, wrong line again, again!”

“voter registration card did not match my address east/west”

Another voter was “not sure the ballots are inserted into the machine correctly.”

We do not know if some voters cast ballots in the wrong precinct, or if some ballots cast in the correct precinct ended up in the wrong stack, or if some ballots were inserted incorrectly into the optical scanners. All we know for certain is that the wrong numbers of ballots are counted, again and again, in Harris Township, Centre County, Pennsylvania, and that this must result in an incorrect vote count for the candidates.

And although we do not know to what extent the vote count is affected, we do know that the official results for the 2008 election differ substantially from our exit poll data, by 7.0% for President and by 8.9% for Congress, even when the exit poll data are properly adjusted to account for party affiliation. It is possible that our exit poll is correct. It is not possible that the official results are correct.

NEW HAMPSHIRE

In New Hampshire, citizen exit polls were conducted at four polling places. There were disparities between the official results and the exit poll data, even when adjusted to account for party affiliation. These disparities appeared in the presidential (ref. Table 4), Senate (ref. Table 14), and Congressional races (ref. Table 10) at all four polling places. The disparities in the margins between the candidates ranged from 3.6% in the Senate race at Manchester 3 to 15.6% in the presidential election at Manchester 5 (ref. Table 15).

For New Hampshire we obtained the data for party affiliation of actual voters, both at the polls and by absentee ballot, by examining the poll books for each of the four polling places. The party registrations of the voters actually appear in the poll books – Republican, Democratic, or Independent. The data show that Republicans were undersampled, and Democrats were oversampled, at each polling place (ref. Table 3), and the exit poll data have been adjusted accordingly. The disparities, derived from the adjusted data, are repeated in Table 33 below.

TABLE 33: DISPARITIES BETWEEN MARGINS OF VICTORY

IN OFFICIAL RESULTS AND EXIT POLL DATA, ADJUSTED

	President	Senate	Congress

NH Hillsborough Manchester 3	8.8%	3.6%	4.2%
NH Hillsborough Manchester 5	15.6%	11.0%	10.4%
NH Hillsborough Nashua 5	6.0%	4.2%	5.1%
NH Hillsborough Wilton	6.7%	8.6%	6.9%

Examination of the poll books in New Hampshire revealed far more than party affiliation. Serious discrepancies became apparent when the total number of people who actually voted was compared to the official number of votes counted. These data, including absentee ballots, are presented in Table 34 below.

TABLE 34: COMPARISON OF OFFICIAL RESULTS

WITH DATA FROM ACTUAL POLL BOOKS, NEW HAMPSHIRE

	McCain	Obama	Others	Votes Counted	Actual Voters

NH Hillsborough Manchester 3	1013 33.2%	2003 65.7%	34 1.1%	3050	3089
NH Hillsborough Manchester 5	992 35.6%	1761 63.2%	32 1.1%	2785	2890
NH Hillsborough Nashua 5	2499 47.2%	2741 51.8%	49 0.9%	5289	5380
NH Hillsborough Wilton	1026 44.6%	1248 54.3%	24 1.0%	2298	2101

When the last two columns in Table 34 above are compared, the data make sense for two of the four polling places. It is entirely credible that there were 39 undervotes (no choice for President), or 1.26% of 3089 ballots cast, in Manchester 3, and 91 undervotes, or 1.69% of 5380 ballots cast, in Nashua 5.

It is less credible that there were 105 undervotes, or 3.63% of 2890 ballots cast, in Manchester 5. And it is impossible that were 197 more votes for President than the number of actual voters in Wilton. This phenomenon is known as “phantom voters” because they are apparitions. They do not actually exist. There can never be more votes counted for any office than the number of actual voters who cast ballots. There could be one or two, if on occasion an actual voter forgot to sign in at the polls, but never 171 (9.41% of votes counted). And this number is a minimum. For every actual undervote, a ballot cast with no choice for president, there must have been yet another “phantom vote,” a vote counted for president with no actual ballot. They cancel each other out.

It is interesting to note that the official results posted on the New Hampshire Secretary of State’s website contain a double asterisk next to the Town of Wilton. The double asterisk indicates “corrections rec’d by town clerk.”

Equally disturbing discrepancies appear in the actual counting of the ballots at the other two polling places. In New Hampshire, absentee ballots are actually delivered to each polling place to be counted right along with the ballots cast at the polls. There are no provisional ballots, because New Hampshire is one of four states with same-day voter registration. So there is only supposed to be one report of the vote count. There is not ever supposed to be a partial report.

In Manchester 3, where optical scanners were utilized, the polling place ran out of ballots, so at least 185 ballots were photocopies on soft paper, not card stock, and were hand counted. No apparent disparity exists between the partial count and the complete count for McCain and Obama. Their percentages changed very little, as shown in Table 35 below. But somehow, 12 votes disappeared from the columns of the third-party candidates.

TABLE 35: COMPARISON OF PARTIAL AND FINAL COUNTS, MANCHESTER 3

NH Hillsborough Manchester 3	McCain	Obama	Others

Partial Count	955 33.2%	1876 65.2%	46 1.6%
Additional Count	58 31.4%	127 68.6%	-12 N.A.
Final Count	1013 33.2%	2003 65.7%	34 1.1%

In Nashua 5, incomplete results were copied by the exit poll coordinator from the “city clerk’s written records.” When compared to the final count, the numbers are utterly impossible to believe. In the partial count, Obama was getting 58.0% of the vote. In the additional count, McCain got 69.9% of the vote (see Table 36 below). At Nashua 5, optical scanners were utilized. None of the ballots were hand counted. They would not have been sorted into piles for each candidate. Also, 30 votes somehow disappeared from the columns of the third-party candidates.

TABLE 36: COMPARISON OF PARTIAL AND FINAL COUNTS, NASHUA 5

NH Hillsborough Nashua 5	McCain	Obama	Others

Partial Count	1637 40.1%	2370 58.0%	79 1.9%
Additional Count	862 69.9%	371 30.1%	-30 N.A.
Final Count	2499 47.2%	2741 51.8%	49 0.9%

In New Hampshire, the 2008 Senate race was a rematch of the 2002 contest between John Sununu, Jr., the Republican, and Jeanne Shaheen, the Democrat. In Manchester, the 2008 Congressional race was a rematch of the 2006 contest between Jeb Bradley, the Republican, and Carol Shea-Porter, the Democrat. Bradley also ran for Congress in 2004 and 2002.

In Nashua and Wilton, the 2006 Congressional race was a rematch of the 2004 contest between Charles Bass, the Republican, and Paul Hodes, the Democrat. Hodes also ran for Congress in 2008. Bass also ran for Congress in 2002. Thus, comparisons with past elections are especially useful in New Hampshire. Rematches are depicted in blue in Table 37 below.

In Manchester 3 and 5, there has been a steady erosion of support for Republican candidates for President and Congress since 2004 and for the Senate since 2002. The declines in 2008, when compared with the elections immediately preceding, range from 4.8% for President in Manchester 3 to 10.0% for Senate in Manchester 5.

But in Nashua 5 and Wilton, for all three offices, the decline in 2008 was much smaller when compared to the elections immediately preceding. The declines in 2008 ranged from 0.6% for Congress in Nashua 5 to 2.8% for Congress in Wilton.

TABLE 37: COMPARISON OF OFFICIAL RESULTS, NEW HAMPSHIRE

	President		U. S. Senate		U. S. Congress
Manchester 3	Republican	Democrat	Republican	Democrat	Republican	Democrat

2008	1013 33.2%	2003 65.7%	1078 36.3%	1794 60.4%	996 34.5%	1809 62.7%
2006					584 41.3%	822 58.1%
2004	1079 38.0%	1744 61.5%			1367 49.9%	1366 49.8%
2002			739 44.3%	889 53.3%	765 46.5%	800 48.7%

	President		U. S. Senate		U. S. Congress
Manchester 5	Republican	Democrat	Republican	Democrat	Republican	Democrat

2008	992 35.6%	1761 63.2%	993 36.1%	1650 60.0%	925 34.9%	1632 61.6%
2006					563 40.9%	809 58.8%
2004	1217 43.2%	1581 56.2%			1461 54.4%	1218 45.3%
2002			822 46.1%	922 51.7%	866 49.5%	828 47.3%

	President		U. S. Senate		U. S. Congress
Nashua 5	Republican	Democrat	Republican	Democrat	Republican	Democrat

2008	2499 47.2%	2741 51.8%	2337 45.2%	2639 51.0%	2239 45.3%	2580 52.2%
2006					1266 45.9%	1445 52.3%
2004	2443 48.1%	2603 51.2%			2816 58.2%	1838 38.0%
2002			1590 47.6%	1657 49.6%	1775 53.4%	1467 44.2%

	President		U. S. Senate		U. S. Congress
Wilton	Republican	Democrat	Republican	Democrat	Republican	Democrat

2008	1026 44.6%	1248 54.3%	1010 45.2%	1148 51.4%	931 42.8%	1203 55.3%
2006					630 45.6%	737 53.3%
2004	1032 45.8%	1209 53.6%			1242 56.3%	865 39.2%
2002			720 46.8%	753 49.0%	875 57.4%	589 38.6%

A similar pattern appears for the entire cities of Manchester and Nashua. In 2008, compared to the election immediately preceding, Republican support for President, Senate and Congress declined by 5.9%, 8.1% and 5.6%, respectively, in Manchester, and by 2.6%, 3.7% and 2.0%, respectively, in Nashua (see Table 38).

TABLE 38: COMPARISON OF OFFICIAL RESULTS, CITYWIDE, MANCHESTER AND NASHUA

	President		U. S. Senate		U. S. Congress
Manchester	Republican	Democrat	Republican	Democrat	Republican	Democrat

2008	21,192 44.0%	26,526 55.1%	21,236 44.8%	24,799 52.3%	19,675 42.5%	25,471 55.0%
2006					12,827 48.1%	13,819 51.8%
2004	23,286 49.9%	23,116 49.5%			27,408 61.0%	17,457 38.8%
2002			16,581 52.9%	14,118 45.1%	17,386 56.1%	12,509 40.4%

	President		U. S. Senate		U. S. Congress
Nashua	Republican	Democrat	Republican	Democrat	Republican	Democrat

2008	17,325 42.6%	22,902 56.4%	16,765 42.3%	21,273 53.7%	15,956 42.0%	21,006 55.3%
2006					9,523 44.0%	12,114 55.9%
2004	18,016 45.2%	21,587 54.1%			21,132 57.8%	15,382 42.1%
2002			11,511 46.0%	12,947 51.8%	13,222 52.9%	11,187 44.7%

But Ward 5 stands out even among the nine wards in Nashua, being the only ward in which McCain actually received more votes in 2008 than Bush had gotten in 2004. Nashua 5 also had the smallest decline in the percentage of Republican support, with McCain receiving 47.2% of the vote compared to 48.1% for Bush. Elsewhere in Nashua, the decline in Republican support for President ranged from 1.7% in Ward 3 to 4.6% in Ward 4 (see Table 39 below).

TABLE 39: COMPARISON OF OFFICIAL RESULTS, WARD BY WARD, NASHUA

	2008			2004
	McCain	Obama	Others	Bush	Kerry	Others

Nashua Ward 1	2475 45.1%	2952 53.8%	55	2594 47.5%	2825 51.8%	37
Nashua Ward 2	2072 44.5%	2535 54.4%	49	2198 48.1%	2333 51.1%	39
Nashua Ward 3	1789 40.5%	2594 58.8%	30	1922 42.2%	2594 57.0%	34
Nashua Ward 4	945 31.3%	2038 67.6%	34	1081 35.9%	1904 63.2%	27
Nashua Ward 5	2499 47.2%	2741 51.8%	49	2443 48.1%	2603 51.2%	35
Nashua Ward 6	1637 40.3%	2370 58.3%	56	1773 42.8%	2347 56.7%	22
Nashua Ward 7	1673 43.1%	2166 55.8%	43	1712 45.6%	2012 53.6%	32
Nashua Ward 8	1845 40.8%	2632 58.2%	47	1903 44.3%	2366 55.1%	25
Nashua Ward 9	2390 45.0%	2874 54.1%	46	2390 47.6%	2603 51.9%	25

The reader will recall that partial results in Nashua 5 had Obama getting 58.0% of the vote, and McCain getting 40.1% of the vote. This would have represented a gain of 6.8% for Obama compared to Kerry, and a loss of 8.0% for McCain compared to Bush.

These numbers would be quite out of line with the ward by ward results for the rest of Nashua. But they would not have been out of line with the ward by ward results for Manchester, where the decline in Republican support for President ranged from 3.6% in Ward 1 to 8.4% in Ward 12 (see Table 40 below).

TABLE 40: COMPARISON OF OFFICIAL RESULTS, WARD BY WARD, MANCHESTER

	2008			2004
	McCain	Obama	Others	Bush	Kerry	Others

Manchester 1	2480 45.7%	2913 53.7%	29	2739 49.3%	2776 50.0%	39
Manchester 2	2055 43.2%	2668 56.0%	39	2209 48.4%	2334 51.2%	18
Manchester 3	1013 33.2%	2003 65.7%	34	1079 38.0%	1744 61.5%	14
Manchester 4	1336 39.7%	1982 59.0%	44	1591 46.8%	1791 52.7%	18
Manchester 5	992 35.6%	1761 63.2%	32	1217 43.2%	1581 56.2%	17
Manchester 6	2324 48.0%	2489 51.4%	32	2263 53.4%	1954 46.1%	21
Manchester 7	1652 44.6%	2016 54.5%	33	1795 49.7%	1788 49.5%	30
Manchester 8	2472 50.6%	2372 48.5%	42	2613 56.6%	1983 42.9%	22
Manchester 9	1819 45.1%	2191 54.3%	26	2022 51.1%	1912 48.3%	21
Manchester 10	1832 45.1%	2181 53.7%	48	2056 51.5%	1921 48.1%	18
Manchester 11	1249 43.1%	1622 56.0%	26	1509 50.4%	1471 49.1%	16
Manchester 12	1968 45.4%	2328 53.7%	42	2193 53.8%	1861 45.7%	19

The differences between the voting patterns in New Hampshire are party explained by demographics. Viewed as a percentage of the electorate, Democrats are stronger Manchester, while Republicans and Independents are stronger in Nashua and Wilton.

The differentials with respect to party affiliation vary according to age bracket. Among exit poll responders, Independents were stronger in Nashua and Wilton than in Manchester among all age brackets, and were stronger among younger voters than among older voters at all four polling places.

Republicans were weakest among younger voters at all four polling places, and were stronger among younger and middle-aged voters in Nashua and Wilton than in Manchester; among older voters, Republican strength was essentially the same at all four polling places.

Democrats were weakest among middle-aged voters at all four polling places, although the difference was substantial only in Nashua (see Table 41 below).

TABLE 41: PARTY AFFILIATION OF EXIT POLL RESPONDERS BY AGE, NEW HAMPSHIRE

	18-29			30-59			60+
	Rep.	Dem.	Ind.	Rep.	Dem.	Ind.	Rep.	Dem.	Ind.

Manchester 3	71 13.2%	240 44.5%	228 42.3%	163 21.1%	333 43.0%	278 35.9%	53 25.1%	101 47.9%	57 27.0%

Manchester 5	49 13.8%	178 50.3%	127 35.9%	121 18.7%	320 49.5%	205 31.7%	32 23.5%	71 52.2%	33 24.3%

Nashua 5	46 16.1%	110 38.6%	129 45.3%	299 27.1%	355 32.2%	450 40.8%	97 26.1%	147 39.5%	128 34.4%

Wilton	37 18.2%	74 36.5%	92 45.3%	188 25.8%	255 34.9%	287 39.3%	44 25.9%	72 42.4%	54 31.8%

Total	203 14.7%	602 43.6%	576 41.7%	771 23.7%	1263 38.8%	1220 37.5%	226 25.4%	391 44.0%	272 30.6%

Support for political candidates also varied by age bracket among exit poll responders. In Manchester, younger voters were the least likely, and older voters the most likely, to vote Republican, and the differentials were greatest in the presidential election. In Wilton, younger voters were the least likely to vote Republican for any office, but there was little difference between middle-aged and older voters.

In Nashua, younger voters were the least likely to vote Republican for President and Congress, but there was not much of a differential in the Senate election (see Table 42). The age differential was most striking for the presidential election in Manchester 3, where Obama received 61.1% among voters over 60, 68.4% among voters aged 30 to 59, and 80.4% among voters under 30. The age differential was also large in Manchester 5, where Obama received 66.2% among voters over 60, 70.7% among voters aged 30 to 59, and 79.6% among voters under 30.

TABLE 42: BREAKDOWN OF EXIT POLL DATA BY AGE, NEW HAMPSHIRE

	18-29		30-59		60+
President	McCain	Obama	McCain	Obama	McCain	Obama

Manchester 3	95 17.8%	430 80.4%	220 29.3%	513 68.4%	77 37.9%	124 61.1%
Manchester 5	65 18.7%	277 79.6%	166 26.9%	437 70.7%	43 33.1%	86 66.2%
Nashua 5	95 33.6%	180 63.6%	476 43.5%	603 55.1%	152 42.2%	197 54.7%
Wilton	62 30.5%	132 65.0%	277 38.1%	437 60.1%	64 38.1%	101 60.1%

	18-29		30-59		60+
U. S. Senate	Sununu (R)	Shaheen (D)	Sununu (R)	Shaheen (D)	Sununu (R)	Shaheen (D)

Manchester 3	149 28.5%	344 65.8%	250 33.1%	471 62.3%	77 36.8%	129 61.7%
Manchester 5	85 24.7%	244 70.9%	175 28.0%	416 66.5%	45 34.4%	79 60.3%
Nashua 5	110 39.6%	149 53.6%	450 41.5%	588 54.2%	144 39.2%	211 57.5%
Wilton	59 31.4%	118 62.8%	272 37.9%	415 57.9%	58 35.4%	104 63.4%

	18-29		30-59		60+
U. S. Congress	Republican	Democratic	Republican	Democratic	Republican	Democratic

Manchester 3	136 26.9%	340 67.2%	223 30.4%	478 65.2%	65 31.9%	132 64.7%
Manchester 5	79 24.2%	232 70.9%	166 27.1%	420 68.5%	40 31.3%	80 62.5%
Nashua 5	78 30.2%	165 64.0%	436 42.5%	550 53.6%	147 40.9%	200 55.7%
Wilton	55 31.3%	113 64.2%	250 36.1%	426 61.6%	60 37.5%	100 62.5%

These differences among age groups did affect the outcomes of our exit polls, because voters under 30 were oversampled, and voters over 60 were undersampled, at three of four polling places (we lack the relevant data for Nashua). Voters were asked to identify their age (18-29, 30-59, or 60+), and their race and gender, on the exit poll questionnaire. The exit pollsters recorded the estimated age of the “refusals” – the voters who declined to participate in the exit poll. This “refusal data” was collected at Wilton and at both polling places in Manchester (but not at Nashua). By adding the numbers from the “refusal data” to the numbers from the questionnaires, we derive a very close estimate of the relative strength of each age bracket among all voters at the polls (see Table 43 below).

The proper procedure is to adjust the exit poll data to account for any variables that caused the sample of voters not to be representative of voters at the polls. For example, in Wilton, where 18.4% of the exit poll responders were under 30, but only 16.0% of the voters at the polls were under 30, the data for voters under 30 must divided by 1.15 (or multiplied by 1/1.15) in order to give this group its proper weight in the final calculations, and the other age brackets must also be adjusted by the proper ratios. When the data are properly adjusted to account for the age of the voters, the largest changes in the exit poll percentages are, not surprisingly, for the presidential election in Manchester 3 and 5, where the margins between the candidates are reduced by 1.8% and 1.1%, respectively (see Table 44 below). The calculations are set forth in detail in the Appendix.

These adjustments do not come close to accounting for the 8.8% and 15.5% disparities that still remained in Manchester 3 and 5 after the exit poll data were adjusted to account for party affiliation (ref. Table 33), so there must be another reason for the disparities.

TABLE 43: EXIT POLL RESPONDERS AND REFUSAL DATA, NEW HAMPSHIRE

Manchester 3	18-29	30-59	60+	Totals

Responders	542 35.5%	774 50.7%	211 13.8%	1527
Refusals	150 19.6%	419 54.8%	195 25.5%	764
Totals	692 30.2%	1193 52.1%	406 17.7%	2291

Manchester 5	18-29	30-59	60+	Totals

Responders	354 31.2%	646 56.9%	136 12.0%	1136
Refusals	110 17.3%	404 63.5%	122 19.2%	636
Totals	464 26.2%	1050 59.3%	258 14.6%	1772

Wilton	18-29	30-59	60+	Totals

Responders	203 18.4%	730 66.2%	170 15.4%	1103
Refusals	69 11.6%	326 54.8%	200 33.6%	595
Totals	272 16.0%	1056 62.2%	370 21.8%	1698

TABLE 44: EXIT POLL DATA ADJUSTED WITH RESPECT TO AGE, NEW HAMPSHIRE

	Unadjusted		Adjusted		Change
President	McCain	Obama	McCain	Obama

Manchester 3	401 26.1%	1106 72.0%	414.7 27.0%	1091.2 71.1%	1.8%
Manchester 5	286 25.0%	832 72.8%	291.8 25.6%	824.4 72.3%	1.1%
Wilton	416 36.7%	692 61.1%	417.9 36.9%	690.6 61.0%	0.3%

	Unadjusted		Adjusted		Change
U. S. Senate	Sununu (R)	Shaheen (D)	Sununu (R)	Shaheen (D)

Manchester 3	484 31.7%	975 63.8%	490.4 32.0%	973.3 63.6%	0.5%
Manchester 5	315 27.7%	762 67.1%	318.4 28.0%	757.3 66.7%	0.7%
Wilton	401 36.6%	650 59.4%	401.1 36.6%	652.9 59.6%	0.2%

	Unadjusted		Adjusted		Change
U. S. Congress	Republican	Democratic	Republican	Democratic

Manchester 3	432 29.3%	972 65.9%	436.2 29.5%	971.9 65.8%	0.3%
Manchester 5	294 27.0%	745 68.3%	296.9 27.2%	742.6 68.1%	0.4%
Wilton	371 35.5%	651 62.2%	373.8 35.7%	652.3 62.2%	0.2%

Our exit poll data in New Hampshire were also unrepresentative with respect to gender. Of the 5511 exit poll responders who identified their gender, 3074 (55.8%) were women, and 2437 (44.2%) were men (see Table 45).

TABLE 45: GENDER OF EXIT POLL RESPONDERS, NEW HAMPSHIRE

	Men	Women

Manchester 3	702 46.1%	822 53.9%
Manchester 5	498 44.0%	635 56.0%
Nashua 5	751 42.8%	1003 57.2%
Wilton	486 44.2%	614 55.8%

Total	2437 44.2%	3074 55.8%

In New Hampshire there was a substantial “gender gap.” Unlike Pennsylvania, the voting patterns among women were very different than the voting patterns among men. Women were more likely than men to vote for all Democratic candidates. The differentials in the margins between the candidates averaged 15.0%, and ranged from 6.9% for the Congressional race in Nashua 5 to 20.3% for the Senate race in Manchester 5 (see Table 46). It is not surprising that the differential was greatest in the Senate race, as the Democratic candidate was a woman.

TABLE 46: BREAKDOWN OF EXIT POLL DATA BY GENDER, NEW HAMPSHIRE

	Men		Women		Difference
President	McCain	Obama	McCain	Obama

Manchester 3	198 29.0%	466 68.2%	189 23.6%	602 75.2%	12.4%
Manchester 5	138 28.8%	330 68.8%	136 22.2%	466 76.1%	13.9%
Nashua 5	332 44.8%	388 52.4%	389 39.3%	589 59.4%	12.5%
Wilton	198 40.9%	271 56.0%	207 33.9%	394 64.5%	15.5%

	Men		Women		Difference
U. S. Senate	Sununu (R)	Shaheen (D)	Sununu (R)	Shaheen (D)

Manchester 3	248 36.2%	399 58.2%	226 28.3%	545 68.1%	17.8%
Manchester 5	160 33.1%	294 60.7%	147 23.9%	442 71.8%	20.3%
Nashua 5	339 45.6%	367 49.4%	361 36.9%	576 58.9%	18.2%
Wilton	194 41.4%	253 53.9%	196 32.8%	381 63.7%	18.4%

	Men		Women		Difference
U. S. Congress	Republican	Democratic	Republican	Democratic

Manchester 3	220 33.0%	412 61.9%	203 26.1%	539 69.4%	14.4%
Manchester 5	139 29.0%	310 64.6%	147 25.1%	420 71.8%	11.1%
Nashua 5	295 41.5%	378 53.2%	361 39.0%	533 57.6%	6.9%
Wilton	185 40.5%	258 56.5%	178 31.5%	377 66.7%	19.2%

Again, the proper procedure is to adjust the exit poll data to account for any variables that caused the sample of voters not to be representative of voters at the polls. To account for gender bias, the simplest way to do this is to adjust the men and women to an even 50%-50% split. For example, in Manchester 5, where 44.0% of the exit poll responders were men and 56.0% were women, the data for men must be divided by 0.88 and the data for women divided by 1.12 in order to give both men and women equal weight in the final calculations. In actuality this is overcompensation, because women do tend to outnumber men at the polls. But it does have the advantages of standardizing the adjustment procedure, and of allowing adjustments to be made for Nashua 5 where we lack “refusal data.”

When the data are thus adjusted to account for gender, the largest changes in the exit poll percentages are, not surprisingly, for the Senate election, for which the margins between the candidates are reduced by 1.1% in Wilton, 1.2% in Nashua 5, and 1.3% in Manchester 5 (see Table 47 below). Again, these adjustments do not account for the 8.9%, 4.2%, and 11.0% disparities that still remained after the exit poll data were adjusted to account for party affiliation (ref. Table 33), so there must be another reason for the disparities.

TABLE 47: EXIT POLL DATA ADJUSTED WITH RESPECT TO GENDER, NEW HAMPSHIRE

	Unadjusted		Adjusted		Change
President	McCain	Obama	McCain	Obama

Manchester 3	401 26.1%	1106 72.0%	404.1 26.3%	1101.8 71.7%	0.5%
Manchester 5	286 25.0%	832 72.8%	290.2 25.4%	827.1 72.4%	0.8%
Nashua 5	746 41.4%	1022 56.7%	752.9 41.8%	1013.2 56.2%	0.9%
Wilton	416 36.7%	692 61.1%	420.5 37.1%	686.6 60.6%	0.9%

	Unadjusted		Adjusted		Change
U. S. Senate	Sununu (R)	Shaheen (D)	Sununu (R)	Shaheen (D)

Manchester 3	484 31.7%	975 63.8%	488.6 32.0%	969.4 63.4%	0.7%
Manchester 5	315 27.7%	762 67.1%	321.1 28.3%	754.7 66.4%	1.3%
Nashua 5	722 40.4%	983 55.0%	733.6 41.0%	972.2 54.4%	1.2%
Wilton	401 36.6%	650 59.4%	406.1 37.1%	643.6 58.8%	1.1%

	Unadjusted		Adjusted		Change
U. S. Congress	Republican	Democratic	Republican	Democratic

Manchester 3	432 29.3%	972 65.9%	435.9 29.5%	967.9 65.6%	0.5%
Manchester 5	294 27.0%	745 68.3%	297.3 27.2%	742.3 67.9%	0.6%
Nashua 5	667 39.7%	943 56.1%	671.2 39.9%	939.5 55.8%	0.5%
Wilton	371 35.5%	651 62.2%	376.8 36.0%	645.7 61.7%	1.0%

Even if the adjustments for party affiliation, age and gender are compounded, disparities of 5.0% or more remain in 8 of 12 cases, as shown in Table 48.

TABLE 48: DISPARITIES BETWEEN MARGINS OF VICTORY

IN OFFICIAL RESULTS AND EXIT POLL DATA, AFTER TWO OR THREE ADJUSTMENTS

	President	Senate	Congress

NH Hillsborough Manchester 3	6.5%	2.8%	3.4%
NH Hillsborough Manchester 5	13.7%	9.0%	9.4%
NH Hillsborough Nashua 5	5.1%	3.0%	4.6%
NH Hillsborough Wilton	5.5%	7.3%	5.7%

But it is not clear that all three adjustments (or two, in the case of Nashua 5) can be added together, because we might, more often than not, be adjusting three times for the same non-responding voters. As shown in Table 49 below, an oversampling of women would almost certainly be an undersampling of Republicans. Among exit poll responders, 24.4% of the men were Republicans, and only 19.9% of the women were Republicans.

The same phenomenon appeared in Table 40 above, in which it is shown that an oversampling of voters under 30 would almost certainly be an undersampling of Republicans. Among exit poll responders, 25.4% of those over 60 and 23.7% of those between 30 and 59 were Republicans, but only 14.7% of those under 30 were Republicans. Thus, adjusting the data to account for party affiliation may in and of itself be accounting for age and gender.

TABLE 49: PARTY AFFILIATION OF EXIT POLL RESPONDERS BY GENDER, NEW HAMPSHIRE

	Men			Women
	Republican	Democratic	Other / None	Republican	Democratic	Other / None

Manchester 3	145 20.7%	254 36.2%	303 43.2%	140 17.0%	415 50.5%	267 32.5%
Manchester 5	98 19.7%	238 47.8%	162 32.5%	107 16.9%	327 51.5%	201 31.7%
Nashua 5	217 28.9%	234 31.2%	300 39.9%	227 22.6%	378 37.7%	398 39.7%
Wilton	135 27.8%	154 31.7%	197 40.5%	138 22.5%	243 39.6%	233 37.9%

Total	595 24.4%	880 36.1%	962 39.5%	612 19.9%	1363 44.3%	1099 35.8%

The one remaining variable that could have skewed the exit poll results would be the race of the voters. However, in these four polling places in New Hampshire, of the 5490 exit poll responders who identified their race, 4749 (86.5%) were white, 182 (3.3%) were black, 264 (4.8%) were Hispanic, and 295 (5.4%) were mixed or “other” (see Table 50 below).

In Wilton, 96.6% of the exit poll responders were white. In Nashua 5, we have no “refusal data,” and where we do have it, there was no separate category for Hispanic. In Manchester 3, where 61 (4.0%) of 1521 exit poll responders were black, 27 (3.4%) of 774 refusals were black, so 88 (3.8%) of 2295 voters at the polls were black. In Manchester 5, where 81 (7.2%) of 1123 exit poll responders were black, 27 (4.2%) of 638 refusals were black, so 108 (6.1%) of 1761 voters at the polls were black.

These differences between the percentage of blacks responding to the exit poll and the percentage of blacks among the entire electorate (4.0% and 3.8% at Manchester 3, 7.2% and 6.1% at Manchester 5) were not large. Clearly, oversampling of black voters in the exit polls at Manchester 3 and 5 could not have accounted for the disparities of 8.8% and 15.5%, respectively, which remain for the presidential election even after the data are adjusted to account for party affiliation.

TABLE 50: RACE OF EXIT POLL RESPONDERS, NEW HAMPSHIRE

	White	Black	Latino	Mixed/Other	Total

Manchester 3	1284 84.4%	61 4.0%	81 5.3%	95 6.2%	1521
Manchester 5	836 74.4%	81 7.2%	131 11.7%	75 6.7%	1123
Nashua 5	1566 89.7%	37 2.2%	44 2.5%	99 5.7%	1746
Wilton	1063 96.6%	3 0.3%	8 0.7%	26 2.4%	1100

	4749 86.5%	182 3.3%	264 4.8%	295 5.4%	5490

In summary, we know that the official results in Wilton are wrong. There could not have been 197 more votes counted for President than the number of actual voters. The official results for Manchester 5 are doubtful. It is questionable whether 3.63% of the voters made no choice for President.

The official results for Nashua 5 are highly suspicious. There is no way that ballots fed randomly through an optical scanner would produce a count of 58% for Obama in a partial count and 70% for McCain in a subsequent count. Even in Manchester 3 the official count is impossible because, as in Nashua 5, third-party candidates managed to lose votes between the partial count and the final count.

We know that there are huge disparities between the exit poll data and the official results at all four polling places in New Hampshire, and that adjusting the raw data to account for party affiliation does not explain them. Disparities remain in the margins between the candidates for all three offices, ranging from 6.0% to 15.6% for President, 3.6% to 11.0% for the Senate, and 4.2% to 10.4% for Congress (ref. Table 33).

Adjusting the exit poll data to account for age and gender cannot explain the disparities, but can only reduce them slightly at best. Adjusting the exit poll data on the basis of race would have almost no effect at all. We have exhaustively analyzed the exit poll data and accounted for every apparent variable and cannot explain the official results. We are forced to conclude that the official results in New Hampshire are not true and correct.
______________________________________________

APPENDICES

The appendices for this report are available for download in the attachments panel at the foot of this article.

The raw EVEP data for each of the polling locations discussed in this report are available for download here [3].

______________________________________________

Richard Hayes Phillips, Ph.D., is one of the leading election fraud investigators in the United States. His book on the 2004 Ohio election, Witness to a Crime: A Citizens’ Audit of an American Election [4], based on examination of some 30,000 photographs of actual ballots, poll books, and other election records, is available here [5].

Attachment	Size
EDA-2008-EVEP-Citizen-Exit-Polls.pdf [1]	469.49 KB
Congressionial-Adjusted-Party-Affiliation.pdf [6]	60.01 KB
Senate-Adjusted-Party-Affiliation.pdf [7]	31.27 KB
New-Hampshire-Adjusted-Age.pdf [8]	51.4 KB
New-Hampshire-Adjusted-Gender.pdf [9]	52.55 KB
Pennsylvania-Adjusted-Age-Gender.pdf [10]	29.57 KB