THE TRUE ELECTORATE A CROSS-VALIDATION OF VOTER REGISTRATION FILES AND ELECTION SURVEY DEMOGRAPHICS

Transcription

1 Public Opinion Quarterly, Vol. 71, No. 4, Winter 2007, pp THE TRUE ELECTORATE A CROSS-VALIDATION OF VOTER REGISTRATION FILES AND ELECTION SURVEY DEMOGRAPHICS MICHAEL P. MCDONALD Abstract I cross-validate the 2004 general election electorate s demographic profile available from voter registration files, the media consortium s National Election Poll or exit poll, and the Current Population Survey within selected states. I find voter files and Current Population Survey to be in general agreement, but the exit poll reports an electorate that is younger and composed of more minorities. All three confirm a pronounced pro-woman turnout gap. Much of what is known about the electorate s demographic composition is derived from election surveys. The components of survey measurement error are well studied and researchers have devoted considerable effort to reduce or control it. A much-studied validity threat to election surveys is over-report bias: that a greater percentage of respondents reports voting than actual election statistics indicate (e.g., Burden 2000; Bernstein, Chadha and Montjoy 2001; Clausen ; Traugott and Katosh 1979; Presser, Traugott and Traugott 1990; McDonald 2003). A method to probe over-report bias is to validate individuals self-reported vote with a record of their voting history on voter registration files (e.g., Kitt and Gleicher 1954; Traugott, Traugott and Presser 1992; Abelson, Loftus and Greenwald 1992), a criterion validation exercise that has been performed in other contexts (e.g., Foley et al. 1991; Hessing, Elffers and Weigel 1988; Kuklinski, Cobb and Gilens 1997). While this procedure may detect respondent misreporting, it cannot correct for validity threats that produce a sample unrepresentative of the intended universe, which may also plague another primary source of the electorate s demographics, the national exit poll (National Election Poll or NEP). Thus, even the best measures of the electorate s demographics may contain biases. MICHAEL P. MCDONALD is with George Mason University Dept of Public and International Affairs, George Mason University, 4400 University Drive 3F4 Fairfax, VA , USA. I would like to thank Moshe Haspel for providing registration data for Kentucky and North Carolina. Joe Lenski, Scott Keeter, John Krosnick, Sam Popkin, Warren Mitofsky, and two anonymous reviewers graciously provided comments. All errors remain my own. Address correspondence to MICHAEL P. MCDONALD; mmcdon@gmu.edu doi: /poq/nfm046 C The Author Published by Oxford University Press on behalf of the American Association for Public Opinion Research. All rights reserved. For permissions, please journals.permissions@oxfordjournals.org

2 Voter Registration Files and Election Survey Demographics 589 The purpose of this research note is to take another look at validation of election surveys vis-à-vis statewide voter registration files in Delaware, California, the District of Columbia (DC), Florida, Iowa, Kentucky, Maryland, North Carolina, Ohio, Oklahoma, and South Carolina. 1 Where previous efforts have validated individuals self-reported vote, this analysis is at an aggregate level. The overall demographics from two 2004 general election surveys with large state samples, the Current Population Survey s Voter Supplement File (CPS) and the National Election Poll, are compared with demographics of registered voters with a 2004 general election voting history available from statewide voter registration files. This exercise triangulates the validity of these data sources, thereby providing an assessment of the validity of voter registration files and a clearer picture of the true demographic composition of the electorate. Voter Registration Files Green and Gerber (2005, 7) assert voter registration files present an important advance over survey research, which often relies on respondents self-reports. Besides the American National Election Survey (ANES) vote validation efforts, scholars have used voter files to explore other important questions, such as the frequency of voting (Sigelman and Jewell 1986), and others are discovering anew their research potential by geocoding voter and polling place addresses to explore polling place location affects on turnout (Dyck and Gimpel 2005; Gimpel and Schuknecht 2003; Haspel and Knotts 2005), by examining registration timing (Gimpel, Dyck and Shaw 2007), and by conducting field experiments of voter mobilization (e.g., Green and Gerber 2000). Green and Gerber (2006) further advocate registration-based survey (RBS) sampling in midterm elections, and one prominent survey house employed by academic surveys, Polimetrix, uses registration lists for their sample matching technique (Rivers 2007). The promise of these data is not tempered by a sober assessment of their validity. In evaluating the inability of ANES interviewers to locate the voting records of a sizable percentage of ANES respondents, Presser, Traugott, and Traugott (1990, 9) warn, There might be errors in the records and Green and Gerber (2006, 199) note, The quality of registration records varies. The validity issue is not simply a consequence of data entry errors, which are endemic to any large-scale data collection effort. Election administrators are challenged continually to keep registration rolls accurate. There are persons who are 1. The scope of this project may appear ad hoc, but is driven by data availability and self-financing realities. In 2004, 35 states including DC maintained and allowed academic use of statewide voter registration databases. The median price was $1,000, the least expensive was Arkansas ($2.50, although voting history is maintained at the county level), and the most expensive was Alabama (estimated cost: $26,176.94). Ideally, a nationwide analysis would be conducted; however, this is impossible due to no 2004 voter registration in North Dakota and small Wisconsin towns. Subsequent Help American Vote Act compliant statewide voter files are now available in these two states.

3 590 McDonald registered at but no longer live at an address, what is termed deadwood, and there are persons who have moved since the last election and no longer appear on the registration rolls, who are removed through a process known as purging. To manage deadwood, jurisdictions purge registrants if they are unresponsive to a mailed inquiry generated because they have not voted in a recent election or if their name appears on lists of felons, deceased, or U.S. Post Office change of address lists. 2 The presence of deadwood likely contributes to more people being registered than respondents to election surveys claim, an ironic twist to over-report bias. For example, the total number of registered voters in the 2004 general election was 81.2 percent of the citizen-voting-age population or CVAP (Brace and McDonald 2005), whereas the Current Population Survey reported 72.9 percent of CVAP registered to vote. Persons with a voting history in a recent election are not a component of deadwood. However, purging often continues even as voting records are updated. As a consequence, the number of registrants with a voting history in an election steadily declines over time. Ideally, jurisdictions would maintain a complete record of all registrants voting history on a given election date. South Carolina follows this practice, although all other states analyzed here reported fewer registrants with a 2004 vote than the total number of ballots cast, even when the postelection file most proximate to the 2004 general election was obtained. 3 Table 1 presents the total number of registrants with a 2004 general election vote history contrasted with the total number of reported ballots cast. A total of 38.9 million, or 30.9 percent, of the million 2004 general election voters are examined. The difference between the total number of reported ballots cast and the total number of registrants with a 2004 general election vote history as a percent of the 2004 general election total ballots cast is presented in the third column. 4 The number of purged records as a percent of the total ballots cast was greatest in DC (5.07 percent) and lowest in South Carolina (0.0 percent), followed by North Carolina (0.06 percent). Green and Gerber (2006, 202) claim that missing and erroneous contact data...is the main source of slippage in the RBS sampling process. Depending on the jurisdiction, purging presents another significant issue affecting turnout models in RBS samples. DC s July 2005 voter file lost over 5 percent of the 2. Federal law sets minimum purging standards (The National Voter Registration Act of 1993, 42 U.S.C. 1973gg-6) and states additionally apply their own laws. 3. Florida provides an example of the difficulty in working with these data. All local jurisdictions reported 2004 vote history in February, except Palm Beach County which reported in August and was merged into the earlier database so as not to drop other counties purged records. Data were obtained directly from Lee County as they were not entered correctly into the statewide voter registration file. 4. Oklahoma did not report total number of ballots cast, and the lower total number of votes cast for president is presented in its place. If the approximately 1 percent difference between total ballots cast and vote for president in the 47 states that reported these two election statistics is extrapolated, then the percent of records with a 2004 vote history purged in Oklahoma is about 1.43 percent.

4 Voter Registration Files and Election Survey Demographics 591 Table 1. Overview of Available Information on Statewide Voter Registration Files Total Ballots Cast in Total Registered Voters Percent of 2004 General Available Demographics State 2004 General Election with 2004 Vote History Election Voters Purged Age Gender Race California 12,589,367 12,171, Y Delaware 377, , Y District of Columbia 230, , Florida 7,640,319 7,431, Y Y Y Iowa 1,521,966 1,497, Y Kentucky 1,816,867 1,796, Y Y Maryland 2,395,791 2,365, Y Y Y North Carolina 3,552,449 3,550, Ohio 5,722,443 5,641, Y Oklahoma 1,463,758 1,457, Y South Carolina 1,631,148 1,631, Y Y NOTE. Total Ballots Cast drawn from official state election administration statistics. Oklahoma Total Ballots Cast for president only, all others include ballots with and without presidential vote. denotes majority of 2004 voters had missing data, no reliable analysis possible. DC, IA, and SC do not provide birth date, only internally computed age ranges.

5 592 McDonald November 2004 voters. In a two-year interval between elections, the number of purged records may grow significantly. For those with access to statewide voter files, comparing the overall number of registrants with a vote history against actual turnout numbers serves as a reasonable data integrity check. Demographics of Voter Registration Files and Election Surveys A demographic profile of the 2004 electorate from voter files can be constructed from demographic characteristics of registrants with a record of voting in the 2004 election. These profiles can be compared to election surveys with large state samples, such as the NEP and self-reported voters from the CPS, to triangulate the validity of measurement of the electorate s demographics (for similar analysis comparing election survey demographics, see Merkle and Edelman 2000; Popkin and McDonald 1998). Table 1 lists the varied demographic information available on voter files analyzed here. At a minimum, a record typically contains a person s name, address, birth date, registration date, and voting history. The age distribution of the electorate can be generated from birth dates and sometimes other demographic information is available, such as a registrant s gender and race. As noted in table 1, in some cases the data s validity is questionable due to a high amount of missingness, a problem prevalent on other unanalyzed voter files, such as Connecticut, Missouri, and New Jersey. 5 A word of caution: all three data sources have error and employ different methodologies. Sample frames are not consistent: the NEP samples the universe of voters, the CPS samples all individuals and includes a self-report voting question, and the voter files are an enumeration of all registered voters. Question wording and item response may not be consistent: one hopes that questions regarding age and gender would produce reliable responses, though race can vary in question wording and item responses. AGE Calculating age at the date of an election from a person s birth date is straightforward, although care should be taken in interpreting birth dates where recorded values are impossible, highly suspect, or simply missing. 6 Table 2 reports the 5. The Connecticut voter file recorded 2004 vote history equal to two-thirds of the total ballots cast. Missing data did not appear a purging issue, as some townships had too few recorded voters to be credible. A number of rural counties in Missouri had still not reported 2004 vote history at the time of this writing. A fifth of 2004 New Jersey voters were missing birth date and over half were missing gender. Missing data and reporting issues may be resolved when states adopt truly Help America Vote Act compliant statewide databases. 6. A small number of records clearly contain errors. Where a year is likely missing a leading 1 a literal interpretation would mean a registrant is several hundred years old. Some people would

6 Voter Registration Files and Election Survey Demographics 593 Table 2. Age Distribution of Voters from Voter Files, NEP Exit Poll, and CPS for Selected States Age-4 categories Age Missing < Missing California Voter file NEP exit poll (1.0) (1.0) (1.0) (0.9) (0.8) (0.8) CPS (0.6) (0.8) (0.8) (0.8) (0.7) (0.7) Delaware Voter file 14.0 N/A 28.0 N/A N/A NEP exit poll (1.3) (1.8) (1.7) (1.8) (1.5) (1.5) CPS (1.3) (1.5) (1.5) (1.5) (1.4) (1.4) District of Columbia Voter file N/A NEP exit poll (1.7) (1.7) (1.5) (1.5) (1.3) (1.3) CPS (1.7) (1.8) (1.7) (1.4) (1.2) (1.2) Florida Voter file NEP exit poll (0.8) (0.9) (1.0) (1.0) (0.8) (0.8) CPS (0.7) (0.9) (0.9) (1.0) (0.9) (0/9) Iowa Voter file N/A N/A NEP exit poll (0.9) (1.0) (1.1) (1.1) (0.9) (0.9) CPS (1.1) (1.2) (1.3) (1.2) (1.2) (1.2) Kentucky Voter file NEP exit poll (1.4) (1.5) (1.6) (1.5) (1.3) (1.3) CPS (1.4) (1.5) (1.5) (1.4) (1.3) (1.3) North Carolina Voter file N/A N/A NEP exit poll (0.8) (1.1) (1.1) (1.0) (0.8) (0.8)

7 594 McDonald Table 2. (Continued.) Age-4 categories Age Missing < Missing CPS (1.2) (1.3) (1.3) (1.3) (1.3) (1.3) Ohio Voter file NEP exit poll (1.0) (1.1) (1.1) (1.1) (0.9) (0.9) CPS (0.9) (1.0) (1.1) (1.0) (0.9) (0.9) Oklahoma Voter file NEP exit poll (1.1) (1.2) (1.2) (1.3) (1.1) (1.1) CPS (1.4) (1.6) (1.8) (1.7) (1.5) (1.5) South Carolina Voter file N/A N/A NEP exit poll (1.0) (1.1) (1.2) (1.1) (0.9) (0.9) CPS (1.3) (1.6) (1.6) (1.5) (1.3) (1.3) NOTE. Exit poll and CPS row percentages standard errors appear in parentheses. age distribution available from the voter files for four age categories (18 29, 30 44, 45 59, and 60+) and for persons less and greater than age 65. Similar demographics of CPS respondents are presented for those who self-reported voting. State-defined age categories in the DC, Iowa, and South Carolina s permit only uniform presentation of the distribution over and under age 65 for these states. A pattern emerges immediately: whereas the voter files and the CPS are in general agreement, the NEP reports a younger electorate. The pattern is clearest for the percentage of voters under age 65. In every state the age distribution of the 2004 electorate drawn from the voter files has a lower percentage of voters in the under age 65 category by an average of 4.3 percentage points. The largest difference of 7.4 percentage points is in South Carolina and the lowest is in Delaware, at 2.5 percentage points. Random sampling error is unlikely to be a have traveled from the future to vote. There are also placeholder dates for missing data, such as thousands of registered voters with a birth date of January 1, 1900, with only a small number of other persons with birth dates in the same year. I code all suspect birth dates as missing, even though a few people were likely legitimately born on these dates.

8 Voter Registration Files and Election Survey Demographics 595 primary source of the error. Not only is the difference outside the 95 percent confidence interval of the NEP for all state except Delaware, if sampling error were the cause it is unlikely that all 10 states would show a bias outside the margin of error in the same direction. For the CPS, none of the states is outside the margin of error for persons under age 65. The CPS percentage relative to the voter files is an average 0.47 percentage points, with a largest difference of 4.1 percentage points in Iowa and the largest positive difference of 5.8 percentage points in the District of Columbia, which has a small sample size. These two states are also the only two where the voter file percentage is outside the Current Population Survey s 95 percent confidence interval; otherwise these numbers appear to fluctuate within a reasonable sampling error. Similar differences are apparent when the four age categories are examined (Delaware only reports similar ranges to the NEP for and 60+). In every state except North Carolina, the voter files record fewer persons in the age category than the NEP, and in all the cases, the voter files record fewer persons in the age categories than the NEP. For the older age categories of and 60+, in every instance the voter files record a greater number of persons in these ranges than the NEP. Ten of the 26 pairwise comparisons are significantly different at the p <.05 level, and the uniformity of the deviations direction further suggests a systematic bias between the two data sources. The voter files and the CPS report remarkably consistent percentages; the only cell outside the 95 percent confidence interval is in Ohio, where the CPS reports 3.6 percentage points higher among the age 60 and older category. Given the 26 comparisons, one difference outside the margin of error is to be expected. Voter registration file errors may explain the observed differences between the voter files and the election surveys. There are two sources of error on the voter files: missing birth dates and purged records. Bounds on these error sources provide some guidance on whether the discrepancies can be attributed to the measurement error. Other data available on voter files provide clues to the distribution of missing birth dates. The date a person was registered to vote plus 18 years provides a lower bound on a registered voter s age for those with missing birth dates. The distributions for California, DC, Ohio, and Oklahoma are reported in table 3. Precautions were taken to identify impossible, highly suspect, and missing registration dates in the same manner as birth dates. Little can be divined from Oklahoma, which had 86.8 percent missing registration dates. For California and Ohio, the distribution strongly suggests that missing birth dates are found among older voters and thus do not fully resolve the age difference between the voter files and NEP, though it may help explain the one deviant CPS cell. For DC, the pronounced age skew on the voter files in contrast to the surveys may be partially a result of missing age data since 91.3 percent were placed in the age category. However, some care should be taken in interpreting

9 596 McDonald Table 3. Registration Date Plus 18 Years for 2004 Voters with Missing Birth Date, Age-4 Categories Age-4 registration categories State Missing age Missing California District of Columbia Ohio Oklahoma the patterns in table 3, as they represent only a lower bound on a registrant s age. Purging represents a different source of error, and it is plausible that younger voters who have a higher mobility rate are more likely to be purged due to a change of address. However, even if all purged or missing records, as reported in table 1, were assigned to the under age 65 category an unlikely assumption since records of older, deceased registrants are also purged only California, the DC, and Oklahoma would have an age distribution taken from the voter files within the margin of error of the NEP. GENDER The voter registration files of Florida, Iowa, Kentucky, North Carolina, and South Carolina record gender (half of California s records contained missing gender data). Among these states, Florida has the largest amount of missing data, a relatively small 1.5 percent. The demographic distribution of gender drawn from the voter files, NEP and CPS are reported in table 4. Generally, we see the three sources of data confirm a significant turnout gender gap in the 2004 election. The two polls generally deviate within the range of sampling error from the voter files, except in North Carolina where the NEP reports a greater percentage of women than the other two sources, roughly 58.8 percent to 55.5 percent. The general agreement between the three data sources lends validity to the information available on the voter files, and it is unlikely that purging or missing data issues significantly further confound these findings as there is no theoretical expectation that gender is correlated with missing data or purging. RACE Florida, North Carolina, and South Carolina maintain a record of a registrant s race (half of Kentucky s records contained missing data). Jurisdictions covered by the Voting Rights Act use the racial composition of registered voters to measure progress toward a nondiscriminatory electoral system. Perhaps because of

10 Voter Registration Files and Election Survey Demographics 597 Table 4. Gender Distribution of Voters on Voter Files, NEP Exit Poll, and CPS for Selected States. Gender State Female Male Missing Florida Voter file NEP exit poll (1.1) (1.1) CPS (1.7) (1.7) Iowa Voter file NEP exit poll (1.2) (1.2) CPS (1.4) (1.4) Kentucky Voter file NEP exit poll (1.7) (1.7) CPS (1.8) (1.8) North Carolina Voter file NEP exit poll (1.2) (1.2) CPS (1.5) (1.5) South Carolina Voter file NEP exit poll (1.3) (1.3) CPS (1.7) (1.7) NOTE. Exit poll and CPS row percentages standard errors appear in parentheses. the legal ramifications of these data, a small percentage of records had missing data: 1.9 percent in Florida, less than 0.1 percent in North Carolina, and none in South Carolina. The detail of racial information varies among states, but at a minimum, the files provide classifications commensurate with the 2004 NEP s five racial categories of white, African American, Latino, Asian, and Other. These categories are found on the CPS, with the caveat that persons may be identified as single or multiple racial classifications and Hispanic self-report designation is drawn

11 598 McDonald Table 5. Race Distribution of Voters on Voter Files, NEP Exit Poll, and CPS for Selected States. State Five race categories Two race categories Non- White Black Latino Asian Other Missing White white Missing Florida Voter file NEP exit poll (1.0) (0.7) (0.8) (0.3) (0.4) (1.0) (1.0) CPS (0.9) (0.7) (0.7) (0.2) (.02) (0.9) (0.9) North Carolina Voter file NEP exit poll (1.1) (1.1) (0.3) (0.1) (0.4) (1.1) (1.1) CPS (1.4) (1.3) (0.2) (0.3) (0.3) (1.4) (1.4) South Carolina Voter file N/A NEP exit poll (1.2) (1.2) (0.3) (0.2) (0.3) (1.2) (1.2) CPS (1.5) (1.5) (0.3) (0.2) (1.5) (1.5) NOTE. Exit poll and CPS row percentages standard errors appear in parentheses. from a separate ethnicity question. 7 South Carolina reports only white and non-white classifications. A comparison of the race demographics for Florida, North Carolina, and South Carolina presented in table 5 shows that the NEP reports a statistically significantly fewer white voters than the NEP or CPS. The deviations between voter files and NEP for the white category (4.9, 7.5, and 6.8 percentage points, respectively) are consistently outside the margin of error in a pro-minority direction. In two of the three cases, voter files report more whites than the CPS ( 0.6, 3.9, and 0.9 percentage points, respectively), with only North Carolina outside the 95 percent confidence interval. Even for North Carolina, the CPS white percentage is closer to the voter files, and there is a statistically significant difference between the NEP and CPS percentages. The five-category 7. All CPS respondents reporting Hispanic ethnicity are scored as Hispanic. All non-hispanics reporting a single race only are reported as that race. Asian and Hawaiian-Pacific Islander are grouped into an Asian category. For multiple-race categories, non-hispanics reporting black in any other combination are scored as black. Among the remainder, non-hispanics reporting Asian or Hawaiian-Pacific Islander in combination with any other remaining race are identified as Asian. Those remaining are classified as Other.

12 Voter Registration Files and Election Survey Demographics 599 classifications provide clues to the source of the deviation of the NEP with the voter files. In Florida, more Latinos and African Americans were represented on the NEP than the voter files, with a deviation about four times larger for Latinos than African Americans. In North Carolina, few Latinos voted; the deviation comes almost entirely from over-representation of African Americans on the NEP. The general agreement between the voter files and CPS and disagreement between these two data sources and the NEP weakly suggests a bias in the NEP toward minorities. Simply put, with only three states to draw inferences from, more data are needed. Missing data are few in North and South Carolina, and are only large enough to explain the Florida s difference if the 2.7 percent of purged records are combined with the 1.9 percent of records with missing race data are all assigned to the non-white category. Conclusion The recent election administration innovation of statewide voter registration files provides a new, easy to collect, and in some cases relatively inexpensive data source to investigate political behavior. Working with these data poses certain challenges, but also provides new avenues of research. However, researchers are wise to follow the advice of Traugott, Traugott and Presser (1992, 13) who caution,... administrative records should be treated with some care. The information is not error free and researchers should carefully check for significant purging of records and missing data. Given this caution, I might have found that registration files are not reflective of election survey demographics. The good news is that voter registration files are generally consistent with the CPS, which lends validity to their information and should comfort scholars as they investigate these data. However, the NEP, in comparison to these two other sources, reports an electorate that is younger and perhaps containing more minorities. It is possible that the NEP is more valid, though logic dictates that it is more likely that voter files and the CPS more accurately measure the electorate s demographics. Some claim that the bias toward Senator John Kerry on the NEP compared to official election results provides evidence of bias in the 2004 election results favoring President George W. Bush (Bleifuss and Freeman 2006). The findings here suggest a true age bias in the NEP consistent with an internal report by the firms that ran the 2004 NEP, which found that younger interviewers were more likely to sample younger voters while older interviewers drew an age-unbiased sample (Mitofsky and Lenski 2005). The NEP reported that younger voters were the only age category that voted as a majority for Kerry, which partially explains the NEP s pro-kerry bias. Indeed, the analysis presented here weakly suggest that the bias goes deeper, as minorities may also be over-represented

13 600 McDonald on the NEP, who are another pro-democratic group that would bias the NEP in favor of Kerry. Others have analyzed the exit polls to provide evidence of voting trends among youth and minorities (e.g., Lopez and Kirby 2005). The evidence presented here suggests that care should be taken in interpreting the results of such analyses as they apply to the 2004 election. Unfortunately, these results are only suggestive of similar bias in unanalyzed states and provide no evidence of persisting bias over time. It is also unfortunate from a scholarly perspective that the NEP exit polls are often the first source of data on the election, and the results here should caution NEP consumers to await confirming evidence before reaching conclusions about the age and race composition of the electorate. Like Green and Gerber, I am optimistic that voter registration files have significant, untapped research potential. More data are available on voter files, such as registered party, voting method, voting history in other elections, and the address and voting precinct a registrant calls home. Systematic collection of voter files will enable longitudinal studies and permit further research, such as investigations of purging practices. My survey of state election offices in the course of data collection indicates that only a few states have use restrictions. An ongoing coordinated data collection effort available to academic and survey communities would enable more academic RBS surveys and data archiving would provide necessary vote history for better predictive turnout modeling (Green and Gerber 2006, 202). It would again enable the ANES to validate respondents self-reported votes, as advocated recently by Bernstein, Chadha and Montjoy (2001). I am confident that as more scholars become aware of the potential of voter registration files, new and innovative projects will follow. References Abelson, Robert P, Elizabeth F. Loftus, and Anthony G. Greenwald Attempts to Improve the Accuracy of Self-Reports of Voting. In Questions About Questions, ed. Judith M. Tanur. New York: Russell Sage Foundation. Alessi, Ryan State Cuts Thousands from Its Voter Rolls: Check with Tenn., S.C. Finds Dual Registrations. Lexington Herald-Leader 23(April) 2006:A1. Bernstein, Robert, Anita Chadha, and Robert Montjoy Over-Reporting Voting: Why It Happens and Why It Matters. Public Opinion Quarterly 65(1):1 22. Bleifuss, Joel and Steve Freeman Was the 2004 Presidential Election Stolen? Exit Polls, Election Fraud, and the Official Count. New York: Seven Stories Press. Brace, Kimball and Michael P. McDonald Report to the Election Assistance Commission on the Election Day Survey. Washington DC: U.S. Election Assistance Commission. Burden, Barry C Voter Turnout and the National Election Studies. Political Analysis 8(4): Clausen, Aage R Response Validity: Vote Report. Public Opinion Quarterly 32(4): Dyck, Joshua J., and James G. Gimpel Distance, Turnout, and Convenience of Voting. Social Science Quarterly 86(3):

14 Voter Registration Files and Election Survey Demographics 601 Edison Media Research, Mitofsky International, and National Election Pool National Election Pool Democratic Presidential Preference Primary Exit Polls, 2004 [Computer file]. ICPSR04183-v1. Somerville, NJ: Edison Media Research/New York, NY: Mitofsky International [producers], Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], Foley, Mary A., Francis T. Durso, Alice Wilder, and Rebecca Friedman Developmental Comparisons of Explicit Versus Implicit Imagery and Reality Monitoring. Journal of Experimental Child Psychology 51:1 13. Gimpel James, Joshua J. Dyck, and Daron R. Shaw Election Year Stimuli and the Timing of Voter Registration. Party Politics 13(3): Gimpel, James G. and Jason E. Schuknecht Political Participation and Accessibility of the Ballot Box. Political Geography 22: Green, Donald P. and Alan S. Gerber The Effects of Canvassing, Direct, Mail, and Telephone Contact on Voter Turnout: A Field Experiment. American Political Science Review 94(3): Green, Donald P. and Alan S. Gerber Recent Advances in the Science of Voter Mobilization. The Annals of the American Academy of Political and Social Science 601(September):6 9. Green, Donald P. and Alan S. Gerber Can Registration Based Sampling Improve the Accuracy of Midterm Election Forecasts? Public Opinion Quarterly 70(2): Haspel, Moshe and H. Gibbs Knotts Location, Location, Location: Precinct Placement and the Costs of Voting. The Journal of Politics 67(2): Hessing, Dick J., Hank Elffers, and Russell H. Weigel Exploring the Limits of Self-Reports and Reasoned Action: An Investigation of the Psychology of Tax Evasion Behavior. Journal of Personality and Social Psychology 54: Kitt, Alice S., and David B. Gleicher Determinants of Voting Behavior: A Progress Report ontheelmira Study. InPublic Opinion and Propaganda, eds. Daniel Katz, Cartwright Dorwin, Eldersveld Samuel, and Alfred McClug Lee. New York: The Dryden Press. Kuklinski, James H., Michael D. Cobb, and Martin Gilens Racial Attitudes and the New South. Journal of Politics 59(2): Lopez, Mark Hugo, and Emily Kirby Electoral Engagement among Minority Youth. Center for Information and Research on Civic Learning and Engagement Fact Sheet, July University of Maryland School of Public Policy: College Park, MD. McDonald, Michael P On the Over-Report Bias of the National Election Study. Political Analysis 11(2): Merkle, Daniel M., and Murray Edelman A Review of the 1996 Voter New Service Exit Polls from a Total Survey Error Perspective. In Election Polls, the News Media and Democracy, eds. Paul. J. Lavrakas and Michael W. Traugott. New York: Chatam House. Mitofsky, Warren, and Joe Lenski Evaluation of Edison/Mitofsky Election System New York: Edison Media Research and Mitofsky International. Popkin Samuel, and Michael P. McDonald Who Votes? A Comparison of NES, CPS, and VNS Polls. Democratic Leadership Council Bluebook. Sept Presser, Stanley, Michael W. Traugott, and Santa Traugott Vote Over Reporting in Surveys: The Records or the Respondents? ANES Technical Report Series. No. nes Rivers, Douglas Sampling for Web Surveys. Presented at the 2007 Summer Political Methodology Conference, Pennsylvania State University. Schachter, Jason P., Rachel S. Franklin, and Marc J. Perry Migration and Geographic Mobility in Metropolitan and Non-Metropolitan American: 1995 to Census 2000 Special Report 9. Washington, DC: U.S. Census Bureau. Sears, David O. and Richard R. Lau Inducing Apparently Self-Interested Political Preferences. American Journal of Political Science 27(2): Sigelman, Lee, and Malcolm E. Jewell From Core to Periphery: A Note on the Imagery of Concentric Electorates. The Journal of Politics 48(2):440 9.

15 602 McDonald Silver, Brian D., Barbara A. Anderson, and Paul Abramson Who Overreports Voting? The American Political Science Review 80(2): Traugott, Michael W., and John P. Katosh Response Validity in Surveys of Voting Behavior. Public Opinion Quarterly 43(3): Traugott, Michael W., Santa Traugott, and Stanley Presser Revalidation of Self-Reported Vote National Election Study. Technical Report Number: 42.