Controversies in Exit Polling: Implementing a Racially Stratified Homogenous Precinct Approach

Matt A. Barreto, University of Washington Fernando Guerra, Loyola Marymount University Mara Marks, Loyola Marymount University Stephen A. Nuño, University of California, Irvine Nathan D. Woods, Welch Consulting, Inc.

n November 2000, interviews counts and lawsuits, it was clear that the unreliable because the members of the I with voters in indicated that Al exit polls were wrong; Bush had won the demographic subgroups interviewed for Gore won the state. As a result, many state by the narrowest of margins. As a the poll are not necessarily representative television networks declared Gore the result of the flawed exit poll1 the media of all members of their demographic winner of Florida, a pivotal state to win- and pollsters scoured and reanalyzed the subgroup. What’s more, with a growing ning the presidency in 2000. Only a few methodology used in 2000 to prepare number of Americans via absentee hours later, the first vote tallies from the and correct for the 2004 presidential ballot, Day-only exit polls2 Florida Secretary of State’s office re- election. The old system, Voter News could miss a large segment of the vealed that George W. Bush was in fact Service ~VNS! was scrapped entirely, electorate. leading in Florida. After 45 days of re- and Edison-Mitofsky Research was cho- Since the November 2004 presidential sen to implement a new and more accu- election considerable media coverage has rate national exit poll in 2004 by a focused on the exit poll controversy. A consortium of news organizations re- December 2004 New York Times article Matt A. Barreto is assistant professor of tained by the called the noted that Congressman John Conyers political science at the University of Wash- ~NEP!. What hap- ~D-MI! asked Edison-Mitofsky “to turn ington. His research interests include Latino pened? Exit poll results from Edison- over raw data collected in Election Day political participation and public opinion Mitofsky showed John Kerry ahead in exit polls, for investigation of any dis- and has appeared in several journals in- cluding the American Political Science Re- Ohio, Florida, and New Mexico—all crepancies between voter responses and view and Public Opinion Quarterly. states which he lost to Bush in 2004. certified election results” ~Associated In addition to the overall exit poll re- Press 2004!; a January 2005 Washington Fernando Guerra is associate professor sults being skewed, comparative vote Post headline read, “Report Acknowl- of political science and Chicano studies at results for subgroups, such as that for edges Inaccuracies in 2004 Exit Poll” Loyola Marymount University and director Latino voters, also appeared to be wrong. ~Morin and Deane 2005!; and CNN of the Center for the Study of Los Angeles, The NEP reported on November 2, 2004, noted in January 2005 that the Kerry a leading think tank on Los Angeles and that Bush won 45% of the Latino vote, a numbers were “overstated,” and claimed urban politics. His research interests include 10-point gain from 2000. In contrast, an that “CNN did not air those inaccurate urban politics, political leadership, and only racial/ethnic politics. exit poll of Latino voters conducted results or post them on its website.” by the William C. Velasquez Institute Thus, many scholars and pundits reached Mara Marks is assistant professor of reported that Bush won only 32% of the the conclusion that new alternatives to urban studies at Loyola Marymount Univer- Latino vote. Moreover, a pre-election the traditional exit poll may be sity and a research scholar at the Center survey of Latino voters by the Tomás warranted. for the Study of Los Angeles. Her research Rivera Policy Institute, a non-partisan Specifically, we pose two important interests include urban politics, racial/ethnic think tank with more than 10 years expe- methodological questions pertaining to politics, and land-use politics. Her articles rience polling Latino voters, reported the science behind exit polls: ~1! what is have appeared in the Urban Affairs Re- Bush garnering just 30% of the vote ~see the most accurate sampling technique view, among other journals. Leal et al. 2005!. for polling racial and ethnic voters in a Stephen A. Nuño is a doctoral candi- What explains such discrepancies? diverse setting, and ~2! how should exit date in political science at the University of One possibility is the methodology used polls account for early and absentee California, Irvine and a research associate to select the precincts where exit poll votes not cast on Election Day? To an- at the Center for the Study of Los Angeles. interviews are conducted is faulty. Ide- swer these questions, we implemented His research interests include Latino par- ally, the respondents in the exit poll sur- an alternative sampling exit poll in the tisanship, voting behavior, and political vey will be accurate representatives of City of Los Angeles during the 2005 participation. the entire city or state in which the elec- mayoral election and compared our re- tion is being held. However, if the exit sults to the exit poll implemented by the Nathan D. Woods is an economist with poll interviews respondents that are too Los Angeles Times. We then compared Welch Consulting in Santa Monica, Califor- nia. His research focuses on the political conservative or too liberal, too young or both polls to the actual election results. participation of racial and ethnic minorities too old, too poor or too rich, or too In short, the different methodologies and has appeared in the American Political White, it could skew the overall results accounted for different results, suggest- Science Review, Journal of Politics, and by a wide margin, even after weights are ing that new approaches to exit polling other journals. employed. Existing exit polls are often are welcome.

PSOnline www.apsanet.org 477 The Good, The Wrong, and While the current debate regarding ~NEP!. The media affiliates that pooled The Ugly sampling continues, practical discussions resources for VNS wanted the most reli- regarding the relationship between the able data possible, to report on election The academic literature regarding exit respondent and surveyor have a long night in 2004. While Edison-Mitofsky polls focuses broadly on two areas, tradition. Common sense tells us that Research sought to address the decline in methodology and human interactions human interactions will produce certain the reliability of exit polls, the results effects. Generally speaking, exit poll biases in any scientific experiment. Drop were disappointing. The 2004 presiden- methodology is comprised of two com- those interactions into a given social tial election exit poll overstated projec- ponents, proper sampling techniques and context, and those biases are likely to tions for John Kerry within precincts the logistical practicalities involved in multiply. The social context can range on average by over 6 points ~Mitofsky administering an exit poll. The conven- from the respondent’s likelihood to an- 2005, 31! and each of the six press re- tional wisdom is that election projections swer sensitive questions honestly ~Aqui- leases issued by Warren0Mitofsky during should be made where precincts are lino 1994; Bishop and Fisher 1995; the election wrongly placed Kerry ahead randomly selected ~Mitofsky 1989!. Benson 1941! to the influence socioeco- in the race ~Morin 2004!. Some tried to However, there is growing belief in the nomic status and ethnicity contribute to attribute the skewed polling numbers to literature regarding general polling tech- the validity of the survey results ~Free- voter fraud ~Baiman et al. 2005!, but the niques that non-probability sampling, man 1969; Welch et al. 1973; Weeks and most compelling explanation has been where respondents are recruited and not Moore 1981; Hurtado 1994!. There are that differential non-response rates by randomly selected, might yield more ac- also limited contributions on the environ- Democrats and Republicans have signifi- curate results ~Couper 2000; Fiorina and mental context of poll taking, such as the cantly skewed the predictive power of Krosnick 2005!. Non-probability sam- time a poll is administered ~Busch and exit polls ~Mitofsky 2005; Liddle 2005!. pling, or quota sampling, has been Lieske 1985! or whether the climate af- This could be based on non-response sharply criticized by the establishment fects the accuracy of polls ~Doob 2001!. patterns, or could be based on precinct ~Mitofsky 1999!, but declining response The business of predicting , selection. If too few “Republican” pre- rates ~Groves and Couper 1998! have led however, goes beyond social and envi- cincts were selected to represent the true to questions about the true randomness ronmental contexts. The political circum- result in the state, Bush’s numbers might of conventional random sampling tech- stances of calling elections present exit appear low. Thus, selecting the best pre- niques. And although lower response polling with a substantial public relations cincts to represent an entire state is ex- rates do not seem to substantively affect problem. The difficulty is that incorrect tremely important. the results of the random polls ~Keeter predictions can be satisfactory from a Predicting elections has never been an et al. 2000! the successful use of Internet statistical standpoint, yet quite unsatisfac- insular affair, but the difference between polls has provoked thoughtful discussion tory when those results are inserted into a good call and a correct one can be at on sampling techniques which are appro- a politically charged atmosphere. George the mercy of political context. For in- priate for the web, but not necessarily for Gallup’s career was catapulted by cor- stance, the Edison-Mitofsky regime a conventional exit poll ~Fiorina and rectly predicting that Franklin D. Roose- overstated John Kerry’s national num- Krosnick 2005!. velt would win the election in 1936 even bers by 2.5 percentage points in 2004, This discussion, however, is largely in though most straw polls predicted a win which was not much larger than the the arena of survey polls, not exit polls, for Alfred Landon ~Fiorina and Krosnick overestimation of Bill Clinton’s numbers but there is no telling what the future 2005, 1!. Twenty years later, Gallup pre- in 1992 by the television networks might hold. There are a growing number dicted that Dewey would defeat Truman, ~Morin 2004!. The only difference is of reasons why contemporary polling and even though the Dewey prediction Bill Clinton won in 1992 and so it mat- techniques would be required. For was statistically more sound3 and within tered little. But the calls for instance, the Edison-Mitofsky 2004 the margin of error than the Roosevelt and John Kerry have brought attention presidential election exit poll could not prediction, some of the public reacted to to previous overstatements for one party conduct a true exit poll in Oregon be- the failed prediction with charges of over another, and with some justifica- cause Oregonians cast their ballots fraud ~Committee on Analysis of Pre- tion. Warren Mitofsky ~2003, 51! gath- through the mail. The growing discus- election Polls 1948–1949, 599!. ered that within-precinct error in the exit sion, and actual use in government spon- After the 2000 presidential election, a polls for senate and governor races in sored pilot programs, regarding voting report produced for CNN by Joan Kon- 1990, 1994, and 1998 showed an under- with eBallots will also increase the ner, James Risser and Ben Wattenberg statement of the Democratic candidate move to conduct exit polling over the ~Konner et al. 2001! on television’s per- in 20% of the 180 polls during that web. The sheer number of elections and formance on the night of the election time period and an overstatement 38% surprise elections, such as the California concluded that exit polls had “lost much of the time. The CNN’s report on the recall, will also require greater flexibility of the value it had for projecting election network’s performance on Election in polling techniques. Should Internet results in close elections.” People are not Night 2000 ~Konner et al. 2001, Appen- polling produce more reliable results only less inclined to respond to exit polls, dix 4! found that the exit polls over- than pen and paper exit polling and but the inability of the exit polls to deal stated the Gore vote in 22 states and should more states follow Oregon’s vot- with shifts in the number of absentee vot- overstated the Bush vote in nine states. ing method, or witness an increase in ers and early voters were further eroding While the history of predicting elec- early voting ~such as in Florida, Califor- their reliability with each succeeding tions has seen its share of the good, the nia, Texas!, the non-probability versus election cycle ~Konner et al. 2001, 3!. wrong, and the ugly, it is growing in- random sampling debate will grow. The Following the 2000 presidential elec- creasingly vital that new sampling tech- use of absentee ballots and expanded tion polling debacle, Edison Media Re- niques are used to accurately depict the election periods, as opposed to election search and Mitofsky International were electorate. Incorrect predictions could days, will further lead to questions about charged with conducting a more accurate foment an erosion of public confidence the validity of traditional sampling exit poll for the 2004 presidential elec- in exit polls and the in techniques. tion by the National Election Pool general, and decrease the response rates

478 PS July 2006 of those who view polling regimes as designing and implementing their meth- are racially homogenous. Thus, the serving a political agenda. Incorrect exit odology. While a telephone survey has White, Latino, Black, or Asian voters in polls could also be used by policy mak- the advantage of randomly calling any that North Hollywood precinct may not ers and pundits to shape public policy. registered voter within the state, an exit be representative of the “typical” White, For example, when President Bush an- poll is economically limited to a small Latino, Black, or Asian voter in Los An- nounced the nomination of then White number of sites. Ideally, exit pollsters geles. According to an analysis of geo- House Counsel Alberto Gonzalez for would set up stations at every single pre- graphic segregation by the University of attorney general, many pundits noted it cinct within a jurisdiction, so that no Michigan Population Studies Center, Los was because Bush did so well among voter is left out. Of course, it is not real- Angeles racial groups are still very much Latino voters, winning an estimated istic to recruit 20,000 volunteers to staff divided ~Farley 2001; see also Logan 45%. Since then, numerous media orga- each of the 20,000 precincts in a state 2002!. On a racial residential segregation nizations and several scholars have re- like California. Thus, the key is picking index of dissimilarity, where a value of 0 vised the numbers downward, likely in a select number of precincts that accu- is perfect integration and a value of 100 the mid-30s ~see Leal et al. 2005!. Exit rately represent the full universe of is extreme segregation, Los Angeles is polls hold considerable value for our po- 20,000 throughout a state. If the “wrong” viewed as highly segregated. Farley’s litical system. In principle, the use of precincts are selected, the results may be analysis provides an index of dissimilar- exit polls represents an important ac- biased. Therefore, exit poll research ity for each racial group vis-à-vis one knowledgement that the interests and teams take considerable care to select another and reported a White-Black opinions of the electorate are an impor- precincts. In fact, this is the most impor- value of 77, a White-Latino value of 71, tant component of our political system. tant step in exit polling. However, the a White-Asian value of 55, and a Latino- In practice, they play an important role selection criteria may still be flawed, as Black value of 61 ~2001!. Given these in the strategic decisions of politicians recent presidential elections have re- residential distinctions, we wonder and political elites, and they provide the vealed. Pollsters may rely on two meth- whether Latinos who vote at the 28th media with an objective resource for ods for choosing their critical sample Street YMCA in East Los Angeles, evaluating the quality of the bonds be- precincts: first, they may put all the pre- which resides in a precinct which has a tween our representatives and their con- cinct numbers into a hat and randomly population that is 95% Latino, differ stituents ~Lavrakas et al. 1995, 3–22!. choose precincts to include, or, second, from Latinos who voted at the heteroge- Although exit polls are a common occur- they may purposely choose precincts to neous precinct in North Hollywood. And rence in the , they are sur- fit the size, , and racial what if those Latinos at the 28th Street prisingly uncommon in other modern specifications that fit a given election. YMCA precinct are excluded from the democracies, such as Canada ~Brown Because pollsters want a large sample exit poll? Will the overall Latino sample et al. 2004!. In elections in the Philip- size and good cross-sections of different be flawed? Similarly, questions may arise pines, Central America, and Russia, exit types of voters, they almost always rely about Asian Americans who vote at the polls are often used by third parties to on a purposive random sample that al- Korean Resource Center ~61% Asian provide a check against voter fraud and lows them to hand pick the precincts to population! and Blacks who vote at the to gauge the underlying sentiments of the include. Crenshaw United Methodist Church citizenry ~Mitofsky 1989!. It is important that the methodologist ~88% Black population!. In a city like Since 1948 and the birth of the is familiar with the universe they are Los Angeles, most citizens vote in pre- modern-day exit poll in 1967 ~Levy interested in sampling, including the geo- cincts where their racial group is a ma- 1983, 54!, the stakes of blown calls have graphic distinctions and racial and ethnic jority. To this end, Los Angeles is not grown steadily. The “early call” of the differences within the universe. For an unique. Data from the University of 1980 presidential election by the media exit poll in the City of Los Angeles, Michigan Population Studies Center, re- may have helped to reduce turnout pollsters would want to capture an accu- ported in Table 1, reveal that residential among Democrats ~Crespin and Vander rate representation of all Los Angeles segregation is still a significant issue in Wielen 2002; Carpini 1984; Jackson voters. For example, if 50% of voters are all of America’s largest cities, an impor- 1983! which could have had an impact White, 25% Latino, 17% Black, and 8% tant consideration in exit poll precinct on close congressional races. The early Asian, it is important that these same selection. call for Al Gore in Florida probably led ratios are reflected in who gets inter- Is the precinct that looks like a mi- to a loss of votes for Bush in the Florida viewed. Because there are not enough crocosm of the city, really a microcosm panhandle ~Sobel and Lawson 2001!. resources to set up exit polling stations at of the city, or is it an anomaly? A more Most recently, the 2004 presidential elec- all 1,700 precincts in Los Angeles, poll- accurate representation of racial and tion exit poll was marred by overstated sters select a sample of about 50 to 60 ethnic voters, and therefore the city at projections which surpassed one standard precincts, while also keeping the geogra- large, might be found if we conducted error in more than half the states ~Mitof- phy and demographics of the city in most of the exit poll interviews in high sky 2005, 3!. These missteps will cer- mind. The easiest way to do this would concentration racial precincts instead tainly not contribute positively to be to pick precincts that most closely of mixed-race precincts. This sampling correcting the differential non-response resemble the overall demographics of strategy is supported by data from the rates which are in turn contributing to city voters. For example, voting precinct Los Angeles city clerk precinct list. the inability to make more accurate calls, # 9007129, situated in North Hollywood, In 2005, only 18% of precincts—less and new sampling techniques must be has a population that is 49% White, 24% than one in five—had no racial or eth- experimented with to increase the preci- Latino, 15% Black, and 7% Asian— nic group as a majority, leaving 82% sion, and credibility, of exit polling. almost a microcosm of the entire City of of precincts in Los Angeles comprised Los Angeles. Or is it? mostly of one racial group or another. Exit Polling Methodology The reality is that the great majority of In full, 38% of precincts were majority voters do not live in such racially inte- Latino, 37% majority White, 7% Practitioners and consultants of exit grated neighborhoods. Instead, most vot- majority Black, and 1% majority poll projects spend considerable time ers reside—and vote—in precincts that Asian.

PSOnline www.apsanet.org 479 students were used in heavily Latino and ferent approach to implementing their Table 1 Asian communities. Exit polling was exit polls. Index of Racial Segregation conducted from 7:00 am to 8:00 pm, the First, both exit poll surveys were only by City, 2000 entire time that the polls were open. conducted of voters on Election Day, Prior to the May 17, 2005, election, which means that absentee voters are not White- White- students attended two training sessions included in the full survey results. To City Black Latino and received instructions on recruiting account for these missing voters, both New York 77 64 participants, skip pattern, and the confi- LMU and the Los Angeles Times Los Angeles 77 71 dentiality of the data. The exit poll im- weighted their data to incorporate absen- Chicago 88 64 plemented a traditional skip pattern and tee voting patterns when the official data 5 Houston 78 66 replacement strategy. The interviews became available from the city clerk. Philadelphia 82 70 were self-administered, meaning that vot- However, considerable differences in Phoenix 63 63 ers filled out a quick survey card on their how the absentee vote was incorporated San Diego 67 64 own, while the students concentrated on into each exit poll potentially bias the San Antonio 60 55 recruiting voters to participate as they results ~explained in detail below!.Inthe Dallas 75 69 left the voting precinct. In total, 100 stu- May 17, 2005, election, absentee voters Miami 86 51 dent exit pollsters were needed to carry made up 27% of the Los Angeles elec- Detroit 68 65 out the project on May 17, 2005. An ad- torate and voted 51.4% to 48.6% in favor Washington 84 65 ditional 20 students assisted with data of Antonio Villaraigosa. However, it is Boston 78 65 entry and project implementation. the Election Day voters that we are inter- Denver 71 63 ested in examining closer. Seattle 69 51 Different Polls, Different The LMU exit poll was carried out in Results 50 precincts across Los Angeles, and Note: Value of 0 reflects pure inte- administered in five languages: English, gration and value of 100 reflects In May 2005, Los Angeles elected its Spanish, Chinese, Korean, and Tagolog. pure segregation. first Latino mayor in more than 130 The Los Angeles Times poll was carried Source: Population Studies Center, years. Once a Mexican city in Alta Cali- out in 59 precincts and available only in University of Michigan, 2001. fornia, the city of angels has the largest English and Spanish. In addition, the Latino population of any city in the LMU poll specifically chose precincts in United States. However, voters in Los homogenous racial communities. For The 2005 Los Angeles Exit Angeles are still predominantly White example, the precincts selected by LMU Poll Pilot Project ~50% of the electorate!, with Latinos consisted of: constituting about 25%, Blacks 17%, and In an effort to address the exit poll Asian Americans 7%. Thus, electing a 10 heavily White communities controversy, a team of researchers at the Latino mayor is not a Latino-only phe- 10 heavily Latino communities Center for the Study of Los Angeles at nomenon. While Antonio Villaraigosa 10 heavily Black communities Loyola Marymount University ~LMU! won an estimated 85% of the Latino vote 10 heavily Asian communities designed an alternative exit poll method- in 2005, he also captured a majority of 10 mixed-race communities ology.4 The new method, described as a White and Black votes to win the elec- “racially stratified homogenous precinct tion 59% to 41% over incumbent Mayor These final 10 precincts were in experiment,” interviewed voters in pre- James K. Hahn. While it takes a broad “mixed” or heterogeneous neighbor- dominantly racially concentrated neigh- coalition to win most public offices— hoods where no group comprised a clear borhoods, and then weighted the final especially mayor of a diverse city—it majority. According to our analysis of results with respect to each racial and was Latino voters and Latino candidates precincts in Los Angeles, about one-fifth ethnic group as necessary. A critical who received most of the focus. A week of polling places are located in “mixed” component to this exercise was the re- after the 2005 Los Angeles election, Vil- communities, with over 80% of polling cruitment and training of student exit laraigosa graced the cover of Newsweek places in racially homogenous communi- poll interviewers. Participating students magazine with the headline, “Latino ties. For the most part, Angelinos continue received a cash stipend, lunch, and mile- Power!” An overlooked aspect of this to live and vote in racially segregated pre- age expenses for their participation in the election was the ability of Villaraigosa to cincts. Therefore it is important that the project. No incentives were given for win among Latinos, Blacks, and Whites respondents to the exit poll come from completing a higher number of inter- in Los Angeles. However, this may have such precincts. In comparison, the Los views; instead, students were instructed been overlooked because the mainstream Angeles Times interviewed an over- to closely follow the interviewing guide- media reported that Villaraigosa did not whelming majority of its respondents in lines. Given the current problems sur- win a majority of Black or White votes. mixed-race precincts, as noted in Table 2. rounding the 2004 exit poll, this project The Los Angeles Times exit poll noted While both datasets were weighted to re- was also an opportunity for students to that Villaraigosa only captured a majority flect the correct percentage that each ra- make a visible and meaningful impact on of Latino votes. In contrast, the exit poll cial or ethnic group accounted for within the future of exit polling in American conducted by Loyola Marymount Uni- the electorate, weighting merely replicates elections. versity found Villaraigosa won among the data already gathered, which may al- In order for the LMU exit poll project Latinos, Blacks, and Whites. ready be invalid. Leal et al. ~2005! argue to be accurate, it was implemented in a The Loyola Marymount University that in 2004, Edison-Mitofsky chose the rigorous and scientific manner. LMU and Los Angeles Times exit polls show wrong precincts in which to interview researchers identified 50 precincts to be many similar patterns and results. How- Latinos in Texas, something that weight- included in the exit poll survey, and two ever, there are also some notable differ- ing could not address. students were assigned to each precinct ences. This might be expected given that Second, given that both studies did not to carry out the interviews. Bilingual each organization used a somewhat dif- include absentee voters, it is important to

480 PS July 2006 In contrast, Table 3b shows Finally, when we compare the results Table 2 the Los Angeles Times data for racially homogenous and racially het- Distribution of Precincts in Los and support for Villaraigosa erogeneous precincts considerable differ- Angeles Exit Polls for each racial group. The ences emerge. Table 4 tabulates vote totals same formula results in a for each racial0ethnic group in Los Ange- Precinct Type L.M.U. L.A. Times 57.7% vote share for Villarai- les based on whether or not they voted in 80% White 10 = 20% 10 = 17% gosa among the exit poll a precinct where their racial0ethnic group 80% Latino 10 = 20% 04 = 7% sample, about 4 points too was the majority ~in the LMU dataset!.A 70% Black 10 = 20% 04 = 7% low. Given that the Election Black voter that voted in a predominantly 45% Asian 10 = 20% 01 = 2% Day sample accounted for Black precinct is counted in the first row, Mixed-Race 10 = 20% 40 = 67% 73% and absentee votes for racially homogenous precinct, while a 27% of all the votes cast, Black voter that voted in a predominantly Total Precincts Selected n = 50 n = 59 we can determine the final Latino, Anglo, Asian, or a mixed-race outcome for each poll by in- precinct is captured in the second row, “all corporating exit poll data and other” precincts. This division allows us approximate the Election Day total that absentee data ~which is known from the to determine whether racially homo- both candidates received. Since we can Los Angeles City clerk’s office!. geneous precincts, where most voters live not know the racial composition or vot- The simple calculations reveal that and vote, actually report different vote ing preferences of each absentee voter, the LMU exit poll, plus absentee voting patterns. we only focus on Election Day voters for patterns, results in an estimated Villarai- Significant vote differences are found this study. If Villaraigosa received 58.7% gosa vote share of 59.2%, about 0.5 with respect to homogeneous and het- of the overall vote, and 51.4% of the ab- points too high. In comparison, the erogeneous precincts for every ethnic sentee vote, it is easy to ascertain his Los Angeles Times exit poll data, plus group in Los Angeles. For example, Lat- share of the Election Day vote, given absentee voting patterns, results in an inos who voted in heavily Latino pre- Tϭtotal vote; Pϭprecinct vote; Aϭabsen- estimated Villaraigosa vote share of cincts demonstrated an 88.5% vote tee vote; and Sϭabsentee share of elector- 56.0%, about 2.7 points too low. We preference for Villaraigosa, compared to ate ~which the city clerk notes was 27%!. argue that the Los Angeles Times 80.3% among Latinos in non-Latino pre- results are too low overall, because cincts. Similarly, White voters in White ~1.1! T ϭ ~A ϫ S! ϩ ~P ϫ ~1 Ϫ S!! they underestimate Villaraigosa’s vote precincts were more likely to vote for share among Blacks and Whites, which Villaraigosa. For Blacks and Asians the 58.7 Ϫ ~51.4 ϫ 0.27! LMU estimates at 58% and 57%, inverse pattern was found. Black voters ~1.2! ~P ϫ .73! respectively. in heavily Black precincts voted for Vil- laraigosa at 56.2%, com- Quite simply, Equation 1.2 tells us that pared to 66.7% for Black Villaraigosa must have won 61.4% of the voters in non-Black pre- precinct vote cast in order to have won Table 3a cincts,6 and Asian voters 58.7% citywide, once the absentee votes Loyola Marymount Exit Poll by Race residing outside Asian pre- were included. cincts were also more likely Thus, the key is to come to a solution Villaraigosa % of Total Contribution to vote for Villaraigosa. The that adds up to 61.4%, and not 58.7%, of White 57 0.50 28.2 clear result from this find- the vote for Antonio Villaraigosa when Black 58 0.17 9.5 ing is that racially homog- using exit poll data. Any attempts to Latino 86 0.25 21.3 enous precincts do matter weight the data by final vote percentages Asian 41 0.08 3.1 in influencing vote patterns. ~which include absentee votes! inappro- Further, this demonstrates priately assign absentee voting patterns LMU exit 62.1 0.73 45.3 that a Black voter is not to Election Day voters. The LMU poll Absentee 51.4 0.27 13.9 just a Black voter, but that weighted its data based on race0ethnicity LMU Total 59.2 racial geography is impor- to avoid this problem and incorporated Actual 58.7 tant in implementing an absentee voters using Equation 1.1. The Difference +0.5 accurate exit poll. While Los Angeles Times weighted its precinct- national exit polling outlets only data on the final vote tally, 58.7% such as the Los Angeles to 41.3%, which included absentee Times or Edison-Mitofsky voters. Table 3b may error in picking too Table 3a below depicts this simple Los Angeles Times Exit Poll by Race few racially homogenous formula for the Loyola Marymount Uni- precincts, it is also possible versity exit poll data. First, the LMU Villaraigosa % of Total Contribution that ethnic-based exit poll- data shows somewhat higher rates of White 50 0.50 25.0 ing outlets such as the support for Villaraigosa among Whites Black 48 0.17 8.2 Velasquez Institute or the Los Angeles and Blacks than does the Latino 84 0.25 21.0 Asian American Legal De- Times data. However, this is consistent Asian 44 0.08 3.5 fense and Education Fund with Villaraigosa winning 61.4% of the may error in focusing only Election Day vote total. If we multiply LAT exit 57.7 0.73 42.1 on racially homogenous the percent support for Villaraigosa times Absentee 51.4 0.27 13.9 precincts. In fact, a propor- the percent of the electorate that each Times Total 56.0 tionate amount of both ra- group comprised on May 17, we arrive Actual 58.7 cially homogenous and at an LMU estimate of 62.1% of the vote Difference −2.7 mixed-race precincts are won by Villaraigosa. necessary to reflect

PSOnline www.apsanet.org 481 and account for minority voters, and sec- Table 4 ond, how to accurately include and ac- Vote for Villaraigosa for Racially Homogenous Precincts count for absentee voters. Both issues speak to sample design, and precinct se- White Black Latino Asian lection in particular is a crucial issue for Precinct Type Voters Voters Voters Voters pollsters to consider. Racially Homogenous 65.0% 56.2% 88.5% 34.7% In 2005, two very different exit poll- All Others 56.4% 66.7% 80.3% 50.0% ing methodologies were employed to Difference +8.6% −10.5% +8.2% −15.3% answer the same question: what percent of the vote did mayoral candidates in *Racially homogenous precincts represent areas where each racial/ethnic group is Los Angeles win, and how did this vary the majority. For example, Villaraigosa won 88.5% of the Latino vote in heavily by racial group? The Loyola Marymount Latino precincts while he won 80.3% of the Latino vote in precincts where Latinos University exit poll implemented a ra- were not a majority. cially stratified homogenous precinct approach that specifically designated pre- the geographic distribution of the voters after the infamous 1954 Gallup Poll that cincts in predominantly White, Black, in the geography being polled. incorrectly called the election for Dewey Latino, and Asian neighborhoods, with a over Truman, political scientists continue handful of mixed-race precincts. The Los Angeles Times poll included mostly Implications to debate the methodology behind pre- dicting elections. Building on the errors mixed-race precincts and only a few ra- Exit polling is both a science and a in the 2000 exit poll, the National Exit cially homogenous precincts. The result business, creating many challenges that Poll was launched in 2004, state sample was different results. We argue that the are not always easy to address. Since the sizes were increased and deemed more racially stratified homogenous precinct inception of presidential straw polls, po- accurate, and, in the end, many of the approach is more accurate because it is a litical parties, candidates, analysts, and same errors were made. In our opinion, more natural, or realistic, approach to academics have all devoted considerable more debate and research are needed for exit polling in a diverse, and residentially time to improving the accuracy and relia- exit polling to address two fundamental segregated city. bility of predicting elections. Fifty years issues: first, how to accurately include

Notes * Author names are listed alphabetically. had been instituted in Florida, Al Gore may have to attempt to interview every fifth voter that left The co-authors were also the co-principal inves- won the state ~Keating and Balz 2001!. the precinct. tigators of the Loyola Marymount University 2. In the state of Oregon, which is entirely 6. Interestingly, the LMU exit poll finds that 2005 Los Angeles Mayoral Exit Poll. Thanks vote-by-mail, pollsters must conduct telephone both groups of Black voters—those in predomi- to Salvador Paniagua and Haven Perez for surveys of confirmed voters to ascertain how nantly Black precincts and those in mixed-race their tremendous research assistance in imple- people in the state voted. For more on vote-by- precincts—voted in support of Villaraigosa, in menting this project and to the more than mail in Oregon, see Karp and Banducci 2000. contrast to the findings by the Los Angeles 120 student researchers who participated in 3. The prediction for Truman, however, was Times poll. This makes the Black sample in the exit polling and data entry. Robert Aguinaga outside the margin of error. See www.gallup. the Times poll even more suspect given that and Antonio Gonzalez of the Southwest Voter com0poll0content0?ciϭ1234 for a historical run- most of their Black sample came from mixed- Registration and Education Project also pro- down of Gallup Poll’s predictions. race precincts. Another explanation for the vided valuable assistance in implementing the 4. To view full poll results, please visit: comparatively lower Villaraigosa support among poll. Mark Blumenthal, of mysterypollster.com www.lmu.edu0csla0press0releases_20050Runoff. Blacks is the Times weighting system which was instrumental in tracking down exit poll html. Or, to download the Los Angeles 2005 exit uses final vote totals including absentee voting. archives. poll data, please visit: http:00faculty.washington. As a point of fact, Black absentee voting 1. While many pundits agree that the 2000 edu0mbarreto0data0 rates and vote preference are unknown by exit poll contained many errors, some analysis 5. Given the number of registered voters and both polls. does suggest that if a statewide manual recount anticipated voter turnout, pollsters were required

References Aquilino, William S. 1994. “Interview Mode 0US0Exit_Polls_2004_Mitofsky-Edison. dian Political Science Association, Winnipeg, Effects in Surveys of Drug Use: A Field Ex- pdf Manitoba. periment.” Public Opinion Quarterly 58~2!: Benson, Lawrence E. 1941. “Studies in Secret- Busch, Ronald, and Joel Lieske. 1985. “Does 210–240. Ballot Technique.” Public Opinion Quarterly Time of Voting Affect Exit Poll Results?” Associated Press. 2004. “Congressman Seeks 5~1!: 79–82. Public Opinion Quarterly 49 ~spring!: Election Polling Data.” New York Times, Bishop, George F., and Bonnie S. Fisher. 1995. 94–104. December 22, A22. “‘Secret Ballots’ and Self-Reports in an Exit- Carpini, Michael X. 1984. “Scooping the Voters? Baiman, Ron, Kathy Dopp, Steven F. Freeman, Poll Experiment.” Public Opinion Quarterly The Consequences of the Networks’ Early Brian Joiner, Victoria Lovegren, Josh Mittle- 59~4!: 568–588. Call of the 1980 Presidential Race.” Journal dorf, Campbell B. Read, Richard G. Borrelli, Stephen, Brad Lockerbie, and Richard G. of Politics 46~3!: 866–885. Sheehan, Jonathon Simon, Frank Stenger, Niemi. 1987. “Why the Democrat-Republican CNN. 2005. “Report suggests changes in exit Paul F. Vellman, and Bruce O’Dell. 2005. Partisanship Gap Varies from Poll to Poll.” poll methodology.” CNN Inside Politics, “Analysis of the 2004 Presidential Election Public Opinion Quarterly 51~1!: 115–119. January 19. Poll Discrepancies.” US Count Votes’ Brown, Steven, David Docherty, Kimberly Ellis- Couper, Mick P. 2000. “Web Surveys: A Review National Election Data Archive Project. Hale, Ailsa Henderson, and Barry Kay. 2004. of Issues and Approaches.” Public Opinion http:00electionarchive.org0ucvAnalysis Presented at the annual meeting of the Cana- Quarterly 64 ~3!: 464–494.

482 PS July 2006 Crespi, Irving. 1989. Public Opinion, Polls, and Night 2000: A Report for CNN.” Report Media Research and Mitofsky International Democracy. Boulder, CO: Westview Press. sponsored by CNN. http:00archives.cnn.com0 for the National Election Pool ~NEP!. http:00 Crespin, Michael H., and Ryan J. Vander Wielen. 20010ALLPOLITICS0stories0020020cnn. exit-poll.net0election-night0Evaluation 2002. “The Influence of Media Projections report0cnn.pdf Jan192005.pdf on Voter Turnout in Presidential Elections Ladd, Everett Carll. 1996. “The Election Polls: Moore, David W. 1995. The Superpollsters: From 1980–2002.” Presented at the 2002 An American Waterloo.” Chronicle of How They Measure and Manipulate Public Midwest Political Science Association Con- Higher Education, November 22, A56. Opinion in America. New York: Four Walls ference, Chicago. Lavrakas, Paul J., Michael W. Traugott, and Eight Windows. Doob, Leonard W. 2001. “Tropical Weather and Peter V. Miller, eds. 1995. Presidential Polls Morin, Richard. 2004. “Surveying the Damage.” Attitude Surveys.” Public Opinion Quarterly and the News Media. Boulder, CO: West- Washington Post, November 21. 32 ~3!: 423–431. view Press. Morin, Richard, and Claudia Deane. 2005. “Re- Farley, Reynolds. 2001. “Racial Residential Seg- Leal, David, Matt Barreto, Jongho Lee, and Ro- port Acknowledges Inaccuracies in 2004 Exit regation: Census 2000 Findings.” Working dolfo de la Garza. 2005. “The Latino Vote in Poll.” Washington Post, January 20, A6. Paper. University of Michigan Population the 2004 Election.” PS: Political Science Robinson, Claude E., Robert E. Chaddock, and Studies Center. enceladus.isr.umich.edu0race0 and Politics 38 ~January!: 41–49. Columbia University Council for Research in racestart.asp Levy, Mark R. 1983. “The Methodology and the Social Sciences. 1932. Straw Votes, a Fiorina, Morris, and Jon Krosnick. 2005. “The Performance of Election Day Polls.” Public Study of Political Prediction: A Study of Economist0YouGov Internet Presidential Opinion Quarterly 47 ~1!: 54–67. Political Prediction. New York: Columbia Poll.” Liddle, Elizabeth. 2005. “Edison0Mitofsky Exit University Press. Freeman, Donald M. 1969. “A Note on Inter- Polls 2004: Differential Non-response or Sobel, Russel S., and Robert A. Lawson. 2001. viewing Mexican-Americans.” Social Science Vote-count Corruption?” Available at “The Effect of Early Media Projections on Quarterly 49 ~4!: 909–918. www.geocities.com0lizzielid0WPEpaper.pdf Presidential Voting in the Florida Panhandle.” Groves, Robert, and Mick Couper. 1998. Non- Logan, John. 2002. “Hispanic Populations and www.be.wvu.edu0div0econ0work0pdf_files0 response in Household Interview Surveys. Their Residential Patterns in the Metropolis.” 01-07.pdf New York: Wiley-Interscience Working Paper. Lewis Mumford Center for Sudman, Seymour. 1996. “Do Exit Polls Influ- Hurtado, Aida. 1994. “Does Similarity Breed Comparative Urban and Regional Research. ence Voting Behavior?” Public Opinion Respect: Interviewer Evaluations of Mexican- May 8. mumford1.dyndns.org0cen20000 Quarterly 50 ~3!: 331–339. Descent Respondents in a Bilingual Survey.” HispanicPop0HspReportNew0 Traugott, Michael W., and Vincent Price. 1992. Public Opinion Quarterly 58~1!: 77–95. MumfordReport.pdf “A Review: Exit Polls in the 1989 Virginia Jackson, John E. 1983. “Election Night Report- Mitofsky, Warren. 1989. “Presidential Address. Gubernatorial Race: Where Did They Go ing and Voter Turnout.” American Journal of Methods and Standards: A Challenge for Wrong?” Public Opinion Quarterly 56 ~2!: Political Science 27 ~4!: 615–635. Change.” In Proceedings of the Forty-Fourth 245–253. Karp, Jeffrey, and Susan Banducci. 2000. “Going Annual Conference of the American Associa- Traugott, Michael W., and Paul J. Lavrakas. Postal: How All Mail Elections Influence tion for Public Opinion Research. Public 1999. The Voter’s Guide to Election Polls. Turnout.” Political Behavior 22 ~3!: 223–239. Opinion Quarterly 53 ~3!: 443–466. New York: Chatham House Publishers. Keating, Dan, and Dan Balz. 2001. “Florida Re- _. 1998. “Review: Was 1996 a Worse Year Weeks, Michael F., and R. Paul Moore. 1981. counts Would Have Favored Bush: But Study for Polls than 1948?” Public Opinion Quar- “Ethnicity-of-Interviewer Effects on Ethnic Finds Gore Might Have Won Statewide Tally terly 62 ~2!: 230–249. Respondents.” Public Opinion Quarterly 45 of All Uncounted Ballots.” Washington Post, _. 1999. “Pollsters.com” Public Perspec- ~2!: 245–249. November 12. tive ~June0July!. Welch, Susan, John Corner, and Michael Stein- Keeter, Scott, Carolyn Miller, Andrew Kohut, _. 2003. “ After the man. 1973. “Interviewing in a Mexican- Robert M. Groves, and Stanley Presser. Fall.” Public Opinion Quarterly 67 ~1!: American Community: An Investigation 2000. “Consequences of Reducing Non Re- 45–58. of Some Potential Sources of Response sponse in a National Telephone Survey.” _. 2005. “Evaluation of Edison0Mitofsky Bias.” Public Opinion Quarterly 37 ~1!: Public Opinion Quarterly 64~2!: 125–148. Election System 2004.” Prepared by Edison 115–126. Konner, Joan, James Risser, and Ben Wattenberg. 2001. “Television’s Performance on Election

PSOnline www.apsanet.org 483