Government Department
Cornell University 214 White Hail Ithaca, New York 14853-7901 College of Arts and Sciences t. 607.255.3549 f. 607.255.4530 [email protected] falcon.arts.cornell.edu/Govt 308 White Hall Cornell University Ithaca NY 14853 garcia.riosScornell.edu
September 6, 2017
Anthony M. Cerreto Village Attorney Village of Port Chester 222 Grace Church Street Port Chester, NY 10753 Email: tcerretoOportchesterny. com
Dear Mr. Cerreto, Please find our response to the Request For Qualifications (RFQ) in response to Village of Port Chester RFQ #2017-02 Alternative Governance Options for Consideration in Future Village Trustee Elections. We are a group of political scientists with current affiliation at U.S. research universities who have expertise and experience in the ability to collect, ana lyze, and interpret vote records and election results; work with U.S. Census data; analyze racial and ethnic voting patterns; develop and make recommen dations for policy and local election systems. We have worked on redistricting research both as consultants and expert witnesses to asses polarized voting and to draw districts in Washington State, California, Texas, and Florida. We have worked on behalf of plaintiffs and defendants so we appreciate and understand both sides of the VRA issues,
Our extensive research experience and consulting expertise ensure we are well positioned to evaluate the current system of cumulative voting instituted by the consent decree in 2009 and provide insight into what, if any, governance options would be better suited for the Village of Port Chester while still ensuring compliance with Section 2 of the Voting Rights Act. This team is lead by Dr. Matt Barreto, professor of Political Science and Chicana/o studies at UCLA. Dr. Barreto's research examines the political participation of racial and ethnic minorities in the United States. He has published 57 academic articles and book chapters on the topics of race, eth nicity and politics and his work has received over 2,250 citations according to Google Scholar. At the University of Washington, Dr. Barreto taught a course on the Voting Rights Acts in the Law School. In addition to his
Cornell University is an equal opportunity, affirmative action educator and entploycr research on Latino voting patterns, Dr. Barreto has conducted extensive re search on voting rights, and has been an expert witness in numerous Voting Rights Act lawsuits. In 2012, he was qualified as an expert witness in Ro driguez V. Harris County, a Section 2 voting rights lawsuit regarding County Commission redistricting, where he provided a report and testimony on vote dilution and racially polarized voting with respect to Latino candidates and he has testified many times in court about racially polarized voting in a va riety of cases. He also served as an expert witness in the 2012 Pennsylvania voter identification lawsuit Applewhite v. Commonwealth of Pennsylvania. In 2014 Dr. Barreto provided an expert report and testify in Veasey v. Ferry in a challenge to the Texas voter ID law, and a Federal Court struck down the Texas ID law as unconstitutional, in part basing her decision on the ev idence presented by Dr. Barreto. In 2011, Dr. Dr. Barreto was retained as the lead expert consultant for the State of California's Citizen Redistricting Commission, and was specifically asked to advise the Commission on Section 2 and Section 5 of the Federal Voting Rights Act and conduct research on polarized voting and vote dilution. He continues to actively research voting rights in California in Latino, Asian American and immigrant communities. In 2011 Dr. Barreto served as the lead expert on the State of California Citizens Redistricting Commission where he helped the commission ensure compliance with Section 2 of the Voting Rights Act.
Also included in this team is Dr. Sergio 1. Garcia-Rios, assistant professor of Government and Latino Studies at Cornell University where he teaches courses on race, ethnicity and statistical methods for the social sciences. As a faculty member at Cornell University, Dr. Garcia-Rios is very interested and well equipped to handle questions related to New York state. Dr. Garcia- Rios has served as an expert witness for voting rights cases. His substantive research focuses on Latino participation and voting behavior, in particular examining immigrants and second generation Latinos. Dr. Garcia-Rios's dissertation and book manuscript extensively examines voting behavior and practices among Latinos, with a focus on naturalized immigrants. He has published numerous articles on voting behavior of Latinos. He is also one of the lead developers of eiCompare, an open-source software package to detect, and understand racially polarized voting. He has worked as a research consultant on voting rights cases. Dr. Garcia-Rios along with Dr. Barreto are currently serving as expert witnesses in a voting rights case in the Southern U.S.'
Assisting Dr. Barreto and Dr. Garcia-Rios is Bryan Wilcox-Archuleta, a Ph.D. candidate in Political Science and M.S. candidate in Statistics at
'Specifics are redacted due to confidentially. UCLA. Mr. Wilcox-Archuleta has assisted on numerous voting rights cases where he analyzed and interpreted election results, voter records, and racially polarized voting in California and Texas with advanced statistical and quan titative methods. Dr. Barreto and Mr. Wilcox-Archuleta recently advised an advocacy group in El Cajon, CA working with the city of El Cajon to help draw new districts that were in the interest of the entire city and made sure all subgroups were well represented. In 2016 and 2017, he was part of a small team of researchers who wrote extensively to determine Latino voting patterns in the 2016 presidential election across numerous states. In 2017 he and Dr. Barreto published an article where they analyzed 39,118 individual voting precincts across 10 states. On these projects, he used his expertise in statistics, big data, voting records, and ecological regression to better under stand individual level voting patterns from aggregate data. As a consultant, he is currently working on two active voting rights cases where he applies his research and statistical skills. He has taken coursework in spatial statistics, machine learning, GIS, and other advanced quantitative methods. Included in this RFQ is a statement of qualifications which contain 1) current Curriculum Vitae for each member of the team. These document all relevant skills, publications, and consulting experience; and 2) Work samples that demonstrate the skills and experience of the team. PLEASE NOTE: The inclusion of work samples is limited since much of the work we have done is protected under attorney client privilege (cases that were settled) or part of ongoing litigation efforts and not yet publicly available. As such, we cannot include the most recent examples of work we have conducted. We included four examples from past settled cases where Dr. Barreto was the lead author. Dr. Sergio Garica-Rios worked as an assistant on these cases. We also include two recent peer reviewed published papers that highlight our competence and experience. Other publications and the discussion of recent work as permitted is available on request. Finally, we include the contact information for two references who are familiar with our work and can comment on our qualifications and experience.
Our team is fully capable of performing the work outlined in the RFQ and working closely with the Village of Port Chester, the County of Westchester, and the State of New York to ensure compliance with section 2. We want to ensure that the next governance option in the Village of West Chester ensures a fair and proportional voting system for all residents. Please direct all correspondence to Dr. Sergio Carcia-Rios. Best,
ergiol Garcia^Rios, Ph
Matt A. Barreto, Ph.D. Bryan Wilcox-Archuleta References Village of Port Chester RFQ #2017-02 Alternative Governance Options for Consideration in Future Village Trustee Elections Reference 1: Chad Dunn, Esq. Brazil and Dunn Attorneys at Law chadQbrazilanddunn.com Tel: (281) 580-6310 Fax: (281) 580-6362 4201 Cypress Creek Pkwy, Suite 530 Houston, Texas 77068
Reference 2: Justin Levitt, JD/MPA. Professor of Law Associate Dean for Research justin.levittOlls.edu Tel: (213) 736-7417 Fax: (213) 380-3769 919 Albany St. Los Angeles, CA 90015
September 2, 2017 Contributed research article
eiCompare: Comparing Ecological Inference Estimates across El and ELRxC by Loren Collingwood, Kassra Oskooii, Sergio Garcia-Rios, and Matt Barreto
Abstract Social scientists and statisticians often use aggregate data to predict individual-level behavior because the latter are not always available. Various statistical techniques have been developed to make inferences from one level (e.g., precinct) to another level (e.g., individual voter) that minimize errors associated with ecological inference. While ecological inference has been shown to be highly problematic in a wide array of scientific fields, many political scientists and analysis employ the techniques when studying voting patterns. Indeed, federal voting rights lawsuits now require such an analysis, yet expert reports are not consistent in which type of ecological inference is used. This is especially the case in the analysis of racially polarized voting when there are multiple candidates and multiple racial groups. eiCompare was developed to easily assess two of the more common ecological inference methods: the El method developed by King (1997), and the EI:RxC method developed by Rosen et al. (2001); Lau et al. (2006). The package facilitates a seamless comparison between these methods so that scholars and legal practitioners can easily assess the two methods and whether they produce similar or disparate findings.
Introduction
Ecological inference is a widely debated methodology for attempting to xmderstand individual, or micro behavior from aggregate data. Ecological inference has come under fire for being unreliable, especially in the fields of biological sciences, ecology, epidemiology, public health and many social sciences. For example, Freedman (1999) explains that when confronted with individual level data, many ecological aggregate estimates in epidemiology have been proven to be wrong. In the field of ecology Martin et al. (2005) expose the problem of zero-inflation in studies of the presence or absence of specific species of different animals and note that ecological techniques can lead to incorrect inference. Greenland (2001) describes the many pitfalls of ecological inference in public health due to the nonrandomization of social context across ecological units of analysis. Elsewhere, Greenland and Robins (1994) have argued that the problem of ecological confounder control leads to biased estimates of risk in epidemiology. Related, Frair et al.(2010) argue that while some ecological analysis can be informative when studying animal habitat preference, existing methods of ecological inference provide imprecise information on variation in the outcome variables and that considerable improvements are necessary. Wakefield (2004) provides a nice comparison of how ecological inference performs across epidemiological versus social scientific research. He concludes that in epidemiological applications individual-level data are required for consistently accurate statistical inference. However, within the narrow subfield of racial voting patterns in American elections ecological inference is regularly used. This is especially common in scholarly research on the voting rights act where the United States Supreme Courts directly recommended ecological inference analysis as the main statistical method to estimate voting preference by racial group (e.g. Thomburg v. Gingles 478 U.S. 30, 1986). Because Courts in the U.S. have so heavily relied on ecological inference, it has gained prominence in political science research. The American Constitution Society for Law and Policy explains that ecological inference is one of the three statistical analyses that must be performed in voting rights research on racial voting patterns.^ As ecological inference evolved a group of scholars developed the eiPack for the software R (R Core Team, 2015) and published an article in R News announcing the new package (Lau et al., 2006). This article does not conclude that ecological inference is appropriate or reliable outside the specific domain of American elections. Indeed,scholars in the fields of epidemiology and public health have correctly pointed out the limitations of individual level inference from aggregate date. However,its application to voting data in the United States represents one area where it may have utility, if model assumptions are met(Tam Cho and Gaines, 2004). Indeed, the main point of our article is not to settle the debate on the accuracy of ecological inference in the sciences writ large, but rather to assess the degree of similarity or difference with respect to two heavily used R packages within the field of political science, ei and eiPack. Our package, eiCompare offers scholars who regularly use ecological inference in analyses of voting patterns the ability to easily compare, contrast and diagnose estimates across two different ecological methods that are recommended statistical techniques in voting rights litigation. Today, although there is continued debate among social scientists (Greiner, 2007, 2011; Cho,1998) - 'http:// www.acslaw.org/sites/default/files/VRI_Guide_to_Section_2_Litigation.pdf
The R Journal Vol. XX/YY, A A A A 20ZZ ISSN 2073-4859 Contributed research article
the courts generally rely on two statistical approaches to ecological data. The first, ecological inference (El), developed by King (1997) is said to be preferred when there are only two racial or ethnic groups, and ideally only two candidates contesting office. However, Wakefield (2004) notes that El methods can be improved with the use of survey data as Bayesian priors. The second, ecological inference R x C(RxC) developed by Rosen et al. (2001), is said to be preferred when there are multiple racial or ethnic groups, or multiple candidates contesting office. However,it is not clear that when faced with the exact same dataset, they would produce different results. In one case, analysis of the same dataset across multiple ecological approaches found they tend to produce the same conclusion (Grofman and Barreto, 2009). However, others have argued that using King's El iterative approach with multiple racial groups or multiple candidates will fail and should not be relied on (Ferree, 2004). Still others have gone further and stated that El cannot be used to analyze multiple racial group or multiple candidate elections, stating that "it biases the analysis for finding racially polarized voting," going on to call this approach "problematic and no valid statistical inferences can be drawn"(Katz, 2014). As with any methodological advancement, there is a healthy and rigorous debate in the literature. However, very little real election data has been brought to bear in this debate. Ferree (2004) offers a simulation of Black, White, and Latino turnout and voting patterns, and then examines real data from a parliamentary election in South Africa using a proportional representation system.(Grofman and Barreto, 2009) compare an exit poll to precinct election data in Los Angeles, but only compare Goodman's ecological regression against King's El, using the single-equation versus double-equation approach, and do not examine the RxC approach at all.
Debates Over Ecological Inference
The challenges surrounding ecological inference are well documented. Robinson (2009) pointed out that relying on aggregate data to infer the behavior of individuals can result in the ecological fallacy, and since then scholars have applied different methods to discern more accurately individual correlations from aggregate data. Goodman (1953,1959) advanced the idea of ecological regression where individual patterns can be drawn from ecological data under certain conditions. However Goodman's logic assumed that group patterns were consistent across each ecological unit, and in reality that may not be the case. Eventually, systematic analysis revealed that these early methods could be unreliable (King, 1997). Ecological inference is King's (1997) solution to the ecological fallacy problem inherent in aggregate data, and since the late 1990s has been the benchmark method courts use in evaluating racial polarization in voting rights lawsuits, and has been used widely in comparative politics research on group and ethnic voting patterns. Critics claim that King's El model was designed primarily for situations with just two groups (e.g., blacks and whites; Hispanics and Anglos, etc.). While many geographic areas (e.g., Mississippi, Alabama) still contain essentially two groups and hence pose no ^eat to traditional El estimation procedures, the growth of racial groups such as Latinos and Asians have challenged the historical biracial focus on race in the United States (thereby challenging traditional El model assumptions). Rosen et al.(2001) suggest a rows by columns(RxC) approach which allows for multiple racial groups, and multiple candidates; however, their Bayesian approach suffered computational difficulties and was not employed at a mass level. Since then,computing power has steadily improved, making RxC a realistic solution for many scenarios and accessible packages now exist in R that are widely used. These two methodological approaches are now both regularly used in political science; however, there is no consistent evidence how they perform side-by-side, and are different. Ferree (2004) critiques King's El model, arguing that the conditions for iterative estimation (e.g., black vs. non black, white vs. non-white, Hispanic vs. non-Hispanic) can be considerably biased due to aggregation bias and multimodality in the data. In a hypothetical simulation dataset, Ferree shows that combining blacks and whites into a single "non-Hispanic" group in order to estimate Hispanic turnout can vastly overestimate Hispanic turnout, for example. However, the analysis did not provide any clues as to the specific conditions when and how RxC is significantly better or preferred to El. For example, if there are three racial groups in equal thirds of the electorate, does aggregation bias create more error in El than a scenario in which two dominant groups comprise 90% and a small group is just 10%? Likewise, is El's iterative approach to candidates more stable when analyzing three candidates and far less stable when eight candidates contest the election? These questions have not been considered empirically. Instead, the existing scholarship uses simulation data to prove theoretically that El might create bias and that RxC is preferred. We argue that real election data should be considered in a side-by-side comparison. Despite some critiques, other political scientists have defended ecological inference and even ecological regression using both simulations and real data. Owen and Grofman (1997) assess whether or not ecological fallacy in ecological regression is a theoretical problem only, a real problem for
The R Journal Vol. XX/YY, AAAA 20ZZ ISSN 2073-4859 Contributed research article
empirical analysis. In an extensive review, Owen and Grofman conclude that despite the valid theoretical concerns, linear ecological regression still holds up and provides meaningful and accurate estimates of racially polarized voting. A decade later, Grofman and Barreto (2009) again take up the question of how ecological models compare to one another using a combination of simulation, actual election precinct data, and an accompanying individual-level exit poll. Their analysis argues that there is general consistency across all ecological models and that once voter turnout rates are accounted for, ecological regression and King's El lead scholars to the same results. However, Grofman and Barreto did not consider RxC in their comparison. Greiner and Quinn (2010) combine RxC methods with individual level exit poll data, and argue that this hybrid model can be preferable to a straight aggregation model. However, using exit poll data is not always available to all researchers and practitioners. Indeed,in most county or city elections, exit poll data does not exist which is why scholars often attempt to infer voting pattems through aggregate data. Herron and Shotts (2003) also criticize El estimates when used for second-stage regression - given that error is baked into the second-level regression estimation. However Adolph and King (2003) respond by adjusting the El procedure to reduce inconsistencies when estimating second-stage regressions. Nevertheless, these issues with El do not speak specifically to RxC methods. Greiner and Quinn (2009) extend the 2x2 El contingency problem to 3x3 and estimate voting preferences simultaneously for three candidates across three racial groups (but using counts instead of percentages). We extend this work by analyzing real-world datasets with sizes greater than 3x3 (multiple candidates and at least three racial groups). In all of this, our main goal is to assess whether using iterative El or simultaneous RxC approaches change the conclusions social scientists can make from the data. Finally, some have gone even further in arguing that El is ill-equipped to handle complex datasets with multiple candidates and multiple racial groups, and that only RxC can produce reliable results (Katz, 2014). In explaining the theoretical reasons why El cannot accurately process such elections Katz argues "adding additional groups and vote choices to King's (1997) El is not straightforward," and also adds "given the estimation uncertainty, it may not be possible to infer which candidate is preferred by members of the group." The argument against El in multiple racial group, or especially multiple candidate elections is that El takes an iterative approach pitting candidate A versus all others who are not candidate A. If the election features four candidates(A, B, C,D) critics state that you cannot accurately estimate vote choice quantities if you compare the vote for candidate A against the combined vote for B, C,D. The iterative approach would then move on to estimate the vote share for candidate B against the combined vote for A,C, D and so on,so that four separate equations are run. Katz(2014) claims that El biases the findings in favor of bloc-voting stating "this jerry rigged approach to dealing with more than two vote choices stacks the deck in favor of finding statistical evidence for racially polarized." Given these debates, our package allows scholars to quite easily make side-by-side comparisons and evaluate these competing claims. While important advancements have been made in ecological inference techniques by King (1997) and Rosen et al.(2001) there is no consistency in which technique is used and how results are presented. What's more, legal experts and social scientists often argue during voting rights lawsuits that one technique is superior to the other, or that their results are more accurate. There is no question that both social scientists and legal experts would greatly benefit from a standardized software package that presents both ecological inference results (El and RxC)simultaneously and metrics to compare each set of results. Thus, eiCompare was designed to compare the most commonly used methods today. El and RxC,but also incorporates Goodman methods. The package lets analysts seamlessly assess whether El and RxC estimates are similar(see King (1997) and Rosen et al.(2001) for a methodological description of the techniques). It incorporates functions from ei (King and Roberts, 2013) and eiPack (Lau et al., 2012) into a new package that relatively quickly compares ecological inference estimates across the two routines. The package includes several functions that ultimately produce tables of results from the different ecological inference methods. Thus, in the case of racially polarized voting, analysts can quickly assess whether different racial groups preferred different candidates, according to the El, RxC,and Goodman approaches. eiCompare wraps the ei procedure(King and Roberts, 2012) into a generalized function, has a variety of table-making functions, and a plotting method that graphically depicts the difference between estimates for the two main El methods(El and RxC). Below, we use a working example of a voter precinct dataset in Corona, CA. To use the package, the process is simple: 1) Load the package, the appropriate data, run the El generalized function, and create an El table of results, 2) Run the RxC function (from eiPack) and create a table of results, 3) Run the Goodman regression generalized function if the user chooses,4) Combine the results of all the algorithms together into a comparison table, and 5) Plot the comparison results. Before we conclude, we also compare El and RxC findings against exit poll data from a 2005 Los Angeles mayoral run-off election. The rest of the paper follows this aforementioned outline.
The R Journal Vol. XX/YY, AAAA 20ZZ ISSN 2073-4859 Contributed research article
1. El Generalize
To begin, we install (install, packages(^ ^eiCompare'')) and load the eiCompare package (library(eiCompare))from theCRAN repository. First, we load the aggregate-level dataset(data(cor_06)) into R, in this case a precinct (voting district) dataset from a 2006 election in the city of Corona, CA. Table 1 below displays the first five rows and column headers of the dataset. This dataset includes all the necessary variables to run the code in the eiCompare package. The first column is precinct, which essentially operates as a unique identifier. The second column, totvote, is the total number of votes cast within the precinct. Columns three and four are the two racial groups of whom we seek to determine their mean voting preference. The rest of the columns are the percent of the total vote for each respective candidate.
precinct totvote pctjatino pct_other pct_breitenbucher pct_montanez pct_spiegel pct_skipworth 1 22000 942 0.21 0.79 0.20 0.21 0.29 0.30 2 22002 1240 0.16 0.84 0.22 0.22 0.29 0.27 3 22003 1060 0.21 0.79 0.22 0.22 0.30 0.26 4 22004 1280 0.45 0.55 0.18 0.27 0.30 0.24 5 22008 1172 0.31 0.69 0.23 0.25 0.30 0.22 6 22012 1093 0.21 0.79 0.20 0.24 0.32 0.24
Table 1: Precinct dataset of Corona, CA, used for ecological inference. Each row is a precinct, the dataset must have a total column, racial/ethnic percentages of people living in the precinct, and vote percent for each candidate.
We are interested in how the four candidates (Breitenbucher, Montanez, Spiegel, Skipworth) performed with Latino voters and non-Latino voters (mostly non-Hispanic white), so we can asses whether racially polarized voting exists. The process begins with the ei_est_gen() function, which is a generalized version of the ei function from the ei package. Instead of having to estimate El results for each candidate and each racial group separately, ei_est_gen() automates this process. The ei_est_gen() function takes a vector of candidate names (e.g., c("pct_breitenbucher", "pct_montanez","pct_spiegel", "pct_skipworth")), a character vector of racial group names with a tilde in front of the variable name (e.g., c("~pct_latino","~pct_other")), a character string of the name of the total column ("totvote") representing the total number of people in the jurisdiction (e.g., registered voters, ballots cast) that is passed to the ei function, a data call for the data.f rame() object where the data are stored, and a character string of table_names (e.g., c("EI: Pet Lat","EI: Pet Other")) that are used to display the results. The function also has four default arguments, rho, sample, tomog, and density_plot. The former two can be used to adjust the parameters of the ei algorithm. These are especially useful when the initial run does not compile or warnings are produced. The latter two plot out tomography and density plots, respectively into the working directory but are default set to off. These plots can be used to assess the stability - and thus veracity - of the EI procedure (see King and Roberts (2012) and King (1997) for details). Finally, the ... argument passes additional arguments onto the ei() function from the ei package. One final note, given its iterative nature, the ei_est_gen() function can take a while to execute. This typically depends on features unique to the dataset, including the number of candidates and groups, the amount of racial/ethnic segregation within the city/area, as well as the number of precincts. This particular example does not take especially long, executing in about a minute on a standard Macbook pro.
# LOAD DATA data(cor_06) # SET SEED FOR REPRODUCIBILITY set.seed(294271) # CREATE CHARACTER VECTORS REQUIRED FOR FUNCTION cands <- c("pct_breitenbucher","pct_montanez","pct_spiegel", "pct_skipworth") race_group2 <- c("~ pct_latino", "*• pct_other") table_names <- c("EI: Pet Lat", "EI: Pet Other") # RUN EI GENERALIZED FUNCTION results <- ei_est_gen(cand_vector=cands, race_group = race_group2, total = "totvote", data = cor_06, table_names = table_names) # LOOK AT TABLE OF RESULTS results The call to the results object produces a table of results indicating the mean estimated voting preferences for Latinos and non-Latinos within the city of Corona (see Table 2). The results strongly suggest the presence of racially polarized voting, as Latinos prefer Montanez as their number one choice, whereas non-Latinos do not.
The R Journal Vol. XX/YY, AAAA 20ZZ ISSN 2073-4859 Contributed research article
Candidate El: Pet Lat El: Pet Other pct_breitenbucher 19.68 21.12
se 0.75 0.13 pct_montane2 35.95 20.13
se 0.03 0.08 pct_spiegel 28.43 31.01
se 0.57 0.23 pct_skipworth 18.64 26.84
se 0.71 0.23 Total 102.69 99.10
Table 2: El mean estimates for Latino and Non-Latino candidate vote preferences in Corona, 2006
2. EI:RxC
The RxC builds off of code from the eiPack package, where eiCompare simply takes the former's results and puts them into a similar data.frame()/table() object similar to the results from the ei_est_gen() function. First, the user follows the code from the eiPack package (here we use the ei. reg. bayesO function), and creates a formula object including all candidates and all groups. The user must ensure that the percentages on both signs of the ~ symbol add to 1. Thus,the initial table() code is a simple data check to ensure that this rule is followed. The RxC model is then run using the ei. reg. bayes() model. Users can read the eiPack documentation to familiarize themselves with this procedure. Depending on the nature of one's data, the RxC code can take a while to run. Finally, the results are passed onto the bayes_table_make() function, along with a vector of candidate names, and a vector of table names,similar to what was passed to ei_est_gen().
# CHECK TO MAKE SURE DATA SUMS TO 1 FOR EACH PRECINCT with(cor_06, pct_latino + pct_other) with(cor_06, pct_breitenbucher + pct.montanez + pct_spiegel + pct_sklpworth) # SET SEED FOR REPRODUCIBILITY set.seed(124271) #RxC GENERATE FORMULA form <- formula(cbind(pct_breitenbucher,pct_montanez, pct_spiegel, pct_skipworth) ~ cbind(pct_latino, pct_other)) # RUN EI:RxC MODEL ei_bayes <- ei.reg.bayes(form, data=cor_06, sample=10000, truncate=T) # CREATE TABLE NAMES table.names <- c("RxC: Pet Lat", "RxC: Pet Other")
# TABLE CREATION ei_bayes_res <- bayes_table_make(ei_bayes, cand_vector= cands, table_names = table_names) # LOOK AT TABLE OF RESULTS ei_bayes_res
Candidate RxC: Pet Lat RxC: Pet Other pct_breitenbucher 18.22 21.58
se 1.62 0.53 pct_montanez 34.96 20.44
se 1.72 0.56 pct_spiegel 28.24 31.05
se 1.08 0.35 pct_sklpworth 18.61 26.91
se 1.73 0.56 Total 100.03 99.99
Table 3: ELRxC mean estimates for Latino and Non-Latino candidate vote preferences in Corona, 2006
The results are presented in Table 3, and look remarkably similar to those presented in Table 2. Indeed, the exact same conclusions would be drawn from an analysis of both tables: Latinos prefer Montanez as their first choice and non-latinos prefer Spiegel as their top choice.
The R Journal Vol. XX/YY, AAAA 20ZZ ISSN 2073-4859 Contributed research article
3. Goodman Generalize
While many users will skip over the Goodman regression when conducting ecological inference, given the documented issues with the method (Shively, 1969; King, 1997), eiCompare nevertheless has a Goodman regression generalized function, similar to the ei_est_gen() function. This function takes a character vector of candidate names, a character vector of racial groups, the name of the column, a data object, and a character vector of table names. Because Goodman is simply a linear regression, the execution is very fast.
table_names <- c("Good: Pet Lat", "Good: Pet Other") good <- goodman_generalize(cands, race_group2, "totvote", cor_06, table_names) good
Table 4 shows the Goodman regression results. In this particular case, these results align quite closely with results from the two El models. All three approaches essentially tell us the same thing.
Candidate Good; Pet Lat Good: Pet Other pct_breitenbucher 17.51 20.34
se 3.18 3.74 pct_montanez 35.00 20.48
se 3.41 4.01 pct_spiegel 28.52 31.61
se 2.16 2.54 pct_skipworth 18.97 27.57
se 3.45 4.05 Total 100.00 100.00
Table 4: Goodman regression estimates for Latino and Non-Latino candidate vote preferences in Corona, 2006
4. Combining Results
The last two sections address the comparison component of the package. The function, ei_rc_good_table(), takes the objects from theEl, RxC,and Goodman regression, and puts them into a data,frame ()/table () object. To simplify comparison, the table adds an El-RxC column differential for each racial group. This format lets the user quickly assess how the El and RxC methods stack up against one another. The function takes the following arguments: ei results object (e.g., results), an RxC object (e.g., ei_bayes_res), and a character vector groups (e.g., c("Latino","Other")) argument. The good argu ment for the Goodman regression is set to Null, and the include_good argument defaults to FALSE. If the user wants to include a Goodman regression in the comparison of results they need to change the latter to TRUE and specify the the good argument as the object name from the goodman_generalize() call.
Candidate EI: Pet Lat RxC: Pel Lat ELDiff EI: Pet Other RxC: Pet Other ELDiff pct_brcitcnbucher 19.68 18.22 -1.46 21.12 21.58 0.46
se 0.75 1.62 0.13 0.53 pct_montanez 35.95 34.96 -0.99 20.13 20.44 0.31
sc 0.03 1.72 0.08 0.56 pct_spiegel 28.43 28.24 -0.19 31.01 31.05 0.04
se 0.57 1.08 0.23 0.35 pct_skipworth 18.64 18.61 -0.02 26.84 26.91 0.07
se 0.71 1.73 0.23 0.56 Total 102.69 100.03 -2.66 99.10 99.99 0.88
Table 5: EI and RxC comparisons for Latino and Non-Latino candidate vote preferences in Corona, 2006
The results of ei_rc_good_tableO is a new class ei_compare, which includes a data.frame() and groups character vector. This output is ultimately passed to plot().
ei_rc_combine <- ei_rc_good_table(results, ei_bayes_res, groups= c("Latino", "Other")) ei_rc_combine@data
The R Journal Vol. XX/YY, AAAA 20ZZ ISSN 2073-4859 Contributed research article
ei_rc_g_combine <- ei_rc_good_table(results, ei_bayes_res, good, groups= cC'Latino", "Other"), include_good=T) ei_rc_g_combine Table 5 displays the output of a call to the ei_rc_good_table() function for the first line of code above. The user must include the code ©data onto the outputted table name to extract just the table. This table basically summarizes the results of the El and RxC analyses. Clearly, very little difference emerges between the two methods in this particular instance.
5. Plotting Results
Finally, users can plot the results of the El, and RxC comparison to more visually determine whether the two methods are similar. Plotting is simple, as plot methods have been developed for the ei_compare class. The code below produces the plot depicted in Figure 1.
# PLOT COMPARISON — adjust the axes labels slightly plot(ei_rc_combine, cex.axis=.5, cex.lab=.7)
Latino: El Diff Comparison
&ci sWpwofth—
;g pCl_SpiegOI —
pel momanoz—
pcl.brolionbochor—
RxC - El
Other: El Diff Comparison
pct_sklpwonh—
pct_#plogol—
pcLmonlanez—
pct.bfcitonbuchor—
RxC - El
Figure 1: Comparison of El and RxC methods for Corona 06 precinct data
The R Journal Vol. XX/YY, AAAA 20ZZ ISSN 2073-4859 Contributed research article
Comparing Ecological and Individual-Level Data
One possible question remains, whether or not ecological estimates line up with individual level estimates. Many studies have pointed out that ecological fallacy and aggregation bias can produce ecological inference results that are highly questionable. In this section we implement the eiCompare package for a mayoral election in a multiethnic setting in which an individual-level exit poll survey was also administered. The eiCompare package provides El and RxC results for the 2005 Los Angeles mayoral runoff election between Antonio Villaraigosa and James Hahn,and we also add results for the Los Angeles Times exit poll. Results arc displayed in Table 6.
El: AV EI:JH RxC: AV RxCJH Exit: AV Exit: JH MOE White 45 54 48 52 50 50 +/- 2.5 Black 58 40 50 50 48 52 -H/-4.2 Latino 82 17 81 19 84 16 +/-3.6 Asian 48 51 47 53 44 56 +/-6.1 Table 6: Percent voting for Antonio Villaraigosa (AY)and James Hahn QH)by ethnic group. Compari son between El, RxC,and exit poll methods, Los Angeles mayoral election nanoff. May 2005. Exit poll taken from Los Angeles Times.
The results presented in Table 6 demonstrate that not only do El and RxC produce remarkably consistent results, but they very closely match the individual level estimates for the Los Angeles Times. The El RxC estimates are all with the confidence range of the individual level data reported by the exit poll.
Summary
eiCompare is a new package that builds on the work of King and others that attempts to address the ecological inference problem of making individual-level assessments based on aggregate-level data. As we have reviewed above, there is considerable debate in the sciences about the utility and accuracy of ecological techniques. Despite these well documented questions, ecological inference is widely used in political science and will continue to grow in importance when the constitutionally mandated redistricting in 2021 occurs. The redistricting cycle will bring with it extensive academic, legislative, and legal research using ecological inference to assess racial voting patterns across all 50 states. While this new package does not develop a new method, per se, it improves analysts' ability to quickly compare different commonly used El algorithms to assess the veracity of the methods and also produce tables of their findings. While RxC has been touted as the method necessary in situations with multiple groups and multiple candidates, the results do not always demonstrate face validity. In these scenarios - and others - analysts may want to incorporate original El methods so they can compare how the two approaches stack up. Ultimately, this approach provides a needed assessment between two commonly used methods in voting behavior research.
Bibliography
C. Adolph and G. King. Analyzing second-stage ecological regressions: Comment on Herron and Shotts. Political Analysis, \1(\):65-76, 2003. [p3] W. K. T. Cho. Iff the assumption fits?: A comment on the King ecological inference solution. Political Anfl/ys/s,7(1):143-163,1998. [pi] K. E. Ferrec. Iterative approaches to RxC ecological inference problems: where they can go wrong and one quick fix. Political Analysis, 12(2):143-159, 2004. [p2] J. L. Frair, J. Fieberg, M. Hebblewhite, F. Cagnacci, N. J. DeCesare, and L. Pedrotti. Resolving issues of imprecise and habitat-biased locations in ecological analyses using GPS telemetry data. Philosophical Transactions of the Royal Society of London B: Biological Sciences, 365(1550):2187-2200, 2010. [pi] D. A. Freedman. Ecological inference and the ecological fallacy. International Encyclopedia of the social & Behavioral sciences, 6:4027-4030,1999. [pi] L. A. Goodman. Ecological regressions and behavior of individuals. American sociological revieiv, 1953. [p2]
The R Journal Vol. XX/YY, AAAA 20ZZ ISSN 2073-4859 Contributed research article
L. A. Goodman. Some alternatives to ecological correlation. American Journal of Sociology, pages 610-625,1959. Ip2] S. Greenland. Ecologic versus individual-level sources of bias in ecologic estimates of contextual health effects. International journal of epidemiology, 30(6):1343-1350, 2001. [pi] S. Greenland and J. Robms. Invited commentary: ecologic studies — biases, misconceptions, and counterexamples. American Journal of Epidemiology, 139(8):747-760,1994. [pi] D. J. Greiner. Ecological inference in voting rights act disputes: Where are we now,and where do we want to be? /wnmefncs, pages 115-167, 2007. [pi] D. J. Greiner. The quantitative empirics of redistricting litigation: Knowledge, threats to knowledge, and the need for less districting. Yale Law & Policy Review, 29(2):527-542,2011. [pi] D. J. Greiner and K. M. Quinn. Exit polling and racial bloc voting: Combining individual-level and RxC ecological data. The Annals of Applied Statistics, pages 1774-1796,2010. [p3] J. D. Greiner and K. M. Quinn. Rx c ecological inference: bounds, correlations, flexibility and transparency of assumptions. Journal of the Royal Statistical Society: Series A (Statistics in Society), 172 (1):67-81,2009. [p3] B. Grofman and M. A. Barreto. A reply to Zax's (2002) critique of Grofman and Migalski (1988) double-equation approaches to ecological inference when the independent variable is misspccified. Sociological Methods & Research, 37(4):599-617,2009. [p2,3] M. C. Herron and K. W. Shotts. Using ecological inference point estimates as dependent variables in second-stage linear regressions. Political Analysis, ll(l):44-64,2003. [p3] J. N. Katz. Expert report on voting in the city of Whittier. March 5,2014. [p2,3] G. King. A solution to the ecological inference problem, 1997. [pi, 2,3,4,6] G. King and M. Roberts. Ei: a (n r) program for ecological inference. Harvard University. Retrievedfrom http://gking. harvard, edu/files/ei. pdf, 2012. [p3,4] G.King and M. Roberts, ei: ei,20l3. URL http://gking.harvard.edu/zelig. R package version 1.3. [p3] O. Lau, R. T. Moore, and M. Kellermann. eipack: RxC ecological inference and higher-dimension data management. Hew Functionsfor Multivariate Analysis, 18(1):43, 2006. [pi] O. Lau, R. T. Moore, and M. Kellermann. eiPack: eiPack: Ecological Inference and Higher-Dimension Data Management, 2012. URL http://CRAN.R-project.org/package=eiPack. R package version 0.1-7. [p3] T. G. Martin, B. A. Wintle, J. R. Rhodes, P. M. Kuhnert, S. A. Field, S. J. Low-Choy, A. J. Tyre, and H. P. Possingham. Zero tolerance ecology: improving ecological inference by modelling the source of zero observations. Ecology letters, 8(11):1235-1246, 2005. [pi] G. Owen and B. Grofman. Estimating the likelihood of fallacious ecological inference: linear ecological regression in the presence of context effects. Political Geography, 16(8):675-690,1997. [p2] R Core Team. R: A Language and Environmentfor Statistical Compitting. R Foundation for Statistical Computing, Vienna, Austria, 2015. URL https://www.R-project.org/. [pi] W. S. Robinson. Ecological correlations and the behavior of individuals. International journal of epidemiology, 38(2):337-341, 2009. [p2] O. Rosen, W. Jiang, G. King, and M. A. Tanner. Bayesian and frequentist inference for ecological inference: The RxC case. Statistica Neerlandica, 55(2):134-156, 2001. [pi, 2,3] W. P. Shively. "Ecological" inference: the use of aggregate data to study individuals. American Political Science Revieio, 63(04):1183-1196,1969. [p6]
W. K. Tam Cho and B. J. Gaines. The limits of ecological inference: The case of split-ticket voting. American Journal of Political Science, 48(1):152-171,2004. [pi] J. Wakefield. Ecological inference for 2x 2 tables(with discussion). Journal of the Royal Statistical Society: Series A (Statistics in Society), 167(3):385-445, 2004. [pi, 2]
The R Journal Vol. XX/YY, AAAA 20ZZ ISSN 2073-4859 Contributed research article 10
Loren Collingwood Department ofPolitical Science University of California, Riverside 900 University Avenue Riverside, CA 92521 USA loren.collingwoodSucr.edu
Kassra Oskooii Department ofPolitical Science University of Washington 101 Gowen Hall Seattle, WA 98195 USA kassrao0uw.edu
Sergio Garcia-Rios Department of Government Cornell University 212 White Hall Ithaca, NY 14853 USA garcia.rios0cornell.edu
Matt A. Barreto Department ofPolitical Science University of California, Los Angeles Bunche Hall 3284 Los Angeles, CA 90095 USA barretom0ucla.edu
The R Journal Vol. XX/YY, AAAA 20ZZ ISSN 2073-4859 Survey Methodology and the Latina/o Vote Why a Bilingual, Bicultural, Latino-Centered Approach Matters
Matt A. Barreto, Tyler Reny, and Bryan Wilcox'Archuleta
In March 2013, after the GOP lost what it expected to be a much closer presidential race, party chair Reince Priebus led the Republican National Committee(RNC) in conducting an autopsy of the 2012 election that they titled the "Growth and Opportunity Project." The resulting 100-page report prioritized outreach to Latino, African American, and Asian American voters. With regard to Hispanic voters, the guide suggested more welcom ing rhetoric: "If Hispanic Americans perceive that a GOP nominee or candidate does not want them in the United States (i.e. self-deportation), they will not pay attention to our [policies]. .. . We must embrace and champion comprehensive immigration reform"(RNC 2013, 8). It came as a shock, then, when Priebus and the RNC backed candidate Donald Trump, the real estate mogul whose career began with being sued for housing discrimination against African Americans (Mahler and Eder 2016); who suggested that the Central Park Five be executed for their crimes even after they were cleared of wrongdoing (Burns 2016); who led the birtherism charges against Barack Obama throughout his presidency (Barbara 2016); and who kick-started his own presidential campaign in 2015 by calling Mexicans murderers and rapists (Ye Hee Lee 2015). Trump broke with the decades-long COP strategy of implicit racial appeals, opting instead for explicitly hostile and xenophobic statements about minority groups throughout his campaign. As a result, he became the first modern Republican candidate to win the party's nomination based on racial preju dice (Tesler 2016).
Ajildn: A Joumai o/Chicano Studies 42:2 Fall 2017 © University of California Regents 209 Barreto, Reny, and WilcoX'Archuleta
Given Trump's racially insensitive rhetoric, particularly toward Latinos and immigrants, it was widely expected that Latino voter backlash would be enormous and crippling for the GOP candidate in the general election. Poll ing and reporting throughout the campaign season suggested that Hispanic voter enthusiasm was at sky-high levels and that registration was spiking (Bemal 2016; Gross 2016; O'Keefe 2016). According to pre-election polling by Latino Decisions, a firm specializing in Latino political opinion research. Democrat Hillary Clinton was situated to win a record high 79 percent of the Hispanic vote and Trump a record low 18 percent (Gross 2016). A sizable body of academic research in Chicano studies and political science similarly suggested that such blatant racial appeals would be detrimental to Trump's chances, particularly in Latino-heavy swing states. Yet on November 8, 2016, as polling place lines dwindled and Edison Research, exclusive provider of exit poll data to a consortium of media outlets known as the National Election Pool, began tallying and releasing results, a new narrative emerged.' Trump, according to the Edison Exit Poll, actually did better with Latino voters than Republican candidate Mitt Romney had done in 2012 (CNN 2016). The findings sparked a public debate between pollsters, pitting those who specialize in measuring Latino political attitudes and who estimated that Clinton's margin of victory among Latinos would exceed that of Obama's, against Edison, whose exit polling found the opposite. Complicating the picture is the fact that Edison does not immediately release its sampling methodology to the public; the release of such information can take years after an election.
Matt A. Barreto is professor of political science and Chicana/o studies at UCLA and co- founder of the research and polling firm Latino Decisions, which is widely acknowledged as the leading Latino political opinion research group in the United States. He is author of Et/inic Cues; The Role of Shared Ethnicity in Latino Political Behavior (University of Michigan Press, 2010) and co-author of Change They Can't Believe In: The Tea Party and Reactionary Politics in America (Princeton University Press, 2013) and Latino America: How America's Most Dynamic Population Is Poised to Transform the Politics of the Nation (PublicAffairs, 2014). Tyler Reny is a graduate student in political science at the University of California, Los Angeles. He studies the impact of demographic change on political attitudes and behaviors in the United States. He has recently had articles accepted for publication at Social Sciences Quarterly, Politics Research Quarterly, and Comparative Political Studies. He may be reached at [email protected]. Bryan Wilcxdx-Arculeta is a graduate student in political science and statistics at the University of California, Los Angeles, where he studies the role of identity in politics among minority populations in the United States. He has work forthcoming in Journal of Race, Ethnicity, and Politics. He can be reached at [email protected].
210 Survey Met/iodology and the Latina/o Vote
In this article, we first briefly summarize the academic literature on Latino political behavior and explain why understanding the attitudes of subgroups requires that pollsters be culturally sensitive to the populations they study. We then present a novel analysis of real vote data suggesting that Clinton did, as expected, surpass Obama's margin of victory among Latino voters. Analyzing 29,045,522 votes from 39,118 electoral precincts across ten states, we show that Latino Decisions polling was far closer to the actual vote returns than the Edison Exit Poll. We conclude by looking to the future of the Latino electorate and polling in US elections.
Awaking the Sleeping Giant Despite the size of the Latino population in the United States, geographic clustering, national origin diversity, immigration and citizenship status, and low levels of participation have long kept Latinos out of the national political spotlight (de la Garza and DeSipio 1996; DeSipio 1996; Pachon and DeSipio 1994). Yet by the early 2000s, media began to speculate on the potential political impact of the then 35 million Latinos residing within the United States, a population Time magazine and others dubbed the "sleeping giant"(Tumulty 2001). The growing Latino population in swing states like Virginia, Nevada, and Colorado, together with the now explicit and active courting of Latino votes by presidential candidates, has ensured steady coverage of Latino voters throughout every recent presidential campaign cycle (Barreto et al. 2008; Collingwood, Barreto, and Garcia-Rios 2014; Fraga and Leal 2004; Garcia and Sanchez 2008). The sheer diversity of the Latino population in the United States has prevented the emergence of a cohesive pan-ethnic voting bloc comparable to the African American vote (Barreto and Segura 2015; Sanchez and Pita 2006). However, Latino voters have long been supportive of the Democratic Party (Alvarez and Garcia Bedolla 2003; Bowler, Nicholson, and Segura 2006; Pantoja, Ramirez, and Segura 2001; Segura 2012; Tolbert and Hero 2001; Uhlaner and Garcia 2005). With the exception of 2004, when George W. Bush was able to garner about 40 percent support from Latino voters. Democratic presidential candidates have received roughly 65 to 70 percent of the Latino two-party vote in each election cycle (Barreto and Segura 2015). There are multiple reasons for this strong Democratic support. First, surveys have revealed that, despite common tropes of Latinos as "natural conservatives," Latinos generally favor a large and active federal
211 Barreto, Reny, and Wilcox-Archuleta government and arc in many ways natural Democrats (Barreto and Segura 2015). Second, there is evidence that a Latino pan-ethnic group identity may be emerging to influence both attitudes and behaviors (Sanchez 2006b; Sanchez and Masuoka 2010). For African Americans, their sense of"linked fate," a product of specific social and historical circumstances, contributes to more homogenous policy preferences and voting behaviors(Dawson 1994). In 1989, investigators of the Latino National Political Survey (LNPS) found little evidence of a similar pan-ethnic identity among Latinos in the United States (de la Garza 1992). Recent survey research, however, finds increasing identification with pan-ethnic terms (Fraga et al. 2010). While there are still questions about the durability of this incipient Latino political consciousness (Beltran 2010), scholars have found that at specific times and under certain circumstances, pan-ethnic identity can be activated and can shape political beliefs and spur mobilization (Sanchez 2006a, 2006b). Finally, as Latinos are socialized into the US political system. Democratic candidates are simply more likely than Republican candidates to reach out to and mobilize Latino voters (Collingwood, Barreto, and Garcia-Rios 2014; Nuho 2007). This consistent outreach by one party can inculcate a sense of belonging that can shape subsequent partisan attachments (Green, Palmquist, and Schickler 2002). As social psychology would predict, a number of studies have found that certain forms of threat can mobilize Latinos into a cohesive voting coalition by increasing the influence of ethnic identity on political evalu ations and behaviors (Michelson and Pallares 2001; Pantoja, Ramirez, and Segura 2001). Indeed, Efren Perez (2014) finds that "high-identifying Latinos" exposed to xenophobic rhetoric become more ethnocentric and more likely to support policies that support in-group pride. Matt Barreto and Gary Segura (2015) find that messages stressing discrimination, harassment, and racial profiling toward Latinos are among the most highly motivating (Schildkraut 2005) Numerous real-world examples show how vitriolic anti-immigrant and anti-Latino rhetoric can mobilize Latino voters and push them toward the Democratic Party. California's experience is a case in point. Throughout the 1980s and early 1990s, California's immigrant population expanded as the non-Hispanic white population shrank. For decades, California anti- immigrant groups had been pushing elites to debate immigration, and by the late 1980s their efforts began to bear fruit. As grassroots conservative movements whipped up anti-immigrant hysteria, activists gathered signa tures for a punitive anti-immigrant ballot measure (HoSang 2010, 164).
212 Survey Methodology and the Latinalo Vote
In 1994 these efforts culminated in the passage of Proposition 187, which imposed restrictions on public education, housing, and public services for undocumented Californians through changes to the state's penal code, welfare and institutions code, health and safety code, education code, and government codes. The measure's unofficial title, the "Save Our State" initiative, helped it garner wide support from the public. Postelection poll ing confirmed that anti-immigrant campaigning was particularly successful in mobilizing white voters. Fully 63 percent of white voters, 62 percent of independent voters, and 55 percent of moderates ultimately supported the measure (HoSang 2010). These policies and political rhetoric did not go unnoticed by the state's Latino population. By 1996, California Latinos were naturalizing, registering to vote, and turning out in record numbers. Adrian Pantoja, Ricardo Ramirez, and Gary Segura (2001)found that after the Proposition 187 fight, newly naturalized Latinos in California turned out at higher rates than Latinos in other states that lacked such an intensely nativist political climate. Latinos were also voting increasingly for the Democratic Party. Shaun Bowler, Stephen P. Nicholson, and Gary Segura (2006) find that Proposition 187 (together with the equally racial Propositions 207 and 229) nearly doubled the probability that California Latinos would vote Democratic (see also Barreto and Woods 2005). The Latino share of the state electorate increased from 7 percent in 1990 to 14 percent in 2000 with the addition of more than 1 million Latino voters to the rolls, accord ing to the California Field Poll.^ By 1998, Democrats had won back the California statehouse, state assembly, and state senate. By 2002, Democrats held every statewide office. This backlash—sometimes called the "Pete Wilson Effect," after the California governor who championed the punitive anti-immigrant propositions—is not limited to California. In Nevada, Sharron Angle's racially charged and vitriolic anti-immigrant appeals destroyed any chance she had of unseating the incumbent senator and majority leader Harry Reid during her 2010 bid for the US Senate. Latino voters in Nevada turned out almost unanimously for Reid (Barreto 2010). In Virginia's 2013 guber natorial race. Ken Cuccinelli's record of hostile anti-immigrant rhetoric and actions mobilized Latino and Asian support for his opponent, Terry McAuliffe, providing McAuliffe just enough votes to beat back Cuccinelli's otherwise promising bid for governor (Segura 2013). It is clear that anti- immigrant political appeals contribute to ethnic solidarity and organized political activity among Latino voters (Martinez 2008; Perez 2014).
213 Barreto, Reny, and Wilcox-Archuleta
Based in part on these experiences in California, Nevada, and Vir ginia, academics and practitioners expected that Latinos would turn out en masse against Donald Trump. Pre-election polls suggested a blowout for Clinton among Latino voters, including polls by Latino Decisions, Univi- sionp^ashington Post, NBC/Telemundo, NALEO/Telemundo, and Florida International University/New Latino Voice. Indeed, Latino registration skyrocketed and early voting in Latino-heavy counties ran at all-time highs (Gamboa 2016). It was therefore astonishing when Edison Exit Poll results suggested that Latinos did not just support Trump, but gave him more support than they had given Romney four years earlier. TTe postelection political narrative shifted from one of predicting Latino backlash against Trump to blaming Latinos for his victory (Brammcr 2016). How could all the pre-election polling have been so wrong? Or was it? Using real election returns at the electoral precinct level together with demographic data from Catalist, a campaign data vendor, we estimate how Latinos really voted in the 2016 election. We find strong evidence that the Edison Exit Poll overestimated Latino support for Trump by 15 percent age points and that pre-election pollsters were far more accurate in their assessment. Indeed, Trump received the smallest share of the Latino vote of any presidential candidate in recent political history. Before we present the findings ofour analysis, we examine why the Edison Exit Poll so badly overestimated support for Trump among Latino voters.
Different Approaches to Polling Latinos Yield Different Results There are a number of reasons to distrust Edison Research's exit poll estimates for Latinos. First, the polling firm does not select enough high- density Latino precincts in its sampling. Second, it does not conduct enough Spanish-language interviews. As a result, the findings do not accurately represent Latinos in the United States, skewing instead toward higher socioeconomic status and more conservative voters. Here we outline each shortcoming and then show how culturally competent methods can overcome these limitations.
Edison Exit Poll Methodology: Sampling and Language Issues The Edison Exit Poll was never designed to capture sub-populations, like Latinos or African Americans. Instead, it was designed to offer one national estimate and to help news organizations predict outcomes. Because it does
214 Survey Met/iodology and the Latina/o Vote
not oversample with sub-populations in mind, it falls short on a number of fronts. Using demographic information from past Edison Exit Polls, we show that the methodology employed is insufficient to capture the complexities of the Latino population. First, Edison does not select many high-density Latino (or African American) precincts. Despite very high levels of segregation in the United States, the Edison poll actually has very few precincts with large numbers of minority voters. The reason, of course, is that minority-heavy precincts are not close in outcome and thus are less helpful to pollsters in predicting the shifting preferences of the electorate. For example, Edison recently admitted that its exit poll had only eleven total precincts with sizable Latino popula tions. In 2014 they admitted they had selected zero precincts in the Texas Rio Grande Valley, where 25 percent of all Texas Latinos reside(Nuno 2014). Second, the Edison poll is primarily conducted in English, not Spanish. According to US Census Bureau data, about 30 percent of Latino voters are foreign-born. Most of those voters are more comfortable being interviewed in Spanish. In past cycles, only 6 or 7 percent of Edison poll interviews with Latinos are in Spanish,^ while the population numbers suggest that it should be closer to 30 percent. Spanish-dominant Latinos are far more heavily Democratic than those who are English-dominant, which suggests that Edison estimates of Latino voting for the Republican candidate could be heavily biased upward. Third, past Edison Exit Polls demonstrate a substantial skew toward minorities with higher income and education than the average for those populations. When compared to the Current Population Survey's Novem ber supplements (official estimates of who voted, compiled by the Census Bureau), the Edison poll has between 11 percent and 12 percent more col lege graduates and 5 percent more respondents with above-median incomes (CNN 2016). That held true in 2016 as well. In the current Edison poll results, 44 percent of nonwhite respondents have college degrees (CNN 2016). The actual proportion of college graduates among all nonwhites in the voting electorate is around 30 percent. As for income by race, though this has been reported in all previous year Edison polls, we cannot find that breakout on any network presentations of the 2016 Edison poll. Historically, Edison poll respondents have had significantly higher income than the aver age among nonwhite voters as indicated by the Current Population Survey. Edison has acknowledged these shortcomings. In 2005, the pollsters wrote that the Edison Exit Poll "is not designed to yield very reliable esti mates of the characteristics of small, geographically clustered demographic
215 Barreto, Reny, and Wilcox'Archuleta groups. These groups have much larger design effects and thus larger sam pling errors. ... If we want to improve the National Exit Poll estimate for Hispanic vote (or Asian vote, Jewish vote or Mormon vote etc.) we would either need to drastically increase the number of precincts in the National Sample or oversample the number of Hispanic precincts" (Edison Media Research and Mitofsky International 2005, 62). Despite their self-critique, it appears that they have made few adjustments to their methodologies and continue to misrepresent minority subgroup voting. Polling firms like Latino Decisions rely on more culturally competent methods, yielding a far more accurate picture of the Latino electorate on election day.
Latino Decisions Methodology: Culturally Competent Methods The Latino Decisions 2016 Election Eve Poll surveyed 5,599 extremely high-propensity Latino voters in the nights immediately prior to the election. It found that 79 percent of Latino voters supported Secretary Clinton, 18 percent supported Donald Trump, and 3 percent chose some other candidate. Latino Decisions takes a culturally competent and rigor ous social science approach to polling US Latinos, taking care to ensure a representative sample of this population. First, respondents were randomly selected from the voter rolls to match a statewide representative sample of Latinos. The sample was pre- screened, based on vote history in previous presidential elections and date of registration, to include a mix of new registrants and Erst-time voters. All respondents confirmed their Hispanic identity at the start of each survey, and non-Hispanic respondents were screened out. Respondents were asked if they had already voted early, and if not, if they were 100 percent certain they would vote on November 8. Any respondent who was not certain was excluded from the poll. In past cycles, thanks to this careful methodology, over 90 percent of respondents were validated subsequently as having voted in the election, and the distributions on variables of interest did not vary between the total and those validated. Representativeness was further ensured by offering a fully bilingual option to respondents. Interviews were conducted either online or by telephone with live callers, all of whom were bilingual, and both phone and web interviews were completed in the language preferred by the respondent. The resulting national sample for the 2016 Election Eve Poll carries an overall margin of error of 1.8 percent. This margin is adjusted to account for the design effect resulting from twelve unique sample strata of varying
216 Survey Met/iodology and the Latina/o Vote size, mode differences, and post-stratification weighting used to derive the national estimate. Florida has 804 completed interviews and carries a margin of error of 3.5 percent. The other individual states sampled—Arizona (417), California (414), Colorado (404), Illinois (406), Nevada (404), New York (405), North Carolina (410), Ohio (403), Texas (409), Virginia (407), and Wisconsin (411)—have a margin of error of 4.9 percent. The remaining 405 respondents are from other states and the District of Columbia.
Cultural Competence Is a Must Exit polls derive estimates from a small, nonrepresentative sample of a handful of precincts, significantly biasing subgroup estimates. By contrast, culturally competent methods, like those employed by Latino Decisions, are necessary to estimate accurate Latino vote outcomes. In particular, Latino Decisions randomly samples a sufficiently large number of Latino registered voters in each state, conducts bilingual surveys, and weights the final results to match the census for correct geographic dispersion, age, education, nativity, and gender of Latino voters. For all these reasons, Latino Decisions results differ significantly from those of the Edison Exit Poll and, from a social science perspective, are more accurate and reliable. Despite the methodological rigor employed by Latino Decisions, its results, like those of Edison, are derived from a single cross-sectional survey that necessarily has a margin of error. To validate these findings, we merge real voting data, collected at the precinct level in ten states with large Latino populations, with precinct demographic estimates. This dataset represents the official votes cast, tallied, and verified by counties around the country in 2016, not survey estimates. We can then use a statisti cal technique called ecological inference (El), which allows us to infer individual-level behavior from aggregate data, to estimate how Latinos voted in the 2016 election (King 1997).
Precinct Analysis The Edison Exit Poll estimates that 28 percent of Latino voters nationwide cast their votes for Trump,one percentage point higher than the estimated 27 percent that cast their votes for Romney in 2012. When broken out by state, as we show in the first column of table 1, the numbers are similarly higher than would be expected. In Colorado, North Carolina, Arizona, Texas, New Mexico, and Florida, the Edison Exit Poll estimated that Trump won more than 30 percent of the Latino vote.
217 Barreto, Reny, and Wilcox-Archuleia
Table 1. Comparing Latino Support for Trump in Ten States across Edison Exit Poll, Latino Decisions Election Eve Poll, and Precinct Vote Returns j State Trump support: Trump support: Trump vote Difference Edison Exit Poll ; Latino Decisions returns: precinct 1 precinct-Edison estimate(%) 1 estimate (%) ecological Exit Poll estimate(%) i (percentage points) 1 Colorado 30 1 8 * 1 -22 i North Carolina 40 15 20 1 -20 1 Nevada 29 16 10 -19 i Arizona 31 15 i -16 j ! 12 ! Texas 34 16 18 i -16 : New Mexico 33 i — 19 i -14 1 California 24 16 11 i -13
1 New York 23 10 10 i -13 i Florida 35 31 31
1 Illinois — ! 10 6 t — 1 AVERAGE 31 i 15.78 14.8 i -15.22 Note: Cells display estimated percentage of Trump support among Latino voters by state and method of estimation. Final column shows difference between ecological inference precinct analysis and the Edison Exit Poll. The Edison Exit Poll did not sample enough Latinos in Illinois to estimate Latino support for Trump, despite the sizable Latino population in the state. Similarly, Latino Decisions did not sample New Mexico Latinos.
Edison estimates for Latino support were also in sharp contrast to the leading polling conducted right up until the election. The Latino Decisions 2016 Election Eve Poll, conducted in the days before November 8,sampled 5,600 Latino likely voters in eleven states. The results from this poll showed that 18 percent of Latinos nationwide supported Trump. State-level results from the Latino Decisions poll, shown in the second column of table 1, are much closer to what we would expect based on the theories outlined above. Averaged across states we see that the Edison Exit Poll estimated Trump support among Latinos as being over 15 percentage points higher than Latino Decisions polling. Who was right? To answer this question we turn to real election data. For each state, we collected 2016 precinct-level election data from each county's board of elections website. We then merged this data, by state, with precinct-level demographic estimates from Catalist, a firm that compiles data, including race, on 240 million voting-age individuals in the United States. These
218 Survey Methodology arxd the Latina/o Vote precinct'level estimates of the Latino registered voter population, together with precinct-level electoral returns, allow us to statistically estimate how Latinos voted in each state. In total, we have 29,045,522 voters in over 39,000 precincts in ten states. This represents approximately 92 percent of the Latino voting population in the United States. We are confident that our results are not driven by more liberal states such as California or by specific regions of the country. We start with a simple scatter plot that plots the share of the vote for each candidate against the proportion of Latino registered voters in the precinct (fig. 1). We then use locally weighted regression curves(LOESS)
100
Obama 75 /
o ^ Clinton (0 £ v>
o > 50
c * > ^ (U T3 '55 <0 * <. Romney 25 Tnifnp
0-
0 25 50 75 100 Latino share of registered voters in precinct(%)
Figure I. Estimated Presidential Vote among Latinos m 20/2 and 20/6, Based on Official Vote /Returns from 39,118 Precincts Source: Authors' calculations based on 39,118 precinct-level election results. Note: TTie lines correspond to local weighted regression curves (loess) and show a weighted average of support for Clinton and Trump in 2016 compared to Obama and Romney in 2012. Each precinct is weighted by the total number of votes cast. The 39,118 precincts are in Colorado, North Carolina, Nevada, Arizona, Texas, New Mexico, California, New York, Florida, and Illinois.
219 Barreto, Reny, and Wilcox-Archuleta to highlight the trend in the data for each of the major party candidates in 2012 and 2016. What is immediately clear is that as the proportion of Latinos in the precinct increases, overall support for the Democratic can didates increases. Comparing the two lower lines, it is clear that Romney does better than Trump on average in the precincts that arc more heavily Latino. It is only in the precincts where very few Latinos live that Trump outperforms Romney. If Trump did in fact do better than Romney among Latinos overall, as the Edison Exit Poll suggests, it is unclear where those votes would have come from. Next, we use a statistical method called ecological inference (El), developed by Harvard political scientist Gary King (1997). This method uses aggregate data to infer behavior at the individual level. While there are issues inherent in estimating individual-level behavior from aggregate data (see King, Rosen, and Tanner 2004 for a discussion), El has been the gold standard in academic applications and has been used extensively in voting rights court cases (Grofman and Merrill 2004; King 1997). El is beneficial because it provides exact statistical estimates as opposed to a general pattern, as we showed in figure 1. Column 3 of table 1 contains the El estimates from our analysis, and column 4 displays the difference between our precinct analysis and the Edison Exit Poll. We show that Edison consistently overestimated sup port for Trump by very large margins. In Colorado, the Edison Exit Poll estimated Latino support for Trump 22 percentage points higher than our results; in North Carolina the excess was 20 points, and in Nevada 19 points. Averaged across the ten states, our estimate of 14.8 percent Trump support among Latinos is much closer to the averaged Latino Decisions esti mate of 15.78 than to the averaged Edison Exit Poll estimate of 31 percent. In table 2, we aggregate all the state-level data together into a single dataset and run a final ecological inference. We report results for Trump and Clinton together with estimates of uncertainty. Using the real election returns data, we find that an estimated 79.2 percent of Latinos voted for Hillary Clinton and 15.8 percent of Latinos voted for Donald Trump. These estimates are almost identical to the predictions of the Latino Decisions Election Eve Poll.
Conclusion In this article we highlight the importance of taking a culturally compe tent approach to collecting accurate data among Latinos. Scholars and
220 Survey Methodology and the Latina/o Vote
Table 2. Precinct Vote Return Analysis for Ten States Pooled i Candidate | Estimated support {%) 1 Standard error of estimate(%) i Clinton 79.2 14.6 i Trump j 15.8 ' 11.7 Note: Cells display the average estimated percentage of Latinos who voted for Clinton and Trump in the 2016 presidential election based on our ecological inference estimates. Our dataset consists of all pooled precincts in Colorado, North Carolina, Nevada, Arizona, Texas, New Mexico, California, New York, Florida, and Illinois. The second column shows the standard error, a measure that indicates the level of uncertainty in our estimate. practitioners need to put more care into devising a research design and approach that is Latino-centered before starting a data collection effort. What differences exist within the Latino community, and how can our research design take this into account? Are we offering surveys in Spanish? Are we targeting all Latino households with equal frequency? Are we word ing questions in a way that is culturally sensitive? Are our sample sizes large enough to allow for generalizable inferences? These questions and more need to be asked when assessing the accuracy of our research approaches within the Latino community. In the immediate aftermath of the 2016 presidential election, many political pundits and even some scholars suggested that Latino voters had supported Trump by larger than expected margins (Cadava 2016; Enten 2016). The historian Geraldo Cadava (2016) released an analysis of selected counties in New Mexico and Texas in an attempt to show that Trump did better than expected among Latinos, notably among rural Latinos, who, he argued, were similar to rural white voters. Similarly, Alejandra Matos of the Washington Post and Harry Enten of FiveThirty-Eight.com both released articles suggesting that Trump did better than Romney among Latino voters (Enten 2016; Matos 2016). These reports coupled with the Edison Exit Poll subgroup results were surprising on many levels. Not only did they come to conclusions that were vastly different from bilingual and bicultural survey research of Latinos during the election, they were at odds with a wide body of scholarship on Latino political behavior. One possible reason the mainstream survey results were unexpected could simply be because they were inaccurate. To assess this possibility, we gathered vote and demographic data from over 39,000 individual voting precincts across ten states and used reli able statistical modeling to infer how Latinos voted. Our findings suggest that Trump not only did worse than Romney with Latino voters but also received the lowest Latino vote share of any candidate in recent presidential
221 Barreto, Reny, and Wilcox-Archuleta election history. Our real election data estimates were nearly identical to the estimates from multiple pre-election polls that took a Latino-centered approach, such as those from Latino Decisions, Univision, NBC/Tel- cmundo, and NALEO. These findings highlight the urgent need for all polling firms to adopt culturally competent methods in future elections.
Notes 1. Members of the National Election Pool include ABC,CBS, CNN, Fox, NBC, and the Associated Press. 2. Based on California Field Polls from various years. California Field Poll data are distributed by UC DATA,University of Califomia, Berkeley, http://ucdata. berkeley.edu/data_record.php ?recid=3. 3. Based on national election day exit polls from various years, available on the Roper Center website, https://ropercenter.cornell.edu/polls/us-elections/ exit-polls/.
Works Cited Alvarez, R. Michael, and Lisa Garcia Bedolla. 2003. "The Foundations of Latino Voter Partisanship: Evidence from the 2000 Election." Journal of Politics 65, no. 1:31-49. Barbaro, Michael. 2016. "Donald Trump Clung to 'Birther' Lie for Years, and Still Isn't Apologetic." New York Times, September 17. Barreto, Matt. 2010. "Proving the Exit Polls Wrong: Harry Reid Did Win Over 90% of the Latino Vote." Latino Decisions blog, November 15. http://www.latinodecisions.eom/blog/2010/l 1/15/proving-the-exit-poll s-wrong-harry-reid-did-win-over-90-of-the-latino-vote/. Barreto, Matt A., Luis R. Fraga, Sylvia Manzano, Valerie Martinez-Ebers, and Gary M. Segura. 2008. "Should They Dance with the One Who Brung 'Em? Latinos and the 2008 Presidential Election." PS: Political Science and Politics 41, no. 4: 753-60. Barreto, Matt, and Gary Segura. 2015. Latino America: How America's Most Dynamic Population Is Poised to Transform the Politics of the Nation. New York: Public Affairs. Barreto, Matt, and Nathan Woods. 2005."Metropolitan Latino Political Behavior: Turnout and Candidate Preference in Los Angeles." Journal of Urban Affairs 27, no. 2:71-91.
222 Survey Methodology and the Latina/o Vote
Bernal, Rafael. 2016. "Hispanic Voter Registration Spikes." The Hill, April 27. Beltran, Cristina. 2010. The Trouble with Unity: Latino Politics and the Creation of Identity. New York: Oxford University Press. Bowler, Shaun, Stephen P. Nicholson, and Gary Segura. 2006. "Earthquakes and Aftershocks: Race, Direct Democracy, and Partisan Change." American]oumal of Political Science 50, no. 1: 146-59. Brammer,John Paul. 2016."The Latino Vote' Didn't Overwhelm Trump, Because We're Not All the Same." Guardian, November 9. Burns, Sarah. 2016."Why Trump Doubled Down on the Central Park Five." New York Times, October 17. Cadava, Geraldo L. 2016. "Rural Hispanic Votes—Like White Rural Voters— Shifted toward Trump. Here's Why." Was/imgtorr Post, November 17. Collingwood, Loren, Matt A. Barreto, and Sergio 1. Garcia-Rios. 2014. "Revisit ing Latino Voting: Cross-Racial Mobilization in the 2012 Election." Political Research Quarterly 67, no. 4: 632-45. CNN. 2016. "Election 2016: Exit Polls." CNN Politics, November 23. Dawson, Michael C. 1994. Behind the Mule: Race and Class in African-American Politics. Princeton, NJ: Princeton University Press. de la Garza, Rodolfo O. 1992. Latino National Political Survey: Summary of Findings. Chicago: Inter-University Program for Latino Research, http://files.eric.ed.gov/ fulltext/ED354281.pdf. de la Garza, Rodolfo, and Louis DeSipio, eds. 1996. Ethnic Ironies; Latino Politics in the 1992 Elections. Boulder, CO: Westview. DeSipio, Louis. 1996."After Proposition 187, the Deluge: Reforming Naturalization Administration While Making Good Citizens." Harvard Journal of Hispanic Policy 9: 7-24- Edison Media Research and Mitofsky International. 2005. Evaluation of Edison/ Mitofsky Election System 2004. Prepared for the National Election Pool. Available on ABC News website, https://abcnews.go.com/images/Politics/ EvaluationofEdisonMitofskyElectionSystem.pdf. Enten, Harry. 2016."Trump Probably Did Better with Latino Voters Than Romney Did." FiveThirtyEight.com, November 18. Fraga, Luis R., and David L. Leal. 2004."Playing the 'Latino Card': Race, Ethnicity, and National Party Politics." Du Bots Review 1, no. 2: 297-317. Fraga, Luis R., John A. Garcia, Rodney E. Hero, Michael Jones-Correa, Valerie Martinez-Ebers, and Gary M.Segura. 2010. Latinos in the New Millennium: An Almanac of Opinion, Behavior, and Policy Preferences. New York: Cambridge University Press. Gamboa,Suzanne. 2016."Trump Targeted Mexicans, Now Latino Vote Surge May Wall Him Out." NBC News, November 8. Garcia, F. Chris, and Gabriel R. Sanchez. 2008. Hispanics and the U.S. Political System: Moving into the Mainstream. Upper Saddle River, NJ: Prentice Hall. Green, Donald, Bradley Palmquist, and Eric Schickler. 2002. Partisan Hearts and Minds: Political Parties and the Social Identities of Voters. New Haven, CT: Yale University Press.
223 Barreto, Reny, and Wilcox-Archuleta
Grofman, Bernard, and Samuel Merrill. 2004- "Ecological Regression and Ecological Inference." In King, Rosen, and Tanner 2004, 123-43. Gross, Justin. 2016."Latino Electorate On Track for Historic Turnout in 2016." Latino Decisions blog, November 3. http://www.latinodecisions.com/blog/2016/ll /03/latino-electorate-on-track-for-historic-turnout-in-2016/. HoSang, Daniel. 2010. Racial Propositions: Balbt Initiatives and the Making of Postwar California. Berkeley: University of California Press. King, Gary. 1997. A Solution to the Ecological Inference Problem: Reconstructing Indi' vidual Behavior from Aggregate Data. Princeton, NJ: Princeton University Press. King, Gary, Ori Rosen, and Martin A. Tanner. 2004. Ecological Inference: New Methodological Strategies. New York: Cambridge University Press. Latino Decisions. 2016. "2016 Election Eve Poll." http://www.latinodecisions. com/2016-election-eve-poll/. Mahler, Jonathan and Steve Eder. 2016.'"No Vacancies' for Blacks: How Donald Trump Got His Start, and Was First Accused of Bias." New York Times, November 28. Martinez, Lisa. 2008. "The Individual and Contextual Determinants of Protest Among Latinos." Mobilisation; An International Quarterly 13, no. 2: 189-204. Matos, Alejandra. 2016. "On the El Paso Border, Trump's Appeal with Latinos Defies Expectations." Washington Post, December 9. Michelson, Melissa R., and Amalia Pallares. 2001."The Politicization of Chicago Mexican Americans: Naturalization, the Vote, and Perceptions of Discrimina tion." Aztldn: AJourrtal of Chicano Studies 26, no. 2: 63-85. Nuno, Stephen A. 2007."Latino Mobilization and Vote Choice in the 2000 Presi dential Election." American Politics Research 35, no. 2: 273-93. Nuno, Stephen A. 2014. "Did the GOP Make Inroads with the Latino Vote?" Washington Post, November 19. O'Keefe, Ed. 2016. "Latino Voter Group Reports Registering More than 100,000 New Voters." Washington Post, October 4. Pachon, Harry, and Louis DeSipio. 1994. New Americans by Choice: Political Perspec tives of Latino Immigrants. Boulder, CO: Westview. Pantoja, Adrian D., Ricardo Ramirez, and Gary M. Segura. 2001. "Citizens by Choice, Voters by Necessity: Patterns in Political Mobilization by Naturalized Latinos." Political Research Quarterly 54, no. 4: 729-50. Perez, Efren O. 2014- "Xenophobic Rhetoric and Its Political Effects on Immigrants and Their Co-Ethnics." American Journal of Political Science 59, no. 3: 549-64. RNC (Republican National Committee). 2013. Grouith & Opportunity Project. http://goproject.gop.com/rnc_growth_opportunity_book_2013.pdf. Sanchez, Rosaura, and Beatrice Pita. 2006."Theses on the Latino Bloc: A Critical Perspective." Aztldn: A Journal of Chicano Studies 31, no. 2: 25-53. Sanchez, Gabriel R. 2006a. "The Role of Group Consciousness in Latino Public Opinion." Political Research Quarterly 59, no. 3: 435-46. . 2006b."TTie Role of Group Consciousness in Political Participation among Latinos in the United States." American Politics Research 34, no. 4: 427-51.
224 Survey Met/iodology and the Latinalo Vote
Sanchez, Gabriel R., and Natalie Masuoka. 2010. "Brown Utility Heuristic? The Presence and Contributing Factors of Latino Linked Fate." Hispanic Journal of Behavioral Sciences 32, no. 4: 519-31. Schildkraut, Deborah J. 2005. Press One for English: Language Policy, Public Opinion, and American Identity. Princeton, NJ: Princeton University Press. Segura, Gary. 2012."Latino Public Opinion and Realigning the American Elector ate." Daedalus 141, no. 4: 98-113. . 2013. "New Race Politics and the Virginia Election." Latino Deci sions blog, November 6. http://www.latinodecisions.eom/blog/2013/ll/06/ new-race-politics-and-the-virginia-election/. Shepard, Steven. 2016. "Latino Voting Surge Rattles Trump Campaign." Politico, November 6. Tesler, Michael. 2016."Trump Is the First Modern Republican to Win the Nomina tion Based on Racial Prejudice." Wos/u'ngton Post, August 1. Tolbert, Caroline, and Robert Hero. 2001. "Dealing with Diversity: Racial/Ethnic Context and Social Change." Political Research Quarterly 43, no. 3: 571-604. Tumulty, Karen. 2001. "La Nueva Frontera: Courting a Sleeping Giant." Time, June 11. Uhlaner, Carole J., and F. Chris Garcia. 2005. "Learning Which Party Fits: Expe rience, Ethnic Identity, and the Demographic Foundations of Latino Party Identification." In Diversity in Democracy: Minority Representation in the United States, edited by Gary Segura and Shaun Bowler, 72-101. Charlottesville: University of Virginia Press. Ye Hee Lee, Michelle. 2015."Donald Trump's False Comments Connecting Mexi can Immigrants and Crime." Washington Post, July 8.
225 To: Chad Dunn, Brazil & Dunn Attorneys at Law
From: Francisco I. Pedraza, Texas A&M University; Matt A. Barreto, University of Washington, Seattle Subject: Harris County voter suppression Date: August 2, 2011
We have sufficient data to provide a sample of charts and tables that illustrate below-average voter registration rates and voter polarization in Harris County. The charts will make more sense once we explain some of the details to you in our next phone meeting. We can present the infor mation in many ways (tables, charts, etc), and we welcome your feedback to produce the most effective presentation for your objectives.
Latino voter registration rates in Harris County relative to other Texas counties is below average. Voter registration rates represent the conversion of citizens who are of voting age into registered voters. Figure 1 shows that Harris county was below average in 2000 (Panel A)and in 2010 (Panel B)in Latino voter registration rates. The pattern of below average voter registration rates for Harris County holds whether using raw numbers or percentages (Panels D and E).
There are other Texas counties with below-average Latino voter registration rates, however, Harris County is unique because it has the most Latinos eligible to vote who are not included in the voter rolls. Panel C of Figure 1 plots the county-level ratio of Latino registered voters and Latino citizen voting age population (y axis), and sorts counties by the total number of Latino citizen voting age population (x axis). The horizontal line represents the average ratio, and Harris County is below the average. Harris County's 84% Latino voter registration rate means that about 86,000 of the 540,000 Latino eligible voters were not registered to vote.
Panel F of Figure 1 plots the growth in Latino registered voters from 2000 to the end of 2009 against Latino citizen voting age growth from 2000 through 2009. The diagonal represents perfect correlation between the two measures. Harris County grew 25% in Latino citizen voting age pop ulation and 18% in Latino registered voters, representing a 7% unrealized gain in Latino registered voters during the last decade. While the 7% unrealized Latino voter registration is not unusual in terms of percentage, the actual number of Latino voters that this unrealized gain impacts is far above average, as indicated in Panel C of Figure 1.
Harris County votersfor Commissioner Precinct 2 exhibit strong racial polarization patterns in Novem ber 2010. We selected the November 2010, Harris County Commissioner Precinct 2 contest to illus trate racial polarization in the electorate. Table 1 lists the estimated distribution of votes among Latino, White and Black voters at the precinct level. In the aggregate 82.6% of the Latino vote was cast in favor of the Latino incumbent Sylvia Garcia and a remaining 17.4% for the challenger Jack Morman. The vote distribution for the White vote is the inverse, with 2.3% for Garcia, and 97.7% favoring Morman. Figures 2 and 3 further illustrate the distinctive voting bloc patterns at the precinct level by plotting the vote for Garcia against the share of Latino and White voters in a precinct. (A) County Voter Registration Rate among Latinos 2000 (B) County Voter Registration Rate among Latinos 2009 (C)Ratio of county Latino RV and CVAP sliares 2009
BexarQ Bexafl^ •Camer(5i^'*^®'9° /A. • • El Paso yHarris *,* • Nueces ••• BexaiQ • •• • • • El Paso y El Paso ••• A • AHarris ^ Travis* *TatTant Harris Hidalgo • Hidalgo/ •
#
©Dallas • Dallas Camerpn Cam^m TravJiC • NueaSs ©Dallas ... .. /^Tarrant Webb* X • "^'"'XTravis Njiieces ytarrant
100 200 300 400 100 200 300 400 500 100 200 300 400 500
Latino CVAP 2000 {in 10OO's) Latino CVAP 2009(in 1000's) sorted by Latino CVAP 2009 (in 10OO's)
N3 (D)County Voter Registration Rate among Latinos 2000 (E) County Voter Registration Rate among Latinos 2009 (F)Percent Latino CVAP growth vs. Percent Latino RV growth
oBexar
arns •/A*' * ' Hergs
isxar*
0.2 0.4 0.6 0.8
Percent Latino CVAP 2000 Percent Latino CVAP 2009 Percent Latino CVAP growtti 00-09
Figure 1: Texas county voter registration rates by county compare how much of the citizen voting age population is converted to registered voters. Counties can have above or below average registration rates. Harris County is represented by a triangle, and is below average in converting the citizen voting age population to registered voters in 2000 and 2010. Table 1: Estimated vote distribution by race/ethnicity group for the November 2010, Harris County Com missioner, Precinct 2 contest Garcia Morman Estimated Latino Vote 82.6% 17.4% Estimated White Vote 2.3% 97.7% Estimated Black Vote 97.5% 2.5%
Harris County - Nov 2010 Vote for Sylvia Garcia by Percent Latino in precinct
§ .6
0 .2 .4 .6 .8 % Latino in precinct
Figure 2: Latino voter polarization at the precinct level in the November 2010 Harris County Commissioner Precinct 2 contest.
Vote for Sylvia Garcia by Percent White in precinct
.4 .6 % Wtilte In Precinct
Figure 3: White voter polarization at the precinct level in the November 2010 Harris County Commissioner Precinct 2 contest. Galveston County Section 5 Memo, JP/Constable Districts
To: Neil Baron, N.G., Baron Law; Chad Dunn, Brazil & Dunn Attorneys at Law
From: Dr. Francisco 1. Pedraza, Texas A&M University; Dr. Matt A. Barreto, University of Washington, Seattle
Re: Section 5 Memo, Galveston County
Date: December 11, 2011
1. Voting Rights Act, Section 5 Overview
Section 5 of the Federal Voting Rights Act requires that specific jurisdictions, including anywhere in the state of Texas, to pre-clear any changes to practices, procedures or district boundaries with the United States Attorney General. Section 5 administrative review is important to provide a check against jurisdictions that previously had discriminatory procedures in place, as outlined in Section 4 of the Voting Rights Act. Because of such a history, the burden is on the covered jurisdiction to prove that the changes they are proposing, in no way decrease, dilute, or diminish the opportunity for minority voters to meaningfully participate in the electoral process.
Though it is not common for Section 5 submission to be overturned by the Department of Justice, in the case of the Galveston County Proposal, there are verv clear retrogressions that can not be ignored. Below we provide clear and succinct data that points to decreased electoral opportunities for Latino and African Americans citizens in Galveston County, Texas. First, we provide evidence that discriminatory outcomes result from redrawing Galveston County Justice of Peace (JP) and Constable districts, and that in the context of extraordinarily polarized racial bloc voting the discriminatory outcomes are particularly sharp. Second, between 2000 and 2010, the growth in the total population and the citizen voting age population (CVAP) in Galveston County has been driven primarily by increases in the Latino community population. Third, the proposed County plan reduces JP/Constable districts from 8 to 5, and does this by collapsing several majority-minority districts into majority-Anglo districts, significantly reducing minority influence. The proposed plan effectively reduces the political voice and representation of citizens who live in the existing majority minority districts, while increasing the political voice and
Pedraza & Barreto Memo //pg 1 Galveston County Section 5 Memo, JP/Constable Districts representation of citizens who live in existing majority-Anglo districts, indicating a clear and strong pattern of retrogression. Further, this pattern in minority retrogression is occurring in the context of a county-wide decline in the Anglo population. There is no question that the proposal submitted by Galveston County is retrogressive.
2. Racial Bloc Voting - Galveston County
While an assessment of retrogression does not require any evidence of racial bloc voting, it can provide a very important backdrop into the degree of racial differences or coalitions in the jurisdiction. In the absence of racial bloc voting changes to the existing electoral jurisdiction boundaries are less of a concern because the geographic location of minorities is less consequential in securing electoral outcomes consistent with minority preferences. However, if Anglos and minorities demonstrate divergent voting preferences, then it is much more important that no reduction in minority voting strength occurs under newly proposed district boundaries. Figures 1, 2 and 3 illustrate extraordinary levels of racial bloc voting patterns, indicating that voters in Anglo majority districts express political preferences that are consistently in opposition to the candidate preferences expressed by citizens in majority-minority districts - in this case Hispanic and African American voters. For example. Figure 1 shows that as the White percentage of voters increased in voting precincts in Galveston County, the percent of the vote won by Wilson, an African American candidate for District Clerk in 2010, decreased precipitously. At the same time, in precincts with almost no White voters, Wilson won an overwhelming majority of the vote. In looking at the 2008 election for President with Barack Obama, and the 2002 election for Governor with Tony Sanchez, the exact same patterns of racial bloc voting are revealed. Across almost any election in Galveston County, White voters take the 180 degree opposite stance than do minority voters. For this reason, it is very important that observers evaluate possible retrogressions with an extremely close eye.
Pedraza & Barreto Memo //pg 2 Galveston County Section 5 Memo, JP/Constable Districts
Figure 1: 2010 District Clerk election Wilson - African American(Dem) vs. Murray - Anglo (Rep)
Vote for Wilson by % White in Precinct
••
•• \ • • • •
is* *
.4 .6 .8 % White in Precinct
Figure 2: 2008 Presidential election Obama - African American(Dem) vs. McCain - Anglo (Rep)
Vote for Obama by % White in Precinct
S O «o • • • • 5
•• • • • i>'
.4 .6 % White in Precinct
Pedraza & Barreto Memo //pg 3 Galveston County Section 5 Memo,JP/Constable Districts
Figure 3: 2002 Gubernatorial election Sanchez- Hispanic (Dem)vs. Perry - Anglo(Rep) Vote for Sanchez by % White in Precinct
: i-: B <0. • •
A A . •••! >• •M•K' . •.A*
.4 .6 .8 % White In Precinct
3. Growth in Latino and African American Population, Galveston County, TX.
In Galveston County the total citizen voting age population increased from 171,966 in 2000 to 199,660 through the end of 2010, representing an estimated total growth of 27,694 or a 16% change in total citizens of voting age. During that same time period the Latino citizen voting age population changed by 37%, or three times the 12% CVAP growth among Anglos. In the decade of the 2000's the Galveston County Anglo population has declined from 63% to 59%, and the Anglo citizen voting age population has declined from 69% to 67%. As the Anglo population is declining and the Latino population growing rapidly, there is no scenario in which opportunities for Latino representation should decrease, or that Anglo opportunities for representation should increase.
Table 1: Summary of Population Growth by Racial Group 2000-2010
2000 2010
Galveston Co. Total Pop 250,158 291,309 41,151 16.5%
Latino Pop 45,153 65,270 20,117 44.6% African American Pop 39,226 40,727 1,501 3.8%
Anglo Pop 157,545 172,652 15,107 9.6%
Pedraza & Barreto Memo //pg 4 Galveston County Section 5 Memo, JP/Constable Districts
Summary of CVAP Growth by Racial Group 2000-2010
2000 2010 Chg %Chg
Galveston Co. CVAP 171,966 199,660 27,694 16.1%
Latino CVAP 21,619 29,544 7,925 36.7%
African American CVAP 26,578 28,681 2,103 7.9% Anglo CVAP 119,418 134,494 15,076 12.6%
Table 2: Summary of Share of the Population by Racial Group 2000-2010 Pop - 2000 Pop-2010 CVAP-2000 CVAP-2010
Latino 18.0% 22.4% 12.6% 14.8%
African American 15.7% 14.0% 15.5% 14.4%
Anglo 63.0% 59.3% 69.4% 67.4%
3. Retrogression in Minority Population under Proposed Plan
When comparing how the Latino and African American population has changed from 2000 to 2010 across the eight current JP/Constable districts, we note that, as compared to the current arrangement, the proposed boundaries reduce the overall number, and proportion of majority- minority districts. As indicated in Table 3, reducing the number of districts from eight to five create a considerable difficulty for maintaining majority-minority districts in Galveston County. The reduction in district from eight to five was not required, and only serves the purpose to reduce minority electoral influence. Under the old 2001 plan, three of the eight districts were majority-minority, accounting for 37.5% of all districts. Under the new 2011 plan, just one of five districts has a majority minority population, or 20% of all districts. This stands in stark contrast to the overall minority population in Galveston. According to Table 2 above, Anglos represent 59.3% of the population in 2010 and minorities represent 40.7% of the population, yet they maintain just one JP/Constable district.
Even though the number of districts was reduced from eight to five, it is possible to match the old district boundaries to the new district boundaries in an effort to determine how minority voters will see their circumstances change under the new plan. Table 4 matches the eight districts under the 2001 plan and reports the total Latino, Black and White population under the
Pedraza & Barreto Memo //pg 5 Galveston County Section 5 Memo, JP/Constable Districts old and new configuration of districts. Districts 2 and 3 under the old plan were very high percentage minority districts with 73% and 71% minority each. Under the new plan they are combined into one district and merged with a majority Anglo district, creating an overall minority percentage of61% minority, resulting in a decrease of 12 and 10 percentage points of the minority population for those voters in Districts 2 and 3 respectively. Likewise a similar trend is found with respect to District 5, which was 55% minority under the old plan, but under the new plan, it is merged with District 7 which is 66% Anglo, and results in a new district which 62% Anglo, thereby erasing a majority-minority district. Looking to Table 4, overall, the Latino population decreased in two of three majority-minority districts, and the Black population decreased in all three of the majority-minority districts. In contrast, the percent White increased by 10, 12, and 17 points across the three districts which used to be majority-minority, through the process of merging them in with large population majority-Anglo districts.
Finally, Table 5 provides additional evidence of discriminatory intent based on the population size of the districts. There is no current requirement that the JP/Constable districts maintain the exact same population size. If districts needs to be smaller in size to meet certain representation needs, it is allowed. However, section 5 retrogression standards apply to all possible measures of retrogression in electoral influence for minority citizens. Under the 2001 plan, each of the eight districts could have expected an average population size of 36,414 people, however the two most heavily minority districts were underpopulated, with -65% deviation in district 2, and -36% deviation in district 3. Not only is the overall minority population decreased for residents of these districts, but their overall population grows to 84,740 in the 2011 district 3. At 84,740, district 3 is now overpopulated by 52% from the expected average population of 58,262 under a five district scenario. Under the new plan proposed by the County, the three majority-minority districts all see a change in deviation for the worse, while the largest majority-Anglo districts see their deviation decrease.
Pedraza & Barreto Memo //pg 6 Galveston County Section 5 Memo, JP/Constable Districts
Table 3: Comparison of Percent Minority in Current and Proposed Districts
2001 OLD BOUNDARIES 2011 NEW BOUNDARIES DATA AS OF 2010 DATA AS OF 2010 District % Lat % Blk % Wht District % Lat % Blk % Wht 1 30.22 11.16 52.71 1 17.94 111 67.7 2 32.37 38.46 26.63 2 24.96 7.96 62.62 3 20.19 48.69 28.96 3 27.64 29.26 39.08 4 14.63 3.21 80.17 4 14.93 4.18 78.82 5 29.22 22.39 45.49 5 14.6 0.66 81.18 7 23.12 5.35 66.35
8 17.93 7.26 67.74 % 9 14.6 0.66 81.18
Number of Maioritv-Minoritv Districts
2001 Plan 3 of 8 37.5% of all districts 2011 Plan lof5 20.0% of all districts 2010 Black + Latino Pop 36.4% of total population
Pedraza & Barreto Memo //pg 7 Galveston County Section 5 Memo, JP/Constable Districts
Table 4: Comparison of Percent Minority in Proposed Districts Matched to Current Districts
Old 2001 Lines New 2011 lines Change in Population Dist % Lat % BIk % Wht Chg to % Lat % BIk % Wht Lat chg BIk chg Wht chg 1 30.22 11.16 52.71 3 27.64 29.26 39.08 -2.58 18.1 -13.63 2 32.37 38.46 26.63 3 27.64 29.26 39.08 -4.73 -9.2 12.45 3 20.19 48.69 28.96 3 27.64 29.26 39.08 7.45 -19.43 10.12 4 14.63 3.21 80.17 4 14.93 4.18 78.82 0.3 0.97 -1.35 5 29.22 22.39 45.49 2 24.96 7.96 62.62 -4.26 -14.43 17.13 7 23.12 5.35 66.35 2 24.96 7.96 62.62 1.84 2.61 -3.73 8 17.93 7.26 67.74 1 17.94 7.27 67.70 0.01 0.01 -0.04 9 14.6 0.66 81.18 5 * 14.6 0.66 81.18 0 0 0 -1.97 -21.37 +20.95
Table 5: Comparison of Population Deviation in Proposed Districts Matched to Current Districts
Old 2001 Lines New 2011 Lines Dist % Wht Pop Dev. Chg to % Wht Pop Dev. Dev Chg 1 52.71 36018 -1.09% 3 39.08 84740 52.50% 53.59% 2 26.63 12708 -65.10% 3 39.08 84740 52.50% 117.61% 3 28.96 23163 -36.39% 3 39.08 84740 52.50% 88.89% 4 80.17 30305 -16.78% 4 78.82 32208 -44.72% -27.94% 5 45.49 37350 2.57% 2 62.62 83092 42.62% 40.05% 7 66.35 60107 65.07% 2 62.62 83092 42.62% -22.45% 8 67.74 89241 145.08% 1 67.70 88852 52.50% -92.57% 9 81.18 2417 -93.36% 5 81.18 2417 -95.85% -2.49% Avg: 36414 Avg: 58262
4. Retrogressive Political Demographics in Precinct 2
In addition to demonstrating an actual decrease in the Hispanic and African American populations from 2001 to 2011-based boundaries, the proposed district represents considerable retrogression in the political opportunities for minorities. Stated simply, the new plan proposes to merge the majority-minority districts with majority-Anglo districts that demonstrated very different vote choice patterns, making it more difficult for minorities to have a real influence on
election outcomes in Galveston JP/Constable elections.
Pedraza & Barreto Memo //pg 8 Galveston County Section 5 Memo, JP/Constable Districts
To demonstrate this point with two basic examples, we have compiled the election results by district for the eight former districts and the five new districts.
Although we requested precinct-by-precinct election result data from Galveston County, they did not provide us with this data in an electronic format that we could analyze. Thus, we were provided with Galveston County election results data in the form of a Microsoft Excel workbook from Neil Baron, and this data is necessary for a thorough analysis of discriminatory intent toward minority communities. The data includes Galveston County voter district election results for various county, state and federal contests since 2002. In order to analyze the information we identified for each voting district the JP/Constable precinct to which it was assigned under the existing 2001 plan and the proposed 2011 County plan. We then tabulated the electoral outcomes for each voting district, and aggregated these results for the existing and proposed JP/Constable precincts to determine whether or not minority citizens in Galveston see an increase, or a decrease in their opportunity to have meaningful influence in election outcomes.
For example, in Table 7 we analyze the election returns for the 2004 U.S. Presidential contest, and provide a side-by-side comparison of results for existing and proposed Galveston County JP/Constable precincts. In the left column of figures existing precincts 1, 2, 3 and 5 gave a majority of electoral support for the minority preferred candidate, Barack Obama. Recall in the racial bloc voting analysis presented in Figure 2 that minority precincts voted overwhelmingly for Obama, while majority Anglo precincts voted strongly against Obama. Under the new proposed plan which collapses the existing precincts 1, 2, and 3 into proposed precinct 3, and collapses existing precincts 5 and 7 into proposed precinct 2, Obama goes from winning 4 of 8 districts to just winning 1 district. In Tables 6-10 which report election results, the percentage for the minority preferred candidate in the three majority-minority districts would decrease significantly under the new 2011 proposed plan, ranging from a decrease of 9.3 points on the low end, to a decrease of 29.3 on the high end. In any possible way to measure political influence, minority voters see a drastic and consistent decrease in voting strength under the new plan.
Pedraza & Barreto Memo //pg 9 Galveston County Section 5 Memo, JP/Constable Districts
Table 6: Election results for U.S. President 2004 by District Change in election influence Existing Boundaries Kerry (D) Bush (R) Proposed Boundaries Bush (R) Kerry (D) Bush (R) PCT1 35.1 64.9 PCT3 48.3 +16.6 -16.6 PCT2 81 19 PCT3 -29.3 +29.3 PCT3 74.5 25.5 PCT3 -22.8 +22.8 PCT4 34.8 65.2 PCT4 ^ +0.3 -0.3 -16.8 +16.8 +5.1 -5.1 ^PCTS"" ^ 27.3 72.7 PCT1 " 27.3 '72.7 0.0 0.0 PCT9 35.6 64.4 PCT5 37.8 62.2 +2.2 -2.2 Minority preferred candidate wins 3/8 or 38% of Minority preferred candidate wins 1/5 or 20% of county county districts districts
Table 7: Election results for U.S. President 2008 by District Change in election influence Existing Boundaries Obama (D) McCain (R) Proposed Boundaries Obama (D) McCain (R) Obama(D) McCain(R) PCT1 54.4 45.6 PCT3 66.9 33.1 +12.5 -12.5 PCT2 81.8 18.2 PCT3 66.9 33.1 -14.9 +14.9 PCT3 76.2 23.8 PCT3 66.9 33.1 -9.3 +9.3 PCT4 2^9 72.1 PCT1 27.9 72.1 0.0 0.0 |PCT? -16.7 +16.7 +3.2 -3.2 POTS 27.9 72.1 PCT4 28.4 71.e +0.5 -0.5 PCT9 30.5 69.5 PCT5 30.5 69.£ 0.0 0.0 Minority wins 4/8 or 50% of county districts Minority wins 1/5 or 20% of county districts
Pedraza & Barreto Memo //pg 10 Galveston County Section 5 Memo, JP/Constable Districts
Table 8: Election results for Texas Governor 2002 by District Change in election influence Sanchez Existing Boundaries Sanchez(D Proposed Boundaries Sanchez (01 (D) Perry (R) PCT1 5 4 4 PCT3 66 34 +12.6 -12.4 PCT2 8 7 1 PCT3 66 34 -16.7 +16.7 PCT3 7 5 2 PCT3 66 34 -10.5 +10.5 P 9 6 PCT1 26.2 73.8 -12.7 +12.7 P -21.0 +21.0 g6-jhaaaa!63'4i +7.0 -7.0 ^8 ^.2 39.2 60.8 +13.0 -13.0 PCT9 46.7 PCT5 46.7 53.3 0.0 0.0 Minority wins 3/8 or 38% of county districts Minority wins 1/5 or 20% of county districts
Table 9: Election results for Texas Governor 2010 by District Change in election influence Existing Boundaries White(D Proposed Boundaries White D PCT1 58.1 69.6 30.4 69.6 30.4 PCT3 83.7 PCT3 69.6 30.4 +14.1 PCT4 32.1 PGT1 30.1 69.9 +2.0 fPCTS' 56.7 rPCT2 36,1 :A 63.1 +20.6 ipCT7 : 31.5 |PCT2 : 36.1 631 -4.6 PCT8 30.1 PCT4 33 67 -2.9 PCT9 42.8 PCT5 42.8 57.2 0.0 Minority wins 4/8 or 50% of county districts Minority wins 1/5 or 20% of county districts
Pedraza & Barreto Memo //pg 11 Galveston County Section 5 Memo, JP/Constable Districts
Table 10: Election results for Galveston Clerk 2010 by District Change in election influence Old Wilson (D) Murray (R New Wilson (D) Murray (R Wilson (D) Murray(R PCT1 59.7 40.3 PCT3 69 31 +9.3 -9.3 PCT2 83.4 16.6 PCT3 69 31 -14.4 +14.4 PCT3 80.2 19.8 PCT3 69 31 -11.2 +11.2 PCT4 ^ 31^^ ^ 26 -5.6 +5.6 -23.2 +23.2 +5.8 -5.8 ^pctr 26 74 PCT4 32.3 67.7 +6.3 -6.3 PCT9 38.3 61.7 PCT5 38.3 61.7 0.0 0.0 Minority wins 4/8 or 50% of county districts Minority wins 1/5 or 20% of county districts
Pedraza & Barreto Memo //pg 12 To: Chad Dunn, Brazil & Dunn Attorneys at Law
From: Francisco I. Pedraza, Texas A&M University: Matt A. Barreto, University of Washington, Seattle
Subject: Harris County Precinct 2 supports minority candidates in statewide contests
Date: September 23, 2011
Summary: Under the 2001 benchmark boundaries Latinos candidates are electable by Harris County, Precinct 2 voters. Under the proposed changes represented in the "Revised A-1 Plan," Latino candidates would do worse in one ofthree ways: lose by a larger margin, win by a smaller margin, or lose the majority ofPrecinct 2 voters secured under benchmark boundaries.
Table 1 below lists examples where Latino candidates win a majority of Precinct 2 votes, but lose that majority under the proposed plan. In the 2008 general election, Democratic candidate Rick Noriega gamered 50.9% of Precinct 2 votes besting the Republican candidate, John Comyn by 3.8%. By contrast, Noriega loses by 0.9% of the vote under the proposed plan. Similarly, under the proposed plan in 2002 Linda Yanez, the Democratic candidate for the State Supreme Court, Place 1, would have lost by 1.8%, rather than win by 0.2% of votes cast.
Table 1: Examples where Latinos won under Benchmark but would lose under Proposed Plan
Status Quo Harris Proposed 2008 U.S. Senate 2001 POT 2 2011 POT 2 ChQ Cornyn - GOP 47.1% 49.5% +2.4% Noriega - OEM 50.9% 48.6% -2.3% Schick - LIB 2.0% 2.0% 0.0% Overall margin +3.8% -0.9% -4.7%
2002 Sup Ct #1 Schneider - GOP 49.9% 50.9% +1.0% Yanez - OEM 51.1% 49.1% -1.0% Overall margin +0.2% -1.8% -2.0%
2006 Primary Lt Gov Grant 37.9% 38.2% +0.3% Alvarado 38.3% 36.1% -2.2% De Leon 23.8% 25.7% +2.0% Overall margin +0.4% -2.1% -2.5%
2010 Primary Lt Gov Earle & Katz(combined) 49.2% 50.1% +0.9% Chavez-Thompson 50.8% 49.9% -0.9% Overall margin +1.6% -0.2% -1.8%
2010 Primary Land Com Burton 49.4% 51.1% +1.7% Uribe 50.6% 48.9% -1.7% Overall margin +1.2% -2.2% -3.4% Pedraza &. Barreto Memo—Supplement //pg 1 The general elections are not the only opportunities for voters to express their preferences. Table 1 also lists primary contests that Hispanic candidates would have lost under the proposed plan. In the Democratic Party primary election for Lieutenant Governor there were three candidates: Benjamin Grant, Maria Luisa Alvarado, and Adrian DeLeon. In that primary contest, Maria Luisa Alvarado won with 38.3% of the vote, and went on to be the Democratic candidate in the general election. Within Precinct 2 the margin of victory for Alvarado was less than 1%. Under the proposed "Revised A-1 Plan," Alvarado would have lost the plurality of votes in the primary election by less than 1%, and a non-Hispanic candidate would have received the plurality of votes.
In the 2010 Primary election for Texas Lieutenant Governor the Latino candidate, with 50.8% of the vote Linda Chavez-Thompson bested the two non-Latino candidates combined, by a margin of 1.6% of the votes cast. Under the proposed boundaries, a majority of Precinct 2 voters cast ballots against the Latino candidate, reducing the vote share for Chavez-Thompson to 49.9%. In another statewide primary contest, the Democratic candidate Hector Uribe won 50.6% of the vote to best Bill Burton by 1.2%. Under the proposed plan the Latino candidate Uribe collects only 48.9% of the vote and would have lost to the non-Latino candidate by 2.2% of the vote.
The proposed "Revised A-1 Plan" negatively impacts Latinos in Precinct 2, in general, because it engenders a loss in vote share for Latinos candidates. Table 2 lists general election contests and compares the Harris County Precinct 2 vote-share for Latino candidates under the benchmark, status quo boundaries, and under the proposed plan. In every case the overall margin increases and further disadvantages Latino candidates.
Table 2: Examples where Latinos do worse under the Proposed Plan
Status Quo Harris Proposed 2006 Judicial 2001 PCI 2 2011 POT 2 Chg Keller-GOP 50.4% 51.5% +1.1% Molina - OEM 49.6% 48.5% -1.1% Overall margin -0.8% -3.0% -2.2%
2002 Governor Perry - GOP 50.3% 51.3% +1.0% Sanchez- OEM 47.8% 46.9% -0.9% Others - OTH 1.9% 1.8% -0.1% Overall margin -2.5% -4.4% -1.9%
2002 Cnty Treasurer Cato - GOP 51.2% 52.2% +1.0% Garcia - OEM 48.8% 47.8% -1.0% Overall margin -2.4% -4.4% -2.0%
2002 Sup Ct #4 Smith - GOP 46.7% 47.7% +1.0% Mirabal - OEM 53.3% 52.3% -1.0% Overall margin +6.6% +4.6% -2.0%
Pedraza & Barreto Memo—Supplement//pg 2