Spatiotemporal Analysis of the 2012 General Election

Transcription

1 Edison Research Working Paper Spatiotemporal Analysis of the 2012 General Election Clint W. Stevenson Edison Research Abstract There is an expansive amount of literature relating to Election Day forecasting during presidential elections. Most of the forecast models rely on approaches such as ordinary least squares regression and Bayesian models. This is particularly true among the pre-election polling. In some cases models use aggregate state level data to simulate the outcome of the electoral college while others will take national samples to estimate the popular vote. However, due to the way the Electoral College operates the state samples are critical to determine the winner of an election. This paper will examine the 2012 National Election Pool Exit Poll conducted by Edison Research on Election Day (November 6, 2012) and will take the spatial information into account. Geostatistical procedures are used to develop a spatial model of voting patterns using actual vote results as well as demographic and other information obtained only from the Election Day exit poll. Various geospatial techniques are used to improve the spatial visualization of these voting patterns. These results from 2012 are compared to the historical exit poll data from the 2008 general election. These data presented here will show how the elections can be visualized, understood, and forecast at both the national and state level. Though this paper will focus specifically on Election Day exit polling these same techniques can easily be extended to address-based sampling. Keywords Geostatistics Exit Poll Kriging Election Author Contact: cstevenson@edisonresearch.com Contents Introduction 1 1 Methods Design of the 2012 Exit Poll The Spatial Method Simple, Ordinary, and Universal Kriging Results and Discussion Visualizing Spatial Data Analyzing Spatial Data Complete Spatial Randomness Subsubsection 2.3 Subsection Acknowledgments 5 References 5 Introduction Each presidential election year brings with it states where there is an expectation that the race will be close and states where there is effectively no contest. There are states where the final election results could bring with them late hours on Election Night and are often of particular interest to individuals actively watching the vote returns. However, states where the presidential race is not as hotly contested also provide additional information on the outcome of the election simply due to its geographic location. The National Election Pool (NEP) exit poll provides the unique opportunity to sample individual states as well as the United States as a whole. What makes this process unique is its ability to take these sampling measurements at known geographic locations on such a large scale and to accomplish this every election cycle. This paper will focus specifically on Election Day voter characteristics using spatiotemporal statistics to summarize the data. In this way trends based on election years and areal data across the country can be identified and measured. For this paper the presidential elections in 2008 and 2012 are used. Geostatistical analysis involves estimation and modeling of spatial correlation, and evaluating whether simplifying assumptions such as stationarity can be justified or need refinement[1]. Geospatial interpolation makes it possible to establish estimates on a random field. This paper will address estimating spatial correlation and spatial prediction in an election exit poll context. Depending on the state and polling location a different set of questionnaires may have been administered to the voters. However, there is some overlap on the questions on the questionnaires, namely the questions on the candidate the respondent voted and basic demographic questions. Aside from that any given precinct may have had anywhere from one to three questionnaire versions with a different set of questions on each. Therefore, using all available polling locations across all states limits the geospatial data to the common questions, namely voting behavior and general demographics. However, since the number of questionnaire versions is decided at a state level there is a rich array of voter data that is available when evaluating specific states.

2 Spatiotemporal Analysis of the 2012 General Election 2/5 Figure 1. Percent with College Eduction from 2012 General Election 1. Methods 1.1 Design of the 2012 Exit Poll The national exit poll is a unique source of data because it does not rely on telephones to reach respondents and it only interviews those who actually voted. In the 2008 and 2012 general elections the National Election Pool (NEP) ABC, The Associated Press, CBS, CNN, FOX, and NBC commissioned a survey of voters in all fifty states and the District of Columbia from Edison Research 1. This consisted of a national and state-specific surveys. Of the surveys in 2012 there were 19 states where the sample size was too small for individual state demographic or other breakouts. The majority of interviews are conducted in-person on Election Day in a probability sample that is stratified based on geography and past vote. In recent years the number of voters who cast their ballot early or by absentee has increased. Consequently, the NEP has supplemented the in-person exit poll with a telephone survey conducted between October 26, 2013 and November 4, This early voter telephone survey includes a landline as well as a cell phone only component. Follow up questions identify the number of landline phones in the respondent s home and if the respondents also has cellphone. In this way the respondent can be classified as landline only, both landline and cell, and cell phone only. In these state (and national) samples a pre-defined target of 30% of the respondents are called directly to cell phones. Random representative samples of in-person polling places are selected in 47 states 2 and the District of was conducted by Edison Research and Mitofsky International 2 Colorado, Oregon, and Washington are not included in this analysis because the voting process in these states are almost exclusively absentee/early Columbia. Each polling location is then geocoded to obtain its corresponding latitude and longitude. On Election Day randomly selected voters within each polling place complete a selfadministered paper questionnaire. Different states have a different number of questionnaire versions ranging anywhere from one questionnaire to three questionnaire versions. cos 3 θ = 1 4 cosθ + 3 cos3θ (1) 4 Lorem ipsum dolor sit amet, consectetuer adipiscing elit. Ut purus elit, vestibulum ut, placerat ac, adipiscing vitae, felis. Curabitur dictum gravida mauris. Nam arcu libero, nonummy eget, consectetuer id, vulputate a, magna. Donec vehicula augue eu neque. Pellentesque habitant morbi tristique senectus et netus et malesuada fames ac turpis egestas. Mauris ut leo. Cras viverra metus rhoncus sem. Nulla et lectus vestibulum urna fringilla ultrices. Phasellus eu tellus sit amet tortor gravida placerat. Integer sapien est, iaculis in, pretium quis, viverra ac, nunc. Praesent eget sem vel leo ultrices bibendum. Aenean faucibus. Morbi dolor nulla, malesuada eu, pulvinar at, mollis ac, nulla. Curabitur auctor semper nulla. Donec varius orci eget risus. Duis nibh mi, congue eu, accumsan eleifend, sagittis quis, diam. Duis eget orci sit amet orci dignissim rutrum. 1. First item in a list 2. Second item in a list 3. Third item in a list voting. In the case of Colorado, a combination of extremely high absentee/early voting and centralized voting.

3 Spatiotemporal Analysis of the 2012 General Election 3/5 1.2 The Spatial Method The approach taken in this paper is kriging. This is a family of estimators used to interpolate spatial data. The common methods include simple, ordinary, universal, and co-kriging. However, for the data collected in these exit polls the method selected is ordinary kriging. Because each state is its own sample and each state has its own covariance structure. In this way each of the 47 states with precinct samples 3 can be treated as its on class with independent covariance structures. Paragraph Sed commodo posuere pede. Mauris ut est. Ut quis purus. Sed ac odio. Sed vehicula hendrerit sem. Duis non odio. Morbi ut dui. Sed accumsan risus eget odio. In hac habitasse platea dictumst. Pellentesque non elit. Fusce sed justo eu urna porta tincidunt. Mauris felis odio, sollicitudin sed, volutpat a, ornare ac, erat. Morbi quis dolor. Donec pellentesque, erat ac sagittis semper, nunc dui lobortis purus, quis congue purus metus ultricies tellus. Proin et quam. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos hymenaeos. Praesent sapien turpis, fermentum vel, eleifend faucibus, vehicula eu, lacus. Paragraph Pellentesque habitant morbi tristique senectus et netus et malesuada fames ac turpis egestas. Donec odio elit, dictum in, hendrerit sit amet, egestas sed, leo. Praesent feugiat sapien aliquet odio. Integer vitae justo. Aliquam vestibulum fringilla lorem. Sed neque lectus, consectetuer at, consectetuer sed, eleifend ac, lectus. Nulla facilisi. Pellentesque eget lectus. Proin eu metus. Sed porttitor. In hac habitasse platea dictumst. Suspendisse eu lectus. Ut mi mi, lacinia sit amet, placerat et, mollis vitae, dui. Sed ante tellus, tristique ut, iaculis eu, malesuada ac, dui. Mauris nibh leo, facilisis non, adipiscing quis, ultrices a, dui. 1.3 Simple, Ordinary, and Universal Kriging Morbi luctus, wisi viverra faucibus pretium, nibh est placerat odio, nec commodo wisi enim eget quam. Quisque libero justo, consectetuer a, feugiat vitae, porttitor eu, libero. Suspendisse sed mauris vitae elit sollicitudin malesuada. Maecenas ultricies eros sit amet ante. Ut venenatis velit. Maecenas sed mi eget dui varius euismod. Phasellus aliquet volutpat odio. Vestibulum ante ipsum primis in faucibus orci luctus et ultrices posuere cubilia Curae; Pellentesque sit amet pede ac sem eleifend consectetuer. Nullam elementum, urna vel imperdiet sodales, elit ipsum pharetra ligula, ac pretium ante justo a nulla. Curabitur tristique arcu eu metus. Vestibulum lectus. Proin mauris. Proin eu nunc eu urna hendrerit faucibus. Aliquam auctor, pede consequat laoreet varius, eros tellus scelerisque quam, pellentesque hendrerit ipsum dolor sed augue. Nulla nec lacus. Reference to Figure 2. 3 Estimates for Colorado, Oregon, and Washington are conducted using absentee/early voter RDD phone surveys. rd[, 1] rd[, 2] rd[, 3] Figure 2. In-text Picture 2. Results and Discussion 2.1 Visualizing Spatial Data One of the many benefits of geospatial analysis is the ability to visualize the data in a way that is intuitive and natural. For this summary and analysis all precincts from 2008 and 2012 are randomly selected based on past voting history. Each polling place is then geocoded to facilitate further geospatial analysis. Two of the closest races in the 2012 presidential election were Florida (Obama won) and North Carolina (Romney won). These two states provide an example how kriging techniques can be applied within individual states. Figure 3 and Figure 5 visually shows the results to the question asked of voters pertaining to which political party (Democrat, Republican, Independent, Other) the voters identify while Figure 4 and Figure 6 provide the percent of the voting population with a college degree. These maps for Florida and North Carolina combine the data from both 2008 and Analyzing Spatial Data The exit poll data can be more fully Table 1. Table of Stuff Name First name Last Name Stuff Clint Stevenson 7.5 Wesley Stevenson Complete Spatial Randomness Due to the nature of the election exit polling complete geographic randomness is not necessary optimal. Consequently, it is better to sample polling places from areas where there is a known higher frequency of active voters rather than sample

4 Spatiotemporal Analysis of the 2012 General Election 4/5 2008, 2012 Percent Party ID Republican 2008, 2012 Percent Party ID Republican Figure 3. Geospatial Party ID Perspective of Florida for 2008 and 2012 Figure 5. Geospatial Party ID Perspective of North Carolina for 2008 and , 2012 Percent with College Education 2008, 2012 Percent with College Education Figure 4. Geospatial Percent with College Degree in Florida for 2008 and 2012 Figure 6. Geospatial Percent with College Degree in North Carolina for 2008 and 2012 from areas where there are very few active voters. This means that there is a very low likelihood that an exit poll state sample will exhibit complete spatial randomness. However, it is worthwhile to measure the distribution of observed locations to determine the geographic areas that may have a higher or lower concentration of sampled polling locations. Word Definition Concept Explanation Idea Text Subsubsection Aliquam lectus. Vivamus leo. Quisque ornare tellus ullamcorper nulla. Mauris porttitor pharetra tortor. Sed fringilla justo sed mauris. Mauris tellus. Sed non leo. Nullam elementum, magna in cursus sodales, augue est scelerisque sapien, venenatis congue nulla arcu et pede. Ut suscipit enim vel sapien. Donec congue. Maecenas urna mi, suscipit in, placerat ut, vestibulum ut, massa. Fusce ultrices nulla et nisl. First item in a list Second item in a list Third item in a list 2.3 Subsection Nulla in ipsum. Praesent eros nulla, congue vitae, euismod ut, commodo a, wisi. Pellentesque habitant morbi tristique senectus et netus et malesuada fames ac turpis egestas. Aenean nonummy magna non leo. Sed felis erat, ullamcorper in, dictum non, ultricies ut, lectus. Proin vel arcu a odio lobortis euismod. Vestibulum ante ipsum primis in faucibus orci luctus et ultrices posuere cubilia Curae; Proin ut est. Aliquam odio. Pellentesque massa turpis, cursus eu, euismod nec, tempor congue, nulla. Duis viverra gravida mauris. Cras tincidunt. Curabitur eros ligula, varius ut, pulvinar in, cursus faucibus,

5 Spatiotemporal Analysis of the 2012 General Election 5/5 augue. Nulla mattis luctus nulla. Duis commodo velit at leo. Aliquam vulputate magna et leo. Nam vestibulum ullamcorper leo. Vestibulum condimentum rutrum mauris. Donec id mauris. Morbi molestie justo et pede. Vivamus eget turpis sed nisl cursus tempor. Curabitur mollis sapien condimentum nunc. In wisi nisl, malesuada at, dignissim sit amet, lobortis in, odio. Aenean consequat arcu a ante. Pellentesque porta elit sit amet orci. Etiam at turpis nec elit ultricies imperdiet. Nulla facilisi. In hac habitasse platea dictumst. Suspendisse viverra aliquam risus. Nullam pede justo, molestie nonummy, scelerisque eu, facilisis vel, arcu. Curabitur tellus magna, porttitor a, commodo a, commodo in, tortor. Donec interdum. Praesent scelerisque. Maecenas posuere sodales odio. Vivamus metus lacus, varius quis, imperdiet quis, rhoncus a, turpis. Etiam ligula arcu, elementum a, venenatis quis, sollicitudin sed, metus. Donec nunc pede, tincidunt in, venenatis vitae, faucibus vel, nibh. Pellentesque wisi. Nullam malesuada. Morbi ut tellus ut pede tincidunt porta. Lorem ipsum dolor sit amet, consectetuer adipiscing elit. Etiam congue neque id dolor. Donec et nisl at wisi luctus bibendum. Nam interdum tellus ac libero. Sed sem justo, laoreet vitae, fringilla at, adipiscing ut, nibh. Maecenas non sem quis tortor eleifend fermentum. Etiam id tortor ac mauris porta vulputate. Integer porta neque vitae massa. Maecenas tempus libero a libero posuere dictum. Vestibulum ante ipsum primis in faucibus orci luctus et ultrices posuere cubilia Curae; Aenean quis mauris sed elit commodo placerat. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos hymenaeos. Vivamus rhoncus tincidunt libero. Etiam elementum pretium justo. Vivamus est. Morbi a tellus eget pede tristique commodo. Nulla nisl. Vestibulum sed nisl eu sapien cursus rutrum. Nulla non mauris vitae wisi posuere convallis. Sed eu nulla nec eros scelerisque pharetra. Nullam varius. Etiam dignissim elementum metus. Vestibulum faucibus, metus sit amet mattis rhoncus, sapien dui laoreet odio, nec ultricies nibh augue a enim. Fusce in ligula. Quisque at magna et nulla commodo consequat. Proin accumsan imperdiet sem. Nunc porta. Donec feugiat mi at justo. Phasellus facilisis ipsum quis ante. In ac elit eget ipsum pharetra faucibus. Maecenas viverra nulla in massa. Nulla ac nisl. Nullam urna nulla, ullamcorper in, interdum sit amet, gravida ut, risus. Aenean ac enim. In luctus. Phasellus eu quam vitae turpis viverra pellentesque. Duis feugiat felis ut enim. Phasellus pharetra, sem id porttitor sodales, magna nunc aliquet nibh, nec blandit nisl mauris at pede. Suspendisse risus risus, lobortis eget, semper at, imperdiet sit amet, quam. Quisque scelerisque dapibus nibh. Nam enim. Lorem ipsum dolor sit amet, consectetuer adipiscing elit. Nunc ut metus. Ut metus justo, auctor at, ultrices eu, sagittis ut, purus. Aliquam aliquam. Etiam pede massa, dapibus vitae, rhoncus in, placerat posuere, odio. Vestibulum luctus commodo lacus. Morbi lacus dui, tempor sed, euismod eget, condimentum at, tortor. Phasellus aliquet odio ac lacus tempor faucibus. Praesent sed sem. Praesent iaculis. Cras rhoncus tellus sed justo ullamcorper sagittis. Donec quis orci. Sed ut tortor quis tellus euismod tincidunt. Suspendisse congue nisl eu elit. Aliquam tortor diam, tempus id, tristique eget, sodales vel, nulla. Praesent tellus mi, condimentum sed, viverra at, consectetuer quis, lectus. In auctor vehicula orci. Sed pede sapien, euismod in, suscipit in, pharetra placerat, metus. Vivamus commodo dui non odio. Donec et felis. Etiam suscipit aliquam arcu. Aliquam sit amet est ac purus bibendum congue. Sed in eros. Morbi non orci. Pellentesque mattis lacinia elit. Fusce molestie velit in ligula. Nullam et orci vitae nibh vulputate auctor. Aliquam eget purus. Nulla auctor wisi sed ipsum. Morbi porttitor tellus ac enim. Fusce ornare. Proin ipsum enim, tincidunt in, ornare venenatis, molestie a, augue. Donec vel pede in lacus sagittis porta. Sed hendrerit ipsum quis nisl. Suspendisse quis massa ac nibh pretium cursus. Sed sodales. Nam eu neque quis pede dignissim ornare. Maecenas eu purus ac urna tincidunt congue. Donec et nisl id sapien blandit mattis. Aenean dictum odio sit amet risus. Morbi purus. Nulla a est sit amet purus venenatis iaculis. Vivamus viverra purus vel magna. Donec in justo sed odio malesuada dapibus. Nunc ultrices aliquam nunc. Vivamus facilisis pellentesque velit. Nulla nunc velit, vulputate dapibus, vulputate id, mattis ac, justo. Nam mattis elit dapibus purus. Quisque enim risus, congue non, elementum ut, mattis quis, sem. Quisque elit. Acknowledgments So Long, and Thanks for All the Fish [2]. References [1] Robert Gentleman, Kurt Hornik, and Giovanni Parmigiani. Applied Spatial Data Analysis With R. Springer. [2] Douglas Adams. So Long, and Thanks for All the Fish. Hitchhiker s Guide to the Galaxy, 4:1, 1984.