The cities included in this analysis were drawn from: 1) the 30 largest U.S. cities (based on the U.S. Census Bureau's 2014 Population Estimates) and, 2) 20 Knight (resident and non-resident) communities (http//www.knightfoundation.org/what-we-fund/engaging-communities).
Of the 50 largest cities in the U.S., more than three-quarters of them hold mayoral elections in odd-numbered years. Many of these cities set their election dates over a century ago, during the Progressive era, when "off year" contests were seen as giving voters more of a chance to focus on municipal-level issues (and in some cases, pay more attention to rooting out corrupt local officials).
Voters in several cities (e.g., Baltimore and Los Angeles) recently changed their election dates to align with presidential elections. Other cities are looking at this change as well, though in most cases it will require a vote of the people to amend the City Charter.
We use census tracts because they are the smallest geography at which we were able to obtain relatively reliable data from the U.S. Census and the American Community Survey (ACS).
Precinct boundaries bear little to no relationship to census tracts. While certain values tied to precincts would be more precise (e.g., actual votes cast as officially recorded by election officials), it would not be possible to analyze such voting patterns alongside key demographic data (i.e., income and education levels, etc.), which is the core purpose of our study.
For some cities, census tracts nest neatly into city boundaries, while for others city boundaries meander and bisect multiple census tracts, effectively splitting tracts at the city's edges. In these cases, we only include tracts with at least 90 percent of the population living within city limits (we calculated the percentage by aggregating census blocks).
You can search for your specific census tract number by using the U.S. Census TIGERweb application (https://tigerweb.geo.census.gov/tigerwebmain/TIGERweb_main.html). Check the Census Tracts and Block Groups layer and zoom into your city. Once you get close enough the tracts and block groups will appear with their labels.
Please report any errors or omissions to firstname.lastname@example.org.
The most common denominator used by elections officials and journalists to calculate voter turnout is to divide the number of ballot cast in a given election by the number of registered voters. However, in addition to registration statistics being sometimes unreliable, this approach also excludes non-registered citizens, and thus artificially inflates voter turnout (higher than it really would be) by not taking Voting Age Citizens taken into account.
At the other extreme are studies that gauge voter turnout based on Voting Age Population (VAP), a measure that also includes individuals who are not U.S. Citizens. In this study, we concluded that the best denominator to use is the Citizen Voting Age Population (CVAP), a special tabulation of American Community Survey (ACS) data provided by the U.S. Census Bureau (https www.census gov/rdo data/voting _age _population _by _citizenship _and _race _cvap html)
CVAP, which we label Voting Age Citizens for more clarity, is the number or percentage of individuals that are 18 years or older and who are U.S. citizens. Unfortunately, no reliable studies exist at the citywide level, that calculate the number of citizens who should be excluded from this list because they are felons or ex-felons barred by state law from voting. Nor do any studies calculate the number of citizens who should be added back in, because while they don't reside in a city they still consider it their home for voting (e.g., military personnel serving overseas).
A 2014 U.S. Census Bureau report on the Midterm elections (https://www.census.gov/content/dam/Census/library/publications/2015/demo/p20-577.pdf)
reveals the national median ages for VAP (46) and CVAP (47) are relatively similar. And while non-citizens of voting age will vary considerably across urban areas, depending on such factors as race/ethnicity, there is less a discrepancy than many believe. For example, among young Hispanic residents 18-34 years of age, national studies show that CVAP, as a percentage of VAP, is close to 90 percent.
The Who Votes for Mayor study examines each city's most recent mayoral election, and therefore includes elections held during the 2011-2015 election cycles. We began collecting voter data in late 2015, when we started the project, and received all requested data by early 2016. In our experience, because most elections offices do not maintain a historical archive of individual voters at the time of a given election, voting data provided by each jurisdiction instead reflect voter registration at the timeof the request. Since population changes (e.g., registered voters moving) between when the election was held and the time of the data request, there are unavoidable discrepancies between the "official" results and the data used in this analysis.
This 'temporal mismatch', combined with states adopting varied approaches for keeping voter rolls current, poses several challenges for this type of research. For example, some census tracts in Chicago (and in other cities) contain more registered voters than Voting Age Citizens. This anomaly is likely the combined result of: 1) population decline; 2) temporal distortion between the Election Date and when the census data was gathered, and; 3) the likelihood that many registered voters still listed on the rolls no longer live within the given geography.
To take one extreme example, Chicago census tract 4605 shows 3,905 voting age citizens, and 4,966 registered voters, which are 2010-2014 (CVAP) and 2015 (Chicago Board of Elections Commissioners) data, respectively. For this census tract, the temporal mismatch is particularly salient given a rapid population decline in this tract (more than a 17 percent decline in population during the 2000s). Since the 5-year ACS acts like an average, the farther away from the midpoint of 2012 the election and voter data are, the greater the likelihood of the temporal distortion.
That said, our pilot project and research for this larger data sets reveals that the temporal mismatch problem is nowhere large enough to affect the overall patterns revealed in our data. However, we do hope that one outcome of this study is that local election officials in the future will decide to capture and preserve “snapshots” of their voter data files after each municipal election, so that future research could be based on individual voter addresses and participation status that is as close as practicable to the actual election date (see FAQ #16 for more details).
Both ACS and Citizen Voting Age Population (CVAP) data from the U.S. Census Bureau contain a corresponding MOE, which the bureau reports at a 90 percent statistical confidence level. Although MOE values should be, and typically are reported alongside ACS and CVAP estimates, in this analysis we do not report MOE values. As such, we employ larger geographic units of analysis (i.e., census tracts), which contain lower MOE values than small census geographies (e.g., block groups).
The Who Votes for Mayor study successfully geocoded over 95 percent of the more than 23,000,000 voter records used in this analysis. Voter records that were not matched to a physical address, and instead were matched to a zip code or city center, are excluded from the analysis. The percent of records considered usable varied by city, and depended on the quality of the data provided, as well as the geography of the municipality. For example, the lowest percent of usable records were in Columbus, OH, largely due to a number of unincorporated areas within the greater city boundary. The highest share of usable records was in Detroit, which can be attributed to data cleaning provided by our neighborhood data partner, Data Driven Detroit.
Depending on its electoral system, a city may have one or two stages of voting. The first round is either called the First Round or primary election, especially if a city has a partisan system with party nominations. The second round is often referred to as a "General Election" in a partisan contest, or a "Run-Off Election" in a non-partisan system should no candidate receive more than 50 percent of the vote in the first contest.
There are some exceptions such as San Francisco and St. Paul, which use the Instant Runoff Voting (IRV) system.
In some cities, the most recent election involved a candidate that ran unopposed. In these cases, there is often very low turnout or the turnout is attributable to other elections (e.g., city council seats or citizen initiatives). For example, in Fort Worth, TX we chose to examine the 2011 election because it was the most recent mayoral election with more than one candidate.
From the U.S. Census Bureau: "Following the Office of Managementand Budget's (OMB's) Directive14, the bureau uses a set of money income thresholds that vary by family size and composition to determine who is in poverty. If the total income for a family or unrelated individual falls below the relevant poverty threshold, then the family (and every individual in it) or unrelated individual is considered in poverty." For more information, see: http://www.census.gov/topics/income-poverty/poverty.html
While political party is often included in registered voter data, we chose not to include it in our analysis for several reasons. First, over 20 states do not register voters by political party; voters in these states can choose to participate in a given party's primary, but they do not “join” them as they do in other states. Second, there are also many inconsistencies in how data are collected across jurisdictions, and complexities about party rules that sometimes allow non-affiliated voters to participate in these contests (and other times, do not allow them to participate). For example, Pennsylvania asks for self-identified political parties, while many states only offer Democrat or Republican as options. Other states and counties apply a party designation based off primary participation.
Yes. All of the data used on this website are available for download here.
Two important reforms in particular would greatly improve and assist this research in the future.
First, we encourage state and local election officials to work together to greatly improve the accessibility and timeliness of basic voter data needed to conduct similar (and even more sophisticated) studies that explore voting patterns and demographics. The current situation around the availability of voter data far too much reflects a situation where some entities (e.g., political campaigns, state and national political parties, and private list vendors) automatically receive (or have the financial resources to purchase) complete voter files on a regular basis. These voter files typically contain far more data than what citizens and researchers need to explore and analyze voting behavior. In a few well-publicized situations, the exchange of these files has led to inappropriate disclosure of especially personal information (such as complete Social Security Numbers).
As noted earlier, what is of most interest to the research community and citizens generally is more basic information (e.g., voter address, age, gender, race/ethnicity [where applicable], and voting history), but these data are not always affordable to purchase or available at all, much less uniformly maintained and updated.
When it comes to voter data, the enormous problem of large qualitative differences between, and even within states, likely will not be solved any time soon. However, there are promising efforts underway, such as the Pew Foundation’s Election Registration Information Center (ERIC), whose goal is to eliminate duplicate voter records and keep them far more accurate across state boundaries. Still, there would be great value in creating some kind of national repository for local election voting records that would be updated on a regular schedule and maintained in ways to foster as much uniformity and accessibility as possible.
The second reform would be for local election officials to simply record “snapshots” of their voter files, as they exist at certain key points in the election cycle -- e.g, as of each major election day. City residents -- and voters -- are constantly in motion, and voting records will never be able to exactly reflect the situation on a given day. But capturing the “closest available” set of data, and preserving it for a period of time, would significantly reduce the “temporal distortion” effect, and allow for even more precise analysis of important voting trends. We believe the costs of doing this would be relatively small -- and the benefits significant.
Essential to the study design was collecting three key pieces of information for each jurisdiction: 1) voter addresses, 2) voters age (or year of birth) and, 3) whether a voter cast a ballot (or not) in one or both rounds of a given mayoral election. In all 50 states, it is mandatory that voters provide their name, address, and age (or date of birth). Gender information is collected in about 40 states (though voters are not required to provide it), and self-identified race/ethnicity information is collected in about a half dozen states – though again, on a voluntary basis, and using race/ethnicity categories (and instructions) that vary considerably.
Although voter registration records are public information,our research team quickly found that state and local laws vary widely as to how readily available these public voting records are for the research community, not to mention citizens generally. Some jurisdictions routinely publish such records on the internet, in easily downloadable form, and others readily grant access to these records, and/or charge nominal or no fees for these data files. At the other end of the spectrum are jurisdictions that only release such public information in aggregate form to certain entities and individuals (e.g., to state residents only, or to certain named state organizations). Some states have explicit, and arguably reasonable prohibitions on using lists for general commercial purposes. Some jurisdictions also charge considerable fees and/or make the data available only in difficult-to-use format .
Even in the vast majority of cases where our research team had to officially request data files, we received great cooperation from state and local election officials, and we greatly appreciate their assistance. We also appreciate the help we received in several cases from other academic institutions and partners within individual states.
The biggest data availability challenge involved three cities in the state of Indiana. Indiana state law automatically provides certain entities, such as political parties, media outlets, and state legislative leaders, with complete voting records. But for citizens, and research institutions outside and even within Indiana, individual counties are given wide discretion.
Officials in Allen County (home to Ft. Wayne) readily provided the voter data files we requested. Lake County (home to Gary) however, could only provide the files in PDF form, which made efficient data entry impossible. In the third case – Marion County, home to Indianapolis – officials said they could only provide us lists of voter names, without addresses or age. Given these limitations, we omitted both Gary and Indianapolis from this study; the latter is the only city among the largest 30 U.S. cities that is not part of this study.