That said, the Election Oracle was not created to be a Democratic or Republican victory scenario generator—although it certainly can be used as one. Our hope is that it will provide viewers with a glimpse into America’s future political landscape and the demographic changes that will reshape presidential politics over the next several cycles.
The simulation presets for 2004, 2008, and 2012 work by bringing together three pieces of information:
• Eligible voter composition: Projections about the future race/age mixture of a state
• Turnout rates: The percentage of eligible voters in a given race/age group that voted
• Support rates: The percent of voters in a given race/age group that will support a Democratic or Republican candidate.
The details of how these values were generated is expanded upon below, but the general idea is that we estimated unique turnout and support rates for relatively detailed demographic groups—1,020 state, race, and age groups to be specific—and then applied these rates to projections of the size of these groups in each state over the next several electoral cycles.
More succinctly: Those presets assume the behavior of the American electorate stays the same as it was in that election year, but the size of different demographic groups changes over time.
The two conditions users are able to alter are the national turnout and support rates among different racial groups. Turnout rates reflect the percentage of eligible voters—people who are at least 18 and a U.S. citizen—that will go to the polls on Election Day. Support rates are the percentage of those voters who vote for either a Democratic or Republican candidate. In short, altering turnout affects who shows up and altering support affects who they vote for.
However, when a user alters one of those values—say, increasing white turnout by 1 percent—it does not suddenly set the turnout rate of all white groups to that value. That would gloss over many of the differences that exist between the various parts of the American electorate. Instead, the unique turnout rates that originally existed for each of the white state, race, and age groups are adjusted so that their collective increases result in a 1 percent increase nationally.
Our States of Change population projections employ a multistate cohort component methodology that begins with the 2010 census and projects ahead in five-year intervals for race- and age-specific populations for each state to 2060 based on the components of domestic migration, international migration, fertility, and mortality. The projections are based on modeling put forth in Andrei Rogers’ Introduction to Multiregional Mathematical Demography. These projections are performed separately for racial groups wherein the states’ domestic migration flows are projected between the state and the remainder of the four census regions—the Northeast, the Midwest, the South, and the West. International migration to the United States for each interval is allocated to states and regions. In both cases, these migration flows and immigration allocations are based on patterns recorded in the 2007 to 2012 multiyear American Community Survey. Race-specific fertility and mortality rates for each state assume national rates specific to age and race.
Using those projections as a baseline, we also employ a demographically based eligibility projection model. The first step in this process was taking data from multiple years of the American Community Survey and dividing up the American population into groups based on state, race, and age—for example, Hispanics ages 30 to 34 in Colorado. Multilevel statistical models were then used to estimate the unique eligibility rates—the rate of citizenship among a given group—and naturalization rates—the rate at which these groups gained citizenship over time—for each state, race, and age group. These groups were then tracked forward in time and had those unique naturalization rates applied to them as they moved into older age groups. Additionally, these estimates account for the influx of immigrants into each state, race, and age group and the effect they have on those groups’ overall eligibility rates. The end result is a procedure that is sensitive to the different rates of naturalization each of these groups experience, as well as the immigration each state is predicted to experience in the future.
For more information about these projections, see our report or the interactive.
To start, the U.S. eligible voter population was broken down in 1,020 groups: five racial categories, four age groups, and 51 geographies. A unique turnout rate was estimated for each state, race, and age group using data from the 2004, 2008, and 2012 November Supplement of the Current Population Survey and multilevel modeling techniques. This approach provides more accurate estimates of support for low-sample populations by partially pooling data across individuals’ geographic and demographic characteristics.
For the 2012 election, we derived Democratic and Republican support rates for each state, race, and age group using data from the 2012 Cooperative Congressional Election Study and multilevel modeling techniques. Data from the 2012 November Supplement of the Current Population Survey, the National Election Pool’s 2012 Exit Polls, and 2012 state-level elections results were then incorporated to harmonize these estimates with other observable features of the 2012 election and electorate.
This harmonization is important because many other popular estimates of support rates either generate election results that deviate from the true election results when combined with plausible turnout rates or propose implausible turnout rates. For example, if we simply combine support rates from the National Election Pool’s exit polls with turnout rates derived from the Current Population Survey—widely considered the gold standard for determining the electorate’s composition—we would find that the results varied significantly from observed election outcomes. The end result of the processes employed here are support rates that are specific down to the state level and completely compatible with the best estimates we have for group turnout rates and election results in the 2012 election.
The support rates for earlier elections are estimated using essentially the same procedures but with different data. The main data sources for 2008 and 2004 are, respectively, the 2008 Cooperative Congressional Election Study and the National Election Pool’s 2004 Exit Polls.