Does Government Dependency Influence Voting Behavior?
Alexis de Tocqueville once prophesied "The American Republic will endure until the day Congress discovers that it can bribe the public with the public's money." The Left vehemently denies that their social spending is anything but altruistic. Any correlation between federal spending on individual aid and voting behavior would reveal this altruism to be politically self-serving. The Left's denial (e.g., Klein, Matthews, Altman, Ponnuru) is often based on state-level observations and Moran has claimed that red states receive greater benefit from federal spending than blue states.The Right (e.g., Bauer and Romney) innately believes that federal largesse buys votes, but has provided little empirical data to back up their belief. Rayne has astutely countered claims of red state welfare, but has not addressed the issue of the influence of social spending on voting behavior.
The 2012 election results have been data mined to determine if a measurable correlation exists between social spending and voting behavior and demonstrate why state-level correlations are misleading. Accurate accounting of the influence of socioeconomic factors on voting behavior will be critical for the 2014 midterm and 2016 election strategies, as well as understanding the political dynamics of current issues including:the fight over federal spending; the expanding welfare, food stamp, and/or disability recipients; immigration reform/amnesty; and ObamaCare.
One key element of obtaining meaningful information from statistical analysis is the determination of the proper sample size. Large sample sizes can obscure trends that are obvious in smaller, more homogeneous subsets. Thus, it should not be unexpected that observations drawn from a statewide basis could mask correlations that exist on a much more local precinct basis. As Tip O'Neill once said "all politics is local."
A glance at any post-election map reveals that the country was sharply divided in 2012; with 21 states solidly for President Obama and 23 solidly for Governor Romney. The winning candidate received less than 52% of the total vote in the remaining six, aptly named, battleground states (Florida, Virginia, Ohio, Colorado, Pennsylvania, and North Carolina). The president won all of these battleground states except for North Carolina.
The voting data for Hawaii (solidly blue), Wyoming (solidly red), and Virginia (battleground) was sorted by the percentage of the vote that President Obama received in each precinct, as shown in Figure 1. As one might expect, almost all of the Hawaii precincts gave President Obama more than 50% of the vote and almost none of the Wyoming precincts did. The Virginia precincts were almost split down the middle. Notice that even the most partisan of states have some precincts that are solidly in the opposition's camp.Virginia had a particularly wide disparity in results, including one precinct that gave 99.5% of the votes to President Obama and another that gave him only 4.6% of the votes.The other five battleground states exhibited trends similar to Virginia's.
The local influence on politics is illustrated by examining the voting characteristics of the city of Richmond Virginia. In the 2012 election, 78% of the Richmond electorate voted for President Obama and a majority of the precincts were won by President Obama, as shown in Figure 2. One precinct gave the president 99.2% of the vote, but another gave him only 24%.The voters in the precincts at opposite sides of the voting spectrum will likely have different socioeconomic characteristics and observations drawn from a city level comparison would not capture these local differences as accurately as a precinct level comparison.
A socioeconomic example of this hierarchy of correlation is the comparison of the voting characteristics of different size regionswith the percentage of households that are "headed by a woman" (Census ID B17012). Figure 3a plots this percentage as a function of the percentage of the vote that President Obama received on a state level for all 50 states. Each black symbol represents an individual state and the line is a linear fit to the data. The value R2 (the coefficient of determination) is used to quantify how accurately the linear fit represents the data. A value of 0 indicates no correlation and a value of 1 indicates perfect correlation. The R2 value of 0.0087 indicates no correlation when compared on a state level. Reducing region size to a county level shows a weak correlation (a R2 value of 0.3), as shown by the blue symbols of Figure 3a. Finally, a strong correlation (an R2 value of0.6) is obtained when the region size is reduced to a precinct level, as shown by the red symbols in Figure 3b. This means that for the six battleground states, every 1% increase in a precinct's percentage of households headed by women resulted in almost a 1.3% increase in the percentage of the vote for President Obama.
One may argue that correlation shown in Figure 3b is biased because the African-American population has a larger percentage of households headed by women and African-Americans voted overwhelmingly for President Obama. That argument merits consideration, but the strong correlation still exists for precincts with populations comprised of less than 10% African-Americans, as shown by the yellow symbols of Figure 3b.
Did people vote to reelect President Obama because they received federal aid? Data mining cannot determine why people voted, but it can reveal how they voted. The percent of the precinct households receiving federal assistance (Census ID B19058) was plotted as a function of the percentage of the vote received by President Obama, as shown in Figure 4. The R2 value was greater 0.5, indicating a strong correlation between the two parameters. On average, every one percent increase in the number of households receiving federal assistance resulted in a two percent increase in the vote received by President Obama.
The data shown in Figure 4 does not mean that everyone receiving federal assistance voted for President Obama, but does indicate a strong correlation between the two parameters. Thus, receiving federal assistance was likely a factor in how some people voted, but perhaps not the only factor. For example, several precincts were dominated by universities and produced strong support for President Obama, but these precincts had low percentages of households receiving federal assistance (at least as counted by the Census database parameter ID B19058).
The precinct level examination of the 2012 election results shows a strong correlation between the saturation of federal assistance received in a community and the voting behavior of its residents. Furthermore, claims that federal assistance does not correlate with voting behavior based on state level comparisons are either intentionally deceptive or statistically worthless. Unsurprisingly to those who understand human nature, the recipients of federal aid have a strong propensity to support the politicians who provide that aid. Thus, the Left has an electoral incentive to do the easy work of increasing federal aid and aid recipients, and a disincentive to do the hard work of moving people off federal aid to become productive, self-supporting members of the Republic. Furthermore, it should not be a surprise that the Left has made no attempt to compromise on budget negotiations; spending reductions will weaken them, but the sequestration and Government shutdown will not affect their base of aid recipients.
David Waciski is a "Big Data" engineer and writer. He can be found on Twitter at @DWaciski.