The presidential election of Donald Trump took many experts and lay individuals alike by surprise. Post-mortem analyses of the election results have attempted to shed light on the unprecedented victory by relying on a host of individual differences. To what extent could “Big Data” predict the results of the 2016 U.S. presidential election better than more conventional sources of aggregate measures? This study compares the predictive value of “Big Data” sources (e.g., Google search trends) with more conventional measures of aggregate, state-level data.
Interestingly, the power of Google search data is a more potent predictor than other aggregate measures, including previously used implicit attitudinal data (i.e., the race IAT) and even behavioral measures of racial bias (i.e., unarmed police shootings of African Americans and hate group activity). Despite the growing number of studies on Google search data as a predictive tool in general, few have directly tested how Google data compare to other forms of aggregate data. In the context of political behavior, these findings show that Google search can be a better predictor of outcomes than aggregated implicit measures or behavioral indices. The role of Big Data sources like Google trends lies in supplementing such traditional sources of self-reported data, particularly in cases where additional factors—like White identity politics—may be difficult to measure directly. Big Data may ultimately prove to be most useful when utilized in conjunction with more conventional forms of survey data.
To what extent could “Big Data” predict the results of the 2016 U.S. presidential election better than more conventional sources of aggregate measures? To test this idea, the present research used Google search trends versus other forms of state-level data (i.e., both behavioral measures like the incidence of hate crimes, hate groups, and police brutality and implicit measures like Implicit Association Test (IAT) data) to predict each state’s popular vote for the 2016 presidential election. Results demonstrate that, when taken in isolation, zero-order correlations reveal that prevalence of hate groups, prevalence of hate crimes, Google searches for racially charged terms (i.e., related to White supremacy groups, racial slurs, and the Nazi movement), and political conservatism were all significant predictors of popular support for Trump. However, subsequent hierarchical regression analyses show that when these predictors are considered simultaneously, only Google search data for historical White supremacy terms (e.g., “Adolf Hitler”) uniquely predicted election outcomes earlier and beyond political conservatism. Thus, Big Data, in the form of Google search, emerged as a more potent predictor of political behavior than other aggregate measures, including implicit attitudes and behavioral measures of racial bias. Implications for the role of racial bias in the 2016 presidential election in particular and the utility of Google search data more generally are discussed.
Using “Big Data” Versus Alternative Measures of Aggregate Data to Predict the U.S. 2016 Presidential Election
Christine Ma-Kellams, Brianna Bishop, Mei Fong Zhang, Brian Villagrana
2018, Vol. 121(4)