Did Big Data Create Trump?
There has been much hand wringing and finger pointing by both parties recently over who is responsible for Republican Presidential Candidate Donald Trump’s seemingly inexplicable rise to front-runner for the presidential nomination. Despite the mutual blame that political elites share in shaping the electorate for good or ill, there is one culprit that has escaped the attention of academics and the political punditry: Data. In particular, the rise and widespread use of data analytics and microtargeting by political campaigns at the national and state level.
Since the early 2000’s, political campaigns have increasingly become more sophisticated in how they disseminate their campaign messaging to the public and mobilize voters on election day through the use of data analytics and algorithms. Much like Amazon or Facebook use consumer data to serve up relevant ads or products to consumers, campaigns have sought to leverage publicly accessible elections data with consumer databases to shape the electorate in favor of their candidate or party. Books such as Sasha Isenberg’s “The Victory Lab” and Eitan Hersh’s “Hacking the Electorate” detail how political campaigns (most notably President Obama’s 2008 and 2012 campaigns) have sought to leverage data science and “moneyball” approaches to the world modern politics. The dark side is that, as I will explain in this post, campaigns are given the ability to create pockets of politically isolated voters who become susceptible to demagoguery and extremism.
Ghost in the Machine
The path to the “Data to Trump” theory started when I viewed the graph below in a paper by David Nickerson and Todd Rogers entitled Big data and political campaigns. This graph plots the number of direct contacts by Democratic Party campaigns in the 2004, 2008, and 2012 election cycles in the state of Ohio (a pivotal state in presidential elections). The vertical axis is the likelihood a voter will turnout on election day based on the proprietary algorithm of the Democratic data vendor Catalist. The horizontal axis is the predicted (or imputed) degree of partisan affiliation for the Democratic and Republican Party as derived from proprietary Catalist algorithms. Within the graph, areas of red and yellow indicate little or no contact with a voter and green indicates heavy campaign contact.
As you can see from the graph above, the increasing availability of information on voter’s turnout likelihood and partisanship dramatically shapes who campaigns seek to persuade and mobilize on election day. Specifically, the campaigns have increasingly come to essentially ignore independent and Democratic voters who appear less likely to vote, and heavily mobilize very partisan and likely voters. While the tactic of mobilizing as many supporters as possible without mobilizing your opponents is a very effective within the scope of a single campaign, it’s shocking and saddening when you think about the potential long term impacts on the body politic.
Political campaigns — especially presidential campaigns — are how most Americans internalize and learn about the salient issues in political life. Despite the hype of cable news channels influencing voters, most average Americans are more likely to be watching Keeping up with the Kardashians or Monday Night Football than Bill O’Reilly and Rachel Maddow (See the work of Markus Prior for more on this). So, when campaigns start to target voters who maximize their electoral gains, they ignore this important civic responsibility and leave millions of potential voters in the dark — leaving them open to demagoguery and populist appeals that are heavy in emotion and light on substance. This may also lead to voters feeling distant from the party “establishment,” making appeals by political elites or rival candidates ineffective counterweights. This is not to say that these supporters are not thoughtful about their policy preferences, but rather that they believe Trump’s tone and approach reflect their own anger at being passed over and forgotten by their parties.
Show me the data!
Does this theory jive with data we have so far on Trump supporters and voters? The early answers say “yes.”
In their analysis of Trump supporters before the primary season, Nate Cohn and Civis analytics found that Donald Trump’s strongest levels of support were among voters with the lowest predicted turnout scores for the general election (based on Civis’ proprietary algorithm) and among unregistered voters and voters registered as Democrats. While this could be viewed as an indication of instability in potential support for Trump, I believe this reflects the impact of disaffected voters from both parties who, for multiple election cycles, have been largely ignored because proprietary forecasting models determined them unlikely to vote.
While the evidence seems suggestive, the data was taken before any ballots were cast in the early primary states. So could it be possible that these marginalized voters failed to show up on election day? Not likely, based on the exit polling from the early primary states. In Iowa, Trump won a plurality among voters who said they were caucusing for the first time (30%). I found similar results were present in New Hampshire, where Trump won a plurality of voters who were voting in a primary for the first time (38%).
The second potential impact I thought would occur from the increased use of micro-targeting would be a strong aversion to the existing political party elites or “the establishment.” I attempted to get at this question by using CNN exit polling from 15 of the early primary states (CNN unfortunately had no data for Alaska, Kansas, and Louisiana), graphing the proportion of voters who say they want a candidate outside of the Republican Party establishment. Looking at the graph below, we see that in nearly every state Trump has won so far, he has won a majority of these voters.
While this question may not fully capture a full rejection of the Republican Party by Trump supporters, CNN did directly ask voters in their South Carolina primary exit polling if they “feel betrayed by Republican politicians” and Trump won a plurality of these voters (36%). Both pieces of evidence also fit with the anecdotal evidence we hear from Republican Party officials who seem frustrated by their inability to persuade or shift supporters away from Trump. Your party is in trouble when former nominees (Mitt Romney and John McCain), a president (George W. Bush), and leaders in Congress cannot sway the support of their own primary voters.
Can we fix it?
The problem I see is as much one of data and statistics as it is one of politics. The widespread use of microtargeting and turnout forecasting models have given parties a huge tool to selectively carve up the electorate to their advantage. While these decisions might be lucrative to campaigns in the short run, they unfortunately appear to have wide ranging impacts on people’s sense of attachment to existing institutions and society. Trump, being the ultimate political entrepreneur, seems to have sensed this dislocation and has exploited it relentlessly for his own personal gain.
Is it possible to fix the damage done? I believe so, but it means re-thinking how we use data in campaigns and elections and considering how to include these voters into the electorate. Right now, political parties see no need to alter their tactics because they have been successful in shaping electoral outcome. However, if we continue to ignore large swathes of voters for multiple election cycles, we may face political upheavals that are too difficult to repair.
Sources
Hersh, Eitan D. Hacking the electorate: How campaigns perceive voters. Cambridge University Press, 2015.
Issenberg, Sasha. The victory lab: The secret science of winning campaigns. Broadway Books, 2012.
Nickerson, D. W., & Rogers, T. (2014). Political Campaigns and Big Data. The Journal of Economic Perspectives, 28(2), 51–74.
Prior, M. (2005). News vs. Entertainment: How Increasing Media Choice Widens Gaps in Political Knowledge and Turnout. American Journal of Political Science, 49(3), 577–592.
