Ryan D. Enos and Anthony Fowler are assistant professors at Harvard University’s Department of Government and the University of Chicago’s Harris School of Public Policy, respectively.
For much of the last half-century, presidential campaigns have been waged over the airwaves of television and radio. However, in the last two elections, fueled significantly by political science research, campaigns have turned to the “ground game,” using get-out-the-vote (GOTV) methods such as door knocks, direct mail, and phones calls to encourage their voters to show up at the polls. The Obama campaign largely embraced these methods in 2008 and 2012, and Republicans, with an eye to the next election, are reportedly now trying to catch up. However, the actual effect of these new GOTV methods remains disputed: many political scientists have argued that these efforts had little effect on the election, and many think that whatever effects there were advantaged Obama. Both conclusions are premature.
We assess the aggregate effects of GOTV in the 2012 presidential election in a new working paper, building upon preliminary results that we reported at The Monkey Cage in May. The 2012 presidential election is particularly ripe for investigation because GOTV was deployed at an unprecedented level, and the Electoral College creates variation in campaigning across states. This variation allows us to compare people who were targeted by the GOTV campaigns to those who were not.
In our research, we utilize information about the operations of the Obama and Romney campaigns and individual-level data on every eligible voter in the United States to estimate the aggregate effects of GOTV. Specifically, we compare individuals in the same television media market but different states, allowing us to account for the effects of news coverage and television advertising. We consistently find that individuals living in states that received concentrated GOTV efforts from the campaigns were much more likely to turn out to vote compared to demographically similar individuals in the same media market who lived in a state receiving less GOTV effort. We also wanted to be sure that our estimates did not also include the effect of previous campaigns, so we examined turnout among these same individuals in 2010 and found that their turnout in that year was nearly identical. This gives us confidence that there were no systematic differences between our comparison groups before the 2012 campaign resulting from chance or from previous campaigns. We conclude that GOTV increased voter turnout by approximately 7 percentage points in the most heavily targeted states. Our back-of-the-envelope calculations suggest that the effects of multiple GOTV interventions are largely additive, meaning that each subsequent intervention is just as effective as the first.
The figure below provides one visualization of our results. Each circle represents the intersection of two states in the same media market, and the sizes are proportional to population. The vertical axis shows the turnout differential between these two regions (where higher values indicate higher turnout in the state that received more GOTV), and the horizontal axis shows the campaign effort differential between the two states as indicated by internal documents from the campaigns. Specifically, we received data from the Romney campaign on the number of door knocks, phone calls, and pieces of mail attempted in each states, and we combined these numbers (weighting each form of contact by its approximate cost) to create a continuous measure of GOTV effort for each state. Further data from the Obama campaign reveals that they prioritized states in virtually the same way. As the difference in GOTV effort goes from 0 to 1 — that is, from the least targeted states like Wyoming and Massachusetts to the most targeted states like Colorado and New Hampshire — turnout increases by 7 percentage points on average. Moreover, this relationship is approximately linear.
One potential concern is that voters in targeted states may be more likely to turn out even in the absence of GOTV. For example, these voters may know that they are in a swing state which may motivate them to turn out in the absence of campaigning. But this seems unlikely for multiple reasons. For example, two previous studies (here and here) attempt to estimate the direct effect of competitiveness on turnout and find little evidence. Moreover, we find very small differences in turnout between swing and non-swing states before 2008, when GOTV efforts were smaller, and we find little difference in political interest and intention to vote between targeted and non-targeted states before the general election campaigning began.
Our empirical design also allows us to partially assess the relative effectiveness of the Obama and Romney campaigns in mobilizing their respective supporters. To do this, we predict each individual’s partisanship using demographic characteristics, and we separately estimate the effect of GOTV for six different categories of predicted partisanship, from the most Republican to the most Democratic.
The figure below presents these estimates graphically along with 95 percent confidence intervals. As expected, we detect the greatest effects of GOTV among strong partisans — those who were more likely to be targeted by the campaigns. Among this subset, we detect GOTV effects greater than 10 percentage points. However, despite the media accounts of the superiority and sophistication of the Obama campaign, we estimate similar effects for the most Republican and Democratic subsets of individuals, as seen on the left and right sides of the figure, respectively. In other words, both campaigns appear to have been very effective in mobilizing their supporters, and there is no evidence that Obama’s campaign was more effective than Romney’s.
Clearly, these results differ from media accounts which praise the Obama campaign for its technological sophistication. They also differ from a recent post by Aaron Strauss who, using similar data and methods, estimates much smaller effects of GOTV and a notable difference between Obama and Romney. What explains these different results? One critical difference is in how the two analyses try to make sure that “all else is equal.” While Strauss’s work uses Catalist’s turnout propensity model to compare individuals across states, our analyses instead employs individual-level demographics to compare individuals in the same media market. As Strauss explains in a follow-up post, analyses that use turnout propensity are likely to be conservative. Our own analyses suggest that Catalist’s turnout propensity scores systematically differ between targeted and non-targeted states, explaining why Strauss’s approach may underestimate the effect of GOTV and overestimate the relative effectiveness of the Obama campaign.
Our results also differ from those of John Sides and Lynn Vavreck, who use a different measure of voter contact (the location of field offices in a county) to estimate a small but non-decisive ground advantage for Obama. To some extent, differences arise because they study a different question than we do. When we conduct a more direct analysis of GOTV and turnout, we obtain much larger estimates of campaign effects but no advantage for Obama.
As it turns out, GOTV plays a very important role in modern elections and the uneven deployment of these interventions means that campaigns are dramatically altering the size and shape of the electorate compared to previous campaigns when such tactics were not used. Moreover, whatever technological advantages the Obama campaign had did not significantly manifest themselves in GOTV effects. Both campaigns deployed significant resources in the traditional forms of GOTV, and both campaigns were extremely effective.