Abstract
Imbued with unprecedented financial resources, the Obama 2008 presidential campaign established more than 700 field offices across the country, mostly in battleground states. To what extend did this form of campaigning actually affect the presidential vote? This article examines the county-level presidential vote in 2008 in eleven battleground states. The findings show that those counties in which the Obama campaign had established field offices during the general election saw a disproportionate increase in the Democratic vote share. Furthermore, this field office-induced vote increase was large enough to flip three battleground states from Republican to Democratic.
Following the historic 2008 presidential election, much of the post-election analysis credited Democrat Barack Obama's victory to his extensive field organization efforts. According to one Republican consultant, “One of the keys to Mr. Obama's success was building an unprecedented ground game manned by a multitude of idealistic, young voters.” (Jarmin 2008). Indeed, the importance of field organizations seems to be one of the key lessons emerging from the contest, no doubt shaping future campaign strategy. As a former Republican officeholder remarked, “If you have the money, you can duplicate the model [Obama]'s got” (Sherry 2008).
Before consultants begin to franchise the Obama 2008 model of campaigning, it is probably worthwhile to determine whether his campaign's innovations actually affected the presidential vote. After all, there is a tendency among consultants and pundits to assume that whatever the winning candidate did in the campaign must have been an effective strategy, but such assumptions rarely hold up to empirical scrutiny. In this article, I test the influence of Obama's local field offices on the county-level general election vote in eleven battleground states. Did Obama's field efforts mobilize voters, or would those voters likely have voted in similar numbers regardless of contact by the campaigns? And did these efforts have an impact on the outcome of the election?
The results suggest that Obama very likely would have won the national contest without these field offices, but that the offices had a measurable impact on the election, likely changing the results of several closely contested states.
Local Campaign Effects
Campaign effects are notably elusive from detection by political scholars. Berelson et al.'s (1954) and Lazarsfeld et al. (1948) finding that campaigns did little more than convince voters to do what they were already likely to do has been followed by a succession of studies showing that neither campaign spending nor advertising have particularly impressive effects on voters (Finkel 1993; Levitt 1994; Polsby et al. 2008). According to this line of research, most of the noise generated by campaigns simply helps bring voters’ preferences in line with the fundamentals of the political environment, including the state of the economy, the conditions of American foreign policy, and the popularity of the incumbent president (Rosenstone 1983; Lewis-Beck and Rice 1992; Gelman and King 1993; Bartels and Zaller 2001).
Nonetheless, some recent evidence has demonstrated that campaigns do shape voter behavior, although often in subtle ways. A well-orchestrated campaign, for example, can frame an election, guiding the public discussion of the campaign and of the political environment in a way that favors its candidate (Iyengar 1991; Vavreck 2009). Additionally, high-profile events by a presidential campaign, particularly convention speeches and visits by a candidate to an area, can produce a short-term boost in that candidate's favorability (Shaw 1999; Hillygus and Jackman 2003).
More directly relevant is recent experimental field evidence showing that campaign contacts can boost voter turnout. Gerber and Green (2000) found that their randomized nonpartisan messages could increase voter turnout by roughly six percentage points (see also Gerber and Green 2005; Imai 2005). This study confirmed the results of earlier research showing that a modest campaign contact (in the form of a phone call or a flyer left at a door) could have a substantial impact on voter participation (Gosnell 1927; Eldersveld and Dodge 1954; Eldersveld 1956). These various studies, while demonstrating convincingly an important campaign effect, all share a common limitation: they use nonpartisan campaign messages to try to affect voters. With very few exceptions (Nickerson et al. 2006), and for obvious reasons, field experiments have generally avoided using the sorts of explicitly partisan messages that campaigns commonly employ. As a consequence, it is less clear if partisan ground campaign efforts are effective at increasing a candidate's vote share.
This study seeks to address this gap in our knowledge with an observational study focused on the location and use of campaign field offices. Little research has focused on field offices, although there are a few exceptions. Nate Silver (2008), for example, in a short postelection analysis, finds that Obama tended to outperform his polling numbers in states where voters reported disproportionate contact by his campaign (although see Sides 2008). Campaign journalists often discuss the “ground game,” as well, although such descriptions tend to laud the organization without testing its efficacy. For example, just prior to the 2004 Iowa Caucus, the New York Times devoted several column inches to describing Howard Dean's and Richard Gephardt's impressive field organizations, noting that “neither Mr. Edwards nor Mr. Kerry can claim the same scale of ground operation” (Purdum 2004, p. 1). Both Edwards and Kerry, of course, finished well ahead of their more organized competitors in that contest (Cohen et al. 2008, p. 294).
Other scholarly research on the ground game has focused more generally on the effects of campaign contact (Kramer 1970; Rosenstone and Hansen 1993; Verba et al. 1995; Hillygus 2005), although such studies tend to rely upon voters’ ability to recall being contacted by specific campaigns, the accuracy of which is questionable (Price and Zaller 1993). There is a rich, emerging strain of campaign literature devoted to understanding the effects of candidates’ ground games (e.g., Magleby et al. 2007), although up until recently, campaign field staff efforts have received little scholarly attention.
To be sure, the considerable attention paid to presidential campaigns’ high profile speeches and use of television advertising is justified given how much money and effort campaigns devote to these activities. However, presidential campaigns also seek to target voters one-on-one. Indeed, recent campaigns are giving renewed attention to ground war activities, including direct mail, telephone calls, and especially, personal canvassing (Monson 2004). More than any other form of political communication, personal canvassing requires extensive coordination at the local level.1 Running a local canvassing operation involves recruiting large numbers of volunteers and sustaining their interest, generating and frequently updating neighborhood walking lists based on state or local electoral information, and making sure not only that volunteers are speaking with the appropriate voters, but that they aren't alienating them. Local campaigns are ideally suited to these tasks. In order to preserve some sense of continuity between the efforts of these local campaigns and the national one, the national campaign often establish field offices in critical areas throughout the state.
The Barack Obama 2008 presidential campaign makes for a particularly interesting subject for a study of local campaign organization. Blessed with enormous campaign coffers, this campaign was unusually aggressive in staffing field offices not just in state capitals, but also at dozens of locations in each of the battleground states. The Obama campaign established more than 700 field offices across the country, compared to fewer than 400 maintained by the McCain campaign (Luo and McIntire 2008). Since it was these offices’ task to mobilize supportive voters in their immediate vicinities, we might be able to detect very localized campaign effects. In theory, that is, a field office in a county could send volunteers to hundreds or thousands of targeted households within the area, but such influence might not be felt far beyond county lines. Volunteers tend to prefer to stay in their own communities, and voters are less likely to be influenced by campaigners from other communities (Nickerson and Feller 2008). An Obama campaign official confirms,
You wanted as many of your local people carrying your message as possible, as opposed to paid field organizers, or even imported volunteers from different parts of the country … . The more offices we had, the easier it was to empower your local organizers and your local volunteers to be part of that effort … . It was more efficient to have more offices (Rodriguez 2009).
Counties themselves are important political entities with which voters identify (Aistrup 1993). Indeed, many local political structures are established at the county level, and it is there that many electoral contests are waged. A presidential campaign established in a county can work alongside an existing county party structure to identify voters and turn them out.
For this study, I examine the influence of campaign activity at the county level. I use the establishment of a county-level field office as a measure of local campaign activity, and I look to see whether those counties with Obama field offices saw a disproportionately higher vote share for the Democratic presidential ticket.
The General Election
When seeking to measure the effect of a campaign's ground game, it is important to take account of national trends that affect voting behavior. We know, for example, that Obama significantly out-performed John Kerry's vote share from four years earlier, but he did so nearly uniformly across states, not just in those that saw campaign activity (Gelman 2008). Figure 1 shows a scatterplot of the two Democrats’ vote shares by state. The diagonal line is the Obama = Kerry line; if a state appears above that line, Obama out-performed Kerry there. The correlation between the two votes is.92, with Obama receiving an average of 5.85 additional percentage points over Kerry's vote in each state.
Consistency of the Two-Party Vote, 2004–2008.
Note.—Each data point is a state, charted by its democratic vote for president in 2004 and 2008. The diagonal line is the obama = Kerry line; if a state is above that line, then Obama outperformed Kerry in that state.
It should be noted that the Democratic vote increase between 2004 and 2008 was not perfectly uniform across states. Arkansas is somewhat below the Obama = Kerry line, perhaps reflecting some Republican voting among disaffected Hillary Clinton supporters in the state where she was first lady for 12 years. Obama underperformed in Louisiana, as well, perhaps due to the exodus of many African-American Democrats in the wake of Hurricane Katrina in 2005. Finally, one notes Obama's overperformance in Hawaii, his home state.
Nonetheless, the overall correlation between the two years’ voting patterns is clearly very high, suggesting that Obama saw a similar improvement in the Democratic vote share in both competitive and noncompetitive states. One could compliment Obama's campaign on the Democratic surge in such heavily contested states as Virginia or Indiana, but it's a stretch to credit the campaign for an increase in the Democratic vote in states like Utah, Idaho, and Vermont that saw essentially no campaign activity.
To look more closely at the extent of variation at the county level—the unit used in subsequent analysis—figure 2 reproduces the same scatterplot within Colorado, one of the primary battleground states of the 2008 contest. Here, the data points represent the Democratic presidential candidates’ vote shares in Colorado's counties in 2004 and 2008. The diagonal line is the Obama = Kerry line; in every county, Obama exceeded Kerry's performance. Within Colorado, the correlation between the two years is even higher than it was nationally, with an r of.99. This suggests remarkable consistency in the county-level vote from one election to the next.
2004 and 2008 Presidential Vote Shares in Colorado Counties.
Note.—Each data point is a Colorado county, charted by its democratic vote for president in 2004 and 2008. The diagonal line is the Obama = Kerry line; if a county is above that line, then Obama outperformed Kerry in that county.
I examined these county-level bivariate relationships within eleven battleground states,2 and the results were virtually identical. The correlation between the Kerry and Obama votes was at least.955 in ten of the states; the lowest (Indiana) was.90. It is not terribly surprising to find a great deal of consistency in the vote from election to election, particularly within a state. Even as people move in and out of regions, states and counties preserve much of their partisan identity over time, particularly in an era of strong party polarization.
That said, there are variations between the two years that deserve notice. In Pennsylvania, for example, Obama improved on Kerry's performance by a full ten points in the southeastern county of Lancaster and yet fell a point short of Kerry's performance in the western county of Armstrong. Did field offices explain such differences? Obama pursued an unusually aggressive field office effort in the 2008 election. Of the 877 counties under examination in these eleven states, Obama had opened at least one field office in 377 (43 percent) of them. Just in Pennsylvania, forty (60 percent) of the state's sixty-seven counties hosted Obama field campaign offices. And it is worth noting that the Obama campaign had a field office in over-performing Lancaster County but not in the under-performing Armstrong County.
Reorganizing the previous graph of Colorado counties into boxplots of those counties with and those without Obama field offices, shown in figure 3, seems to suggest that field offices mattered. The figure shows the distribution of the increases in the Democratic presidential vote from 2004 to 2008 for counties with and counties without Obama field offices. Although there is a good deal of variance in the degree of this Democratic vote increase, the counties with Obama offices had, on average, a significantly higher increase (6.3 percent as opposed to 4.5 percent), and no county with an Obama office saw less than a three-point increase in the Democratic vote.
Democratic Vote Increase in Colorado Counties.
Of course, there are plenty of reasons why the counties might have varied in their relative embrace of Obama over Kerry, campaign offices being just one of them. It is possible, for example, that Obama saw his greatest Democratic vote increases in counties that have experienced high levels of population growth. That is, perhaps it is new residents, departing left-leaning coastal areas in search of less expensive housing options, who have helped to turn states blue (Frey and Muro 2008). It is also possible that Obama's share of the vote increased disproportionately in counties with high percentages of minority voters who participated at higher rates due to the presence of an African American on the national ballot (Wamsley 2008).
To account for these alternative explanations, I estimate a regression model predicting the change in the Democratic proportion of the two-party presidential vote between 2004 and 2008, using each of the 877 counties in these eleven battleground states as units of analysis. (Two of these counties had insufficient demographic information and had to be dropped from the analysis.) The dependent variable ranges from −.084 to.171 with a mean Democratic vote increase of.046. I include a dummy variable that equals 1 for counties that hosted an Obama field office during the fall campaign and 0 otherwise.3 I additionally employ a dummy variable equaling 1 if the Kerry/Edwards campaign had a field office in that county in 2004.4 The regression also includes an interaction term of the Obama and Kerry dummies. This specification allows us to distinguish any Democratic vote boost in counties where only Kerry had a field office, where only Obama had a field office, and where both candidates had established offices.
Economic conditions can, of course, influence vote choice both at the national and local level (Cho and Gimpel 2009). I have thus included the growth in the unemployment rate in each county between July and October of 2008 as a variable.5 Also included in the equation are variables measuring the percent of the county that is African American, the percent that is Latino, the median age, the median income,6 the total population of the county, and the county's population growth between 2003 and 2007. Additionally, I include the Kerry percentage of the two-party vote in 2004 to control for the possibility that the Obama campaign might have selected disproportionately liberal (or conservative) counties to host its offices. A conversation with Matt Rodriguez (2009), the western states director for Obama-Biden 2008, confirmed that this was an exhaustive list of factors that the campaign used to select the location of field offices.7 I ran a fixed-effects regression, controlling for state. The results of this regression can be seen in table 1.
Variables Predicting Democratic Vote Increase, 2004–2008
| Coefficient | |
|---|---|
| Variable | (Standard error) |
| Obama county field office, 2008 | 0.008* |
| (0.002) | |
| Kerry county field office, 2004 | 0.005 |
| (0.010) | |
| Obama office × Kerry office | −0.013 |
| (0.011) | |
| Increase in unemployment (July–October, 2008) | 0.006* |
| (0.001) | |
| Percent population growth, 2003–07 | 0.006 |
| (0.020) | |
| Percent African American | 0.068* |
| (0.011) | |
| Percent Latino | 0.125* |
| (0.015) | |
| Median age | −0.001* |
| (0.000) | |
| Median income (in thousands) | 0.001* |
| (0.000) | |
| County population (in thousands) | 0.00001 |
| (0.000) | |
| Kerry share of two-party vote, 2004 | −0.043* |
| (0.011) | |
| Constant | 0.060* |
| (0.013) | |
| Observations | 875 |
| R-squared | 0.146 |
| Coefficient | |
|---|---|
| Variable | (Standard error) |
| Obama county field office, 2008 | 0.008* |
| (0.002) | |
| Kerry county field office, 2004 | 0.005 |
| (0.010) | |
| Obama office × Kerry office | −0.013 |
| (0.011) | |
| Increase in unemployment (July–October, 2008) | 0.006* |
| (0.001) | |
| Percent population growth, 2003–07 | 0.006 |
| (0.020) | |
| Percent African American | 0.068* |
| (0.011) | |
| Percent Latino | 0.125* |
| (0.015) | |
| Median age | −0.001* |
| (0.000) | |
| Median income (in thousands) | 0.001* |
| (0.000) | |
| County population (in thousands) | 0.00001 |
| (0.000) | |
| Kerry share of two-party vote, 2004 | −0.043* |
| (0.011) | |
| Constant | 0.060* |
| (0.013) | |
| Observations | 875 |
| R-squared | 0.146 |
Note.—Cell entries are fixed-effects regression coefficients, controlling for state, predicting the increase in the Democratic share of the two-party presidential vote between 2004 and 2008. Statistically significant (p ≤.05) coefficients are indicated by an asterisk.
Variables Predicting Democratic Vote Increase, 2004–2008
| Coefficient | |
|---|---|
| Variable | (Standard error) |
| Obama county field office, 2008 | 0.008* |
| (0.002) | |
| Kerry county field office, 2004 | 0.005 |
| (0.010) | |
| Obama office × Kerry office | −0.013 |
| (0.011) | |
| Increase in unemployment (July–October, 2008) | 0.006* |
| (0.001) | |
| Percent population growth, 2003–07 | 0.006 |
| (0.020) | |
| Percent African American | 0.068* |
| (0.011) | |
| Percent Latino | 0.125* |
| (0.015) | |
| Median age | −0.001* |
| (0.000) | |
| Median income (in thousands) | 0.001* |
| (0.000) | |
| County population (in thousands) | 0.00001 |
| (0.000) | |
| Kerry share of two-party vote, 2004 | −0.043* |
| (0.011) | |
| Constant | 0.060* |
| (0.013) | |
| Observations | 875 |
| R-squared | 0.146 |
| Coefficient | |
|---|---|
| Variable | (Standard error) |
| Obama county field office, 2008 | 0.008* |
| (0.002) | |
| Kerry county field office, 2004 | 0.005 |
| (0.010) | |
| Obama office × Kerry office | −0.013 |
| (0.011) | |
| Increase in unemployment (July–October, 2008) | 0.006* |
| (0.001) | |
| Percent population growth, 2003–07 | 0.006 |
| (0.020) | |
| Percent African American | 0.068* |
| (0.011) | |
| Percent Latino | 0.125* |
| (0.015) | |
| Median age | −0.001* |
| (0.000) | |
| Median income (in thousands) | 0.001* |
| (0.000) | |
| County population (in thousands) | 0.00001 |
| (0.000) | |
| Kerry share of two-party vote, 2004 | −0.043* |
| (0.011) | |
| Constant | 0.060* |
| (0.013) | |
| Observations | 875 |
| R-squared | 0.146 |
Note.—Cell entries are fixed-effects regression coefficients, controlling for state, predicting the increase in the Democratic share of the two-party presidential vote between 2004 and 2008. Statistically significant (p ≤.05) coefficients are indicated by an asterisk.
As this table shows, most of the suggested causes of increased Democratic vote shares were positive and statistically significant (p ≤.05). Most critically, even accounting for these alternative predictors, the presence of an Obama field office was associated with a 0.8 percentage point increase in the Democratic vote share in the county. Although this is not an enormous effect, it is worth noting that the presidential contests in North Carolina and Missouri were settled by margins smaller than this, and Indiana's margin was only slightly greater.
Notably, neither the Kerry office coefficient nor the interaction variable was statistically significant. This suggests that the Obama campaign succeeded where it went beyond what the Kerry campaign had done. In counties where both campaigns had set up offices, there was no net boost to the Democratic vote in 2008. However, Kerry had only established 125 county-level offices in these eleven states compared to Obama's 377. Obama saw the boost to his vote share, these figures tell us, when he set up an office in places that Kerry never did.
Unsurprisingly, economic conditions appeared to affect vote choice. Each percentage point increase in the local unemployment rate was associated with a 0.6-point increase in the Democratic vote share over the 2004 baseline, a result that was statistically significant. While county growth and population size seemed largely irrelevant, counties with high numbers of African Americans and Latinos saw disproportionate rises in the Democratic vote share. The coefficient for Latinos was actually twice that of African Americans, suggesting considerably greater activation of Latino Democratic voting by Obama. Age had a negative and statistically significant relationship with the Democratic vote increase, consistent with other evidence that younger voters became substantially more Democratic between 2004 and 2008 (Gelman and Sides 2009). Income, interestingly, had a positive effect, suggesting that Obama was able to make gains in some wealthier counties that had eluded Kerry. Finally, the Kerry share of the 2004 vote had a negative relationship with the Democratic vote increase, suggesting that Obama saw a greater Democratic vote increase within more conservative areas than within more liberal ones, perhaps reflecting a possible ceiling effect.
Obama, of course, wasn't actually competing against John Kerry so much as he was against another senator, John McCain. In table 2, I have specified the regression equation somewhat differently, using Obama's share of the two-party vote as the dependent variable. I have replaced the Kerry county field office variable with a McCain county field office one, and I have interacted it with the Obama field office variable. Otherwise, the two equations are specified identically.
Variables Predicting Obama's Share of the Two-Party Vote, 2008
| Coefficient | |
|---|---|
| Variable | (Standard error) |
| Obama county field office, 2008 | 0.006* |
| (0.002) | |
| McCain county field office, 2008 | 0.010 |
| (0.007) | |
| Obama office × McCain office | −0.004 |
| (0.007) | |
| Increase in unemployment (July–October, 2008) | 0.006* |
| (0.001) | |
| Percent population growth, 2003–07 | 0.007 |
| (0.020) | |
| Percent African American | 0.068* |
| (0.011) | |
| Percent Latino | 0.128* |
| (0.015) | |
| Median age | −0.001* |
| (0.000) | |
| Median income (in thousands) | 0.001* |
| (0.000) | |
| Total number of voters (in thousands) | 0.0000008 |
| (0.000) | |
| Kerry share of two-party vote, 2004 | 0.953* |
| (0.011) | |
| Constant | 0.059* |
| (0.013) | |
| Observations | 875 |
| R-squared | 0.895 |
| Coefficient | |
|---|---|
| Variable | (Standard error) |
| Obama county field office, 2008 | 0.006* |
| (0.002) | |
| McCain county field office, 2008 | 0.010 |
| (0.007) | |
| Obama office × McCain office | −0.004 |
| (0.007) | |
| Increase in unemployment (July–October, 2008) | 0.006* |
| (0.001) | |
| Percent population growth, 2003–07 | 0.007 |
| (0.020) | |
| Percent African American | 0.068* |
| (0.011) | |
| Percent Latino | 0.128* |
| (0.015) | |
| Median age | −0.001* |
| (0.000) | |
| Median income (in thousands) | 0.001* |
| (0.000) | |
| Total number of voters (in thousands) | 0.0000008 |
| (0.000) | |
| Kerry share of two-party vote, 2004 | 0.953* |
| (0.011) | |
| Constant | 0.059* |
| (0.013) | |
| Observations | 875 |
| R-squared | 0.895 |
Note.—Cell entries are fixed-effects regression coefficients, controlling for state, predicting the Democratic share of the two-party presidential vote in 2008. Statistically significant (p ≤.05) coefficients are indicated by an asterisk.
Variables Predicting Obama's Share of the Two-Party Vote, 2008
| Coefficient | |
|---|---|
| Variable | (Standard error) |
| Obama county field office, 2008 | 0.006* |
| (0.002) | |
| McCain county field office, 2008 | 0.010 |
| (0.007) | |
| Obama office × McCain office | −0.004 |
| (0.007) | |
| Increase in unemployment (July–October, 2008) | 0.006* |
| (0.001) | |
| Percent population growth, 2003–07 | 0.007 |
| (0.020) | |
| Percent African American | 0.068* |
| (0.011) | |
| Percent Latino | 0.128* |
| (0.015) | |
| Median age | −0.001* |
| (0.000) | |
| Median income (in thousands) | 0.001* |
| (0.000) | |
| Total number of voters (in thousands) | 0.0000008 |
| (0.000) | |
| Kerry share of two-party vote, 2004 | 0.953* |
| (0.011) | |
| Constant | 0.059* |
| (0.013) | |
| Observations | 875 |
| R-squared | 0.895 |
| Coefficient | |
|---|---|
| Variable | (Standard error) |
| Obama county field office, 2008 | 0.006* |
| (0.002) | |
| McCain county field office, 2008 | 0.010 |
| (0.007) | |
| Obama office × McCain office | −0.004 |
| (0.007) | |
| Increase in unemployment (July–October, 2008) | 0.006* |
| (0.001) | |
| Percent population growth, 2003–07 | 0.007 |
| (0.020) | |
| Percent African American | 0.068* |
| (0.011) | |
| Percent Latino | 0.128* |
| (0.015) | |
| Median age | −0.001* |
| (0.000) | |
| Median income (in thousands) | 0.001* |
| (0.000) | |
| Total number of voters (in thousands) | 0.0000008 |
| (0.000) | |
| Kerry share of two-party vote, 2004 | 0.953* |
| (0.011) | |
| Constant | 0.059* |
| (0.013) | |
| Observations | 875 |
| R-squared | 0.895 |
Note.—Cell entries are fixed-effects regression coefficients, controlling for state, predicting the Democratic share of the two-party presidential vote in 2008. Statistically significant (p ≤.05) coefficients are indicated by an asterisk.
The Obama county field office coefficient is again positive and statistically significant, and the coefficients on the other control variables are essentially the same as they were in table 1. One surprising finding, however, is that McCain's field office presence had a positive impact on Obama's vote share, although this coefficient is not statistically significant. Somewhat less surprising is that the interaction term is negative, suggesting that McCain was able to check Obama's gains in counties where both candidates had a field office presence. Statistically, however, this result is indistinguishable from zero. The overall lesson of this table is that Obama's field offices were helpful to their candidate while McCain's were not.8
These findings are consistent with the notion that local campaign organizations may be pivotal in elections. Interestingly, however, while these findings are robust in the aggregate, they lose their statistical significance in many of the battleground states, even though the coefficients remain generally positive. When the regressions are run within states, the county campaign office variable only remains statistically significant for Florida, Indiana, and North Carolina. The second column in table 3 shows the county field office coefficient (as derived from table 1) for each of the eleven battleground states.9 Standard errors appear in parentheses next to the coefficients.
State-Level Field Office Coefficients and Impact on Election
| Field office coefficient | Actual Obama share | |
|---|---|---|
| State | (Standard error) | of two-party vote |
| Colorado | 0.006 | 0.544 |
| (.008) | ||
| Florida | 0.033* | 0.514 |
| (.008) | ||
| Indiana | 0.031* | 0.505 |
| (.012) | ||
| Iowa | 0.010 | 0.547 |
| (.007) | ||
| Missouri | 0.005 | 0.499 |
| (.005) | ||
| Nevada | 0.011 | 0.564 |
| (.010) | ||
| New Mexico | −0.014 | 0.574 |
| (.011) | ||
| North Carolina | 0.014* | 0.502 |
| (.005) | ||
| Ohio | −0.008 | 0.519 |
| (.007) | ||
| Pennsylvania | −0.005 | 0.542 |
| (.008) | ||
| Virginia | −0.0002 | 0.531 |
| (.006) |
| Field office coefficient | Actual Obama share | |
|---|---|---|
| State | (Standard error) | of two-party vote |
| Colorado | 0.006 | 0.544 |
| (.008) | ||
| Florida | 0.033* | 0.514 |
| (.008) | ||
| Indiana | 0.031* | 0.505 |
| (.012) | ||
| Iowa | 0.010 | 0.547 |
| (.007) | ||
| Missouri | 0.005 | 0.499 |
| (.005) | ||
| Nevada | 0.011 | 0.564 |
| (.010) | ||
| New Mexico | −0.014 | 0.574 |
| (.011) | ||
| North Carolina | 0.014* | 0.502 |
| (.005) | ||
| Ohio | −0.008 | 0.519 |
| (.007) | ||
| Pennsylvania | −0.005 | 0.542 |
| (.008) | ||
| Virginia | −0.0002 | 0.531 |
| (.006) |
Note.—Statistically significant (p ≤.05) coefficients are indicated by an asterisk.
State-Level Field Office Coefficients and Impact on Election
| Field office coefficient | Actual Obama share | |
|---|---|---|
| State | (Standard error) | of two-party vote |
| Colorado | 0.006 | 0.544 |
| (.008) | ||
| Florida | 0.033* | 0.514 |
| (.008) | ||
| Indiana | 0.031* | 0.505 |
| (.012) | ||
| Iowa | 0.010 | 0.547 |
| (.007) | ||
| Missouri | 0.005 | 0.499 |
| (.005) | ||
| Nevada | 0.011 | 0.564 |
| (.010) | ||
| New Mexico | −0.014 | 0.574 |
| (.011) | ||
| North Carolina | 0.014* | 0.502 |
| (.005) | ||
| Ohio | −0.008 | 0.519 |
| (.007) | ||
| Pennsylvania | −0.005 | 0.542 |
| (.008) | ||
| Virginia | −0.0002 | 0.531 |
| (.006) |
| Field office coefficient | Actual Obama share | |
|---|---|---|
| State | (Standard error) | of two-party vote |
| Colorado | 0.006 | 0.544 |
| (.008) | ||
| Florida | 0.033* | 0.514 |
| (.008) | ||
| Indiana | 0.031* | 0.505 |
| (.012) | ||
| Iowa | 0.010 | 0.547 |
| (.007) | ||
| Missouri | 0.005 | 0.499 |
| (.005) | ||
| Nevada | 0.011 | 0.564 |
| (.010) | ||
| New Mexico | −0.014 | 0.574 |
| (.011) | ||
| North Carolina | 0.014* | 0.502 |
| (.005) | ||
| Ohio | −0.008 | 0.519 |
| (.007) | ||
| Pennsylvania | −0.005 | 0.542 |
| (.008) | ||
| Virginia | −0.0002 | 0.531 |
| (.006) |
Note.—Statistically significant (p ≤.05) coefficients are indicated by an asterisk.
Even if the presence of a campaign office didn't seem to matter in some states, however, this table suggests that it was determinative in others. The third column in this table shows Obama's actual share of the two-party vote in each of the states. To better understand the impact of the field offices, I conducted two simulations that estimate the election outcome in the state in the counterfactual condition in which Obama did not have a field office. In the first simulation, I estimate the statewide vote assuming those mobilized by the Obama field offices never turned out to vote. In other words, I remove a share of Obama's vote commensurate with the size of the field office coefficient from each of the counties that hosted such an office. In the second simulation, I assume that those mobilized for Obama instead voted for McCain. For example, the field office effect was an estimated 0.033 in Florida. Removing that share of the vote from Obama's vote in the counties with a field office brings Obama's two-party vote share in Florida down from 0.514 to 0.5003, a bare win. Placing that share of the vote into McCain's column brings Obama's statewide vote share down to 0.487, a loss.
In three of the states under analysis—Florida, Indiana, and North Carolina—Obama won the actual election but would have lost if the mobilized voters had instead voted for McCain. McCain would also have won Indiana and North Carolina had the mobilized voters simply chosen to stay home on Election Day. These three states were worth a total of fifty-three electoral votes—not enough to actually cost Obama the White House, but certainly enough to make it a much closer election.
It is curious that the field office effect is not statistically significant within each of the 11 battleground states. This is due at least in part to the smaller number of cases within states—it is easier to detect modest but important effects among 875 counties than among just a few dozen (Nevada has only seventeen counties). But there are likely other forces at work here. One hesitates to make strong inferences from a dataset of eleven states, but it is possible to conjecture about the relative influence of field offices.
In particular, it seems possible that the effect of field offices might decline as a state becomes saturated with them. Figure 4 shows a scatterplot of the state-level Obama field office coefficients (as shown in figure 3) plotted against the proportion of counties with such field offices. We see that those states in which the counties were most inundated with field offices showed a smaller field office effect. This relationship is statistically significant (p =.014). This trend suggests there might be diminishing returns in effectiveness. Although the exact mechanism for this relationship is unknown, one possibility is that voters might respond negatively to excessive campaigning or potential volunteers might be confused or irritated by the multiple requests for involvement. It may also be that the campaign placed a greater focus on building offices in some states than on developing a strategy for their use.
Effect of Obama Field Offices by Office Saturation.
There are certainly other possible explanations of the state-to-state differences in field office effectiveness. Notably, the field office coefficient was statistically indistinguishable from zero in Colorado, Nevada, and New Mexico—three states that share a border with John McCain's home state of Arizona. It is conceivable that familiarity with McCain in the West muted Obama's effectiveness there. It is also possible that the field offices were more effective in recruiting volunteers in those states with closer elections.
This examination of state-to-state differences in field office effectiveness should be considered preliminary in nature, as the number of cases is so small. It is certainly possible that these differences simply come down to the technique, skill, and organizational abilities of the offices’ directors. Future research may shed light on how local campaign officials were able to boost vote shares in some battleground states but not in others.
Discussion
As the evidence presented in this paper suggests, the establishment of a local field office by a presidential campaign can yield modest but important dividends for a candidate. Obama's decision to establish hundreds of county-level offices helped to boost his vote share by almost one point overall and by more than three points within some states.
When analyzing any campaign effect, it is worth asking whether it matters for the final election outcome. The general election analysis suggests that three states, worth fifty-three electoral votes, may have gone Obama's way because of the effective allocation of field offices. While those electoral votes weren't pivotal in this contest, they were certainly enough to turn a tossup into an Electoral College blowout.
Any observational study, this one included, is potentially vulnerable to the criticism that correlation does not equal causation. That is, perhaps Obama would have received roughly the same vote shares in the counties with field offices had those offices never been erected. Perhaps the campaign simply established offices in those counties that were already looking very promising for the campaign.
This is certainly possible, but it doesn't seem particularly likely. The main rebuttal to this concern is that I have controlled for basically every political variable (racial demography, population age, previous voting behavior, population size and growth) that the Obama campaign used in deciding where to allocate its resources among the 877 counties in these eleven battleground states. Even controlling for all these variables, the presence of a field office was still associated with a significant vote boost over the 2004 baseline and a significant increase in the Democratic share of the two-party vote. It is additionally encouraging for this line of observational research that the findings essentially match those of experimental studies on the topic. Of course, further research, both observational and experimental, into the influence of field offices could increase our confidence in these findings.
Interestingly, the McCain field offices proved considerably less effective than the Obama ones. The McCain coefficient was statistically significant and in the correct direction in only one state (New Mexico), and it was dwarfed by Obama's fifteen-point victory margin there. Again, though, a field office can only be as effective as its volunteers, and as was often reported (Quinn 2008; The Pew Research Center for the People and the Press 2008), the McCain campaign had difficulty attracting enthusiastic supporters. It seems fair to say, however, that the Republican Party faced an unusually daunting array of crises in 2008, so it is unclear if the patterns observed here would show up in other election contexts. In particular, we might expect there to be more parity in the campaign efforts of the two sides, perhaps nullifying some of the Democratic advantage observed in 2008.
What this study ultimately suggests is that, in an era when campaigns sink more and more money into television advertisements with less and less to show for it, investing more in shoe leather may be a wise decision. Not only does it appear to actually move voters, but it also produces a number of positive externalities that we say we want from campaigns: greater individual involvement in politics, increased neighbor-to-neighbor contact, the education of volunteers and contacted citizens about the issues of the day, and increased feelings of efficacy among participants. Indeed, increased staffing of local campaign offices may be the better choice not only for campaigns, but for the nation as well.
References
For an entertaining look at just how complex this task can be, see Bai's (2004) study of the 2004 Bush presidential campaign.
Colorado, Florida, Iowa, Indiana, Missouri, New Mexico, Nevada, North Carolina, Ohio, Pennsylvania, and Virginia. Kerry won only one of these (Pennsylvania) in 2004; Obama won all of them but Missouri. I limited the analysis to these states because of the closeness of the polling and election results there and because this was where the bulk of campaign activity occurred on both sides. The two presidential candidates visited these eleven states a combined total of 338 times between September 1st and election day, accounting for more than three quarters of all the candidate appearances during this time period. Of these states, Iowa received the fewest visits (nine) and Ohio the most (seventy-five) (Washington Post Staff 2009). These eleven states were also home to more than half of the campaign field offices established across the country and much of the advertising and other forms of campaign expenditures and activities; any campaign effects should be found there.
Information on the location of Obama/Biden offices comes from the campaign website, which listed all offices by county and state (http://www.barackobama.com/).
Information on the location of Kerry/Edwards and McCain/Palin offices comes from George Washington University's Democracy in Action websites (http://www.gwu.edu/%7Eaction/2004/kerry/kerrgenstates.html and http://www.gwu.edu/%7Eaction/2008/mccain/mccainorg.html#s).
County-level unemployment figures were collected by the U.S. Bureau of Labor Statistics (http://www.bls.gov/data/home.htm).
Race, age, growth, and income statistics come from the U.S. Census. The most recent complete county income figures, unfortunately, are from 1999. However, they should still prove a reasonable indicator of relative county wealth.
According to Rodriguez, “We modeled out the voters pretty closely as to who we thought, when hearing the Obama message, were likely to come to us.” The primary factors he listed as important to locating field offices were population size, voting history, racial makeup, and proximity to a college or university. The regression model I employ controls for all of these. Proximity to a university is not measured directly, although the age variable is used as a proxy.
Another way of modeling these equations is in the form of a two-stage least squares regression, using the Obama office variable as the instrument, the unemployment rate and the McCain or Kerry office variable as exogenous variables, and the other control variables as endogenous ones predicting the likelihood of Obama establishing an office in the first place. Such models are designed to compensate for endogeneity of variables. However, one of the assumptions for instrumented variable analysis is the exclusion restriction; in this case, the instruments should be uncorrelated with Democratic vote share. This assumption is clearly violated in this analysis, so I have used OLS for the main analysis in the paper. Regardless, the results of the two-stage least squares model are substantively identical to those produced in tables 1 and 2.
Author notes
seth e. masket is assistant professor with the Department of Political Science, University of Denver 469 Sturm Hall, 2000 E. Asbury Avenue, Denver, CO 80208, USA. He wishes to thank David Ciepley, Tom Cronin, James Gimpel, Hans Noel, Michael Potere, John Sides, Wayne Steger, and Nancy Wadsworth for their comments and suggestions.
