Developing Geo-strategic Campaigns for Maximizing Grassroots Support: A Step-by-Step Guide Illustrated by the 2000 New York Senate Election

Summary:  The idea that scarce resources should be deployed strategically in grassroots electoral campaigns is hardly novel. Judgments have long been made about where investments of money and people will pay the greatest dividends in terms of votes. Often, these decisions are based simply on ‘eyeball’ assessments of past performance and political instinct. Such crude practices do not systematically evaluate how well a party or candidate should do as a consequence of the social, economic, demographic, and political characteristics of different parts of the district. As such, traditional campaigns find it impossible to assess with confidence or precision the efficiency of their past efforts and - therefore - their future potential. Barometrics Research, a pioneer in the application of geoscience to grassroots campaigns, addresses this challenge. In this piece, we introduce several of our analytic techniques in the context of a case study of voting in the 2000 New York state Senate election. Using a statistical analysis of the state’s roughly 13,000 Voting Tabulation Districts (VTD), we are able to advance powerful explanations for variations in the levels of support for both Democrats and Republican Senate candidates, and to account for variations across VTDs in voter turnout. We use these general statistical understandings to characterize the political opportunities and constraints presented by diverse VTDs across the state. Specifically, we develop two indexes – the “Campaign Efficiency Index” (CEI) and the “Get-Out-The-Vote Index” (GOTV-I) and use these to classify almost half of all VTDs in the state in terms of their potential for improved electoral performance. We proceed to illustrate how a GIS-enabled smart campaign could use our insights to help maximize the effectiveness of future campaigns. While not exhausting the range of analytic tools Barometrics Research can provide, the case study does highlight some of the enormous potential for grassroots campaigning that can be tapped using geographic information and analysis.

The State of New York is home to one of the nation’s largest and most diverse electorates. Democrats dominated the state’s registered voters (46.4% of all voters on the lists, compared to the Republicans’ 28.9%).  In 2000, this electorate was engaged in one of the more high profile and contentious elections of our time. At the top of the ticket, both the Presidency and one US Senate seat were up for grabs. In Presidential politics New York is solidly blue, and Gore handily defeated Bush in the state by a 65:35 margin. Campaigning for the open US Senate seat involved Rick Lazio for the Republicans and Hillary Clinton for the Democrats. By all accounts the competition was intense and – according to pre-election polls – remarkably close – but in the end Clinton won by a 56:43 margin.

In recent decades, divided government has been the norm within New York. The Assembly has been dominated by Democrats, whose support is concentrated downstate and particularly in New York City. On the other hand, the Republicans have controlled the Senate for all but one year since the Great Depression (the Democrats controlled the chamber in 1965-66). Their support is concentrated in upstate and rural New York and on Long Island. Going into the 2000 campaign, the Republicans held 35 of the (then) 61 Senate seats, while the Democrats had the other 26. Despite the traditional Republican “lock” on this house, Senate Democrats hoped to benefit from Presidential coattails in 2000. However, they entered that campaign at an enormous financial disadvantage. According to FEC reports filed in July of 2000, the Senate Republican Campaign Committee had $3.8 million in the bank, compared to the Democrats’ $260,000. Another indicator of the Republicans’ senatorial advantage in 2000 is the fact that the Democrats failed to run a candidate in 14 of the 61 districts, as compared to the 8 districts where Republicans did not field a candidate.

Today there are 62 senatorial districts in New York, a number that is too small to sustain a meaningful statistical analysis if these districts themselves were the focus of attention. However, each of these districts can be potentially subdivided into an infinite number of territorial subunits. Subdividing them into smaller units can radically increase the number of cases for statistical analysis while simultaneously allowing us to refine our appreciation of the political geography of the state. For example, there are over 13,000 “Voter Tabulation Districts” (VTDs) in the state, and most (all except 313, or 2.4%)[1] of these nest completely within the boundaries of the state senatorial districts (and also into the boundaries of various other units of political competition such as Congressional or Assembly districts). As such, information based on VTDs can be used to understand political patterns shaping electoral outcomes at a variety of contests within the state. In this case study, we use the VTD to fuse the electoral and census data necessary to conduct a thorough assessment of the 2000 NY state Senate race. On this basis, we illustrate how an analysis of this campaign can unlock valuable insights that can be used to plan subsequent campaigns.

Our primary purpose is to illustrate the power of geographic information and analysis for conducting smart campaigns, and to outline in general terms some of our analytic techniques. In this case study we ask: “how effective were the state Senate campaigns run by candidates of the two major parties in 2000?” How can the lessons learned from the 2000 campaign support geo-strategic decisions for future campaigns?  These are complex questions and answering them requires several steps. We first need to identify the sources of electoral support for each party as they are observed across the roughly 13,000 VTDs. The second step is to identify local variations from these general relationships.  Thirdly, we look at the pattern of voter turnout across these same VTDs. Fourth, we look for local deviations from the general patterns of voter turnout. Fifth, we employ the results of these first two steps to draw lessons from the 2000 campaign for future senatorial campaigns in the state. Sixth, we classify VTDs and develop distinct geo-strategic campaigns for each type of VTD we identify. Though we will not demonstrate it here, we can use our analysis to provide a street-by-street guide to assist a micro-targeted mobilization effort.

Step One.  Identifying the Sources of Democratic and Republican Senate Voting:

Our first step is to identify the sources of voter support for the two major parties that dominate Senate races in New York (for simplicity we ignore the role of minor parties – research by others suggests that since the 1940s their interventions have been decisive in only about 3% of all legislative races). Figure One, showing a map of the proportion of the vote going to Senate Democratic candidates in New York, establishes clearly that there are sizeable variations in the party’s performance across the 13,000 VTDs found in New York state. How can these large differences be explained?

Figure One: % Democratic Senate Vote, New York, 2000
(click here for a larger, more detailed view)

For this purpose, we draw on the academic research that conventionally depicts the Republicans and Democrats as drawing support from quite distinct social bases. Democrats tend to do well in areas with concentrations of poor, immigrants, and racial minorities. They do especially well in urban areas (and New City especially). By contrast, Republican support is stronger among whites, rural dwellers, and wealthier groups.  Using data from the 2000 census for the approximately 13,000 VTDs, we can identify with statistical confidence the precise contribution of a wide variety of social characteristics to determine the levels of support for the two major parties. Our statistical analyses for Democratic and Republican senatorial candidates results in a powerful explanation of the sources of each party's successes and weaknesses. Using the combination of variables we are able to statistically explain just under 80 percent of the observed variation in party support. Clearly, party choice in New York’s Senate elections is strongly related to the socio-economic and ethno-cultural characteristics of the state’s population.

Senate campaigns will want to know what kinds of groups are likely to give them support, and the results of these models will themselves be useful information for campaign strategists. In this respect, the patterns in party support we uncovered in our statistical analysis largely confirm our conventional understanding of the sources of partisanship in New York.  For example, our analysis makes it clear that Senate Democrats tend to do better in VTDs with concentrations of black residents – for every 1% increase in the proportion of African Americans in a VTD, the vote for Senate Democratic candidates in 2000 went up .330 percentage points. Similarly, VTDs with large immigrant populations were strongly Democrat – each additional 1% of foreign born residents across VTDs increased the Democrat Senate vote in 2000 by .197 percentage points.  Many other socio-economic and demographic/geographic factors were also included in the statistical analysis, resulting in a very robust and powerful explanation of patterns of Democratic support across the state.

Our statistical analyses also shows that on virtually every measure, the relationship of social characteristics to Republican support was the mirror opposite to that found in the analysis of the Democratic vote. This is precisely as we would expect in a party system with relatively well-defined and distinctive social bases. The portraits of party support that our analysis uncovers is very much in keeping with the conventional and academic wisdom.

Step Two. Calculating Local Deviations from the State-Wide Relationships – The Campaign Efficiency Index (CEI)

The relationships captured in our statistical models of party support reflect the general patterns that can be identified across VTDs statewide. Because of the wealth of data we are consulting to calibrate these models, the relationships uncovered are extremely precise and statistically reliable. These relationships, drawn from our analysis of patterns across all 13,000 VTDs in New York state, generate valuable insights when used to identify areas of the state where a party does particularly well or poorly in mobilizing the support of its social base. We can get this additional insight when we mathematically apply the measures of the general relationships that are captured in the statistical model to the specific situation of each VTD. Because we know the specific scores for each individual VTD on each of the census variables included in the model, and since we have now calculated the general relationship between each of these variables and party support across all VTDs, we are in a position to know what the level of party support should be in every VTD.  In other words, by combining the specific information for each VTD with the general information on the statewide relationships, we can identify a predicted level of party support for every district.[3]

Based on our statistical analyses, Barometrics Research assigns a “Campaign Efficiency Index” (CEI) score for each party in every VTD. This score is mathematically determined, based on our statistical analysis and the characteristics of each district. This figure reveals how much support a candidate received locally above or below that which would have been expected based the statewide relationships depicted in our statistical models. In our work, positive CEI scores indicate that the party's candidate received more votes than our statistical modeling would lead us to expect, and vice versa.

High CEI scores – indicating a local partisan advantage - can arise for any number of reasons, not all of them related to campaign efficiency. For example, a high CEI might arise in a VTD which includes the home residence of the senatorial candidate for that party. In such a case, the candidate might enjoy unusual local popularity as a result of a “friends-and-neighbors” vote that arises because they are especially well-known to other local residents. There are potentially many explanations for these deviations from the general pattern. Detailed local knowledge will be required to interpret these index scores.

Similarly, in some VTDs, CEI scores will be negative. In these areas, the party's candidate polled less well than our statistical model leads us to expect. Again, detailed local knowledge will be required to explain why the party’s promise was not realized by its campaign. Perhaps the area was targeted by a rival candidate for intensive campaigning. Maybe the candidate’s own party overlooked the potential to mobilize support in this part of the district. Certainly, some of these areas with negative CEI scores will represent areas of potential growth for the party, since the VTD has the population characteristics that normally support higher levels of Democratic voting. Knowing where these opportunities can be found is potentially invaluable for waging a smart campaign.

Figure Two visualizes the range of CEI scores for the Democrats in VTDs all over New York. As can readily be seen, despite the powerful statistical explanation we have developed to account for the varied levels of Democratic support, there remain significant variations across VTDs. Though we don’t show it here, essentially the same thing can be said for the campaign of the Republicans. We will return to a discussion and application of these CEI scores shortly.

Figure Two: Democratic CEI scores, NY Senate election, 2000

Step Three. Accounting for Variations in Voter Participation in the 2000 Senate Election

Our third step is to provide an account of the varying level of voter participation in Senate races using information about the VTD and its residents. Figure Three establishes that there are sizeable differences in voter turnout in the 2000 Senate campaign in different parts of New York state. How is this variation to be explained?

Figure Three: Turnout, NY Senate election, 2000

A great deal of research points to some general patterns in the social bases of voter participation, and we can explore for evidence of these relationships in our data for the 13,000 VTDs in New York. In general, conventional understandings of voter participation suggest that participation should be lowest in urban VTDs that are also characterized by high proportions of young people, of poor, single parent and large sized households, the residentially mobile, and the uneducated. Conversely, we expect that more rural, affluent, smaller and better educated, residentially-stable, and white electorates will manifest higher rates of voter participation.

(Detailed Statistical Results available upon request
from Barometrics Research)

The results show that turnout is well explained by reference to characteristics of the VTD: our model explains almost two-thirds of the observed variation across VTDs in turnout. The expected relationships between the social characteristics and the level of voter participation in the Senate race are generally confirmed. Voter turnout is higher in VTDs characterized by older, better educated, less urban, and wealthier electorates. Turnout is generally higher in upstate communities and those with large farming sectors.  Concentrations of African Americans and Spanish-speakers in VTDs also enhance levels of turnout, but immigrants tend to depress them.

Step Four. Identifying VTDs with Turnout Levels that Deviate from Statewide Norms: The GOTV-Index

As with our discussion of party support in 2000, it is also useful to identify VTDs in which the actual level of voter turnout in 2000 deviated from the predicted level based on the general statewide model results. For this purpose, we calculate a GOTV-Index (GOTV-I) that is generally comparable in terms of its calculation to the CEI index discussed above. These GOTV-I scores for VTDs will be negative when the observed turnout in 2000 was lower than our statewide model of the determinants of voter turnout suggests it should be. By the same token, some VTDs will have positive GOTV-I scores, signaling that the level of turnout at the 2000 Senate race was in fact higher than the general model’s relationships suggested it should be for a VTD with the population characteristics it has. Figure Four shows the range of the GOTV-I scores for the roughly 13,000 VTDs across New York.

Figure Four: GOTV Index Scores, NY Senate election, 2000

Step Five. Geo-strategic Choices for Smart Campaigns

Efficient grassroots campaigning involves knowing and adapting strategy and tactics to the characteristics of the local terrain. The 13,000 VTDs in New York state provide both parties with an enormous variety of opportunities and constraints when it comes to maximizing voter support. Smart party campaigns will begin by looking at the competitive situation in all Districts and decide which of those held by their opponents are winnable and which of their own seats are vulnerable. These would become the focus of campaign effort, since even small advantages in these settings can be decisive in winning the seat. The kind of analysis described above can help target campaign efforts to areas where they have the potential to be most effective.

The tools we have developed thus far in this case study provide wonderful guides in this respect. Considered both separately and jointly, the two indices - the CEI and the GOTV-Index  – can be used to identify VTDs that offer parties and candidates particularly promising opportunities to enhance and protect their vote, and on that basis, to improve their chances of winning or holding a district seat. All VTDs in the state can be classified according to their scores on both of these indices, and Figure Five outlines four distinct campaign environments that are defined by the intersection of these two dimensions. Smart campaigns will develop strategic orientations appropriate to the kinds of situations depicted in the figure.

Figure Five: Four Types of Campaign Environments

	Negative GOTV-I  Scores	Positive GOTV-I Scores
Negative CEI scores	Prime growth targets Type (1)	Hostile territory Type (3)
Positive CEI scores	Possible battlegrounds Type (2)	Eclipsed expectations Type (4)

Prime Targets  (top left quadrant):  These areas are “prime targets” for growth in vote share. In these VTDs, the party failed to attract its ‘normal’ level of support, and voters were not turning out at the normal rate. These environments call for a strategy emphasizing both targeted partisan appeals and ‘get out the vote’ drives. These will help the party reach its potential support level in VTDs where it has underperformed, and mobilize this support effectively on election day at a rate closer to the norm for districts with similar population characteristics. It is important to note that these low CEI scores do not mean that the party has done badly in the area. It is entirely possible that it did better than any other in terms of the proportion of the vote it won. However, the negative CEI index scores indicate that there remains untapped potential for further mobilization, if only the party could poll as well in that VTD as it does generally across the state. Therefore, it is likely that some of the strongest potential returns will come from aggressive campaign investment in these terrains.

 

Potential Battlegrounds (bottom left quadrant):  Smart campaigns will attempt to engage supporters in areas where they are already present in greater proportion than would be expected in terms of their population characteristics, but where the overall level of voter turnout is lower than it should be based on general relationships. More than one party may vie for the unharvested votes, however. For this reason, such settings may well constitute the prime battlegrounds for head-to-head partisan clashes. Barometrics Reseach can examine the party registration status of residents in these potential battleground VTDs, and identify those in which one party enjoys an enrollment advantage. When enrollment advantages exist, these potential battlegrounds can become important secondary targets for vote growth. While these secondary targets may not prove to be as productive in terms of new votes won as those in ‘prime target’ areas, effort expended here will also help to diminish possible effects of rival campaigns that may also target these areas

 

Hostile Territory (top right quadrant):  Geo-strategically smart campaigns will avoid wasting scarce resources on hostile territory (in the top right quadrant of Figure 5). In VTDs where a party underperforms in terms of its expected vote share despite higher-than expected local voter turnout, there is little prospect for gain by vigorous campaigning. These areas are very likely the strongholds of your opponent. Campaigns focusing on these terrains will at best diminish your rivals’ potential for growth

 

Eclipsed Expectations (bottom right quadrant):  Finally, smart campaigns will protect their partisan strongholds. These VTDs arise when a party has out-performed expectations in terms of vote share and residents have turned out to vote at higher than expected rates. There is, therefore, relatively little potential for further vote growth in these settings. However, these VTDs will be attractive as potential sites for aggressive fund-raising. Campaign investments should be geared to defending or maintaining the party’s local advantage.

 

The 2000 experience of the Senate Democrats can be used to provide an illustration of these types of campaign environments, and also suggest something of their practical significance.  For them, a VTD could be classed as a “prime target” if they under-performed expectations by at least 5% (CEI_Democrat <= - 5%) voter turnout was at least 5% lower than expected (GOTV-I <= - 5%) based on the VTD’s characteristics – a combination that arose in 1,397 (10.5% of all) VTDs. Figure Six identifies the geographic distribution of these areas across the state. A smart 2002 Senate campaign would have sought to mobilize greater support from these districts. Even in a system in which most seats are decided by large majorities like the NY Senate, some VTDs in this class will be located in competitive or marginal Districts. For example, in 2000, Republican Roy Goodman beat Democrat Liz Krueger in the 26^th Senate District in the Midtown/Upper East Side of Manhatten by a margin of only 200 votes (50.1% to 49.9%). Our analysis shows that there are 14 “prime targets” in the 26^th District for the Democrats (7.2% of all VTDs located there).

Figure 6: Prime Target VTDs

A VTD was classified as a “potential battleground” if the CEI _Democrat was at least + 5%, and if the GOTV-I score was below -5%. Applying these criteria in the state as a whole, 1,814 VTDs (13.6% of the total) fell into this category. Four of these were located in the tightly contested 26^th District (another 2.1% of the VTDs in that District).  From our analysis of the Senate Democrats, a VTD was classed as “hostile” if the CEI _Democrat was less than or equal to -5% and if the GOTV-I score was +5% or higher. Such a combination suggests that the intensely mobilized electorate favors another party. This category includes 2,914 (21.9% of the total) VTDs. In the 26^th District, fully 69.7% of the VTDs (136 of the 195) fall into this category. A smart campaign in 2002 would make use of this information to target its most aggressive efforts in the eighteen VTDs where strong gains could be expected, and avoid concentrating additional attention on the 136 VTDs where the party is weak despite stronger than average turnout. Finally, a VTD was classed as ‘eclipsing expectations’ for the Senate Democrats if the party’s CEI was greater than +5% and the GOTV-I was +5% or better. By these criteria, 1,389 (or 10.4%) of New York state’s VTDs fall into this category. Interestingly, Krueger’s candidacy in the 26^th District in 2000 did not generate any VTD performances of this type.

Applying the above operational definitions, we classify about half of all VTDs in the state – and also in the 26^th Senate District - into one of these four types. These distinctive micro-geographies could provide an important roadmap for strategies to plan and execute a maximally effective campaign in 2002. The remaining VTDs (i.e., the roughly half that do not fall into one of the four types discussed above) are environments in which either the actual observed level of Democratic support or the level of turnout in 2000, or both, was closer to the values that the models of the general determinants of these phenomena predicted. In these remaining VTDs, the general relationships captured in the statistical model provide quite accurate guides to the observed local reality. However, it is worth remembering that these VTDs also have both CEI _Democrat and GOTV-I scores, and appropriate campaign strategies identified for them. In a District as competitive as the 26^th, a smart campaign would be well advised to look carefully within this category for less obvious opportunities and constraints.  Figure 7 shows the geography of the district is quite sharply divided, with the VTDs adjacent Central Park typically being the less supportive of the Krueger candicacy than those nearer the East River.

Figure 7: Democratic CEI, 26th Senate District

Step Six: Mobilizing Partisans within Targeted VTDs

Choosing strategies is therefore a function both of the opportunities and constraints made available within the area defined by the Senate District boundaries and of the resources available for campaigning. Recent research from a series of experiments in real world politics by political scientists Alan Gerber and Donald Green of Yale University has established clear evidence that mail and (especially) personal contacting have substantial impacts enhancing voter turnout. Making repeated contact further strengthens the impact. We have suggested that the effectiveness of these techniques in generating additional votes will vary from place to place. Knowing where to invest in these costly activities is therefore especially critical.

We can illustrate some of the power of GIS-enabled targeting by looking within the 60th Senate district, where incumbent Republican Mary Lou Rath convincingly beat Democratic challenger Mark Doane (with roughly 79,000 to 35,000 votes, respectively). If the Democrats wanted to build on Doane's 2000 performance for the next election, they might well want to identify VTDs within the 60th district where their campaign for him failed to attract its expected level of support and where turnout was lower than expected. Within these prime target VTDs, attention could be given to Democratically-registered households who failed to vote in 2000. Mobilizing these households for in the next Senate race would be the best strategy for building on past support and improving the Democratic share of the vote next time around.

To provide a specific illustration of this targeting strategy at work, we selected one VTD -# 36029000098. Our statistical analysis determined that Doane received fully 15.05% LESS of the vote share in this VTD than was expected based on the general model of Democratic support across the state. Turnout in this VTD was  6.75% LOWER than it should have been, based on the statewide analysis of the determinants of turnout and the population characteristics of this small geographic unit. This suggests that Doane was particularly unsuccessful in mobilizing the Democratic vote in this corner of the Senate district. A smart future campaign might well focus on bringing its harvest of support in this local area up to state-wide levels. In doing this across all such prime targeted VTDs across the 60th District, and maintaining its support in other areas where its performance was closer to the statewide norm, the Democrats would be positioned to improve over Doane's 2000 performance. Supporting this effort, Barometrics Research can lead you, on a street-by-street basis, to precisely those houesholds where registered Democrats failed to vote in the 2000 Senate race.

Figure 8 - Senate 2000 Turnout and Abstention in a Prime Target VTD

The brown dots in Figure 8 illustrates all households in this specific Amherst-area VTD where a resident voted in the 2000 Senate race. However, the blue stars showing in Figure 8 are of special interest because they represent Democratic-registered households where NO ONE voted in the 2000 Senate race. Obviously, had Doane's campaign mobilized these partisans more effectively in this and other prime target VTDs, his vote share would have been considerably higher. Getting these people to the polls next time around  - perhaps by having the candidate canvass the area personally - represents the single best strategy for boosting the Democrat's vote over their 2000 performance. Imagine if the campaigners in the NY 26th, where the Democrats lost the seat by a couple hundred votes, had the benefit of such micro-targeting from 1998 to guide their 2000 campaign. The result might very well have been another Democrat Senator sitting in the state legislature.

Conclusions:

We have shown how geographic information and analysis can extract lessons from past campaigns and can help target areas of potential growth for the next election.

The case study reveals how a geo-strategic campaign could win votes and seats in New York Senate elections. As such, this historical case study illustrates the power of GIS-enabled analysis to guide a smart campaign. Bear in mind, however, that the techniques illustrated here can be adapted to virtually any campaign, at any level, anywhere in the United States. Barometrics Research can update and generalize these strategies and ensure that your ground campaign is as efficient and effective as technologically possible.  Many of the techniques demonstrated here can be used with some modifications to support campaigns in other countries. The techniques can be scaled upwards to guide state-wide races, or downwards to guide Assembly or city, county, or other local races. While the technical details may appear somewhat complicated to those unfamiliar with statistical analysis and GIS, the techniques we employ are adaptations of some of the most powerful and basic statistical techniques that have been employed for decades by academic researchers. We at Barometrics Research take pride in our work, and will be eager to work with our clients both so that they can appreciate the scientific principles underpinning our work, and to ensure that they can extract maximum benefit from the results of our analysis.

Contact us to become the first to put the power of geographic information and analysis to work in your grassroots campaign effort.

paul.belanger@barometrics.org
munroe.eagles@barometrics.org
anne.wadsworth@barometrics.org