Microtargeting is a process that answers the questions that are fundamental to the strategy of any campaign. Which voters support your candidate? Which voters are undecided? Which voters care about a particular issue? Which voters will vote at all? Most important of all, which voters will respond if you reach them?

Microtargeting utilizes a variety of tools that have guided commercial marketing for decades and have been more recently utilized by political campaigns as well. Microtargeting helps campaigns better shape and deliver messages to specific individuals and households by tracking and analyzing information on a person-by-person basis.

Voter files are the raw material of political microtargeting, because they are the way to start developing a person-by-person view of the electorate. Retailing changed in some dramatic ways when “point of sale” systems finally were put into place – high quality voter file systems are their broad political equivalent. Tremendous advances in data technology have made it easier than ever to access and analyze new high quality databases about voters which have rich amounts of detailed information drawn from public sources, commercial data providers, and past campaigns.

Having the basic data is the start – and microtargeting can then go in many different directions. At one end of the spectrum, there are large scale surveys of voters (with six thousand or more interviews), serious data analysis and modeling, detailed scores and profiles of voters, and carefully matched mail, phone, canvass, and even cable media buys based on those scores and profiles. At the other end, there simply is a better list of which doors have already been knocked in a precinct and a tracking program that remembers that a voter asked about a particular issue.

Microtargeting is also closely linked to clinical trial testing – a related set of approaches used to carefully measure the impact of campaign activities. These approaches use standard experimental design -- “treating” one large group of people with a campaign communication (some pieces of mail, a phone call, a visit) but also holding out another small group of statistically indistinguishable people who do not get the treatment (the control group). You can then actually measure how much impact a program has – and a cycle of learning can get started. Good voter files and database systems make running these experiments far easier than ever.

All of these approaches present a set of immediate challenges that can be intimidating to candidates and campaign managers. Microtargeting and testing can be expensive. Different people call different things “microtargeting” and there’s no Good Housekeeping Seal of Approval for vendors. Some people appear more interested in microtargeting because it is cool or new than because it will really help win an election. Intentionally not communicating with a group of voters by holding them out as a control group strikes many as insane (and there are certainly times when it should not be done).

And on top of it all, any new and major project always threatens to distract campaign attention from getting the basics right – having the candidate or campaign make a persuasive case about issues that matter in language that people understand.

We cannot clear up all of these confusions or overcome all of these obstacles. We can on the basis of our experience and the experience that others working in the field have so generously shared with us – lay out the basic steps for using microtargeting and some of the different options available. We also append a list of resources from which you can learn more.

We can also offer some quick rules of thumb. You should consider using the tools of microtargeting if you are going to run a large direct contact program. If voters in your state do not register by party, or if when they do, it is not very predictive of voting behavior, you should explore how you might get access to these tools. Smaller campaigns and organizations can often gain the benefit of these tools by turning to the Party, coalitions like America Votes, or subscription data services (including Catalist). You need not be running a large program to be able to participate in a cost effective manner.

One thing this memo won’t try to do is explain how the analytics behind microtargeting actually work. We’ve put too many people into dazed, confused states in person to branch out into doing it through a broad memorandum. Hal Malchow has been a leader in this field for decades and has literally written the book that describes many of these tools and their proper use in clear detail. The book is called The New Political Targeting and a new edition will be available in the fall of 2007. We’re going to focus on why to do microtargeting, and how to run the process.

Why Microtarget?

Karl Rove was surprised when George Bush lost the 2000 election popular vote. He had expected to win, and concluded afterwards that the Democratic Get Out the Vote operation had outperformed the Republicans. He set out to study systematically how to do it better, launching the 72 Hour Project and executing a set of experiments in the governor’s race between Mark Early and Mark Warner in Virginia in 2001. Republicans lost that race, but they learned more in losing than we on the Democratic side learn in most elections we win. They systematically compared different approaches to voter contact in Virginia and other races. And by 2002, we progressives were waking up to find ourselves doing worse than expected in Georgia and in other states and other races.

Campaigning is becoming more complicated. There are more mediums and more ways that voters are getting information. Measuring what works, establishing best practices and launching a process of continuing improvement is one key to winning campaigns in the years ahead. The only way to do these things is to test.

As television’s audience spreads out over hundreds of channels, it is no longer the dominant medium it once was. That is why many campaigns are investing in more narrowly focused, individually oriented, and carefully tested communications.

This has been going on in the commercial world for quite some time. Targeted advertising is growing, as can be seen in Figure 1 – look at the growth for Direct Mail, Internet, and Cable TV.

Two caveats: First, television advertising remains the central driver of message and strategy in many campaigns, and for good reason. Television is all that is left of mass marketing and voters, especially in a presidential election when turnout is high, represent the biggest mass market in America. TV can still reach voters more comprehensively and effectively than any other medium. It just does not do it as well as it did ten years ago. But while broadcast TV’s impact may be dropping; much of its competition also faces problems. The issue has occasionally been portrayed falsely as TV ads versus field activities – when the real challenge is to link TV, in all its forms, with direct contact efforts into an integrated, layered communications and engagement plan.

Second, message still matters most. Many of the techniques we will discuss below involve finding “targets” – people with whom to communicate. The question that matters most, of course, remains what to say to them – and how to engage them. These tools can indeed help inform message strategy, but they are clearly only one piece of the puzzle.

So what does microtargeting do? First and foremost, microtargeting targets individuals. Every campaign has targets. Microtargeting helps you understand the likelihood -- at an individual level – that a voter will vote, is a supporter, or will be moved by a particular message.

Let’s look at a classic example. One critical campaign task is finding people to turn out to vote in the final weeks of the campaign. The traditional approach is to find neighborhoods where your supporters are concentrated – so called “65% plus” Democratic Performance precincts. Knock on all the doors there, urge every person to vote and know that the majority of those you turn out for Election Day will be supporters.

Many campaigns have been successful using this program but targeting individuals can add value. Let’s look at the problem Lieutenant Governor Tim Kaine’s campaign faced in his 2005 campaign for Governor of Virginia. Precinct targeting did not produce enough targets to provide the Election Day turnout he needed to win. So the Kaine campaign invested in an early microtargeting project designed to give every voter on the Virginia voter file a pair of scores based on statistical modeling – one score that predicted a person’s likelihood to vote and another score predicting how likely he or she would be to support Kaine. This gave the campaign a way to identify likely supporters outside the 65% plus precincts. When combined with other models, the campaign was able to expand its GOTV targets and increase the efficiency of its persuasion efforts. These techniques could also be used to avoid expending resources knocking on the doors of those with low likelihood to support inside 65% plus neighborhoods.

Recently, the basic support and turnout scores have been developed in many Democratic campaigns. By ranking or sorting voters according to how likely they are to vote and how apt they are to vote for one side rather than another, campaigns can have a better baseline view of the playing field on which their campaigns operate. This baseline view may not be enough, however, as it only tells you where people sit, not how likely they are to respond to campaign communications.

Microtargeting matches messages to people

Microtargeting starts with computing probabilities, but it succeeds when it helps campaigns pinpoint ways to change them. One way is to match the right message to the right people. Microtargeting, in a wide range of ways, can help get the right messages to the right people.

One way to find that out is to develop issue models. Issue models use the same techniques as turnout and support models to assign individual scores reflecting the probability that someone is interested in a specific issue, such as education, choice, or strong opposition to the Iraq War. These can be combined with support and turnout models in interesting ways. In 2006, we found a group of voters that models predicted were likely to strongly oppose the war, but were also not yet strongly supporting the Democratic candidate. These voters were good targets for communication.

A deeper and strategically richer set of issues involves finding groups of people with shared values, learning styles, or other “psychographic” characteristics. Finding groups of individuals through use of these tools has been going on in a variety of ways in politics for quite some time. These segmentations can be developed based on original surveys, already developed commercial segmentations, and other hybrid approaches.

Original segmentations are one approach. As part of major microtargeting projects, it is possible to develop customized values segmentations based on the large scale polling. In Ohio, voters were given a set of scores saying how likely they were to be “aging traditional liberals,” “younger engaged agreeables,” “moderate demoralized Christians,” and a half dozen other categories. The clusters were somewhat different in Michigan. Various pollsters have been using these tools productively for more than a decade. The Republicans have their own multiple clustering systems as well.

These are, however, often expensive, and can be difficult to score back to the voter file (connect with confidence to individual voters). Large survey samples are costly; long questionnaires make them even more so. Several strategies have been used to minimize these problems. One approach is to do surveys of people who have answered a very long questionnaire once, and now agree to answer shorter surveys that can be matched to their previous answers for modeling purposes. Another is to use segmentations developed by commercial providers. There are also approaches that combine parts of both.

A more detailed and deeper view of the electorate along values lines is clearly a good idea. All people who regularly attend church are not alike; having models that find those church-goers who have ears to hear a Democratic message of social justice can be of tremendous value to a campaign. These techniques are very widely used in commercial marketing; Wal-Mart recently did a survey of 4,565 of its shoppers and developed seven key segments ranging from “brand aspirationals” to “social shoppers” based on dozens of psychographic measures. According to news reports, this was central to changes in their strategic marketing.

Microtargeting increases the chances of winning

The only overall reason to spend resources on microtargeting is, of course, to increase the likelihood of achieving your campaign’s goals. When linked with clinical trial testing, it is possible to both design and evaluate the impact of campaign activities with a new level of precision and accuracy; choosing the smarter programs thereafter is the same as raising more money.

One portion of the labor movement’s activities in the 2006 campaign was a program of mail and phone contacts run nationally. Extensive modeling, microtargeting, and testing led to a targeted communications program aimed at specific individuals and households. Under the direction of Mike Podhorzer, the political staff of the AFL-CIO was determined to measure the exact impact of this program, so they randomly picked some people they would otherwise have contacted, and did not send them anything. Right after the election, they had Garin-Hart-Yang Research do a survey of 4,291 people who were in the target audience for the program, and the1,117 who had been held out in the “control group” and received no communications. Did the program make a difference? Did the people who received communications vote more Democratic than the people who did not get the communications?

The results are both impressive and interesting. The people who were in the treatment group voted for the Democratic candidate for Congress by a 10% greater margin than those who received no communication. (The margin for Democratic Senate candidates was also 10% greater). Analysis of the final voter file results on a district by district basis is still underway, but it is a safe prediction that had this program not been run, the Senate would be in Republican hands.

Note that this is a different form of argument than “if Labor union members had not voted for Democrats disproportionately, the House and/or Senate would still be in Republican hands” (though both statements are true.) It is not a numerical reflection of something that might have happened for a thousand reasons, many of which are beyond the control of any campaigns or causes. This is a quantitative evaluation of the impact of a specific program aimed at specific people. We need more of this in progressive politics.

The interesting thing is why this 10 point increase in margin arose -- it appears that a major difference between the control group and the treatment group was what they thought the election was about. The control group’s thinking was more closely in line with the electorate as a whole – that Iraq was the most important issue. The men and women who had received the AFL’s program were more likely to think the election was about the economy (and less likely to think it was about Iraq or moral values).

This result may not be generalizeable outside of this experiment, but it again points to the basic idea – run a careful, individually tracked program that has a sensible message aimed at the right people and you can change the outcome of elections.

Six steps towards microtargeting

So, what are the building blocks of microtargeting – how can one get started? These lessons apply to campaigns or causes; we focus primarily on elections, though the same lessons apply to any campaign with clear goals.

1. Build microtargeting into your overall planning and strategy process

The best place to start is determining how many votes you need to win your election -- or whatever specific and measurable change you seek. When setting specific vote goals and budgets to achieve them, try to be as explicit as possible about how your choices accomplish your goals. We are a ways off before we can confidently rank alternative campaign investment expenditures on the basis of the expected increase in net margin per dollar spent. But being systematic is a start.

2. Upgrade your voter files and data systems

Gain access to the highest quality voter file available. We are, of course, partial to Catalist. In any case, look for how recently the file has been cleaned and checked for changes of address. How comprehensive is the vote history on the file? How are absentee and early votes dealt with by the Secretary of State and on the source file? (This is a tremendously important issue and one that is growing). What past ID’s are on the file from previous campaigns? How reliable is household level information? How will you be able to match the data you collect to the files?

The voter file is the basis, but you also need tools with which to use the file. Often you need both “front end” tools that can help gather data in the field (think palm pilots for entering canvass results at the door), and “back end” tools that allow you or your vendors to store, analyze, and manipulate data at ever increasing scale as you build data (and learning) test after test and cycle over cycle.

Good data is useless unless appropriately shared throughout your organization. Many campaigns, in our experience, have data flows that look like the chart below. When a major new poll comes in, all the relevant people may be on the phone for a conference call, but other than that there is little data or strategic integration. Breaking down these data silos brings increasing yields to every endeavor over time.

3. Gather more data

Once you have access to a voter file and systems with which to use it, you can start adding data. Make sure your surveys are drawn from voter files, and use larger samples when you survey. A large sample survey is like a better digital camera – more detail allows you to see smaller objects more clearly.

All the weight need not fall on your shoulders. Lots of data can be found at lower than do-it-yourself costs. You can turn to the Party, coalitions including America Votes, and subscription services from various data providers (including Catalist). You can purchase or rent lists of specific interest to your campaign; for example, specific hunting lists for a mailing on a local environmental issue. Or lists of organizations that support you. By collecting and saving these lists, perhaps in a more systematic manner than before, you can steadily grow the value of your system.

Another critical issue is gathering better basic information about some traditional but key variables. Race and ethnicity. Church attendance and gun ownership. As you run ID or direct contact programs, collect and store this information. But again, as you collect data you need to be prepared with systems that can store and integrate this data with the voter file – and ideally do this over multiple election cycles. Be sure to shop for cost-effective ongoing data services and not just data.

4. Better define your targets;

As we discussed above, individual level data can help you add people you need to your lists of people to contact; it can also help prune people who aren’t yet ready to be moved by your message. Matching messages to targets increases lift as well.

Once you’ve made your plans and started gathering your data, you need to make the decisions about what specific type of microtargeting program to run. Will it be one-directional – applying scores to set initial targets – or are you going to be able to run tests on those targets before you roll out all your contact programs? Do you want original modeling, or can you get by with using scores or data previously developed? Do you want deeper understanding of values themes that might reach groups through tailored messaging and targeted media channels, or do you want to simply identify more supporters? There are no general answers – this is the point where you need to work with your pollster, your data team, your strategic advisors and your budget to determine what is best.

One point here: It is critical that microtargeting is integrated with traditional targeting. The best teams and projects we have seen combine local knowledge and traditional targeting approaches with the newer microtargeting tools. Results that bewilder the statistical analyst with no local knowledge are obvious to someone who knows the area (“why does that race seem so important? – well their candidate ended up in jail two weeks before the election – anyone who voted for him nonetheless was family or an unalterable Republican”). This type of integration is becoming more possible as more political data is shared across elections, and tools for analyzing it are becoming easier to use and more transparent.

5. Test approaches and measure impact

In some senses this is the hardest step. Resistance to testing springs up from every corner, sometimes for good reasons. Reliable consultants who have seen fads and charlatans come and go often start off by distrusting new approaches. Testing can threaten habits and relationships. But again and again, the real deal killer is a problem facing all sorts of placebo control group studies in medicine -- shouldn’t every patient get the medicine? Learning is a nice quality for a candidate, but winning is his or her only job. If learning for later is going to reduce my chances of winning now, isn’t it irresponsible to use these tools?

Well, in many cases, it is what our former President used to call a false choice. And resistance to this form of testing must be overcome if we are to enter a cycle of learning – as our competition has. EMILY’s List and others have done excellent work measuring their programs; these processes need to become more standard.

The simple if somewhat technical truth here is that two facts make the expected cost of holding out a control group small. First, we have limited budgets and seldom if ever treat everyone who we might reach. We cut off lists when our budgets run out. If I have a budget that allows for 10,000 voters to be contacts, we find the best 10,000 targets we can – we rank the electorate to find those we think is most likely to be moveable in our direction or capable of being mobilized to vote.

Second, we are not all that great at ranking people. When you hold out a control group you still mail the same number of people (in this case 10,000). From the first 10,000, for example, you might hold out 1,000 – and you then just loosen the belt a little and add people who would have been just barely contacted before – the same people you would have mailed if you could have afforded 11,000. The only expected vote cost of a control group experiment is the difference between the impact of a communication to the average person in the treatment group and that of the average person in the people added. And that difference is often neither measurable nor significant.

6. Use Repeatedly and Share

At the risk of sounding repetitious, these tools all work best when used repeatedly. “Learning organizations” may be a business buzzword, but the power of these tools grows when time and energy are invested. It is also critical to share and swap data over time. Data and data systems are more like language than widgets – the more people who use them, the more valuable they become. Progressives need to understand that sharing and swapping data strengthens everyone.

Conclusion

Our focus has been on traditional field operations, in part because these are the operations that depend most on voter files. There are, however, serious applications going on and still to be developed in internet communications, fundraising, and targeted TV, especially cable media ad buying. “Microtarget your new media” could be a separate paper -- there is a great deal of work to be done – and tremendous opportunities.

This is a fast growing and changing field. Microtargeting and clinical trial testing are providing measured gains for campaigns and organizations, and we are confident more will come. This makes it an exciting field to work in, but also makes it an important time to use caution. All new tools take time – polling took decades until it became fully functional and trusted.

As this field grows, there are, in our opinion, three key strategic imperatives for progressives if we are to make the most of these opportunities.

Focus on movement not location: these tools develop good profiles of how likely people are to vote one way or another, and how likely they are to vote. They further find people who are likely to say they are interested in an issue or share a set of values. These can be thought of as developing a GPS to identify each voter’s location on the political probability map. But more than location, what we really want to know is what moves people along both axes – toward or away from voting; toward or away from support. We want to be able to model and predict where voters are – but even more we want to model and predict which voters will move in response to campaign communications. A wide range of exceptional testing has been done by Women’s Voices Women Vote showing the responsiveness of single women to various political communications; this is the type of work that needs to be continued and expanded.

Build engagement not one way communications: “Targeting” too often returns to a bombing metaphor – dropping message from planes. Yet the best data tools help build relationships based on observed contacts with people. Someone at the door finds out someone is interested in education; we get back to that person and others like him or her with more information. We measure whether the contact persuades or motivates. We adapt our messaging according to response. Amazon’s recommendation engine is the direction we need to head.

Increase the number of laps around the track: we spoke to a successful business person who had built a multibillion dollar financial services company by using analytics and testing. He spoke of the early days in the company, and how they remained focused on the fact that no particular piece of knowledge or result would be key to their success, because competitors would always be able to repeat their learning. The new company’s abiding competitive advantage would instead be its consistent ability to stay ahead by having more experiments (adding individual level data and understanding about what moves a voter with each round of contact) – more “laps around the track.” Single firms now do thousands of models and tests a year; the progressive movement needs to increase our laps around the track by orders of magnitude.

We hope these ideas help.