By Ken Strasma
Micro-targeting is arguably among the most valuable new technologies in political campaigns. That's saying quite a bit given the number of technologies that have emerged over the last few cycles and transformed the way campaigns are run.
Internet fundraising, blogs, robo-calls, satellite television, and video on palm pilots are just some examples of how technology is changing the way campaigns communicate. Even with all these new ways of reaching voters, resources are limited, putting a premium on using these new technologies as efficiently as possible.
That's the power of micro-targeting. Micro-targeting makes all of these new technologies, and the old standbys like door-knocking and mail, more efficient.
Uses of micro-targeting
Uses include GOTV and persuasion targeting, message selection and fundraising. Small-dollar donor prospecting is an especially valuable use for micro-targeting. Campaigns and party organizations spend a great deal of money prospecting for new donors. Looking at the results of past prospecting mailings or phone calls, it is possible to build statistical models giving the likelihood that any individual will give if asked. In donor prospecting, a tiny improvement in the response rate can make a huge difference once resolicitation of new donors is factored in.
To target for GOTV, a campaign could combine two micro-targeting models, one giving the likelihood that an individual was a supporter and the second giving the likelihood that that individual was going to vote, in order to find likely supporters who are unlikely to vote if not reached.
If there are sufficient IDs where voters are asked their issue priorities, models can be built giving the percent likelihood that an individual voter cares about any given issue so that the campaign can select different messages to target to different audiences.
What exactly is micro-targeting?
Micro-targeting is much like traditional demographic and geographic targeting, except that it works at the individual, or “micro,” level. Geographic targeting might be used to target a GOTV (Get Out The Vote) message to precincts that have traditionally supported one party or another. Demographic targeting might be used to target a persuasion message to a demographic group that polling has shown to have a large concentration of undecideds.
In both these examples, the targeted messages will be more efficient than if they were sent to all voters. However, because these types of targeting target entire geographic regions or demographic groups, they will never be as precise as individual targeting.
Unfortunately, obtaining IDs for every voter is not possible for most campaigns. With more voters having caller ID, unlisted numbers, or using cell phones as their primary phone, it is getting more and more difficult to reach voters with an ID call. Phone IDs can be supplemented by door-to-door canvasses, but there will still be a large segment of the population that can't be reached. Micro-targeting fills in the gaps by using statistical models to predict how the individuals who weren't IDed would have responded if they had been reached.
How it works
Micro-targeting works by taking whatever individual-level information is available (e.g., IDs, contributor information, vote history) and combining it with demographic, geographic and marketing data about those individuals to build statistical models that predict the attitudes and behaviors of voters for whom that individual-level information is not known.
For example, a campaign might have ID information on 10,000 voters indicating whether the voters supported or opposed the candidate or were undecided. The campaign could collect geographic and demographic information about those voters, things like the past performance in the precincts where the voters live, the Census demographics of the blocks where the voters live, the age and gender of the voters from the voter file, and, budget permitting, marketing data, such as the type of car the voters drive.
All of this information can then be combined with the available IDs to build demographic profiles of different types of voters. These demographic profiles can then be applied to voters for whom there is no ID information in order to predict with remarkable accuracy how those voters would have IDed if the campaign had had the time and money to reach them.
There are a great many statistical techniques that can be used in micro-targeting, including regression analysis, segmentation techniques like CHAID and CART, and newer techniques, such as genetic algorithms and neural networks. The ideal methodology to use depends on the circumstances. Often, the best micro-targeting model is yielded by combining several of these techniques.
Until recently, some of the newer techniques were primarily of academic interest and took too long and required too much computer time to be practical in the compressed timeframe of a political campaign. Advancing technology is changing all that. Computers have become less expensive and more powerful in recent years.
[**CUSTOM SCRIPT:8**]At the same time, innovations in clustering or grid computing allow multiple computers to work together, yielding computing power formerly only available in multi-million dollar supercomputers. These innovations have made modeling techniques that were once exclusive to big banks and credit card companies available to political campaigns.
It is vitally important that a micro-targeting model be thoroughly tested before it is used in the field. Anyone can develop a statistical model that exactly predicts the behavior of a sample of voters, but then falls apart when applied in the real world.
This problem is known as “over-fitting.” It is important to set aside a holdout sample or control group when building a model. The control group – usually one-third to one-half of the total number of IDs – is not used in building the micro-targeting model. Once the model is built, it is applied to the control group and the actual IDs can be compared to those predicted by the model.
Only after a model has been tested against the control group and found to accurately predict voter response should a campaign begin to use the model for targeting activities.
Using a micro-targeting model
The high precision of micro-targeting models can lead to the erroneous sense that the model will tell you exactly who to target. That is not the case.
Micro-targeting – as is the case with any other kind of targeting – is a tool for prioritizing targets, rather than a set target universe. When a micro-targeting model is applied to a voter file, each voter gets a score giving the percent likelihood that they exhibit the behavior or characteristic being modeled.
[**CUSTOM SCRIPT:8**]For example, a persuasion model might have a score giving the percent likelihood that each individual voter would say that they were undecided if they were called and asked their candidate preference. This score allows the campaign to focus their persuasion efforts on those voters most likely to be undecided, but it does not provide a cutoff point – there is no magic number above which all voters should be targeted and below which they should be ignored. The micro-targeting score allows the campaign to select however many individuals the budget and campaign plan allows for a persuasion universe, and to know that these individuals are the ones most likely to be undecided.
The value of micro-targeting
As with all new technologies, it is sometimes difficult to separate the true value of micro-targeting from the hype. However, once you cut through the buzzwords and actually have a well-built and tested micro-targeting score applied to a voter file or donor prospect list, the value becomes readily apparent. Phones, mail, door-knocking, even cable, radio and broadcast television can be targeted to the most appropriate target universes, saving the campaign money and delivering the campaign's messages to the most receptive audiences.
Ken Strasma is president of Strategic Telemetry, an
organization specializing in micro-targeting
and other strategic consulting services for
progressive campaigns and organizations.
Ken can be reached at Click here to contact this Author