Articles and Analysis


Alex Gage & Microtargetting: "Putting A Very Fine Point on the Obvious"

Topics: 2008 , Microtargeting , The 2008 Race

An article today by Chris Cillizza continues The Washington Post's "occasional series" on the campaign "gurus" of 2008 profiles Romney pollster Alex Gage. The article actually says less about the Romney's current campaign than about Gage's pioneering efforts "microtargeting" for Romney's election in 2002 and the Bush reelection campaign in 2004:

Describing what he does, Gage, 57, sounds part marketer, part political strategist -- and more than a little Big Brother. "Microtargeting is trying to unravel your political DNA," he said. "The more information I have about you, the better."

The more information he has, the better he can group people into "target clusters" with names such as "Flag and Family Republicans" or "Tax and Terrorism Moderates." Once a person is defined, finding the right message from the campaign becomes fairly simple.

"'Flag and Family Republicans' might receive literature on a flag-burning amendment from its sponsor, while 'Tax and Terrorism Moderates' get an automated call from [former New York mayor] Rudy Giuliani talking about the war on terror, even if they lived right next door to one another," Alex Lundry, the senior research director of TargetPoint -- the firm Gage founded in 2003 -- wrote recently in Winning Campaigns magazine. [link added]

Cillizza's profile -- which is well worth reading in full -- goes on to provide some explanation of how microtargeting works:

[Michael] Wszolek, the Michigan-based direct-mail consultant, has known Gage since 1984 and worked closely with him to fine-tune a theory of political microtargeting. Wszolek acknowledged that "what you're doing is putting a very fine point on the obvious."

But, he added, the key insight of political microtargeting is that, rather than simply determining whether married men are more likely than unmarried women to support a candidate, a campaign can identify segments within larger demographic groups and tailor messages down to the household level -- an extraordinary amount of precision that helps turn a guessing game into a series of targeted strikes. If television advertising is painting with broad brush strokes, microtargeting is political pointillism.

For all the hype, microtargeting applies the same concepts that pollsters use on surveys of hundreds of respondents to surveys of thousands. Why is this a new idea?

Over the last 20 or 30 years, most internal campaign polls have been designed primarily to guide strategic decisions regarding broadcast television advertising. The great power of broadcast television is its ability to reach a mass audience through a uniquely powerful and vivid medium. However, the ability to target television advertising is limited to choosing specific media markets and skewing the advertising buy to emphasize certain demographic groups (based on the ratings of individual programs). Internal campaign poll samples have traditionally been designed to provide a minimum of 100-200 interviews per media market (in states or districts that reach multiple markets).

Even though television advertising reaches a mass audience, political strategists are always most concerned with reaching the 20% or so that appear open to persuasion. Campaign pollsters use their surveys to identify a "target" subgroup and profile its demographics, but since their ability to target television is relatively blunt, they only analyze a handful of demographic subgroups.

The key distinction of political microtargeting is a shift in the survey design to facilitate targeting of direct voter contact (i.e. direct mail, phone calls and door-to-door canvassing). With these media, campaigns can obtain and use individual level data about each voter gleaned from public and commercial databases: gender, age and past voting history (typically available on registered voter lists) as well as "thousands of data points" (from consumer databases as described in the Cillizza article). So the potential number of target subgroups is essentially unlimited. Microtargeting surveys involve thousands of interviews rather than hundreds so that even tiny slivers of the electorate yield the minimum 100-200 interviews necessary to make statistically meaningful distinctions among subgroups.

Microtargeting pollsters like Gage combine the same techniques they have been using for years to identify persuadable voters on surveys with data mining software that facilitates the analysis of those "thousands" of commercial data points. More specifically, they use survey questions and traditional methods to create some sort of numeric index of an ideal target voter. Think of it as a "persuadability index" -- the higher the score, the more persuadable the voter (although the index could be for any sort of target, such as support for a particular candidate or likelihood to vote). Broadly speaking, they use data mining software to guide campaign communication efforts in two ways:

  • Identify a set of very small target subgroups that score high in the survey on that target index and replicate the subgroups (using the same data variables) on the full registered voter file. Once identified, the pollsters can look at how each subgroup responds to messages tested on the survey, so that each subgroup can receive a specifically tailored message.

  • Create a statistical model that predicts the target index, using only the non-survey data available for all voters as "independent" variables. Then take that model and run it on the full registered voter list, giving each voter a "persuadability" score. These scores can be used to determine which individual voters are worth the expense of contacting.

That is the general idea behind a technique that, as Cillizza's article makes clear, is now all the rage in both parties.


Ian Hoffman:

I read that article and I thought it might have run in the "50 Years Ago Today" column. Gage is honing the techniques apparently, but his work didn't sound like a paradigm shift from what direct mail has done for decades, i.e., target individuals.

Also, Cillizza said that New Mexico was the only state Bush won in '04 after losing in '00. That wasn't right as Iowa also fit that description.

All in all, I was underwhelmed with the article.


Dwight McCabe:

There's a challenge with this approach, speaking from a couple of decades of doing customer relationship analytics and targeting in the commercial world. The available data on individuals for these models is either demographic, which is a weak predictor of most behavior, or at best limited behavioral data. I say limited because the behaviors are usually purchasing behavior which are not terribly relevant to how someone decides to vote for a political candidate. The only voting behavioral data we have on individual voters is if they voted in each election. Other than that we just know how their neighborhood precinct voted on average.

To persuade people with a political message you need to understand their attitudes and values. However, attitudinal data covering the entire voting public is hard to get. When we run surveys to collect both behavioral and attitudinal data, to see if we can use a behavioral segmentation to predict a person's attitudinal segment, we usually find that the two segmentation schemes don't match very well.

This means you can not easily take data that is readily available on most adults, and use it to predict where they will place in an attitudinal segmentation.

Now campaigns and parties do contact a lot of voters, so you'd think they would be able to collect attitudinal data on a significant number of target voters. However, in my experience that requires a huge shift in perspective and systems for political field operators and campaign managers. The other choice is to run very large surveys early in the campaign, but only the largest state and federal campaigns can afford this expense.



vote now


Post a comment

Please be patient while your comment posts - sometimes it takes a minute or two. To check your comment, please wait 60 seconds and click your browser's refresh button. Note that comments with three or more hyperlinks will be held for approval.