I’m at Netroots Nation 2012 (#nn12) and will liveblog as possible. The first panel I got to covered the data-driven transformation of American politics, whether that’s testing messaging, identifying groups of persuadable voters, and modernizing political advertising.
David Mermin of Lake Research Partners talks about testing messages.
Traditionally, we do dial testing of targets while watching a video. The viewer toggles a dial up and down over thirty seconds, and we get a nice line graph of their reactions.
Now we have Internet surveys, visual stimuli, and other methods. But public polls online have sample issues — weighting and demographic matching still doesn’t always account for online demographics. Mapping to voter files is a challenge, as well as district boundaries. But unlike phone surveys, you can be interactive.
You can simulate a ballot, and see where people have trouble with the design. (see also http://ballotusability.blogspot.com/).
You can ask people to highlight the parts of messages they like, and create a paragraph of talking points where the key phrases that resonated with voters are larger fonts, like in a tag cloud.
And then there’s online dial testing, where you can see how the base, persuadable, shifters, and opposition each trend in their reception of your message. In this case, you can view not only how each group responds to the message, but also the gaps between groups, which you might actually seek.
Who’s persuadable?
Polling has always been concerned with identifying the persuadable voters, but there are new tools for getting answers to whose mind can be changed and which messages are effective.
Are weak partisans more partisan than independent leaners?
They tested attitude consistency from 2008-2010 on attitudes towards Obama, the NRA, and other hot-button issues. They found that independents who lean Democrat are actually more progressive on a whole range of issues than people who identify themselves as weak Democrats. The same holds true for Republicans. [Could the answer to this be that this group of independents who lean towards one side aren’t, in fact, centrists, but are actually to the left and right, respectively, of the parties?]
Iterative mail testing for the AFL-CIO
They measured the persuasion effect of mailers in 2010 and tested the targets their model predicted versus those the model predicted, and tripled the efficacy of their communications (and mail is expensive). Many of our assumptions don’t actually play out when we look at empirical tests.
Once we have data, what can we do with it?
David Radloff with Clarity Campaign Labs, formerly with ISSI:
There are a lot of terms buzzing about these days – modeling, microtargeting, statistical analysis. But a model is just a statistical tool that gives a probability for each voter’s likelihood of taking an action, voting a certain way, support an issue, and so on. Common types of models include turnout and ballot return dates and the likelihood that the person is a progressive activist.
We’ve been carving universes of voters from the voter file for a long time now. We have many pieces of data about voters – age, race, marital status, voter registration status, where they live past information from previous campaigns, and now, thousands of consumer fields augmenting the existing file. No one uses these fields individually, but rather as parts of complex models that create likelihood scores for each voter.
It’s a three step process:
1. Training data is collected from a poll or previous turnout data or other sources
2. Algorithms are applied to find patterns, usually based on
3. Score the voter file
Once you have the counts, you run them against a turnout model, which projects likelihood of turnout and likelihood of being a Democrat and cross them and see where they meet and talk to those groups. It becomes a fairly easy tool to aggregate large amounts of data you weren’t otherwise able to take into account.
So far, the empirical research has found that it’s really hard to guess, and it usually requires testing for each issue and each campaign. Conventional wisdom is something consultants tell you because they feel like they need to have an answer.
Persuasion targeting best practices
- Cross-pressured voters
- base education or low information models
- intra-survey message test models
- experiment-informed-program models
Targeted Political Advertising
Tim Lim (@PrecisionNet @limowitz):
Tim starts by illustrating the waste that’s happening in online political advertising. The total Democratic vote is much smaller than the US population, voting eligible population, and actual turnout groups.
Innovation in online advertising happens in the commerical sector. Between 15-20% of commercial media budgets are spent online. In the political world, it’s even less, and amongst Democrats, only 5% or lower.
Online advertising is effective. 2% of time spent viewing video online is spent viewing ads, while with television, 25% of your time is watching ads. This leads to better recall amongst targets in the online environment.
Most political advertising is still done primitively, through site-by-site ad buys and vendor-by-vendor deals. Measurement of ads is poor.
Precision takes the voter file, overlays third party data, does precision matching, and serves up ads.
Targeting options range from electoral information (voter ID, party, voting frequency), demographic, economic (including donor history and occupation) and behavioral (online / offline purchases, content subscriptions, online browsing habits, and browser language settings).
They also do custom list matching, so you can target ads on your existing supporters list. It’s a shift from buying media on specific websites to buying media to reach specific people.
aCPM is the actual CPM it takes to reach your target audience. Advertising on mainstream sites like CNN, HuffPo, and others costs more because only 20% of site visitors might be in your target group.
David Radoff: The exciting thing is merging all of these fields together.