Ground-breaking Models

Marital status – in and of itself – is a fundamental determinant of whether or not one registers and votes. Reaching unmarried women regardless of location required using list-based approaches and the ability to identify who was and was not unmarried. And that required VPC to develop its own marital status model because those that existed were inadequate, both in terms of accuracy and in terms of coverage of the file. Over time, VPC realized other groups – people of color and young people – were equally disengaged and under-represented and VPC developed models that improved the accuracy of its ability to target unmarried women, people of color and young voters who were most likely to respond to its programs to register or re-register to vote after a move, to vote by mail, to turnout or to take action on a policy issue.

Today, VPC has moved well beyond traditional support and turnout modeling currently used in political campaigns. VPC programs have been made more effective and efficient by introducing response modeling for registration and vote-by-mail; modeling on issues to target voters who share the same values; advocacy modeling to predict who will and will not take an action in support or opposition to public policy debates and, as mentioned above, a marital status model to improve targeting far beyond what is possible with existing commercial data.

In addition, VPC identified mobility as an important factor in non-voting. In response to this finding, some VPC programs were focused on re-registering movers and a new VPC model was developed to achieve better predictions about movers’ likelihood to return a voter registration application.

The list of models developed by the VPC (and now available for others’ use), include:

  • Marital Status Model: This was the first model built by the VPC and it revolutionized targeting. The model identifies unmarried (or married) individuals and allows groups to mobilize unmarried women. The VPC now has developed three, ever more accurate generations of this model and achieved dramatic improvements over commercially available data.
  • Issues Model: This model helps target VPC programs toward voters and potential registrants who share the same value sets as the overwhelming majority of the Rising American Electorate. This issues model is updated in each election cycle as the focus of the debate changes and the issue operating environment shifts.
  • Registration and Vote By Mail Response Models: These models define who is most likely and least likely among the RAE to respond to either a registration or a vote by mail application. These models are updated constantly in real time with each registration or VBM mailing VPC conducts.
  • Drop-off Model: This model is able to predict the likelihood a voter will not turn out in a midterm election. For the first time, the model was highly predictive for those who cast their first-ever vote in the most recent presidential race and had no previous voting history. The ability to predict new voting behavior was a breakthrough.
  • Advocacy Model: This model predicts the likelihood someone will take an action on a public policy issue. The model allows for significant savings—it creates a 45 percent average gain in efficiency—and can be used by grassroots groups to educate the public and generally facilitate civic engagement or by advocacy-oriented groups that focus on generating action on particular policy topics.

The VPC is also examining possible new models, model refinements and tests that could produce even more cost-effective programs in 2016 and, just as importantly, position partner organizations to be more effective. These tests will add to the VPC’s knowledge about voting behavior and contribute to developing tools to motivate individuals to take action on key issues. These include:

  • Voter Information model: The VPC has conducted very preliminary research and modeling on voter information levels – predicting the relative level of information a citizen has—and is working to document the links between information levels and RAE members likelihood to vote. The VPC is proposing to do a broad-based survey, model information levels and match these levels to participation rates, with level of information being the dependent variable. Finding the tipping point for how much information is needed, when it is needed, and determining how best to deliver it to increase participation offers the potential of dramatic advances.
  • Net Registration and Vote By Mail Models: The VPC is working on modeling people who respond to a treatment by becoming a net additional registrant or net additional voter using VBM and not just a responder. In other words, these models will try to identify those responders least likely to have taken the desired action without VPC intervention or treatment, generating cost efficiencies and greater ROI.
  • RAE-specific turnout model: The VPC plans to develop a model that will provide a very granular look at the emerging demographic groups and their particular patterns of turnout. The VPC will explore if additional inputs (like information level) not traditionally used in turnout models impact the accuracy of results.
  • Voting Age Population model: The VPC is in the process of building a VAP model to increase effectiveness of registration efforts targeted to this group. By working to understand what keeps certain groups from registering, we can shape our models and programs to target them specifically.
  • Advocacy model: The literature indicates that personal letters to policy makers are more effective than other means of contact (e.g., phone or email); therefore the VPC is developing an advocacy model for letter writing. The goal here would be to re-create the advocacy model with the dependent variable being the propensity to write personal letters as opposed to patch through calls. This will be an enormous breakthrough.
  • Issue model: The VPC is updating the issues model. This analysis will make it clear which issues need to be added or eliminated to capture the current political environment.