Data Science & Modeling

Experienced data scientists at Change Research can create models that make it easy to target specific types of voters and/or constituents in specific geographies, such as people who are most likely to support a candidate or ballot initiative, be concerned about climate change, or be activated by a proposed policy.

Custom Survey-Based Modeling – Using responses to a specific survey question, we match respondents back to the voter file and train custom models that predict responses to that question for all voters. With each wave of polls, data scientists at Change Research can train the model to give your team a good idea of how every single voter would respond to a survey question if they were interviewed. Change Research will provide a scored voter file of all registered voters in a district or nationally, with a numeric score based upon a specific trait or opinion that you seek to model. To do this, Change Research will need:

• 1500 respondents to a statewide survey(s)
• A specific survey question, such as:
• Who are you voting for in the election for [candidate]?
• How favorably do you view [candidate]?
• What news sources do you consume on a weekly basis?

Custom survey-based models are available for $10,000 for the first model, and $6,000 for all subsequent models built off the same survey. For example, a custom support model for your candidate and a model abortion support would cost a total of $16,000 for both.

Change Research is also constantly accruing data through historic surveys, including over 200,000 responses to questions about approval for Planned Parenthood, approval for pro-life groups, and importance of abortion policy to voters. We could use this data like this to:

1. Target messaging to voters who are pro choice
2. Exclude voters who are strongly pro life from messaging
3. Identify voters who are pro choice, but don’t rank it as one of their top issues

A custom model could also be used to identify support for a candidate who breaks the mold or to identify newly persuadable and/or single-issue voters.

Abortion Modeling – With the Supreme Court’s Dobbs ruling, abortion is in the spotlight in a way it hasn’t been for decades. This barrage of news and of new state restrictions means that people’s feelings on the topic are more defined than they were just months ago. 

On every survey that Change Research fields, we ask standard questions about the changing political environment. Within weeks of the Dobbs decision, we had tens of thousands of responses, and that number continues to grow. Because of this up-to-date data, our modeling speaks to where people are today, rather than where they were last year or in the last decade.

Change Research’s model scores not only estimate each registered voter’s support for abortion rights, but also models more nuanced feelings about abortion in a given region or district. This enables you to tailor language accordingly, and more accurately optimize your digital, mail, and field targeting and outreach. Prices range from $1500 to $4000 for local and statewide models, and $15,000 for a national model.

Off the Shelf Models – Change Research has collected survey responses from over 3 million voters over the last 4 years. Using those responses, we’re able to develop models to predict core voter traits and opinions. When a model is purchased, you will get access to a file of all registered voters in a targeted district or geography, with SOS ID and scores. Change Research currently has 5 models available, for a price of $4,000 each:

• 2016 vote choice
• 2020 vote choice
• Education level
• Ethnicity
• Partisanship

The New Georgia Project

The New Georgia Project (NGP), a nonpartisan effort to register and civically engage Georgians, came to Change Research in 2018, and has since become one of our most valued and innovative research partners who played an instrumental role in flipping Georgia blue for the presidential and US Senate elections.

In 2020, Change Research conducted polls of likely voters of color in Georgia. The research focused on the issues that voters of color – and Black voters in particular – prioritize, their motivation to vote, their faith in the electoral process to bring the change they want to see, and the types of messaging are most persuasive. The data was used to build out two distinct voter models.

The first model was an “Abrams support score” for the presidential election. Change Research provided a rank ordering of all Georgia’s registered voters according to their likelihood to have supported Democrat Stacey Abrams in her gubernatorial candidacy in 2018. Three surveys included two core questions, which were used to estimate the probability of giving similar answers for every registered voter in Georgia:

1. “Did you vote for Stacey Abrams for Governor in November 2018?”
2. “How favorable are your feelings towards Stacey Abrams?”

Change Research also integrated five additional questions on these surveys solely for model improvement, which offered high information value to the modeling process and gathered insights about participants’ living arrangements, community engagement, and employment type.

The final model for the Abrams score, derived from Change Research’s modeling technology, underpins both our internal modeling (turnout, survey weighting), as well as similar engagements with progressive organizations, all of which is based on modern data science and machine learning algorithms.

The second model was a “US Special Election support score” in Georgia for Democratic candidate Raphael Warnock. In anticipation of a runoff election, the New Georgia Project identified two critical voter blocks would need to be activated: 1st-choice Warnock supporters who would need to be mobilized to get to the polls and vote for him again, and not-1st-choice Warnock supporters who were still inclined to vote for Democrats but who would need to be persuaded vote for him in the special election. Change Research developed these targeting audiences based on a model of pre-November surveys. In addition, Change Research supplemented the model with the additional information gathered in a short 16-question statewide survey with a larger sample that collected basic information, along with the Special Election Senate horserace. The results of the survey were used to build likely-supporter models, one for Warnock and another for his Democratic opponent, Joe Lieberman. Change Research delivered the results of these models as both individual voter records and as a Facebook digital audience.

Public Wise