TO

Interested Parties

FROM

Chris Wilson, Ashlee Rich Stephenson, WPA Intelligence
Joe Pounder, Colin Reed, Definers Public Affairs

DATE

August 27, 2018

RE:

Political Peril Facing Red State Democrats On SCOTUS Obstruction

WPA Intelligence and Definers Public Affairs collectively set out to understand the raw political impact
of the nomination of Judge Brett Kavanaugh to the Supreme Court of the United States. With hearings
and a vote pending, it’s clear that obstruction of Judge Kavanaugh will be a political albatross for
Democratic Senate incumbents in the lead-up to November 6.
Bottom line: support for Judge Kavanaugh’s nomination is highest in four states that President Trump
won by double-digits in 2016 and where incumbent Democratic senators are particularly vulnerable:
•  North Dakota (80%)
•  Montana (71%)
•  West Virginia (71%)
•  Missouri (67%)
Senators in those states stand to pay a political price if they join national Democratic party leaders and
oppose Kavanaugh.
In fact, in every state we modeled, more likely voters support the nomination of Judge Kavanaugh than
voted for these Democrats when they were last elected in 2012.
The following chart reports the number of likely voters, along with their respective percentages as
a makeup of the entirety of the anticipated 2018 electorate, in the ten (10) states with Democratic
senators up for re-election that President Trump won in 2016.

Research • Rapid Response • Communications • Tracking

1

 The question we asked more than 7,000 voters in these states was:
Thinking about President Trump’s nominee to the Supreme Court, Judge Kavanaugh, are you more likely to
vote for a U.S senator who supports the President’s Supreme Court nominee, more likely to vote for a U.S.
senator who opposes the President’s nominee, or is a U.S. senator’s position on whether they support the
President’s nominee to the Supreme Court not something that will affect your vote?
These responses were then used to train a predicted model and score every voter in each state on their
likelihood to support a candidate who supports Kavanaugh or a candidate who opposes him.

Research • Rapid Response • Communications • Tracking

2

 Statement of Methodology
Thousands of responses were collected via Interactive Voice Response (IVR) calls. Responses were
then matched to a consumer data enhanced voter file which contains data on vote history, voter
demography, socio-economics, political behavior, and consumer behavior.
A variety of machine learning techniques were tested to find the best fitting model including penalized
regressions, boosted and bagged decision trees, adaptive boosting, polynomial fits using support
vector machines, and random forest ensemble models. In this case, stochastic gradient boosting
provided the best performing model.
Based upon this model we scored the full consumer data enhanced national voter file for the
probability that each voter supports or opposes the nomination of Judge Kavanaugh to the Supreme
Court. Modeled voters totals may not match states’ BOE (Board of Elections) reported totals due to
state’ various “inactive voter” definitions and related policies.

Research • Rapid Response • Communications • Tracking

3