Predicting Voter Preferences

Libraries

• Python 3.4.1
• TensorFlow 2.15.1
• GeoPandas 20.22.7
• Pandas 1.16.0
• Matplotlib

Datasets

• CCES 2016 voter survey
• Official Presidential Election data

Tasks

• In 2016, which major party candidate (Trump or Clinton) would voters of third party candidates (Johnson, Stein, McMullin) likely choose as their second choice if Ranked Choice Voting was allowed?
• Rank all the features from most important to least important when it comes to determining whether third party candidates would put Trump or Clinton as their 2nd choice.
• Run a simulated presidential election with Borda, IRV, and Copeland using the new RCV ballots generated using official election data and second-choice predictions made using ML
• Compare predicted results with original 2016 election results

Feature importances

• Trained a Random Forest Classifier model using Trump and Clinton data

Second choice predictions for Third-party voters

Simulate 2016 Presidential Elections

Original

• Texas has 38 Electoral Votes. 1 was cast for John Kasich and 1 was cast for Ron Paul.
• Hawaii has 4 Electoral Votes and 1 was cast for Bernie Sanders.
• Washington has 12 Electoral Votes. 3 were cast for Colin Powell and 1 was cast for Faith Spotted Eagle.
• Maine has 4 electoral votes. 3 were cast for Clinton and 1 was cast for Trump.

Borda Count

Instant Run-off Voting

Copeland

Limitations

• Third party voter counts for states are approximations based on nationwide percentages since state-wise voter counts for each third party candidate are not provided.
• Results might change if more features are added to the ML model.
• A lot of the features are unanswered on the survey so training data is limited