Performance Metrics for Canvassers in Campaigns

Canvasser PAR

Common ways of assessing how productive canvassers have been when door knocking are:

  1. Counting the number of shifts they've canvassed,
  2. Measuring the canvasser's Contact Rate: how often they get people to answer the door,
  3. and Persuasion rate: the rate at which they get someone to committ to supporting their candidate.

However, we can take all three of these and combine them into a single metric: Positive IDs Above Replacement-level Canvasser, or PAR. This just means we take the total number of positive IDs they've gotten, and subtract what a "Replacement-level" Canvasser would have done in the same situations, i.e. a fresh canvasser with no experience.

Read More

A Geographical Progress tracker for campaigns made in plotly

Tracker Test

Here you can find a tracker meant to demonstrate campaign progress over time, visualized geographically, i.e. an interactive map.

Click through to play around with the tracker in the notebook, or click on the map and get a full screen tracker.

Read More

Solving a Drift-Diffusion model via a Weighted Spectral Method in Julia

Solving a Drift-Diffusion model via a Weighted Spectral Method in Julia

Predicting Voter Affiliation using Machine Learning Classifiers in Python

Predicting Voter Affiliation using Machine Learning Classifiers

In this notebook I'll go through some work I did for a recent political campaign where we needed to predict the affiliation of voters whose ballots were invalidated. We wanted to sue to restore the validity of many ballots - but obviously we wanted to prioritize those ballots who we had good reason to think voted for our candidate.

Our candidate being young, progressive, and Latin american, we had good reason to expect that age, ethnicity, and geography (ie more left-leaning neighorhoods) were reasonable factors to influence voter affiliation. In the following map, brighter yellow areas are the strongest for our candidate relative to the primary opponent, and dark purple areas are strongest for the primary opponent relative to our candidate. Voting preference by geography

Read More

Traveling Waves for the Regularized Long-Wave Equation in Julia

RLW writeup

A Machine Learning model comparison on the 1984 Congressional Voting Records Dataset

A Machine Learning model comparison on the 1984 Congressional Voting Records Dataset

A Customer Segmentation Project

customer_segmentation