wentao yao

2021 · Misconduct, location and income

Citizens' Police Data Project

Is there over-policing in low-income neighborhoods? SQL, maps, graphs and NLP on police misconduct data.

Year
2021
Who
Team project · The Dapper Squirrels
Stack
Data science · SQL · Graph analytics · NLP
Project site ↗Code ↗
From the project.
From the project.

01WhatDoes income shape policing?

With my team, The Dapper Squirrels, I worked on the Citizens' Police Data Project database. Our theme was misconduct by location and community: is there over-policing in low socio-economic neighborhoods, and could a neighborhood's income influence the “victim” narrative in complaint reports?

02QuestionsWhere we started.

  • Which neighborhoods have the highest and lowest incomes?
  • What are each neighborhood's income and complaint records per capita?
  • How many tactical response reports per capita?
  • What is the racial makeup of each community?
  • Which streets have the most allegations in each beat?

03HowFive checkpoints.

  • SQL — the relational questions above.
  • Visualization — officer hours per capita over the years, and complaints versus tactical responses per capita.
  • Interactive visualization — high- and low-income communities on a map, with a time slider.
  • Graph analytics — co-accusal networks of officers, with triangle counts and PageRank to find the most connected officers.
  • NLP — topic modeling of complaint text, from TF-IDF towards graph neural networks and BERT.
From the project.
From the project.

Adapted from my original write-up · more in the links above.