Our group attempts to investigate how pressure stemming from society could play a role in the outcomes of police misconduct investigations. We look at this phenomenon from two complementary viewpoints: from the cases that are picked up by the press (both local and national) and from the social media landscape represented here by Twitter.

Research Question: When a case of police misconduct causes indignation and broad discussion in society, how does it influence the case's investigation?

This single question unfolds over a series of hypothetical, measurable effects on the investigation (part I of section "Feature Space" below); and we can measure pressure exerted by society using a number of digital channels (part II of section "Feature Space" below).

Feature Space:

We hypothesize that, to some extent, the following effects could happen:

Effect 1: It could expedite a disciplinary investigation;

Effect 2: Increase the likelihood of the officer(s) involved being disciplined;

Effect 3: Increase the likelihood of the state granting compensation.

And these effects could depend on the following features:

Feature 1: The number of news stories reporting on the case;

Feature 2: The involvement of certain media outlets (local vs national ones);

Feature 3: The length of the news burst (from the first story to the last);

Feature 4: Social engagement with the case as measured in Twitter.

Methodology:

Our proposed methodology comprises the following steps:

Step 1. Manually analyze a few cases on Google: Try to use some allegation fields to search for stories in Google and map the relevant news sources (both local and national);

Step 2. Manually analyze a few cases on Twitter: Try to use some allegation fields to search for tweets and map the Twitter handles that can be considered sources or good proxies for public sentiment;

Step 3. Collect more data points: Use Python, the Newspaper library [1], and Twitter API [2] to automate the process in Steps 1 and 2 in order to collect a larger sample;

Step 4. Run predictive analysis [3]: Use the features generated by Step 3 to predict the target features (Effects 1, 2, and 3 can be obtained directly from the Invisible Institute's Citizens Police Data Project).

[1] The Newspaper Library for Python — https://newspaper.readthedocs.io/en/latest/

[2] Twitter API — https://developer.twitter.com/en/docs/developer-utilities/twitter-libraries

[3] Hall, Mark, et al. "The WEKA data mining software: an update." ACM SIGKDD explorations newsletter 11.1 (2009): 10-18.