COVID–19 Incarceration
Model

As COVID-19 spreads, state and local government leaders are on the front-lines of managing the response.

Get startedContact

This toolkit was created to help criminal justice decision makers track and anticipate the impact of Covid-19 in incarceration settings. It includes:

Outbreak modeling
System (and facility) level modeling to project cases, hospitalizations, and deaths among both staff and the incarcerated population.

Real-time rate of spread
The model reports the current rate of spread and trajectory of the outbreak in each facility, based on case numbers entered to-date.

Impact analysis
Comparison of before-and-after scenarios for policy changes and early release decisions

What's Changed

The model is updated daily to include new case numbers for each county and state. 

In addition, we update assumptions based on new information and add additional features and capabilities that might be helpful. We recommend re-downloading the model every 1-3 days to ensure you’re using up-to-date information.

5/15/20

  • Added initial impact report functionality - provides scenario comparison between early projection / without mitigation steps taken, and current projection with them
  • Added support for copying the baseline scenario, to create new scenarios to test potential changes against the baseline

5/5/2020

  • Added the ability to add historical case or population information
  • Added real-time Rt support - the model now learns, on a per-facility basis, the current rate of spread and uses this for the projection instead of attempting to derive it from the policies introduced.
  • Added ‘Rate of Spread’ tab to the scenario view, to make it easier to compare rate of spread across multiple facilities

4/21/20

  • Support added for modeling multiple facilities in the same web model
  • Summary view added for a ‘baseline scenario’ / current real-world cases

4/16/2020

  • We recommend all users update to this version or later.
  • Based on feedback, the population turnover ratio can now be entered for the facility being modeled to adjust the total impacted population so that it includes individuals who left and new individuals who entered the system. (Thanks to Bret in PA for submitting this feature.)
  • Based on feedback, the model now has improved handling for mid-outbreak projections. Older versions of the model assumed all cases started roughly on Day 0, which worked well for early-stage outbreaks but underestimated deaths from the virus in the near-term when several dozen cases were already in progress. This version of the model better distributes in-progress cases along the timeline.

4/13/2020

  • Web-based Covid-19 model released
  • Minor parameter updates and bug fixes

4/3/2020

  • Bugfix: The outputs tab in the 4/2 version included an error that left staff totals in the incarcerated population outputs. This is fixed in the 4/3 version.
  • Based on feedback, a facility’s percent over or under capacity (as an indication of density and ability to quarantine) has been added to the model.
  • Based on feedback, we’ve updated the ‘Facilities type’ to allow for modeling facilities which are partially bunked housing and partially cells.
  • Based on feedback, we’ve added a set of controls to add a planned future release of individuals into the model.

4/2/2020

  • Bugfix: A calculation error magnified case numbers when the population was entered by age bracket. This is fixed in the 4/2 version. 
  • Separated outcomes for staff from outcomes for incarcerated individuals to help with staffing models.
  • Tied in-facility case estimate more closely to the facility population.
  • Added case and fatality estimates for ~500 new counties.

3/31/2020

  • Fixed an issue with counting staff as part of the in-facility population. 
  • Added support for modeling rate of spread in bunks vs. cells.

3/30/2020

  • Added support for county jails, and for full incarceration community (prisons + jails) in a state.

3/28/2020

  • Initial state prisons model.

Note: The date on the 'Output' tab reflects the release date of that model. Even if the model was released several days ago, if you just downloaded it today the case and fatality numbers will be current within one day.

Instructions

If you’re using the model for the first time, you’ll be asked to set up your first facility. First, enter your facility’s name, the type of facility (prison or county jail), and its current location.

Then, enter your current cases and population for the facility, and the set of mitigation steps taken to-date.

Once you’ve saved the new facility, you’ll be taken to your baseline scenario (which lists all of your facilities).

To calculate the rate of spread, we’ll need historical data on when cases developed in each facility. Click the red number of cases to add new or historical information.

Click on the date to see the calendar widget, which will tell you which dates are missing data for this facility (the red background).

(Note: Cases should always be entered as a cumulative total - so include individuals who have recovered. Also, remember to hit ‘Save’ each time if you’re adding multiple days’ data at once.)

Once you have facilities set up, and enough historical data, you’ll see the current rate of spread in your facilities. You can click on a facility to see its rate of spread over time:

Or, from your baseline scenario you can click the ‘Rate of Spread’ tab to compare all facilities.

Lastly, to generate an impact report see our User Guide.

Join the Conversation

Questions? Feedback? Need help refining this model or fitting it to your system? Send us a note at covid@recidiviz.org.

Get Assistance

Recidiviz and the Council of State Governments Justice Center are offering assistance to state and county leaders looking to refine and implement the model in their system. Set up a 15 minute initial consultation with the team here and we’ll get you started.

FAQ

What makes this model different from the models I see online or in the news?

This model is similar to models being used publicly, but factors in common criminal justice variables (population size, potential population releases, etc).

It also uses an R0 (basic reproductive number, which estimates the rate of an infection's spread) that is specific to prison and jail populations, where airborne illnesses spread at a significantly faster rate than is seen in the general population. See relevant academic work on the spread of similar illnesses in enclosed populations cited in the 'Variables' tab.

Can I get help expanding this model to better fit my jail or prison system?

Yes! You can set up a 15m consultation with the team that produced this model here.

We also suggest joining the 'Criminal Justice C19' community on Slack (see ‘Join the Conversation’ above). This is a community of criminal justice department leads and research staff sharing knowledge on how they're approaching the challenges posed by COVID-19.

Where did all these numbers come from?

You can see a complete list of sources in the ‘Variables’ tab in the model. 

Where can I learn more about how others in my field are preparing for COVID-19?

Join the COVID-19 Criminal Justice community on Slack (see ‘Join the Conversation’ above). This is a community of criminal justice department leads and research staff sharing knowledge on how they're approaching the challenges posed by COVID-19.

How can I submit questions, concerns, or suggestions about the model?

First, double-check that you have the most recent version of the model, in case your issue has already been addressed (you can download the latest copy above).

If your issue persists, send your question to covid@recidiviz.org or set up time with the Recidiviz team here. Recidiviz is a non-profit organization applying technology to issues in criminal justice, and maintains this model to help agencies anticipate and respond to the crisis.

For general questions or suggestions, we'd recommend bringing them up in the 'Criminal Justice C19' community on Slack, where other criminal justice agencies can contribute and build on your ideas (see ‘Join the Conversation’ above).

Can I make changes to this model, or re-share it with others?

You can modify or re-share this model so long as you keep the license notice in the FAQ tab of the model unchanged.

What other guidelines or recommendations are available to prepare our system for COVID-19?

How can I help other criminal justice systems?

The best way to help the rest of the community in real-time is to join the 'Criminal Justice C19' community on Slack (see ‘Join the Conversation’ above). This is a community of criminal justice agency leads and research staff sharing knowledge on how they're approaching the challenges posed by COVID-19.

Can I still use the Excel version of the model?

The model was initially released in Microsoft Excel, but as we added more functionality we surpassed the calculation complexity that Excel is intended to support (real-time rate of spread calculations, in particular, are very math-heavy).

We’ve removed the Excel model from this page because it was falling behind in functionality, and it became challenging to keep features and epidemiology factors aligned across both. If you’d like a copy of the old Excel version, however, send us a note at covid@recidiviz.org and we can share a copy with you.

The web version of the model is still open source, meaning you can still see how the model is implemented and how calculations are run in our code repository. You can also see an overview of the SEIR compartments used by the model in this diagram.

Contributors

We’d like to thank the following contributors for helping us build and continuously improve the model.

  • Justine Kunz (Senior Data Scientist, Recidiviz)
  • Kristofer (Bret) Bucklen, Ph.D. (Director, Bureau of Planning, Research & Statistics, Pennsylvania Department of Corrections)
  • Ribhav Gupta, Graduate Student, Department of Epidemiology, Stanford University School of Medicine
  • Ben Packer, Ph.D. (Software Engineer, AI and ML Fairness, Google)

You can see a full list of model collaborators on recidiviz.org/collaborators.