R_t Covid-19

Sign up for updates

Calculating R_t

The metric being tracked here (R_t) represents the effective reproduction rate of the virus calculated for each locale. It lets us estimate how many secondary infections are likely to occur from a single infection in a specific area. Values over 1.0 mean we should expect more cases in that area, values under 1.0 mean we should expect fewer.

Download our Jupyter notebook for how we’re calculating R_t

Download calculated R_t values per-state

Case count data from The COVID Tracking Project.

Model Updates

• The current model uses the same math as in the notebook.
• The model is evolving rapidly and while values should not change dramatically, there may be changes to historical values when we find improvements or fix data anomalies. You can download the latest run here and email us with any suggested model improvements.

Known Issues

• South Carolina has had two days in the past week without reporting new cases. An upcoming update to the model will address irregular data like this.
• Connecticut saw a very large increase in cases relative to its baseline on 4/20 due to "true-up"-ing their calculations with the CDC's guidelines for which cases should be counted. Their R_t will normalize in the coming days.
• Washington State had several days from 4/10 to 4/13 without reporting new cases. They've resumed reporting and our model will continue to improve its R_t estimate for WA as more days come in.
• Absolute testing levels should not affect this algorithm much, but a fast ramp or decline in testing will affect numbers.
• There is a delay between onset of sickness and testing positive, this is not reflected in these numbers. Actual R_t values are delayed by some amount.
• While we make a best effort to accurately describe R_t, nothing can do that perfectly - please calibrate accordingly.

Blog Posts

The Metric We Need to Manage COVID-19

The Numbers Behind Social Distancing

The US Just Crossed a Dangerous Threshold

Contact us with any questions.

Data analysis by Kevin Systrom and site built by Mike Krieger, with thanks to Ryan O’Rourke and Robby Stein.
Also thanks to Frank Dellaert and Adam Lerer for their contributions to the model and notebook.
Visualizations built using d3; site built using Next.js.