Social Impact

IHME Models over time, for the United States, for 4 representative dates, made with the interactive tools offered below.

IHME Model Uncertainty, Visualized over Time

The Institute for Health Metrics and Evaluation (IHME) creates, maintains, updates, and publishes an open-source statistical model of the impact of the COVID-19 pandemic, based on open-data resources. As a public service, glue solutions, inc. here offers an online tool for visualizing the evolution of the IHME models over time.

The general public has seen many versions of the IHME "Daily Deaths" plots, including in several White House briefings. Our goal here is to offer a look at how the models change—appropriately, in response to new data and informationover time, and how that affects model updates. In a companion essay online at the Prediction Project site, we offer more context on why this evolution is so interesting.

(Banner above shows sample IHME "Daily Deaths" graphic, from 14 May 2020.)

What's this tool for? Using the interactive graphics below, you can re-create the display of deaths/day akin to what would have been visible at IHME's COVID-19 web site on a range of modeling dates, for any region you select. In addition, you can show more than one model (date) at a time, to make comparisons.

How should I interpret what I see? In each of the panels below: red dots show reported actual deaths per day; solid blue lines show forecasts, and light blue shaded regions show uncertainty bands. Those uncertainty bands indicate ranges of possible outcomes, as forecast on the date when the model was made. The ranges should account for 95% of possible outcomes. As one can see by moving the time slider below each graph, the model and its associated uncertainty band change over time. As more and more models are added, regions where shading appears darkest are regions where models have been most consistent.

There are four versions of the IHME evolution visualization offered below. They are as follows (with source links in [brackets]):

  1. For the United States, showing only 4 representative model dates. [source, GitHub] [mobile site]
  2. For the United States, offering a wide range of model dates [source, GitHub]
  3. For the World, showing only 4 representative model dates [source, GitHub] [mobile site]
  4. For the World, offering a wide range of model dates [source, GitHub]

This content is licensed as CC BY, with attribution "glue solutions, inc." Static graphics can be extracted using the three dots at the upper right of each graphic.

How can I share interesting graphs I create? Join the discussion at the 10QViz.org IHME COVID-19 Model Uncertainty Visualization page to upload your graphic and tell the world what it shows you. (You can download your graphic using the three dots at the top right of each panel below.)

At present, this site's visualization interactions work best on larger screens. We provide links to standalone views of the visualization showing 4 representative model dates that may work better on many mobile devices.

IHME Models over time, for the United States, for 4 representative dates

IHME Models over time, for the United States.

Click the grey bar to show/hide models for dates listed.


IHME Models over time, internationally, for 4 representative dates

IHME Models over time, internationally.

Click the grey bar to show/hide models for dates listed.


TECHNICAL DETAILS The IHME statistical model (see preprint) is based mainly on the Johns Hopkins University CSSE COVID-19 data hosted on GitHub; deaths per day may differ from local government counts. The IHME modeling approach is updated frequently, as new data and information become available. There was a substantial methodology change in the IHME models in the beginning of May (May 4 for US locations and May 12 for all locations) to a hybrid model that combines statistical curve-fitting with an epidemiological (SEIR) model. On this website, we show a subset of all the models released online by IHME for clarity and ease of use, focusing on models that span the time range and ones that provide substantially new predictions. New countries have been added over time, so not all countries have models on all dates. There are additional caveats about the data and interpretation for some of the included countries, detailed at the IHME FAQ page.

CREDITS Jonathan Foster wrote the code for the interactive graphics shown here, using functionality provided by the open-source "Vega" visualization grammar and the Altair library. Alyssa Goodman helped design the graphics and wrote the text. We are grateful to many colleagues for comments, and to Francisco O. for technical assistance.