CDC's Near-Term Deaths Estimates Epitomize Analytical Failures
The CDC’s consensus estimate for near-term deaths from COVID-19 is an ongoing demonstration for why many corporate analytics projects fail to achieve their desired results.
The estimate is derived by averaging the results of 35 models from respected institutions whose models are based on varying assumptions for things like social distancing.
The estimates for the week ending January 2 range from 5k to well over 25k. Those results are depicted in the left-hand chart. From that ball of confusion, the CDC estimated 18,000 deaths for the week (right chart).
Here's the simple and important issue. 98% of the variation in the number of deaths two weeks from now can be explained by the number of people in US hospitals today. Unfortunately, that means the consensus estimate should be centered on 20k, not 18k. It also means the mountain of work to create these charts is a gigantic waste of time and analytical talent.
Managers have been led to believe that accurate predictions require massive amounts of data and algorithms beyond human understanding. Sometimes that is true. Other times it’s just a way to burn through shareholder equity. You need someone on your side who can help you differentiate one from the other.