Analytic Strategy Partners

Improve your analytic operations and refine your analytic strategy

Archives for October 2020

Why Great Machine Learning Models are Never Enough: Three Lessons About Data Science from Dr. Foege’s Letter

by

Figure 1. William H. Foege, MD, MPH standing next to the bust of Hygeia, the Greek goddess of health on the grounds of a CDC facility in Atlanta. Dr. Foege was the Director of the CDC from 1977 until 1983.
Source: https://phil.cdc.gov/Details.aspx?pid=8149.

Foege’s Letter

In September, William H. Foege, MD, MPH sent a private letter to Robert Redfield, the Director of the CDC reminding him that the “best decisions come from the best science” and the “best results come from the best management.” The letter became public on October 6, 2020 in a USA today article written by Brett Murphy and Letitia Stein and it is well worth reading.

In this post, we look at how these insights apply to building analytic and AI models and applying them to challenging real world problems.

Dr. Foege trained in the Epidemic Intelligence Service (EIS) of the Centers for Disease Control and Prevention (CDC) between 1962 and 1964. The EIS is a fellowship program run by the CDC that trains epidemiologists and is famous the quality of the epidemiologists it trains and for the effectiveness of its investigative and emergency response efforts. In the 1970s, Dr. Foege made critical contributions to the global strategy that led to the eradication of smallpox, culminating in the May 8, 1980 declaration at the 33rd World Health Assembly (WHA) that the world was free of this disease. Smallpox is one of only two diseases that the WHA has designated as eradicated. He served as the Director of the CDC from 1977 to 1983.

Great Science Supports Great Management

To say the least, Dr. Foege is well qualified to understand the role of data and science, management and coalitions, and the leadership necessary to tackle challenging problems, such as the COVID-19 pandemic and how to organize and lead the response to it. In the letter he states:

The first thing … [is] to face the truth. We have learned that the best decisions are based on the best science and the best results are based on the best management. William Foege, MD, MPH, in a letter dated Sept 23, 2020.

From an analytics perspective, I would add two more layers

  • the best results are based on the best management
  • the best decisions are based on the best science
  • the best models are based on the best data
  • the best data are based on the best data sharing (or data collection efforts)

The first two are the domain of management; the second two are the domain of data science. The role of analytic governance is to knit these together through an analytic strategy and to develop a strategic implementation plan to produce the best results. See Figure 2. For background information about analytic strategy and analytic governance, my Primer may be helpful.

Lessons for Tackling Challenging Data Science and Analytic Problems

From the perspective of this blog, I would highlight three lessons that Dr. Foege’s letter suggests:

  1. When you have a challenging problem, face the truth and speak the truth.
  2. Clearly separate the data science / analytics from the management, and make sure you have the best of both. It is critical that there is sufficient analytic governance and strong enough leadership to guarantee that the best science supports the best management.
  3. Good models require good data, and one of the best ways to get good data is by through data sharing collaborations. This is especially important in times of national emergencies.
Figure 2. In analytics and AI, the best data sharing and data collection leads to the best data; the best data leads to the best models; the best models lead to the best decisions; the best decisions lead to the best results. The best results require both the best science and the best management. Analytic governance is the governance structure that knits this all together.