The Page 99 Test: Howard Steven Friedman and Akshay Swaminathan's "Winning with Data Science"

Howard Steven Friedman, an adjunct professor at Columbia University, is a data scientist with decades of experience leading analytics projects in the private and

public sectors. His previous books, including Ultimate Price (2020) and Measure of a Nation (2012), have been translated into many languages and featured on national media.

Akshay Swaminathan is a data scientist who works on strengthening health systems. He has more than forty peer-reviewed publications, and his work has been featured in the New York Times and STAT. Previously at Flatiron Health, he currently leads the data science team at Cerebral and is a Knight-Hennessy scholar at Stanford University School of Medicine.

Friedman and Swaminathan applied the “Page 99 Test” to their new book, Winning with Data Science: A Handbook for Business Leaders, and reported the following:

Winning with Data Science is a unique contribution to the data science field. It takes a narrative approach with fictitious characters and dialog to help us understand the trajectories of Kamala and Steve as they work with data science teams in their respective health care and finance companies to help deliver their projects. It focuses on the customer and specifically on guiding the customer to ask good questions and be armed with frameworks that help ensure that they get the most value out of their data science investments.

The process we used to write the book is also unique. We each wrote one character’s experience. We then worked together to finalize the manuscript and make sure all aspects of data science projects are covered.

Page 99 of Winning with Data Science drops us in the middle of a discussion between Kyra, a data scientist, and Kamala who is the data science customer and business expert. From the dialog, the reader immediately understands that they are working on a problem for a healthcare company where they are trying to understand whether the results of a study for a new cancer medication called ClaroMax can be applied to their company.

The coworkers focus on potential weaknesses in the methodology of the study, how patients were recruited, and other factors that might contribute to patient outcomes.
Kyra continued. “Once the inclusion and exclusion criteria are defined, it’s important to consider how participants will be recruited. One of the most important considerations is generalizability. To what extent will the recruitment methodology allow you to extend the conclusions of this study beyond this specific population? Let’s suppose that BioFarm chooses to recruit participants who are being treated at the top cancer institute in the United States. Can you think of any reasons why this recruitment mechanism may affect the generalizability of the results?”
Kamala responds by drawing on her insight into patient care and challenges the generalizability of the study.
Kamala thought for a few seconds. “I can think of a few reasons. For one, patients who are treated at this cancer institute may receive a higher level of care than patients who are treated elsewhere in the country. Second, the patients who are treated at this top cancer center may be different from patients treated elsewhere: they may have better insurance coverage, they may be more well-off, they may have more advanced cancer, etc. The recruitment mechanism would cast doubt on whether the results of the study could generalize to patients who are treated elsewhere in the country. Sure, ClaroMax may work if it’s administered in the top cancer center in the United States, but what about when it’s administered in an underresourced cancer clinic in the rural Midwest? BioFarm should consider recruiting participants from diverse clinical settings to improve the generalizability of its findings.”
Kyra then suggests how they can implement a randomized study to test the impact of the product on some markets.
Kyra was impressed. “You’re sounding like a causal inference expert already! Once a recruitment strategy is in place, the next thing to think about is how to operationalize randomization.
The Page 99 Test works reasonably well in that readers would immediately understand that this is not a traditional data science textbook but rather uses characters and dialog to communicate some fairly complex ideas. It also succeeds because it introduces one of our main characters, Kamala, and shows her modeling what we call a customer mindset toward a real-world example of a data science project. That said, page 99 would not help the reader understand the overall scope of the book, so I would give it a B-.

The main shortcoming is that Kamala is one of the two main characters in the book, and this section does not mention Steve who works at a consumer finance company. It is also limited to a discussion about a single aspect of data evaluation. With only this one page, readers would fail to understand that another industry is discussed in the book, nor could they grasp that the book covers a large assortment of topics from project management to tool selection to supervised machine learning to unsupervised machine learning to ethics.

Winning with Data Science makes data science approachable by allowing the reader to follow the adventures of Kamala and Steve and experience real world scenarios of how data is used by different companies. This approach is unusual, but we are hearing from readers that it is highly effective. The book has generated substantial interest from companies looking to maximize their investments in data science and the authors have been invited to give numerous company presentations on the topics.