Penn IUR Faculty Fellow Spotlight: Ken Steif

Ken Steif is the Program Director for the MUSA program, an Associate Professor of Practice in City and Regional Planning, and Penn IUR Faculty Fellow. At the forefront of data-driven public policy for more than a fifteen years, he combines technical knowledge of Geographic Information Systems (GIS) and applied statistics with an interest in housing policy, child welfare, education, the economics of neighborhood change, transportation policy and more. As MUSA Director, Steif has shifted the program's emphasis from GIS to civic technology, enabling students to develop technology and governance solutions to solve today's most complex public policy problems. His latest book, Public Policy Analytics: Code & Context for Data Science in Government (Routledge, 2021) enables readers to build public-sector analytics in R ranging from simple maps and indicators to complex machine learning algorithms. It also provides context to policymakers on how to trade off a “useful” service delivery algorithm with its potential bias. He recently discussed his book with a panel of experts at a book talk hosted by Penn IUR and the Department of City and Regional Planning.

Your recent book, Public Policy Analytics: Code & Context for Data Science in Government, tackles the challenges of applying data science to public sector decision-making. Data-driven, evidence-based decision-making is common in the private sector. What are the challenges of incorporating data science into governmental decision-making?

Data Science, defined as data-driven decision making, has taken over the private sector because algorithms make businesses more money. While economic bottom lines are paramount in business, negative externalities are often a distant concern. Not so in government. Government programs and the algorithms that increasingly power them must hit on multiple, often disparate bottom lines—like equity, politics, and social cohesion.

This is one of the great challenges of data science in the public sector, and I hope one of the juxtapositions that makes my new book Public Policy Analytics so engaging. Statistics are precise—but their application in the delivery of government services requires a deep domain expertise.

Can you give us an example of how a local decision-maker—a city planner, for example—might shift their traditional approach to program design or service delivery to incorporate a data-driven approach?

The first step is to understand whether a government program is providing value. Next, delve into the demographic, placed-based, and outcome data collected to administer the program. Not only do these data help understand what drives success, but also how to predict success. If prediction is feasible, it may be possible to develop some planning tools to better allocate resources.

You write about potential bias in algorithms. Can you explain how this might come about in a seemingly “objective” algorithm?

The bottom line is that when government discriminates, discrimination is baked into government data. If that data is then used to make decisions, it is possible for those decisions to be discriminatory, even if unintended. This is a legal concept called “disparate impact” and it is a core theme of the book. Readers learn how to “open the black box” of their algorithms, identify discrimination, and then communicate shortcomings and solutions to non-technical policymakers.

You propose the concept of “Algorithmic Governance” as a way of addressing potential bias. Can you describe this framework?

The main theme of the book is that government programs and algorithms are one in the same—they both allocate resources and deliver services. Algorithmic Governance advocates that algorithms be designed in the same way as programs—with transparency and community engagement at the fore.

There are four steps. First, evaluate the current program. Second, engage stakeholders in the existing program and the use of an algorithm. Third, do not procure a black box algorithm from a private company but develop a transparent, open-source solution, in house. Finally, evaluate the algorithmic service delivery against the “business-as-usual” approach to understand the value add.

What made you want to write this book?

This is a text I have been developing over the last three to four years to accompany my course also called Public Policy Analytics. Each chapter is a different use case that students take on in one- to three-week modules. Over time, I have developed some consistent narratives both around the development of analytics and their application in government. The real challenge was to weave these themes throughout the book, again, given that each chapter poses to the reader a unique challenge.

There are lots of data science books, but none written for those working in government. I hope that I have created a resource to fill this gap.

You run the Master’s in Urban Spatial Analytics (MUSA) program, which Penn IUR co-administers with Penn Design, which trains students in the thinking and in the tools and techniques you write about in this book. You are also Associate Professor of Practice in City and Regional Planning and, as a professor of practice, you founded and lead Urban Spatial, a consultancy that combines data science with public policy. How do your teaching, consulting, and writing influence one another?

As an educator and as a business owner, my goal is social impact. The demand for civic technologists in government is greater than the supply and, as an educator, I want to help grow the needed human capital. As a consultant, my goal is to develop and share new best practice widely and with minimal friction. Nearly everything we do at Urban Spatial is open source, meaning we post our computer code online for anyone to use.

Open source brings a social return on investment, not a monetary one. Our greatest victories have been when others pick up our tools and do something amazing with them.

Public Policy Analytics: Code & Context for Data Science in Government, with all the code and data, can be accessed online here. The print version can be ordered here. To watch a video of Ken discussing his book with a panel of experts, visit the Penn IUR website.