Public Policy Analytics Code and Context for Data Science in Government

Preface 

Introduction 

Indicators for Transit Oriented Development 
1.1 Why Start With Indicators? 
1.1.1 Mapping & scale bias in areal aggregate data 
1.2 Setup 
1.2.1 Downloading & wrangling Census data 
1.2.2 Wrangling transit open data 
1.2.3 Relating tracts & subway stops in space 
1.3 Developing TOD Indicators 
1.3.1 TOD indicator maps 
1.3.2 TOD indicator tables 
1.3.3 TOD indicator plots 
1.4 Capturing three submarkets of interest 
1.5 Conclusion: Are Philadelphians willing to pay for TOD? 
1.6 Assignment - Study TOD in your city 


Expanding the Urban Growth Boundary
2.1 Introduction - Lancaster development
2.1.1 The bid-rent model
2.1.2 Setup Lancaster data 
2.2 Identifying areas inside & outside of the Urban Growth Area 
2.2.1 Associate each inside/outside buffer with its respective town
2.2.2 Building density by town & by inside/outside the UGA 
2.2.3 Visualize buildings inside & outside the UGA
2.3 Return to Lancaster’s Bid Rent 
2.4 Conclusion - On boundaries 
2.5 Assignment - Boundaries in your community 

Intro to geospatial machine learning, Part 1 
3.1 Machine learning as a Planning tool 
3.1.1 Accuracy & generalizability 
3.1.2 The machine learning process 
3.1.3 The hedonic model 
3.2 Data wrangling - Home price & crime data 
3.2.1 Feature Engineering - Measuring exposure to crime 
3.2.2 Exploratory analysis: Correlation
3.3 Introduction to Ordinary Least Squares Regression 
3.3.1 Our first regression model
3.3.2 More feature engineering & colinearity 
3.4 Cross-validation & return to goodness of fit
3.4.1 Accuracy - Mean Absolute Error 
3.4.2 Generalizability - Cross-validation 
3.5 Conclusion - Our first model 
3.6 Assignment - Predict house prices 

Intro to geospatial machine learning, Part 2
4.1 On the spatial process of home prices 
4.1.1 Setup & Data Wrangling 
4.2 Do prices & errors cluster? The Spatial Lag
4.2.1 Do model errors cluster? - Moran’s I
4.3 Accounting for neighborhood 
4.3.1 Accuracy of the neighborhood model 
4.3.2 Spatial autocorrelation in the neighborhood model 
4.3.3 Generalizability of the neighborhood model
4.4 Conclusion - Features at multiple scales

Geospatial risk modeling - Predictive Policing 
5.1 New predictive policing tools 
5.1.1 Generalizability in geospatial risk models 
5.1.2 From Broken Windows Theory to Broken Windows Policing 
5.1.3 Setup 
5.2 Data wrangling: Creating the fishnet
5.2.1 Data wrangling: Joining burglaries to the fishnet 
5.2.2 Wrangling risk factors 
5.3 Feature engineering - Count of risk factors by grid cell 
5.3.1 Feature engineering - Nearest neighbor features 
5.3.2 Feature Engineering - Measure distance to one point 
5.3.3 Feature Engineering - Create the final_net 
5.4 Exploring the spatial process of burglary 
5.4.1 Correlation tests 
5.5 Poisson Regression 
5.5.1 Cross-validated Poisson Regression 
5.5.2 Accuracy & Generalzability 
5.5.3 Generalizability by neighborhood context
5.5.4 Does this model allocate better than traditional crime hotspots? 
5.6 Conclusion - Bias but useful? 
5.7 Assignment - Predict risk 

People-based ML models
6.1 Bounce to work
6.2 Exploratory analysis 
6.3 Logistic regression
6.3.1 Training/Testing sets 
6.3.2 Estimate a churn model 
6.4 Goodness of Fit 
6.4.1 Roc Curves 
6.5 Cross-validation 
6.6 Generating costs and benefits 
6.6.1 Optimizing the cost/benefit relationship 
6.7 Conclusion - churn 
6.8 Assignment - Target a subsidy 

People-Based ML Models: Algorithmic Fairness
7.1 Introduction 
7.1.1 The spectre of disparate impact 
7.1.2 Modeling judicial outcomes 
7.1.3 Accuracy and generalizability in recidivism algorithms 
7.2 Data and exploratory analysis 
7.3 Estimate two recidivism models 
7.3.1 Accuracy & Generalizability 
7.4 What about the threshold?

7.5 Optimizing ‘equitable’ thresholds 
7.6 Assignment - Memo to the Mayor 

Predicting rideshare demand
8.1 Introduction - ride share 
8.2 Data Wrangling - ride share 
8.2.1 Lubridate
8.2.2 Weather data 
8.2.3 Subset a study area using neighborhoods 
8.2.4 Create the final space/time panel 
8.2.5 Split training and test
8.2.6 What about distance features? 
8.3 Exploratory Analysis - ride share 
8.3.1 Trip_Count serial autocorrelation 
8.3.2 Trip_Count spatial autocorrelation 
8.3.3 Space/time correlation? 
8.3.4 Weather
8.4 Modeling and validation using purrr::map
8.4.1 A short primer on nested tibbles 
8.4.2 Estimate a ride share forecast 
8.4.3 Validate test set by time 
8.4.4 Validate test set by space 
8.5 Conclusion - Dispatch
8.6 Assignment - Predict bike share trips

Conclusion - Algorithmic Governance 

Index

Biography

Ken Steif Ph.D is the Director of the Master of Urban Spatial Analytics program at the University of Pennsylvania and an Associate Professor of Practice in the City Planning Program. He teaches courses on the application of spatial analysis, statistics, predictive modeling and data visualization to solve traditional and contemporary public policy programs. Dr. Steif also is the founder of a consultancy that develops analytics for both public and private-sector clients. He lives in West Philadelphia with his wife Diana and sons Emil and Malcolm. You can follow him on Twitter @KenSteif.

Finally, a book that connects two parallel lessons. First, how to design geospatial data science workflows in the public policy sector; and second, when applying algorithms in government, there is no free lunch. Rather than making data science in government a bleak challenge suited only the fearless, Ken sustains a tone of optimism and a sense of purpose for readers curious enough to ponder the kernels of wisdom he has sprinkled in each chapter. 

-- Mark Wheeler, Chief Information Officer, City of Philadelphia

Public Policy Analytics is a must-read for creating data-driven urban plans and policies. In crisp and compelling chapters, Dr. Steif steps through real-world problems and links them to critical methods in R. The included assignments are perfect for both self-guided students and educators. There is no better guide to data science in the public realm! 

-- Dr. Allison Lassiter, Assistant Professor of City & Regional Planning, Univ. of Pennsylvania

Ken Steif has written a clever and instructive text to guide students of planning and public policy decision-making. This accessible book brings data science and machine learning into the realm of public policy through a series of common and compelling “use cases,” with an emphasis on the critical role of geospatial analysis. The examples provided are practical, address important social issues, and demonstrate impact. Readers will appreciate the thoughtfulness of the prose and a narration sympathetic to the challenges of doing data science in a policy environment.   

-- Dr. Dennis Culhane, Dana and Andrew Stone Professor of Social Policy, Univ. of Pennsylvania