The importance of data
- 1. What is the importance of data? David Henderson
- 2. It has none.
- 3. “Data is only useful if it informs decision making” - Every Carnegie Mellon University professor of public policy
- 4. There are lots of cool analytic techniques • Econometric modeling • Linear programming • Machine learning • Social network analysis • Etc.
- 5. • Selecting data points • Assigning value to data I will focus on two data prerequisites
- 7. Selecting data points is difficult in the social sector because a lot of data points are proxies Example We trackWe care about Graduation ratesEducation Job placementsEconomic independence
- 8. Tracking the wrong data points leads to the wrong decisions Example We trackWe care about Arrest ratesCrime reduction Public perceptionFeeling safe
- 9. People felt safer in higher arrest areas
- 10. Focusing on the wrong indicators can lead to negative (but rational!) decisions
- 11. Funder wanted a workforce development program to maximize job placements
- 12. Assigning value to data
- 13. Before diving into data, we need to be explicit about our own utility frameworks Likely overly simplistic
- 14. A utility framework is how we value data points, including interactions therein
- 15. The vulnerability index is a good example of a utility framework in homeless services Prioritizes those most likely to die on the street
- 16. Without establishing an organizational utility framework, everyone makes decisions according to their own values Different utility functions + same data = Different decisions
- 17. How do we model a utility framework?
- 18. Utility elicitation is common in the business world Model a decision maker’s risk reward preferences through a series of hypothetical investment scenarios
- 19. The key is to find points of indifference Housing one chronically homeless person is equivalent to placing three people into jobs Points of indifference allow us to compare unlike indicators
- 20. Developing a utility framework enforces consistent investment decisions Makes investment priorities clear Can be tested and refined through time
- 21. • Select (likely a series) of data points that approximate what we really care about • Build a model that assigns value to data points, and update it through time Before diving into data:
- 22. End