Data for Good! Predictive vs. Causal Analysis

Your voice has impact.

Predictive vs. Causal Analysis – Unique Data Science Needs in the Social Sector

Great meetup with the great crew from Data for Good: (Data Scientist and Devs doing GOOD) at headquarters, Toronto.

Heather Krause, PstatDatassist, Founder and CEO of Datassist gave a most excellent talk on some worthy methods and pitfalls to look out for when building your models and algorithms.

Heather talked about her experience using Data Science around the world in often challenging environments for unique impact projects, like one to help a nations rural milk production in a part of the world where food issues matter in lives.

Topics under discussion were predictive vs causal analysis when building your models, when to use each, as well as some of the other snares to look out for when collecting and analyzing data to assess real world phenomena.

Does x=y? or does y = x.

Or , does x=y, because x=z=y  (z being the mediator variable)

Or, does x only look like y, because z = x, y (z being the confounding variable)

Heather also raised poignant philosophic questions (continued with a good remark from an audience member), of bad models (but not necessarily wrong) models, meaning that while the numbers are working well within a chosen analysis, its application to the real world could have bad, if not horribly wrong consequences.

The solution? Apply deeper analysis, critical thinking, and context to ones data, perhaps also with a touch of humanism and morality; the art to ones science.

  • Stephane

(Happy Friday!)

June 9, 2017