Thursday, February 8, 2018

Accurate v Fairness

Do you use a lot of data in your project? Most everyone does.
Things to keep in mind:
  • All data has some bias(es) built in
  • Because it's objectively accurate doesn't make it acceptably fair
Bias in data
There are library shelves full (or virtually full at library websites) of papers explaining biases that get built into data, even if unwittingly. Kahneman and Tversky probably have some of the best reading on this subject. Factors:
  • (Planning) Experiment design flaws
  • (Collection) Measurement and counting method error sources
  • (Collection) Expectations influences on observations
  • (Analysis) Statistical errors and data selection errors (bias)
  • (Interpretation) Framing to make a point
Accuracy v Fairness
You probably ran into this in school: grading on the curve. Grades were objectively one thing, but in fairness--we are told--grades posted were something else.

What's fair isn't what's accurate, necessarily
Fairness is about impact--what the downstream causation and consequences are for having applied data results to an issue. When people are in the loop, all the harder: race, class, win-lose, antimosities.

In the project office, the first instinct is (or should be) objectivity.

Risk management
But then objectivity meets management goals (*), and before you know it, we're back to "grading on the curve". If "fair" and "accurate" are not the same, then there is a gap. And, what is the nature of this gap? Risk! And, who is the risk manager? The PMO, naturally

(*) Political, humanitarian, economic, social justice, or competitive factors enter the picture.

Read in the library at Square Peg Consulting about these books I've written
Buy them at any online book retailer!
Read my contribution to the Flashblog