Biased algorithms

A high integrity way of thinking about our data is to always consider that it is flawed in some way by some kind of a bias. What that bias might be is usually not readily obvious. It's just good to stay skeptical since a bias almost undoubtedly lurks within.

Here's an example of a fairly subtle systematic bias

By way of background. The Emory University area in Atlanta received about 2 inches of snow overnight. It might as well have been two feet. The university is closed. The city is shut down.

Nothing stops Atlanta, except snow.

So a "heavy traffic" alert from google on my phone caught my eye.

Google's traffic algorithm uses GPS to calculate the speed of cell phones as they travel down a road. The premise is that heavy traffic is associated with slower rates of cell phone migration along a path.

That premise is the bias.

It's a very sophisticated and extremely useful app.

But it's obviously biased. The traffic in Atlanta today is not heavy at all. The traffic is very light. What little traffic is out there is just moving very, very slowly.