Choice, Expectations and Predictive Analytics
James March explains that making a decision involves understanding alternatives, forming expectations about what’s likely to happen, thinking about your preferences in terms of your wants, fears, hopes, dreams in relation to those expectations, and then making a choice.
That explanation really resonates. So we’re going to use it here.
There’s an assumption that choice amongst alternatives is cut and dry. It isn’t.
Choice is a form of knowledge – specifically:
- There are choices that you know you know.
- There are choices that you know you don’t know about.
- And there are choices you know you don’t know.
Choices themselves aren’t even really binary. There’s significant ambiguity as to what a choice really means. How many times have you heard a statement in the form: “I thought we agreed on x, that we also agreed on y and z.” People having varying understandings of a choice is standard.
Analytics is a bit hamstrung on what we do and do not say. For instance, we wouldn’t say that just because a Featured Content Area didn’t generate a high clickthrough, that the entire unit should be replaced by a navigation bar. We could. It’s just that you need evidence for the alternative choice.
Analysts make very factual statements about what is, and really fight the good fight on explaining why something happened.
Analytics really hasn’t turned attention to the discovery of alternative choices. I believe that analysts are really well equipped to methodologically engage in alternative discovery, if they could keep themselves from forming expectations so prematurely.
Expectations make alternative discovery look like a trivial problem.
The most obvious predictive analytical method is the plotting of a regression line and projecting it into the future.
To unpack that:
- We build a dataset containing observations about the past and past performance.
- We use a statistical or machine learning process to identify the line that best fits the past performance.
- We take that equation for the line and project it into the future.
That method of forecasting the future assumes that past performance will continue. That trends continue. That history repeats.
Forecasting is not destiny.
The bucket we call predictive analytics has many applications. In the context of supporting decisions, the field has a long way to go in terms of helping people understand their expectations in relation to choice and alternatives.
Existing theoretical frameworks assume too much, and demand too much cognition by the end user.
It shouldn’t just be waved away.
I’m Christopher Berry.
I tweet about analytics @cjpberry
I write at christopherberry.ca