Predictive Modeling Collection
This page collects our philosophical and theoretical work on predictive modeling in one place.
How to Specify and Evaluate Predictive Models
Under journal pre-consideration
Statistical models are rarely well contextualized and evaluating them turns into a laundry list of observations and questions. Here we suggest a structured approach inspired by the law of threes.
Philosophical Stances for Predictive Modeling
Forecasting differs by what it tries to approximate: the world around us, the data as it comes, a set of possible futures, or informed human judgment. Each orientation reflects a distinct view of what makes a forecast valid.
How do I Know If My Model Works? Hypothesis Validation Approaches
There is a tension that runs through modern epistemology: between a science of refutation and a desire for affirmation. We want hypotheses “proven”, but this is impossible. Here we delineate the approaches that can be used to get closer to proof.
Ex-post and Ex-ante Validation of Statistical Models
We summarize and give structure to validation methods in time-dependent forecasts.
Combining Judgmental and Mechanical Predictive Models: The PoluSim P Controller Logic
Adding expert judgment to models improves their accuracy. We discuss how to do this the best way. What you see here is how PoluSim, our strategic forecasting solution, works.
PoluSim Business Impact: Board Room and Operating Level Views
Our PoluSim forecasting solution is widely appreciated by customers around the world. A key reason is that we design it for total business impact instead of narrow gauges like accuracy alone.
The Horns of the Dilemma in Statistical Modeling
In statistical analyses, you cannot get everything. This framework explains the trade-offs. It is based on Runkel and McGrath’s classic framework.