In the following post, D’Amore-McKim School of Business Distinguished Professor of Supply Chain Management Nada Sanders, an expert in business forecasting and risk management, explains the importance of human judgment in the forecasting process and how it impacted the 2016 presidential election polls.
The pollsters, pundits, and churning of complex algorithms completely missed the forecast of the presidential election. While they are surprised, I am not. As an expert on forecasting, I knew from the beginning where the problem was.
What happened? Blind reliance on algorithms.
Analytical forecasts are based on a “system” of historical data and relationships between variables. They assume these relationships will continue into the future. However, if there is a shift in the “system,” these relationships and data points become invalid. This is true of every forecast – from corporate forecasting of markets to climate change to the election. It is up to experts to recognize when such a shift occurs and reassess the algorithm. This did not happen.
What should have been done? The pollsters, pundits, and analysts all worked off of sample polling and data as in the past. Even when indicators showed the system had shifted (nontraditional candidate, many supporters who had not voted in the past, unprecedented tenor and discourse) the pollsters still relied on the same processes.
Instead they needed to recognize that the system was different and apply different sampling methods and assumptions. They needed to consider different predictors and look for analogous relationships from history, other countries, or scenarios. Instead they blindly relied on analytics.
What is the lesson? Huge errors can occur from unchecked reliance on analytics. Yes, there is still an important role for expert judgment.