Instructor Notes

Instructor background

WHile hurricane forecasting has gone a very process-based (theoretically inspired) path towards forecasting, election forecasting has developed very differently. The goal of this unit is to expose students to data-driven approaches to forecasting, compare that numerical weather prediction, and think about how these two approaches apply to ecology.

Discussion Questions

What is the data that 538 uses for its election forecasts? What sources of uncertainty does this data have?

  • Polling data conducted by a variety of independent polling entities with their own methodologies, sampling intensities, etc. Sample sizes may be small. Who is contacted may be biased. Answers may not be truthful.

How does 538’s model attempt to compensate for this uncertainty?

  • Uses many polls, weighted by their accuracy in the past
  • Runs probabilistic simulations to capture the range of possibilities

What do the simulations tell us?

  • How likely certain outcomes are given the uncertainty in the data

How does this approach compare to that used for hurricane forecasting? What are the similarities and differences?

  • Similarities: a main one is the use of simulations to see how sensitive results are to uncertainty in the underlying data. Both approaches do better with more data (and accurate data) collected frequently across space and time.
  • Differences: statsitical vs process-based approaches. 538 does use state-run super computers - so likely a big difference in computational requirements…

Let’s think about how these two approaches to forecasting relate to ecology. 1) Which approach do you think ecology should be pursuing to get the most accurate forecasts? 2) Which approach do you think is most likely to work in ecology?

  • Instructor notes: the goal of this question is to get students to think about how they feel about process-based vs data-driven forecasting approaches generally and in the context of ecology specifically. And to think about what these other approaches tell us about what we may or may not be able to do in ecology given our data limitations.