The post truth world

Faith in experts

We haven't done fabulously well, have they.

• Gordon Brown thought he’d abolished boom and bust in 2005.
• Financial crisis
  • credit rating agencies
  • risk models failed
• Democratic exercises - count the number of experts who got this wrong:
  • polling errors
  • economic projections

Modes of prediction

• Physical science model:
  • Carefully constrained conditions for observation: “laboratory conditions”
    • Extraneous noise removed from the system
    • Theoretical assumptions and experimental conditions are absolutely rigid and by definition cannot change
    • Outcomes are tightly prescribed and you can’t test assumptions themselves the theory: in an experiment with dice you can’t throw a seven, mich less hypothesise what would happen if you did.

• • Model is to observe discrete events whose occurrence:
    • doesn’t depend on observation
    • isn’t affected by your observation
    • neither depends on nor is affected by other events in the sample: If you throw one six, that makes it no more or less likely that you'll throw another. Even though instinctively it seems like it.
    • Isn’t variable in retrospect. A six, once thrown, can’t change its mind.
  • Ordinary probabilistic “Gaussian” models work well here, but note even here the model is better at explaining observations once they've happened rather than accurately predicting how they’ll happen before they do.
    • Try to catch a cricket ball using only scientific modelling.

• Human sociological events are profoundly different character, even though the way we test them is not.
  • We do have memory of previous occurrences, and (outside laboratory conditions) it is almost impossible:
    • to avoid knowing about other relevant events in a sample
    • to avoid changing your behaviour as a result - i.e. reacting to them.
  • This changes the statistical analysis. Now the Gaussian distribution is only suitable where events fall broadly within the standard deviation (being events that conform with most people’s general expectations, and against which they are less likely to react.)
  • When events are significantly outside the standard deviation, people will react, pushing their own reaction outside the standard deviation - the result being the “long tail” phenomenon.