The Trump victory has left many of my friends reeling and in disbelief. It has also already brought out criticism of pollsters and polling data. Some sceptical ones go a step further and condemn all prediction makers, and mock machine learning and artificial intelligence. This condemnation is foolish and tantamount to throwing the baby out with the bathwater.
Prediction models turn on data, collected from the right questions being asked of the right statistical sample of people. Reductive questions generate neat data sets which then provide all the right answers.
But as experienced market research people will tell you, far fewer people, than those who enthusiastically nod and say they will purchase a new product, actually do. It is not that people are lying, it is just that human beings tend to give answers to please the asker. On politics and other emotionally charged matters, this tendency to give socially acceptable answers to minimise confrontation is especially pronounced. People also change their minds over time.
If the outcomes of any large exercise, where people actually make active choices, shock us it is worth remembering that the only truth is revealed preference i.e. what our actual choices reveal about our preferences. Human beings do not always seek to maximise utility, often preferring to use simplifying shorthand or heuristics to make decisions. The heuristics could have encoded in them experience and knowledge, as well as prejudices and received wisdom.
The concept of revealed preference is, of course, flawed too. If I pick Candidate A over Candidate B, it does not say I prefer Candidate A, merely that I prefer Candidate A to Candidate B. In the future, if Candidate A is up against Candidate C, I may pick Candidate C not because of Candidate C’s superiority over Candidate A but because my preferences are not immutable. Faced with more than two options, we have a way to simplify the choice for ourselves as well as I have written here.
It may sound nihilistic to suggest predictive modelling is not really reliable. But if we are relying on flawed and mutating preferences, and treating them as immutable truths in our analysis, how can methodologies and predictive models generate anything reliable?
It would be akin to putting lipstick on a pig. We would have used up lipstick but the pig would still be a pig.
In the last UK general elections, the Brexit campaign, and now the US general elections, predictions have failed to, er, predict anything reliable.
It is time we learnt to judge differently — by expanding our comfort zones, by listening more, by asking and seeking to understand more, by being healthily sceptical, and by bringing critical thinking lenses to all those pursuits.
For now, if your side won, good for you. The advice to try and understand the other point of view applies to you too. But if your side didn’t win, dry your eyes, dust yourself up, and go out and talk to someone who is not cohabiting your comfort zone.
The narratives we hear will have rough edges, and not the cleanliness or reductiveness of survey questions. But that texture is the stuff understanding is made of. Less data, more understanding. That is what we need.