I’m surprised by the election outcome, but unsurprised by a stupendous analytics failure in predicting this result. In fairness: most election models did not certify a Trump defeat. They simply and wrongly indicated it was highly probable.
Amongt the histrionics I read this morning was the statement that “data is dead,” which is a bit much. However, as I’ve remarked before, modern analytics can focus too much on the process- and computer-driven aspects of modeling, rather than good data assessments and whether the questions being asked align with the answers we’re giving. And now we know: the questions being asking of polling data could not predict the true electoral outcome.
Time will tell, but I strongly suspect that the real election result was there all along – people just never bothered with good question and data asssessment, in part because the numbers seemed to deliver the result that everyone expected. And then the numbers took over…
Been there, done that. I’ve often worked with good people and organizations who use analytics largely to confirm what they already believe is true. I understand, but also say that’s not really enough reason for an analytics investment. Surprises make an investment really worthwhile, if we can convince people that those surprises represent reality. Convincing people of the unexpected is less about the apparatus of transformations, reductions and nominal predictions that constitute formal data analysis, and more about good data questions and data assessment. For people have to agree on and have a common understanding of the question at hand, or all the algorithms on the planet are irrelevant – even the most amazing outcomes will simply be disbelieved.
That’s irritating in a way – algorithms and processes are comfortable. But they are not dominant, and analytics assessment is often difficult and frustrating, particularly when we discover we’re not answering the question we really wanted to understand, or do not have the information on hand to do so. I still find many people want to believe that computers will somehow deliver answers without much human intervention. Well, they can’t, and they won’t. Perhaps yesterday’s result was a wake-up call for us data users- we can’t trust the numbers when we don’t know the question they answer. If so, that would be a very good thing.