PhDeac
PM a mod to cement your internet status forever
- Joined
- Mar 16, 2011
- Messages
- 155,209
- Reaction score
- 22,224
It really is a silly thing to criticize people for using the available data at the time to make predictions about the election outcome, that proved to be incorrect. The data were flawed and the models interpreting those data were flawed and therefore the predictions were erroneous, also, events occurred between July and November that swung the polls (e.g., the Weiner letter). Predictions made on July based on data up to and at that point are subject to more uncertainty than predictions made in October. There are stochasticities in any system that are difficult to predict but the best predictive models incorporate the most and the relevant stochasticities and the results are expressed in terms of probabilities of occurrence not specific outcomes. The primary issue with many pre-election prediction models was that they failed to account for the high levels of dislike for each of the main candidates. We'd never had an election in the era of modern statistics, where both candidates had such high unfavorable or unlikeable ratings and most models did not even have those data as part of the prediction. That means they failed to account for an heretofore unimportant model parameter, which introduced significant unmodeled uncertainty into the predictions. The data indicated that Hillary was ahead in the polls right up to the very end but her unlike-ability, made the election outcome much more volatile to late season stochasticities, like the Weiner letter.
Well said. Also if you look at the polls from the summer through October, there was an ebb and flow. For a few weeks, Hillary would have a big lead, then Trump would close the gap. Hillary would take a big lead. Then Trump would close the gap. The polls a week before the election showed a downturn for Hillary. The election took place during a period in which Trump closed the gap.