Well there are two points here that should be focused on I think. The first is, as you rightly note, deduction; and this is really the most important and only point worth taking home from what I have above. Cognitive science, as an information processing discipline, can facilitate deductive predictions from systems of interacting mechanistic components. They’re not clear deductions in the sense that mathematical models can be, but they’re probabilistic deductions. This is very juxtaposed to the statistical or evolutionary approaches that people are advocating neither of which provide mechanistic deductive explanations. Statistical arguments are largely descriptions of the data itself, so it fails as a generative explanation or deduction. Similarly, evolutionary “explanations” are largely reframing of narratives or data on “traits” and so also fail as they cannot offer predictions on how their “traits” come to be, much less change, in any way that would facilitate a deductive prediction (one of the key reasons why Bunge considered both evo. psych and cultural evolution to be pseudo-sciences).
While Silver (and others) do use “probabilistic models” their models are largely based on the assumption of linearity and static data manipulation functions. These are simply not compatible with the richness of human psychology and cultural changes.
The less important point is about the extent to which this was inevitable (the second point, which I find interesting, but isn’t really the point of what I was presenting). I would say that the outcome was not a sure bet. When we look at the greatest thermometer of them all (currency exchange rates) we see that the global markets were bullish on the Euro going into the election with the only adjustment coming with May’s Florence speech.
In fact, the Euro USD exchange, leading into the vote, had a confirmatory bullish engulfing signal in the two trading days prior to the vote, and at the current levels (being 1.18, well above the “priced” in assumption level of 1.11 post Brexit and the fallout support level of 1.05).
https://www.investing.com/currencies/eur-usd-candlestick
Largely, the markets were not seeming to respond to the idea of independence at all. This seems to bolster the fact that collections of the polls (polls of polls as Silver calls them) still seemed to be against the independence vote).
Reliance on polls in complex linear models is problematic. Silver knows this now. While his earlier work is nothing short of inspiring, it falls into the adage in modelling and simulation “all models are wrong, some models are useful”. The fact is that the underlying assumptions of his models were no longer valid, and yet they were applied anyway. Let’s not forget how many news organisations basically called the vote for Clinton well before the election. I’ll link my favourite here for nostalgic purposes:
http://www.slate.com/articles/news_and_politics/politics/2016/10/donald_trump_could_have_been_president_don_t_forget_it.html
And I think that many people are doing Monday-morning quarterbacking on that. The idea that Comey killed Clinton’s chances seems to be more of a scape-goat than anything else in my opinion, but it helps make sense of it for people like Silver and Clinton.
http://thehill.com/homenews/campaign/309871-nate-silver-clinton-almost-certainly-wouldve-won-if-election-were-before
When we look at Silver’s final projection it was wide of the mark in the majority of the key swing states of FL, PA, NC, WI, & MI. The only one’s he got right were CO, NV, and VA.
Silver’s prediction:
http://www.independent.co.uk/news/world/americas/us-elections/us-election-2016-nate-silver-538-chances-of-hillary-clinton-donald-trump-winning-a7404931.html
Final projection:
http://www.politico.com/mapdata-2016/2016-election/results/map/president/
This makes sense given the mathematical bases for his projections. If you assume linear change mechanisms, then plugging in x+1 gets you more of the same. Its the same with system dynamics models in the modelling and simulation literature. As useful as they may be for interpretation, they rarely hold enough explanatory power to be predictive.
