The NY Times compiles a list of “forecasts” of the presidential election. People I know are paying a lot of attention to the NY Times’s forecast (The Upshot) and the 538 forecast. As the reported probability of Hillary Clinton winning varies, so do their hopes and anxieties.
These forecasts are not really forecasts, however. At the very least, they are not what we usually think of when we think of forecasts. Forecasts are predictions about the future. For the most part, the forecasts that the NY Times reports are judgements about who would win if the election were today - or a “now-cast” (as the 538 people call their explicitly non-forecasting non-forecast). Claiming that these are forecasts is like estimating whether it will rain this weekend by looking out the window.
Of those reported on the NY Times site, the Huffington Post “forecast” is the only one that explicitly says it is a now-cast. And there are other forecasts on the site that are, without a doubt, actually forecasts. These are the ones that are not based on statistical models but on the subjective reports of experts (Cook’s Political Report, Rothenberg & Gonzalez, and Sabato’s Crystal Ball) or the aggregated subjective reports of gamblers (Predictwise).
Although the FiveThirtyEight polls-only model is described as being a true forecast, but comparing their now-cast to the polls-only forecast suggests that the forecast is just a smoothed out version of the now-cast. That is, to the degree that the polls-only model is actually a forecast, it is a poor one, since it varies tremendously, suggesting that it has little predictive power beyond just assuming that what is true today will be true on November 8th.
The remaining “forecasts” do not specify how they work, so it is unclear whether they are really forecasts in a meaninful sense, but given that they all (other than the PEC one that has a broken link) look a lot like now-casts.
It is instructive to compare these probable now-casts to the one true quantitative forecast, the one at Predictwise. Note how much smaller the variance is. And note too the large, consistent, and increasing gap between Clinton and Trump.