Posted by Stephen Beban on November 08, 2016
The patterns that are emerging based on the reported votes from states that have been the earliest to close their polls, relative to those states’ partisanship, may offer early indications of where the popular vote will wind up nationally.
Partisanship (driven by the ideological divide between the parties and the country at large) is such a strong motivator that it is a highly predictive indicator; states that have leaned towards one party or the other in the past 4 elections are highly likely to stick with the same one this year. However, we may find that accounting for trends in state partisanship contains more predictive power about the final result, as we experimentally hope to show.
FairVote has measured the partisanship of states cycle-to-cycle – you can see the ratings for all states from 2000-2012 shown in this post. Our Partisanship Index is calculated by comparing the Democratic and Republican vote share in a state to the national vote. The greater a state voted Democratic relative to the nation as a whole, the higher the Democratic partisanship score we derive. For example, Florida voted for Obama in 2012, but by a smaller margin than he earned nationally – this lean towards the Republicans is recorded as a Democratic partisanship of 48.5%.
In addition to that, we have created a simple formula to project how the states are expected to line up in the vote based on trends in partisanship, weighted by recency on the assumption that more recent changes are more indicative. We take the average change in partisanship each cycle (with half the weight for the 2000-2004 shift, and double the weight for the 2008-2012 shift), and apply this change to the 2012 partisanship to give a projected partisanship for 2016 results. We haven’t adjusted these projections based on home-state bonuses of the VP candidates.
By looking at the votes for Hillary Clinton and Donald Trump in states whose polls have closed, and seeing how their reported votes deviates from the projected underlying partisanship, we can extrapolate what the national popular vote will ultimately wind up as. For example, if one candidate is consistently overperforming the expected partisanship by a point or two, that candidate will likely get 51 or 52% of the national two-party vote. More specifically, we average the difference between the actual reported vote share and the state partisanship, weighted by each state’s population. We assume that there has been enough time for the states listed here to have data that is representative enough of the state as a whole to therefore allow reasonable inferences.
Below, we have included a table that shows the states whose polls closed before 8:59pm, along with the aforementioned projected 2016 partisanship; the reported Democratic share of the vote; the deviation; and the projected national Democratic Two-Party Popular Vote share based on it. Will the trends in partisanship hold enough predictive power for this projection to be sound (with those wanting it to be wrong indulging in panglossian hopes) or have Trump and Hillary altered the map enough to defy our expectations? We’ll soon find out.