This is how it works: We start with the 40,000 simulations that our election prediction runs every time it updates. If you choose the winner of a state or district, we will throw out any simulations that failed the outcome you selected and recalculate the candidates’ odds using only the remaining simulations. If you select enough improbable results, we will end up with so few simulations that we cannot get accurate results. In this case, we’ll go back to our full simulations and run a series of regressions to see what your scenario might look like if it came up more often.
Put simply, the regressions start by looking at the percentage of votes for each candidate in each simulation and seeing how the rest of the map has changed in response to big or small wins. So let’s say you picked Trump to win Texas. In some of our simulations, Trump Texas may have narrowly won and also narrowly lost some defect states. But in simulations where he won Texas by a large margin, he may have won in throwing states as well, drawing some democratic states into his column, while the opposite may be the case in simulations where he lost the state. We find out what each other state looked like in this whole series of scenarios, tracking not only whether the candidate has normally won other states, but also how much they have generally won or lost each other.
With that said, we’ll take some representative sample scenarios that include the decisions you made and use what we’ve learned from our regression analysis to adjust all 40,000 simulations and then recalculate the state and national odds of winning. Finally, we mix these adjusted simulations with the original simulations that are still valid and make a final forecast.
* Maine and Nebraska will give two ballots to the national race winner and one ballot to the winner of each congressional district.