By 9 p.m. Central ST, we will be getting some results from the early states which will give us some idea of how the night is going with a strong emphasis on the some.  As noted in Part 1 of this series, every state treats the counting of mail-in votes.  In some states, like Missouri and Texas, mail-in votes are likely to be the first results reported.  In other states, like Michigan, those votes are likely to reported after the election day results.  And for election day results, precinct sizes (more importantly the number of voters per election judge) and other factors have historically resulted in longer lines at closing time in urban area.  When combined with the number of precincts in urban areas, in early states, rural areas are likely to report a greater share of their results in the first couple of hours.  Both of these factors distort the conclusiveness of early vote counts (which is why the best analysts start looking at what vote is still outstanding — both where that vote is and the total number of votes — in forecasting whether it is possible to call the race).    But by this time of the evening, there is some hint at the level of turnout in the areas that tend to vote Democratic and the areas that vote Republican and which way swing areas are swinging.

In turn, this information gives us some idea of the accuracy of pre-election polls.   In viewing pre-election polls, there are three things to remember.  First, in viewing them, you should focus on two things — margins and the size of the “undecided” voters.  In every poll, there will be some undecided voters (and, because voters tend not to want to waste votes, the supporters of third-party candidates should be treated as undecided as a significant share of them will move to one of the two major candidates by election day).  Because undecided voters will not split 50-50, a large pool of undecided voters makes the margin less reliable.  An eight percent lead with ten percent undecided is more likely to hold than a twelve percent lead with twenty percent undecided.  On the other hand, it is likely that both candidates will pick up some undecideds.  So both candidates are likely to end up with something higher than their last poll number.  Second, in looking at the margin, every poll has a margin of error (typically between three percent and four percent).  That margin of error applies to each candidate.   Which means, in theory, that even a well-constructed poll can be off on the margin by six or seven percent..  Part of the error is that every pollster has their model on who is likely to vote and how to weight responders to overcome response bias.  Some years the actual pool of voters is bluer than the model shows and in other years the actual pool of voters is redder than the model shows.  Finally, a poll is a snapshot in time.  Events occurring after the poll is taken will move a small percentage of voters (both undecided voters and voters who were tentatively supporting a candidate).  In short, it is highly probable that the polls will be off by some margin.  And while the direction and size of the error will not be uniform nationally, the early returns can give an idea of the direction and size of the error.

As things stand four days out, the polls seem to be indicating a red ripple which will switch a narrow Democratic majority in the House to a narrow Republican majority in the House.  The Senate could go either way and the hold of state offices could swing either way as well. Continue Reading...