Polling and time to the election


Quote from: pbrower2a on November 12, 2019, 01:18:36 PM

As a rule I do not predict polling results except to expect more of the same. I am at least as much concerned with disapproval as with approval.

Now how important are leads with time? Close to Election Day, electoral leads of even 1% can give the leader nearly 2/3 of a chance of winning the state. Leads that may not look 'that bad' for the nominee behind in polling can go from troubling to ominous to politically lethal over a year even if the lead remains the same.

  I just got my hands on Nate Silver's The Signal and the Noise (why so many predictions but some don't)It relates probability well, and as I have suggested, being up 5% in a binary election a year before means little, being up 5% a month before the election is huge. It is from 2012, and it relates much other than elections (like sports, poker, and even chess). What it says of electoral leads as a campaign approaches its conclusion is telling.

On page 63, Figure 2-4 shows the probability of a Senate candidate winning (1998 to 2008) with a certain lead (1, 5, 10, and 20 points) at one year, six months, three months, one month, one week, and one day. Because statewide races for President are much like statewide races for the Senate -- with the qualification that Presidential nominees do not usually make appearances where they see themselves losing -- unless they really are losing nationwide.

Time to election  |1 point|5 points||10 points|20 points|
one day............. |...64%|....95%|.....99.7%|.99.999%|
one week........... |...60%|....89%|.......98%|...99.97%|
one month......... |...57%|....81%|.......95%|.....99.7%|
three months..... |...55%|....72%|.......87%|........98%|
six months..........|...53%|....66%|.......79%|.......93%|
one year.............|....52%|...59%|.......67%|.......81%|

(I am going to put this back in my "electoral theory" section because it will remain relevant.

So what conclusions can I draw? You might be surprised that a five-point lead one month before Election Day is no less significant than a twenty-point lead one year before the election. Thus one hears things like Democrats saying "We have a chance of winning West Virginia if everything goes right" and Republicans say that they have a chance of winning Massachusetts... yadda, yadda, yadda. Or is it "Yabba, dabba, doo!" Likewise, being one point ahead on the day before the election is worth almost as much as being five points ahead six months before the election or even ten points ahead  a year before the election.

Polling can be surprisingly stable. But a 5% lead that doesn't mean much in November 2019 (59%, which is insignificant in difference between winning and losing, as there is plenty of time to catch up and plenty of time for events to unfold) is decisive (95% chance of winning) on Election Day in 2020. Electoral results are not so random as they might seem.  

OK, we all thought in November 2019 that Donald Trump would be the Republican nominee unless the Grim Reaper took him away from us, but how many  of us could have predicted that Joe Biden would be the Presidential nominee? Many liberals were looking for grounds on which to impeach the President, but nobody predicted what those grounds would be, and what the consequences would be upon the 2020 election even if Trump got away with some impeachable behavior. Above all, who would have predicted  that this fellow

would be the big event in American politics in 2020?

OK. Politicians cannot change direction on a dime. A leader who loses does acts of unpredictable incompetence or gets tripped up in a scandal. Senators may not be as likely to get the blame for military debacles and blunders of foreign policy -- but the President does. Positive events such as military victories and improvements in economic statistics are less swift to change perceptions than are bad events.


[0] Message Index