In 2008 he missed only one. No one else came close to his accuracy
This will be old hat for many (you don’t need to read any further if it is), but hopefully it will be helpful to those who want to know more about polls and polling. For instance I have seen on line many discussions about the validity of recent Quinnipiac University state polls, and some of these discussions were not on a high knowledge plane. If you understand how to use the FiveThirtyEight site, you can answer a lot of your own questions. However, there is a lot of information on the site for those poll and statistic wonks like me that most of you aren’t interested in, so I am going to keep this at a fairly high level. Even so, it gets down in the weeds somewhat.
Note: Nate Silver does not use the national polls to calculate his current probabilities that each candidates has of winning the election. He first uses the recent state polls to calculate the probability of each candidate winning each state and then uses those figures to calculate (by running 10,000 simulated on a computer) to calculate the candidates’ chances of winning the electoral college.
If you want to know about the accuracy and bias of a particular poll, you can go to this page of the FiveThirtyEignt website – the FiveThirtyEight’s Pollster Ratings.
http://projects.fivethirtyeight.com/pollster-ratings/
(This one is pretty much for wonks so you can skip studying this one in detail if you would like, but quickly read the following narrative.) This page shows how Silver evaluates each polling organization. First of all you need to understand there is a difference between the factors of accuracy and bias in polls and Silver corrects for both before using a poll in his calculations. As an example, look for the Quinnipiac University polling organization. Notice Silver must consider their polls consistently accurate because he gives their polls a high weighting factor. That is their poll results will given more importance when combined with other polls. However, Silver has calculated that there is some bias in Quinnipiac sample which favors Republicans – see the last item on the Quinnipiac line. Silver will use that information to raise the score of Democratic candidates for Quinnipiac results before combining them with the results other polls.
Then there is the main website page entitled, “Who will win the presidency?” Note this page is constantly changing with each new poll so it could change multiple times in a single day.
This is a cool page you can check on a regular basis to see how the race is going over time. First it gives the current probabilities that Hillary and Trump have of winning the electoral college. Hillary is currently doing well, though not nearly as well as right after the convention. (Note that a 50.0% – 50.0% score would obviously be a dead heat.)
Then you will see the color coded map. Like most maps of this sort, the darker the red (Trump) or Blue (Hillary), the better the candidate is doing in that state. Very light colors indicate a small lead. If you hoover your cursor over an individual states you will see a snapshot of pertinent information for that state – the current probabilities of each candidate winning the election and the number of electoral votes. If you click on and individual state on the map, say Ohio, it will take you to the individual state page. More about this later.
Then you will see the calculated Electoral votes and popular votes percentages for each candidate; Clinton, Trump and Johnson. (I guess Silver doesn’t consider Jill Stein a significant factor in the election.)
Next you will see a graph of the probability of winning the election for each candidate over time. You can the huge bounce Hillary got and the way that the race has been gradually tightening since. (By the way, some pretty good news there, for the first time in a long time Hillary’s numbers have been improving some over the last day or so.)
Then you will see the “Who’s ahead in each important swing state and by how much” graph which gives a visual picture of who is ahead in Silver’s combined state polls and by how much. In blue font to the right for each state you will see the actual lead for the candidates while the percentage figure in back font give the affect that state can have on the election if it flips from one candidate to the other. Both of those factor into Silver’s calculations of the probabilities of candidates winning the entire election. If you click on a state’s line it will bring you to the individual state’s page in a similar fashion as clicking a stat on the map above.
There is additional information on the page you an explore if you would like.
Individual state pages
Lot of interesting stuff on these kinds of pages for individual states – here I clicked on Ohio. Much like the main page these web pages give similar information, but for the the individual states: Candidate probabilities of winning the state, the current combined poll numbers for the state (after they have been tweeked for bias and weighted for according to their accuracy value and timeliness), etc.
But further down you can see the “Latest polls” section. This is where you can see info about how particular polls are affecting Silver’s calculations for the state. You will see all of the polls which Silver is currently using in his calculations for the state in question.
For instance look at the Quinnipiac University Oho poll which many have been discussing; it is currently the first one in this section. You will note that Silver must consider it consistently accurate because he gives it a very high weight when combining with the other polls, but also note that he has corrected the poll by 2% in favor of Hillary before using it in his calculations because he knows Quinnipiac polls are usually biased in favor of Trump. So instead of using the +4 Trump lead he has reduced it to a +2 lead for calculation purposes. So he stills gives the poll a lot of weight (for instance twice as much weight as the Ipsos poll below it), but using the +2 Trump lead instead of the +4 figure. (That high weight is given because of the polling organization’s consistently accurate results and because the poll was taken recently. As a poll get older – it get less weight in the calculations until it slips of the grid completely.)
Note that the Public Policy Polling Ohio poll taken about the same time gets an even higher weight than the Quinnipiac poll. The Public Policy Ohio poll shows Hillary with an adjusted lead of +4. Silver using this weighting system to calculate the combined poll results for all of the polls shown.
If you are truly wonkish, and I’m not, you can scroll down further and delve into how Silver does his calculations – I wouldn’t suggest it.
Okay, if you have stuck it out this far, and I expect to care that much, hopeful you can now answer a lot of questions for yourself (and without relying on the sometimes tenuous opinions of others) how the election is progressing form day to day and the affect of new polls. Let’s hope that Hillary can expand her lead again so this is a landslide of historic proportions.