Thursday, August 25, 2016

The Ipsos polling flood

The polling company Ipsos, affiliated with the newsgroup Reuters, released information late yesterday on 43 state polls on the Clinton-Trump race. This is by far the largest set of polls I've dealt with in a single day this year, but when stuff gets serious in October, this will be something close to an average day's work. Here are my impressions.

1. Okay, finally a challenge. My polling system does better when there are a lot of polls to work with, but this many polls means maybe an hour of data entry. It kind of amazes how much I don't mind this. At my age, I always look at how long a video is to decide if I want to watch it or not, but with a flurry of data, I don't even blink. When it's done I will know more than I did when I started. That's all the incentive I need.

2. More data is almost always better. There are now only 6 of 51 races that I have to extrapolate from the 2012 race. Three of the unknowns are dark blue races in District of Columbia, Hawaii and Rhode Island, while the others are in deep red Wyoming, South Dakota and North Dakota. These aren't likely to be close and they represent a grand total of 20 electoral colleges. My model always craves fresh data, but these missing contests are small morsels and I don't expect surprises.

3. Almost always better is NOT always better. One problem I have is that Ipsos decided on very small sample sizes in the smaller states. For example, Ipsos decided a sample of 115 New Mexico voters was sufficient. Clinton's lead here was only 3 points, which my system turns into a 63.7% Confidence of Victory value. In contrast, PPP also polled New Mexico, using 1,103 voters, showing a nine point lead, which turns into 99.98% CoV. These are the only two polls that exist, so right now the state looks like a battleground. One more poll in the next few days can either keep it as a battleground state or move it back to Solid Clinton.

4. Ipsos did produce some surprises. As I wrote in the paragraph above, Ipsos thinks New Mexico is close. Ipsos also thinks Pennsylvania is close, but this is a heavily polled state and one poll isn't going to make that much difference. This is what I like about using the median. Almost every surprise Ipsos produced says Trump is doing better than other polling says, including leading in Maine. The one unusual Ipsos result that favors Clinton is a 5 point lead in Missouri. Every other poll in Missouri this month has Trump leading and I doubt the Ipsos result is the start of a trend.

5. This data will be a factor for several weeks to come. Last week's polling results had Trump's chances rounding up to 0.01%, not even one chance in ten thousand. More polls in the next three days could change this, but right now Trump is back up to 0.08%, or about eight chances in ten thousand. Obviously, he has room to improve, but my best guess is that Oregon moving to Solid Clinton will be the death knell for his chances. Let me say that I'm guessing, but I'm guessing based on solid prior knowledge.

Okay, this is a whole lot of words about very few numbers. When I write again about the Senate races on Saturday and the electoral college contest on Sunday, expect a lot more numbers and a lot less words.


  1. Wow! What an eye opener this post has been for me. Very much appreciated, bookmarked, I can’t wait for more!
    Parafoil floodgate manufacturers

  2. I like your way of writing! Keep sharing here! I like your support! useful site

  3. Thanks a lot for sharing! Your post catches up the reader’s attention and certainly best assumptions put here.