We had another batch of primary elections yesterday, the March 8 primary, so it is time to see how well the pollsters interpreted the entrails of goats or black roosters or whatever it is the use to come up with their predictions. Actually, at least reading entrails would have something like a coherent methodology behind it.
There appears to be no polling available for Hawaii. That, ladies and gentlemen, is a blessing.
B****, please! In a season of craptacular polling this is notable for being craptacular. The poll understates Cruz by 26 points. How do you even do that? On the other hand it got Rubio right and was within MOE of Kasich. So if predicting third and fourth place is your goal WIN. Otherwise you knew less about the Idaho race after reading this poll than you did before you started.
The polling trend:
The fact that the RCP average was very close on Trump, Kasich, and Rubio but understated Ted Cruz’s total by 6 points and puts him a weak third rather than in second. The Trump number, however, is a statistical fluke based on averaging a lot of bad numbers. The inability to pick the order of the finishers means there was a lot FAIL active here.
Most of the polls overstated Trump outside the MOE. The average was rescued by the ARG poll that understated Trump outside the MOE. That poll had Kasich winning by 2 points. The Monmouth poll stands out of the crowd in not only getting the order of the finishers correct but calling the numbers very, very close. The CBS/YouGov poll is creditable on the first two finishers but misses Kasich and Rubio by a wide margin.
It is a sign of the times that I when I look at this dog’s breakfast and say, “eh, not so bad.”
Only one poll here and it gives the Idaho poll a run for its money in terms of FAIL. In understates Trump by 6. It misses Cruz by 19. It misses Rubio by 10, or I could say by 200%. This is not a poll if the word is to continue to have meaning.
After doing a couple of these this is my conclusion about polls and poll averages.
The publicly available polls are not able, with any frequency, to forecast the order of finish or the relative magnitude of the final vote tally. This renders them useless for all intents an purposes. I can do the same thing these polls do from my desk. Trump is going to win. He’ll poll about 35%. Check me on this as we go forward. Now when we come to the other finishers or have to give them vote percentages my methodology becomes less accurate, but not necessarily any less accurate than the method used by the pollsters.
As to poll averages, they are bullsh**. All averaging bad numbers give you is an average of bad numbers. You can’t correct lots of wrong data by averaging it. The extent to which the averages are right is reliant upon outlier polls to tug the number away from the median and toward the mean. They remind me of an experience I had in high school chemistry. There was a particularly difficult problem that had stumped the whole class. Except me. So I went to the board to show the morons how to do the problem and left them agog with my brilliance. Actually, I recall the teacher saying, “you can’t solve the equation like that.” As it turned out the problem involved Oxygen (atomic number 8) and Sulfur (atomic number 16). Because their atomic numbers were in a 1:2 ratio, my solution worked. Otherwise, it would have been obvious gibberish.
And gibberish, gentle readers, is what we are looking at here.