The PiRates here at the PiRate Ratings do not believe in editorializing or expressing political opinions. When we publicize political events, this is an extension of our application of statistical calculation.
If you ask us which candidates we ally with, our answer is “none.” Our alliance is trying to be as accurate as possible and to have the best predictions of the outcomes of the races in question. We do not want to influence any voter in any way. That said, here is our mathematical look at the first in the nation Democrat’s Iowa Caucus.
First, for those of you that don’t closely follow the Iowa Caucuses, it is an interestingly unique form of choosing candidate preference. In 2020, there are 1,681 caucus sites. Unlike normal polling precincts, these sites are more localized. Some of the sites are in peoples’ homes. In other areas, libraries, churches, schools, social halls, and community centers may host a caucus site.
Caucus goers sign in and usually have the opportunity for refreshments, much like a school PTA meeting. Representatives from within the caucus footprint for each candidate (sometimes not all candidates have a representative at a caucus site) speak briefly about their candidate.
After speaking, the caucus-goers are then directed to a certain location to show their support for their chosen candidate. For example, at a West Des Moines school cafeteria, the candidates may be divided into each of the four corners, the stage, the tray return area, and in the middle of the room. Let’s say the candidate you support has a sign above a table in the Northwest corner of the Cafeteria. When the time comes, you simply walk over to that table and stand or take a seat in a chair.
After the time period given to make your first choice, multiple officials (of different candidates) come around to count and verify the number of caucus-goers that chose each candidate.
That’s not the end of the festivities. After this first go around, there is a period of time where everybody can then change and choose a different candidate. At this time, a lot of lobbying is done by supporters of other candidates trying to get people to change their minds. This is not an exercise in futility, as many people do change their support.
In order to receive any delegates, a candidate must receive 15% of the count in the location in order to be considered “viable.” If a candidate does not reach 15%, then his/her supporters can then choose to caucus with a candidate that has already reached 15%, or they can simply leave the site and not be counted at all. The lobbying by the supporters of the viable candidates get to play like they are back room brokers. It is an interesting real life game.
Now that you know a little about the process, let us tell you a few facts that we use to predict the outcome. First, past performances in prior years offer minimal bias for this year. In other words, just because something happened in 1972, or 2004, or 2016, it does not hold as much weight as the somethings happening in 2019 and 2020. Past situations in Iowa marginally affect the polling bias we apply to the raw numbers.
To be more specific, the Iowa Caucuses have been a little biased toward the more vociferous voter. In many cases, the elderly, overly shy, and others did not necessarily attend. Bitter cold and inclement weather also has kept people away. This isn’t like an election day in another state, where the polls are open from 7AM to 8PM. They begin at precise times, and you must be there or not participate. In 2020, there has been an added element to allow people unable to attend for valid reasons (like being old like our Captain) to participate electronically via computer. This could slightly affect the outcome, but the prior bias is still there.
How do we predict the outcome of the Iowa Caucus? By no means do the PiRates conduct their own polling. We’re too small to poll our own local council race. We rely on multiple polls by polling companies that have consistency. Notice, we did not say accuracy; we said consistency. If the John Smith National Poll chose Candidate X to beat Y and Z with a predicted percentage of 45 to 35 to 20, and the actual vote count was 42 to 34 to 24, this creates a bias. In this case, the poll bias slightly favored X and discounted Z. If Smith polls consistently overstates the favorite and understates the underdog, we can see the bias. We call that a positive bias. FWIW, The PiRate Football and Basketball Ratings have historically consistently had a small negative bias–we give underdogs a little better credit than we give favorites. Because in sports the underdog covers the spread more than the favorite, the PiRate Ratings have numerous times finished at the top versus the spread.
What we do is to find the most consistent polls. If they are off by a percentage with a low standard deviation, they are consistent. All we have to do is adjust their polling data to erase their bias. So, if Smith consistently has a bias of 1 percent in favor of the third place candidate at the expense of the fourth place candidate, we can factor that into a more accurate prediction.
We aim to find multiple polling sites so that we can form raw percentages every week during the election cycle. The most important cycle to us is the one that encompasses the polls taken in the final month of the race. It is during this time where large numbers of undecided voters make their decision.
What we end up with is a 30-day Moving Average. We can perform simple linear regression and to make it easier simply draw a straight line that best represents the candidate’s support.
For instance, let’s say that George Washington, Thomas Jefferson, Abraham Lincoln, and Theodore Roosevelt jump off the rocks at Mount Rushmore and come back to life. They all decide to run for President in the Granite Party. Polls on January 1 after smoothing for bias show Washington with 29, Jefferson with 27, Lincoln with 25, and Roosevelt with 19. On January 8, it’s Washington 28, Jefferson 27, Lincoln 24, and Roosevelt 21. On January 15, it is Washington 28, Jefferson 26, Lincoln, 23, and Roosevelt 23. On January 22, it is Washington 29, Roosevelt 25, Jefferson 24, and Lincoln 22. On January 29, it is Washington 30, Roosevelt 28, Jefferson 23, and Lincoln 19.
Washington has led the polling for the entire month, but Roosevelt would be our pick to win that poll that mattered on February 3. Looking at his last 30 days in the polls, he has risen 22%, while Washington has risen 3.4%. Between that January 29 poll and Caucus night, it’s not just five days later. The poll released on 1/29 was calculated from data obtained between January 22 and January 26. In actuality, the poll is eight to 12 days old. Since Teddy has been steadily gaining the entire last 30 days of the race, he will win the night with 31 to 32 percent. Washington will be just behind at 30-31 percent. Poor Abe might have to worry about viability in a lot of precincts, and he might perform worse than his polling indicates, when his supporters choose another candidate or go home in precincts where he polls at 10-14%.
Now that you’ve seen how we work, here is how we predict the Iowa Caucus results will finish in statewide percentage in six nights.
Candidate |
% |
Bernie Sanders |
28 |
Joe Biden |
20 |
Pete Buttigieg |
16 |
Amy Klobuchar |
13 |
Elizabeth Warren |
12 |
All Others |
11 |
The Pi-Rate Ratings 