As many of you know, I like to forecast. Many of the forecasts that I have done are trial and error. You come up with a hypothesis, test it, see if it holds, and are either happy with joy for being right or depressed that you were totally wrong. Last election, I tried predicting legislative elections using top-of-the-ticket support, party composition, and money spent by candidates. Guess what, I was WAY off on some seats (though I think the model can be saved using different approaches).
This year, I created a model to forecast the upcoming presidential election. My model was based on two things…pre-Election Day vote totals to determine partisan enthusiasm for presidential candidates (since most people are voting for president first and foremost) and overall voter registration differences between 2012 and 2016 to determine overall voter enthusiasm, which would impact voter turnout.
Before I get into the “why I lied about it” part of the title, I first need to explain my model and show my results. So let’s start! Please visit my new page, Elections Florida to see up to date stats on Florida politics.
Prediction of the Election
Are voters in a state favoring one party over another? This is what we are trying to determine here. Luckily in the State of Florida, voters register by political parties, thus each voter has a partisan identification. However, simply looking at party registration does not tell us if the state is favoring one party over another.
So, how do we determine in a particular election year that Florida is actually making a shift? Since everyone is registered by party, we can see which voters are turning out to vote. Since Florida does have a large early vote and vote-by-mail turnout rate, we can compare the composition of those are registered and and have actually voted. Therefore, to determine partisan enthusiasm, we must first see how each major party performed by starting with the following:
% Major Party Turnout – % Registered Major Party = Major Party Partisan Enthusiasm
In Florida, the percentage of registered Democrats was 37.92%. However, the composition of the pre-Election Day vote that was Democratic was 40.29%. This is a difference of 2.37% for Democrats. As for the Republicans, their percentage of registered voters was 35.37%. As for the pre-Election Day composition, Republicans were 38.23%. This gives us a difference of 2.85%. Since we are looking at major party swing, NPA/Other voters are not examined (and in many case simply mirror partisan voters, according to Shaw (in Kenneth F. Warren’s Encyclopedia of U.S. campaigns, elections, and electoral behavior: A-M, Volume 1; 2008). In order to figure out the the actual swing of the state, we must simply subtract the two percentages, so:
Democratic Enthusiasm – Republican Enthusiasm = Major Party Swing
Using this equation, in the State of Florida, we see a .48% swing to the Republicans. This indicates a favorable election result for Republicans, if only slightly. To determine how this will impact the 2016 Election, we simply do the following:
2012 Obama Percentage + Major Party Swing = 2016 Clinton Party Percentage
The above can give us an idea of how voter enthusiasm looks in comparison to the previous election. In 2012, President Obama received 50.44% of the major party vote. If we subtract the .48% rate for enthusiasm for the Republicans, we see what the projection shows Clinton with 49.96% of the major party vote, which shows a very slight loss for Clinton. When comparing this to the actual election results, with Clinton only receiving 49.34%, the model is only off by .62%.
Predicting Voter Turnout
If people are enthused about an election, they are more likely to participate, and the opposite is true if they do not. Not only is political participation measured by turning out to vote, but participation can be measured by voter registration. If voter registration numbers increase, so does enthusiasm. However, we need to determine what the turnout rate will be in the first place.
Florida usually has a higher turnout rate, with the turnout rate being exactly 71% in 2012.
In order to determine if turnout will increase or decrease, I compare the increase in voter registration total between book closing totals in the primary and general elections for both 2012 and 2016. With that increase being calculated, I simply do the following:
(2016 Registered Voter Increase % – 2012 Registered Voter Increase %)+2012 Overall Turnout = 2016 Projected Turnout.
In 2012, voter registration only increased by 1.05%. In 2016, that number jumped to 3.98%. Therefore, with a increased difference of 2.93%, we see that voter turnout was probably going to increase. (Note: I will be using this hypothesis to see if holds up historically). With that number, I simply add the 2.93% to the 71% from 2012, resulting in 73.93% turnout in Florida. As of this moment, turnout in Florida is at 73% even, a .93% difference at this time.
While I plan on testing these models in other elections (if I can find the data), it seems to have held during this cycle. The model is very easy, and uses actual hard numbers to determine the outcome of the election. So many “mainstream media” forecasting models simple aggregate survey research. But those numbers are not hard numbers. Many in academia use economic indicators, and sometimes mix it with polling. The method presented here takes a new approach, which I think will be beneficial to the field of forecasting.
Why did I lie about my model?
Simply, I couldn’t live with the result. I couldn’t live with the fact of a Trump presidency. And with that, I broke a major rule of political science research…letting your emotions take over and not listening to the data.
In all fairness, I did not make any scientific prediction of Florida. I just had a “gut feeling” and based my numbers off of those feelings. However, whenever I looked at my model, I never saw Clinton take the lead. As each day passed, it become disheartening. Therefore, I switched tactics.
If you remember in my first few articles about pre-Election Day polling in Florida, I constantly mentioned that the swing was benefiting Republicans. Even in my last article (Oct 29th), I said that the Republicans had a 3.02% swing. Because of this, I started making excuses…party ID is not an indication of vote choice (which is still true), Democrats are increasing their number in early vote totals (which was true). Eventually, I would tweet that Democrats had overtaken Republicans, and treated this aggregate total like a victory for Democrats. However, even when the Democrats took the lead, that swing number still favored the Republicans. I was fully in my “it can’t be true” mode, and refused to listen to my own numbers.
In the end, the numbers were correct.
Why mention this?
Over the last few days, I have heard a number of people in the media talk about how the numbers were “wrong”. And in many cases, I am sure they were. But if you create a theory based on hard numbers, and know how those numbers might cause shifts in voting behavior and voter turnout, the numbers can be spot on! In this case, my numbers were spot on….I just refused to except the results.
I’ve learned my lesson…I won’t do that again!
Working the Equations above
Major party swing (Dem)+(Rep)
(37.92-40.29) + (35.37-38.23) = -.48
2016 Clinton Party Percentage
50.44+ (-.48) = 49.96% (actual 49.34%)
(3.98 – 1.05) + 71.0 = 73.95% (actual 73% even)