Florida Vote Projection Model – Day 22: Race still too close to call

As one party moves up, the other party also moves up. This has been the story of the 2018 general election in Florida. Days where it seems that Democrats have made gains, Republicans offset those gains, and vice versa. One thing is for sure, a polarized political climate turns voters out. And yes, this election shows that having a base candidate does make a difference, though I would argue that Andrew Gillum wanted to be Hillary Clinton before the primary, Bernie Sanders during the primary, and Barack Obama after the primary. At least Ron DeSantis has stayed true to his character and issues, which is scary.

Yesterday, the Democrats took the lead, and they continue to have the lead today. However, this weekend might be make or break time for both parties. With a larger percentage of the electorate voting in pre-Election Day polls, there is a possibility of Election Day voting going down.

As I mentioned on Twitter, I have created a new conflated model. What does that mean? Well, let me explain both models I have for vote choice. The first model I created, and the one that has been used to this point, looks at voting behavior by voting method. Basically, it says that, for example, an independent voter in County X has a 63% chance of voting Democrat if they are an early voter. That percentage might change to 56% if they vote by mail. The reason why I did this is because Early Voting usually trends Democratic and VBM usually trends Republican. This was the original model (and I still have it available on Tableau).

The new model, called the conflated model, does not look at method of voting when determining vote choice. It simply says, for example, a NPA voter in County X has a 59% chance of voting Democratic, regardless of voting method. Basically, it assigns all voters the same value. The previous model assigned values based on method.

The reason I have made the change is because of the erratic behavior of Early Voting and VBM. Also, I realized that it might be a problem assigning current VBM, EV, and Election Day voters the values by methods of previous elections because it assumes that the current voters selected the same method as in previous elections. The new conflated model takes that bias away. The model is also more parsimonious. With that, I think the conflated model is a better predictor of voting behavior.

Turnout Summary: As of today, 4,470,548 voters have turned out, which results in a turnout rate of 33.67%. Republican turnout rate is at 39.44%, while Democratic turnout rate is at 36.16%. NPA/Minor party voter turnout is at 22.9%. The Enthusiasm Gap is currently at -3.29%, favoring Republicans.

Bottom Line – Method Model: The voting method model shows a slight improvement for Democrats. A generic Democratic statewide candidate would receive 48.52% of the vote, while a generic Republican candidate would receive 48.43%, a margin of .09%, a .05% increase from yesterday.

Bottom Line – Conflated Model: The conflated model shows that a generic Democratic candidate would receive 48.69% of the vote, while a generic Republican candidate would receive 48.45%, a .24% advantage for the Democrats.

Details: The only detail has to do with turnout. All but four counties in Florida have surpassed their 2014 Pre-Election vote turnout. The four that have not are Washington, Calhoun, Liberty, and Union. The turnout rates in North Florida and the Panhandle (minus the counties between Escambia and Walton) seem to be lacking in turnout. These small counties can decide the margins in many races. But this year, they might not.

What to watch: Will Election Day have lower turnout? With so much Pre-Election polling, I am starting to wonder if Election Day will be lackluster (which is another reason I created the conflated model). If so, what we are seeing now is a real indication of how the vote might look. However, if the turnout is really increasing, this could actually result in a vote that favors Democrats on Election Day. For example, in Lee County in 2014, only 33% of votes cast in that county happened on Election Day.Ā  But in Palm Beach County in 2014, 57% of votes cast were on Election Day. That equates to a lot more votes from Democrats than it does from Republicans. Therefore, don’t be surprised if Election Night is Democratic.

Here is a link to the recent model.


  1. Anonymous · · Reply

    Looks like Madison, DeSoto and Clay counties have yet to report. Would there be an adjustment in your model?

    Any other data missing today?


    1. My model has all the numbers from all counties. So, if you look at the map, the number in these counties are much higher than what is reported on the state site. I do my numbers through the county, not the state. I only used the state for counties that don’t have them online, like Palm Beach, Lake, and Miami-Dade. That is why my model throughout the day always skews Republican, because it is missing Miami-Dade and Palm Beach.


  2. Still thinking the final days will be telling here. Democrats need to smoke Republicans pretty good these final days in EV to be in a good position.


  3. Anonymous · · Reply

    Thanks, Dave. Wow, you’re doing a ton of work.


  4. Amazing work! Thank you. I’m curious if you could give any estimate of what the projections for a generic d and r would have been in 14 and 16 based on the voters? Either after all EV or after election day or both?


    1. I can’t because that is the actual election, and the model is based off of previous elections šŸ™‚ The only thing I could do is try to guess turnout, and that would be an absolute guess (as turnout is not predictable).


      1. Ah ok! Makes sense! Thank you šŸ™‚ Would it be fair to guess that for example Miami-Dade getting to a projected 63% generic Dem… Would actually be more favorable for dems than the 63% that voted for Hillary in 16 there? Due to her over performing the generic dem there in 16? Sorry if I’m not making sense haha.


      2. Well, remember, this is only based on the votes that have come in so far, and the model also takes into account the likelihood of a NPA voter to vote Democratic. In 2016, Hillary won 63.9% of Pre-Election Day voting. My projection has Democrats winning 62.19% of pre-Election Day vote right now, which is under-performing compared to 2016. However, it is important to remember that this is a midterm election, and GOP turnout in midterms in Miami-Dade County is usually better than in presidential years. So, Democrats are performing better compared to 2014 (but I don’t have those numbers right now).


      3. Thanks! Sorry I didn’t explain myself very well. I guess my question is more like… Clinton’s numbers wouldn’t be the baseline for generic Dem 16 in Miami-Dade county I imagine? Just as Murphy’s wouldn’t be? But rather somewhere in between? And same with 14, and years prior…in that ‘generic Dem/repub’ would be a composite of the statewide races that year rather than one specific race?


      4. Actually, the model uses 2016 because we are in a post-Trump world. So for my model, vote choice is based on 2016 results. Both the Senate and Presidential race had similar likelihoods of people voting a certain party. With that, it means that my model would probably skew more Republican (since 2016 was a Republican surprise year). So, when Democrats are leading in my model, they might actually be ahead by more. Chances are that independents didn’t swing to the GOP. However, it is all determined by party turnout.

        Of course, the model can be absolutely wrong. If it is, that means that we have mad a major shift in the electorate. And, honestly, I wouldn’t be surprised if that happened.


  5. oooo ok very interesting! And how does your model handle the 2016 baseline in areas where there was a rather large split between the presidential and senate numbers (Miami-Dade for example Clinton 63.2%, Murphy 54.6% or Osceola Clinton 60.4%, Murphy 54.6%? Or is it based on statewide numbers and not broken down into different trends in different regions from 2016? Sorry for all the pestering questions, feel free to ignore me and get back to work šŸ˜‰ lol


    1. It is done on a county by county level. As far as the vote choice numbers, it is simply based off of the presidential race. The reason why is because it was the of the ticket race, and I feel (yes, just my feeling) that it is a better representation of the electorate. So, using the presidential numbers, it skews Democratic, but the 2016 election skews Republican. So, I thought using the presidential race alone would be a good balance, as a average of the two races might have made it too Republican. So yes, there might be some issues with bias, but it is hard to try to figure out.


      1. Makes sense! Thanks so much for explaining šŸ™‚


      2. No problem. I would have made it more complex but some SoEs never got back to me. Maybe 2020.


  6. […] 2020 continues we will provide analysis and like 2018, numbers to predict and analyze electoral […]


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: