Over the last few weeks, I have been putting together a data set for precinct-level results in Florida. I am still waiting for some final numbers, but I have enough to at least do a preliminary examination of how Florida voted. With the numbers I have, which is over 5,500 precincts (almost all), there is a clear pattern of party registration in relation to vote choice.
Overall, we see that Democratic-leaning precincts were more likely to vote for Trump, while Republican leaning precincts did not break for Clinton. The way that I tested this was by comparing major party voter registration to differences in major party vote total, with positive numbers representing Democrats and negatives representing Republicans. For example, if a precinct had 40% registered Democrats and the Republicans had 30% in that precinct, the number would be +10%. The same applied to the Trump-Hillary vote. The reason that I did major party vote was the lessen the skew that the independent voters might bring. So, in a precinct like Miami-Dade 366, where over 50% of the voters are NPA or minor party, the partisan gap shows only 2.61% difference. Basically, this assumes that NPA and minor party registrants are split in a similar way that a precinct does in major party registration (thus meaning that Democrats would be slightly favored in Miami-Dade 366). I could control for NPAs and minor party voters, but decided not to for right now. Also, precincts with under 25 voters were excluded because they could skew the results as well, since this examination is based on percentages, not raw vote totals. Even with these precincts excluded, there were 5,390 observations, which means only 2.53% of precincts were excluded from the analysis.
With the data being determined, I then ran a simple linear regression to see if the voting patterns were similar. Overall, the model fit (R-squared of .758), and major party registration is still a good indicator of vote choice throughout the state. Still, registered Republicans were much more rigid in voting for Trump than Democrats were voting for Clinton. The best way to see this relationship is by looking at it visually. The scatter plot shows the the results. As we can see, there are quite a few Democratic precincts that voted for Trump, and in some cases quite heavily. In the case of the Republican precincts, very few deviated from the pattern. The scatter plot also shows us that there is heteroskedasticity in the direction of the Trump vote. We see that Trump had broad support among precincts, something that Clinton could not replicate. But what is most important on this scatter plot are the outliers. In the case of Clinton, we hardly see any precincts that are favoring Republicans going her way, but we do see the opposite with Trump. The fact that the top part of the scatter plot is fairly defined but the bottom isn’t shows that Clinton rarely expand outside of her core votes.
I also examined the data on a county by county basis. This analysis, as well as the numbers listed above, exclude five counties. Seminole. Glades and Polk still do not have their early vote and vote by mail numbers distributed by precinct. Jefferson and Desoto still do not have precinct-level results. Therefore, these counties have been excluded. I will conduct another test when all of the numbers are in.
When looking at the counties (as well as the state in general), we are trying to see if partisan identification equates to vote choice. Basically, this means that if a precinct has a 50%-50% split in major party registration, then the election results are likely to be a 50%-50% split as well. The higher to coefficient, the more likely this is the case*. This map shows the results by county. But before getting into the results, it needs to be noted that counties which lacked statistical significance have been excluded from the map.
When we look at the county level, the data tells a different story. In Miami-Dade, Charlotte, and Duval County, registration is a strong predictor of vote choice. This means that Republican precincts are Republican, and Democratic precincts are Democratic. And, as expected, we see the reverse trend in North Florida and the panhandle. In Liberty County, party registration is a poor indicator of vote choice.
The counties that are the most interesting, however, are Pasco and Citrus. As Kartik Krishnaiyer stated on The Florida Squeeze, there is an issue with Pasco. And when we look at vote choice, we see that to be the case. In Pasco County, party registration is becoming less of a predictor of vote choice. In a nutshell, those who are registered as Democrats are voting Republican (since the electoral swing was to Trump). The same holds true in Citrus County, though that might be because Citrus represents North Florida more than Central Florida. The other two counties that Krishnaiyer mentions in his article are Volusia and Brevard. In Brevard, vote choice and party registration still has a strong relationship. Basically, there are not as many Democrats in Brevard, which is why Democrats perform poorly there. As for Volusia, the partisan trend is weakening. While not as bad as Pasco (with a coefficient of .523) Volusia is starting to go the way of Pasco (with Volusia now at .695).
Another county of interest is Alachua County, with a coefficient of .594. In this case, there might be rural areas of the county that still have a lot of registered Democrats who voted for Trump. Again, a characteristic of North Florida voting patterns. Basically, it seems in Alachua County that there is a “Gainesville vs. The Rest of the County” dynamic happening (which might explain Rod Smith’s loss), with traditional conservative Democrats now voting strongly Republican. Still, it is unlikely that Alachua County would see a Volusia type swing, as Alachua isn’t experiencing exurb expansion like Volusia. As long as Gainesville remains a liberal university town with a large chunk of the vote, it should remain Democratic.
In this article, we were able to expand our understand partisanship in Florida. While registered Republicans mainly stood by their candidate, registered Democrats did not. We also see that party registration and vote choice are not always the same. What we should take away from this is that political party registration does have some utility when it comes to understanding vote choice. However, it does depend on geographical location, as not all voters are not the same. The main takeaway…you can’t treat Florida as a monolithic political state, it must be dissected county by county.
Further examination needs to be done on this subject. Again, why are voters in North Florida registered Democrats when they are now voting Republican? Does it have to do with tradition? Does it have to do with the costs (aka, time) associated with changing voter registration? We also need to understand the casual direction of voter registration. Do voters register with a political party to vote in a primary, or is it because they like the top-of-the-ticket candidate? Do candidates cause voters to register with a party or do issues? Again, we have very little understanding of this issue, and expansion is needed.
*Coefficient: With each 1% increase in Democratic registration advantage, the coefficient shows how much of a shift toward Clinton. So, if the coefficient is .800, that means that for every 1% of Democratic registration advantage, that Clinton’s advantage increases by .8%. A coefficient of 1.00 would be high for Clinton with .00 being high for Trump.
(Note: I have run a model with independents, but the coefficients did not change drastically. I will provide more when when all precincts are finalized.)