South Texas’s transition from a deep-blue Democratic stronghold to rapidly reddening competition surprised many. Few saw the magnitude and speed of the shifts before they happened. As 2024’s election is just a few days away - many are asking if these redshifts will repeat. Further, GOP gains with Latinos nationwide can doom Democratic hopes of holding the presidency. And in Texas, they might also seriously stall any project to turn the state blue.
We decided to model the early vote in Cameron County, to see whether these trends had any chance of continuing. Polls have routinely shown race depolarization to hurt Democratic margins with Latinos. But Texas’s strong early voting systems allow most of the trends to be observed before election day.
Modeling the electorate, we estimate that Democrats hold a c. 2,000-2,500-vote lead among mail-in and early votes. Democrats still dominate mail, winning by an estimated 19-20 points. The early in-person vote is modeled between D+2 and D+3.
| results | Freq |
|---|---|
| d | 44167 |
| i | 976 |
| r | 42198 |
| Type | d | i | r | lean |
|---|---|---|---|---|
| EV | 42347 | 964 | 40989 | D+1.61 |
| 1820 | 12 | 1209 | D+20.09 |
| Return Date | Type | r | d | i | lean |
|---|---|---|---|---|---|
| Early Mail | 624 | 1044 | 4 | D+25.12 | |
| 10/21/2024 | EV | 5489 | 5470 | 99 | R+0.17 |
| 10/21/2024 | 111 | 143 | 0 | D+12.6 | |
| 10/22/2024 | EV | 4796 | 4802 | 99 | D+0.06 |
| 10/22/2024 | 92 | 112 | 1 | D+9.76 | |
| 10/23/2024 | EV | 4094 | 4268 | 79 | D+2.06 |
| 10/23/2024 | 87 | 149 | 1 | D+26.16 | |
| 10/24/2024 | EV | 3522 | 3733 | 66 | D+2.88 |
| 10/24/2024 | 51 | 83 | 1 | D+23.7 | |
| 10/25/2024 | EV | 3662 | 3788 | 83 | D+1.67 |
| 10/25/2024 | 26 | 38 | 0 | D+18.75 | |
| 10/26/2024 | EV | 2464 | 2589 | 72 | D+2.44 |
| 10/26/2024 | 8 | 18 | 0 | D+38.46 | |
| 10/27/2024 | EV | 1186 | 1287 | 36 | D+4.03 |
| 10/27/2024 | 5 | 4 | 0 | R+11.11 | |
| 10/28/2024 | EV | 3054 | 3108 | 82 | D+0.86 |
| 10/28/2024 | 79 | 93 | 1 | D+8.09 | |
| 10/29/2024 | EV | 2779 | 2882 | 63 | D+1.8 |
| 10/29/2024 | 32 | 30 | 2 | R+3.12 | |
| 10/30/2024 | EV | 2639 | 2863 | 79 | D+4.01 |
| 10/30/2024 | 23 | 41 | 0 | D+28.12 | |
| 10/31/2024 | EV | 2536 | 2668 | 60 | D+2.51 |
| 10/31/2024 | 71 | 65 | 2 | R+4.35 | |
| 11/01/2024 | EV | 4768 | 4889 | 146 | D+1.23 |
| HD | Dem | Rep | Ind | lean |
|---|---|---|---|---|
| 35 | 6788 | 5847 | 155 | D+7.36 |
| 37 | 17759 | 22882 | 452 | R+12.47 |
| 38 | 19620 | 13469 | 369 | D+18.38 |
In terms of simulating the election day vote, we looked at all the voters who didn’t cast a ballot by mail and then observed their propensities, normalized to a 0-1 scale. Typically, election-day voting is 30% of ballots cast in Cameron County. Being generous to Republicans, we benchmarked 34%. This works out to around 44,500 election-day votes, which are randomly selected from the remaining voter pool. The same random coin-flip process was applied to these. Republicans have a 1-point edge in the election day vote, or roughly 400 votes.
The result: a very narrow Democratic lead of c. 2 percentage points. This would represent a 12-point shift right from the 2020 margin, and a continued collapse for Democrats in what used to be their strongest region statewide. Given that Cameron County is home to competitive races up and down the ballot, this has major implications. In the Senate race, Ted Cruz will likely benefit from increased coattails, making Colin Allred’s path to flipping the state tougher. At a congressional level, this puts Republican Mayra Flores in serious contention to unseat Democratic Rep. Vicente Gonzalez in a rematch.
There are some caveats to this analysis that must be considered. First, the assumption about nonvoters could be entirely wrong. Though there isn’t a ton of polling on nonvoters, R+10 is much redder than what one would anticipate. Furthermore, when assigning propensity scores, young voters could be penalized for only turning 18 in the past cycle, particularly in a state as young as Texas. As many young voters, particularly young women, lean towards the Democrats, this could make the projections redder than reality. Additionally, the attrition rates between primary to general, and between cycles for low-propensity voters that skip cycles are just assumptions. If Democrats in the Valley defect in smaller numbers, and say have only a 98% attrition rate instead of 95%, that meaningfully pushes the margin up. Most importantly, this analysis is primarily turnout-based, and cannot directly discern persuasion between different offices. A presidential result says nothing about down-ballot candidates’ ability to outrun the baseline. Election-day votes are a purely random sample as well, which belies the potential for campaigning deltas between the parties to produce systematic error regarding who shows up. No matter what though, this analysis strongly suggests that Cameron County will swing right, with an outright flip waiting in the wings as a distinct possibility.
| ed_results | Freq |
|---|---|
| d | 21503 |
| i | 627 |
| r | 21196 |
In the legislature, this continues to be bleak for Democrats. Erstwhile competitive House District 37, centered on Harlingen, is looking to be a comfortable Republican hold for incumbent Rep. Janie Lopez, while Senate District 27, the only competitive seat in the chamber, will require massive ticket-splitting for incumbent Senator Morgan LaMantia to overcome the whiter northern parts of the district. A result like this would confirm a secure beachhead for Republicans to consolidate strength in the Valley.
| HD | Dem | Rep | Ind | lean |
|---|---|---|---|---|
| 35 | 3532 | 3496 | 95 | D+0.51 |
| 37 | 6594 | 9046 | 241 | R+15.44 |
| 38 | 11377 | 8654 | 291 | D+13.4 |
| d | r | i | Type | margin |
|---|---|---|---|---|
| 44167 | 42198 | 976 | Mail/Early | D+2.25 |
| 21503 | 21196 | 627 | Election Day | D+0.71 |
| 65670 | 63394 | 1603 | Total | D+1.74 |
| hds | d | r | i | lean |
|---|---|---|---|---|
| 35 | 10320 | 9343 | 250 | D+4.91 |
| 37 | 24353 | 31928 | 693 | R+13.3 |
| 38 | 30997 | 22123 | 660 | D+16.5 |
Running a simulation of the election, we see the following below. Harlingen generally leans red, while Brownsville generally leans blue.
Traditional early-vote modeling as practiced on Twitter usually relies on county-level data and partisan registration. In Texas, neither is particularly applicable since most counties in Texas are deep-red with non-existent Democratic turnout operations, and Texas’s lack of party registration makes the second idea complicated. Gaining access to the individual-level voter file with voter history was a great starting point for this project.
With the variables in the voter file (age, gender, Spanish surname, past elections, and precinct), one can construct a reasonably accurate picture of the electorate. To estimate the future electorate, two variables must be constructed: propensity and partisanship.
Solving for propensity is as simple as weighting the past elections since 2016 in the voter file by recency. Each voter is assigned a propensity score that will be used in later forecasting. Solving for partisanship is a little more complicated. Even though 2016 and 2018 are included in the voter file, we decided to exclude them since pre-2020 elections in the Rio Grande Valley operated under a different set of circumstances. With 2020, 2022, and 2024 cycles, each voter had to have a stated or implied partisan lean.
In 2024, the only real race so far is the March primary. I estimated that 90% of Democratic primary voters will support the party in November. Even though the area may continue trending right, that is primarily driven by general-election-only voters. Conversely, I assumed only 2% of Republican primary voters would consider voting Democratic. Trump’s overwhelming support here in the spring means he is unlikely to suffer defections.
In 2022, having both a general and a primary makes estimating partisanship much easier. The voter file lists each voter’s ballot type in each election, whether by mail, early, or on election day. I made my base partisan assumption for each voter using the mail/early/election day splits for each precinct that the voter lived in. For example, if the voter cast a mail ballot in a precinct where Democrats won 65% of the mail vote, that voter’s chance of voting blue in 2022 was 65%. However, this results in too much crowding around the mean precincts. Each precinct total represents a summary statistic of all the ages, genders, and groups living within. We created a gender-age distribution for Spanish-surname voters and non-Spanish surname voters. Relative to baseline partisanship in each precinct, a woman in the same precinct would have a higher chance of voting Democratic than a man. A 35-year-old would code as redder than a 65-year-old.
These distributions were created in discussion with Texas Republican campaign consultants. Once these gaps were created, above and below the baseline partisanship, they were added to create a more natural and normal-looking distribution. This same procedure was also applied to the 2020 general election as well. For the primaries, the same logic from the 2024 example was used. In 2022, we estimated that 90% of Democratic primary voters supported Democrats in November, as there was no big realignment visible. In 2020, that number was put lower, at 85%, to account for the sea changes visible that fall.
The next big question was how to address nonvoters. Given the high electorate volatility with plenty of low-propensity voters, the initial assumption made matters. The gold-standard poll in Texas, UH-Hobby, had Harris winning statewide non-voters by 16, 56-40. A recent Siena College-New York Times poll had Trump winning Texas non-voters statewide, 44-42.
As a red-leaning assumption, we decided to have Democrats win 45% of nonvoters from past elections. As a base case, these voters’ partisan leans were buffered by age and demographics similar to what was done for 2020 and 2022. Nonvoters who voted last in 2016 or 2018, and not in 2020-2024 went through the same processing, as pre-2020 elections in the Valley were conducted under a different environment. This 45% was spread with each precinct’s relative lean against the countywide margin but diminished slightly, as nonvoters of similar demographic profiles will have more similar preferences across geographies than higher-propensity voters.
With 2020, 2022, and 2024’s partisan leans established, each voter has to have a blended topline partisanship. First, we weighted primary and generals for 2020 and 2022, with 95% on the general and 5% on the primary. If voters only voted in the primary, we estimated that 95% of primary-only voters would be Democrats in those general election years. In 2022, with no realignment observed, that number jumped to 98%.
For blending the leans, we weighted the three cycles, emphasizing 2020 and 2022. If a voter voted in an early cycle but not later, we implied the partisanship for the nonvoting cycle by taking 98% of the previous cycle’s lean. In other words, if a voter had an 80% chance of voting Democratic in 2020, and sat out 2022, then we implied that in 2022, this voter would have had a 78% chance. Jumping from 2022 to 2024, with more potential Democratic slippage, that 98% was downgraded to 95%.
Lastly, we imputed third-party preference as a function of age. Polling indicates younger voters are more disenchanted with either party than older ones. Assembling the voting probabilities, we can conduct random coin flips across the entire electorate and understand.