Early vote

Winning margins use the 2024 Harris–Trump two-party vote: 0–<15, 15–<30, and 30+ points. Each percentage divides summed 2026 votes by summed 2024 votes. † marks fewer than 3 groups or 100 baseline votes. — means no baseline or unavailable data. These are precinct patterns, not voters’ party choices.

Regression uses unweighted ordinary least squares with an intercept: one point per selected comparable precinct group. X is 2024 Harris / (Harris + Trump), or the selected Census population / Democratic primary participation share; Y is cumulative 2026 in-person votes / cumulative 2024 votes through the same early-voting day. Auto scales the vertical view around the main cluster using 1.5 interquartile-range fences with padding. Full range reveals every observation; the fixed 60–150% band is also available. The All/Trump/Harris controls select the 2024 winner of the matched geographic group for demographic and primary regressions. Ties and incomplete presidential returns remain in All only. The fit, sample size, axes and Points CSV use the selected regression cohort; bucket and map data remain independent. Every eligible observation in that cohort remains in the fit and CSV. Zero 2024 baselines are omitted; all other eligible points, including outliers, remain. In partisan mode, quadrants split at 50% of the selected share and 100% of 2024 turnout. Blue/red dots are at or above 2024; teal/ochre dots are below, on the Harris/Trump side respectively. In demographic and primary modes, dots retain their 2024 two-party winner colors: Harris blue, Trump red; ties or missing presidential votes are gray. The selected share and turnout do not change that color. Points on a partisan divider use the at-least-50% or at-least-100% side. The line spans observed vote shares. Bucket selection highlights observations without changing the fit. All counties pools groups without county adjustment; the county selector gives within-county fits. This descriptive fit has no causal interpretation.

The Demographic × partisan matrix crosses Census composition buckets with the existing 2024 presidential-margin bands. Cells divide summed 2026 in-person votes by summed 2024 votes for the same available comparison groups. Both classifications refer to the compared group. Its partisan band uses the combined 2024 Harris–Trump margin, with complete returns required for every old precinct. Different member bands do not make the combined result unknown. Tied, zero-vote or missing returns stay in Tied / unknown. All marginals include that column. Missing or invalid Census counts are omitted, and a zero 2024 denominator remains undefined. The matrix has its own demographic selector and follows the selected day, county and matched/unchanged cohort. These are patterns among places, not the turnout of individual demographic groups.

Maps use the Texas Legislative Council 2026 primary precinct shapefile as a November proxy. A comparison dot may span multiple current precincts; selecting it highlights the whole group. Census values describe residents, not the demographics of people who voted. Demographic matched views can include mixed-partisan groups because no partisan classification is required. Percentages use summed Census counts, never averages of precinct percentages.

Each day compares the largest available set of reported precincts matched in both years. Daily and cumulative coverage can differ and change between days. Missing pairs are omitted, never treated as zero. Future days show the available 2024 benchmark alone.

County/state totals default to mail plus early in-person voting. Mail receipts from before opening day remain in cumulative totals; daily mail uses a preceding-day report. Both methods must be available for the same geography and date in both cycles. Texas precinct charts are explicitly in-person: paired precinct mail has not yet passed source reconciliation. Received, issued and accepted mail categories are not interchangeable.