Mid-Decade Redistricting: The Same Votes,
Different Seats
The Same Votes, Redrawn: Texas in 2025, Florida in
2026
Democracy’s Data
Last updated September 15, 2026 at 4:42p EDT
Texas held 17 statewide elections between 2016 and 2024, and every
ballot in them has been counted. In 2025 the legislature redrew the
state’s congressional districts, four years into a ten-year cycle, with
the stated aim of electing five more Republicans. Those two facts make
an unusual test possible. The votes are fixed; the lines are not. So you
can ask a question that usually has no answer: with the same people
voting exactly the way they already voted, does a new map send different
people to Congress?
The question belongs to you rather than to a mapmaker. You cast one
vote, in one place. Whether it lands in a district your side wins
narrowly, loses narrowly, or wins by forty points is settled by other
people, in a room, after the last election and before the next one. This
brief pours already-counted votes into both sets of lines and reports
what comes out.
The test
Using 17 Texas statewide elections, 2016 through 2024, re-counted
inside two congressional plans: the map Texas enacted in 2021 and the
map that replaced it in 2025. Both re-counts come from Dave’s
Redistricting (https://davesredistricting.org), a free public site. It
rebuilds American elections at the precinct level, the few thousand
voters who share one polling place, then re-sums them inside any
district map you load.
Almost nothing else can answer this question. No agency reports
statewide results inside a map that has never held an election. A
private volunteer project does, which is why this site turns up wherever
a proposed plan is argued about.
Two files meet here, and the meeting is the whole
method. Precinct returns are joined to a district plan by
geography: each precinct falls inside one district, and the district’s
number is the label that comes back. Do that twice, once per plan, and
the two results are then joined to each other on that number. The
assumption is that district 28 in 2021 and district 28 in 2025 name
comparable things.
Join Keys and
Crosswalks works that kind of match on the best-behaved key in
American data. It also says what to do when a key stops behaving: build
a crosswalk, a table that translates one set of areas into another.
Texas needed none. Florida, at the end of this brief, did.
Nothing in federal law stops a state from redrawing between censuses,
which is what mid-decade redistricting means: a new map
in the middle of the ten-year cycle rather than right after a count.
Missouri, North Carolina and Tennessee have written their own bars into
state law, and Texas has not. In League of United Latin American
Citizens v. Perry (2006) the Supreme Court declined to treat a
mid-decade redraw as presumptively partisan. Rucho v. Common
Cause (2019) then closed the federal courts to partisan
gerrymandering claims altogether. So the question is not whether the
2025 map was allowed. It was.
The data
The table underneath everything here is
district_shares.csv, one row per district per election, 646
rows. Each row carries district, election and
year, then share_old and
share_new: the Democratic share of the two-party vote
inside the 2021 lines and inside the 2025 lines. A two-party share is
the Democratic vote divided by the Democratic and Republican votes
together, third parties set aside, so fifty per cent means tied.
seats_by_election.csv collapses those rows to one per
election, with rep_old, rep_new and
rep_gain. district_summary.csv gives each
district its mean_old and mean_new across all
17 races.
Three things happened to the votes before they reached those columns.
Precincts cut by a district line were split, with votes shared out in
proportion to population. Every district total was reduced to a
two-party share. And the identical arithmetic was run twice, once per
map, which is what makes the two columns comparable.
One further file carries a second join.
tx_overlap_pop.csv holds old,
new, pop and weight: how much of
each 2021 district’s population sits inside each 2025 district. It is
built by giving every 2020 census block group, a cluster of city blocks
holding roughly six hundred to three thousand people, to whichever
district contains its centre of population, under each plan in turn.
That match is approximate, reproducing district totals to within about
1.5%, so read its shares to the point rather than to the decimal.
Before you scroll, write down two numbers. Across 17
statewide elections, in how many does the new Texas map elect more
Republicans than the old one, and in how many does it elect fewer?
What it says about democracy
Figure 1. Thirty-eight districts, two maps, one set of
votes. A map, because what the legislature changed was
geography, and no other form lets you watch a boundary move. Every
district is traced with the same number of points under both plans, so
pressing the morph button slides the old shape into the new one. Color
is the Democratic share of the two-party vote in the race you pick.
Click a district to load all 17 of its elections. In print the two plans
sit side by side, colored by the 2024 presidential race, with district
28 below.
The seat counts settle the aggregate question immediately. Count the
districts where the Democratic share tops one-half, election by
election, under each map. Take the 2024 presidential race. Under the
2021 lines, 11 of the 38 districts gave the Democratic candidate more
than half of the two-party vote, so 27 went Republican. Pour the same
precinct votes into the 2025 lines and 8 do, so 30 go Republican: 3
seats change hands with not one vote changed. Do that for all 17 races
and every one of them elects more Republicans under the new lines: at
least 2 more seats, at most 5, never zero and never negative.
Figure 2. The gain, race by race. Each row re-counts
one statewide election under both maps, and the dot never moves left.
One row keeps the collection’s own shorthand: the 2022 race labeled
trea arrives that way in the source export, and this brief
will not guess at an office name the file does not state.
The size of the gain moves with the year. The 2018 races, a strong
Democratic year nationally, hand the new map its smallest gains; the
2024 races hand it some of its largest. A map drawn from recent returns
is tuned to recent returns, and you are watching the tuning.
Figure 3. What the swap did to every district at
once. One point per district per election, 646 of them, with
the old share across and the new share up. Gray points barely move: most
districts were already safe and stayed safe. The colored points are the
exchange the map was drawn to make, a handful of Democratic-held
districts pushed across the line with almost nothing pushed back.
One district argues with that finding, and it is worth
hearing out. In August 2025 Representative Henry Cuellar told
POLITICO that his redrawn 28th district might favor him more
than the one it replaced (Benson 2025). He is reading real
numbers. In the 2024 presidential race the new lines are worse for a
Democrat, moving from 46.3% to 44.8% Democratic. But in 15 of the 17
races, all of them from 2016 through 2022, the new district is the bluer
one.
The overlap table says where those voters came from. The new 28th
keeps 53% of the people who lived in the old 28th and takes in 42% of
the old 15th. Nearly half the district he is arguing about is territory
he has never represented. His staff and the map’s drafter can both be
right, because they are looking at different electorates: voters who
backed Democrats reliably before 2024 and swung hard toward Trump in
2024. Which of those two turns out in 2026 is the bet the whole redraw
rests on.
The same test, run in Florida
Florida ran the same exercise a year later, and it comes out the same
shape. The legislature replaced its congressional map in 2026. The same
site re-counted 13 statewide elections from 2016 through 2024 inside the
2022 plan and the 2026 plan, with the same two-party arithmetic. Every
one of the 13 races elects more Republicans under the new lines, from 1
to 4 more seats. Under the 2024 presidential race, Republican-carried
districts go from 20 to 24 of 28. That is exactly the swing the map’s
own drafters described in public as the point of the exercise.
Two things make Florida the harder case. Its constitution forbids
drawing a district with the intent to favor a party, language voters
added by initiative in 2010, and a state court has already voided one
congressional plan under it. Every number above measures effect rather
than intent, so it is the wrong shape for that rule, however plainly the
drafters named the seat count they wanted.
The second difference is the join. The redraw renumbered 3 of
Florida’s districts as well as redrawing them, so the district number
stopped naming the same thing on both sides. Pairing the two plans took
a crosswalk. Every one of Florida’s 2020 census blocks, the smallest
areas the Bureau reports and often a single city block, was assigned
into both maps, and each new district matched to the old one it drew
most of its people from. That match is only a plurality: the old 23rd
supplies 46.3% of the new 25th’s population, not a majority.
What this brief cannot tell you
These are statewide contests re-counted inside district lines, not
House contests. A House race has its own candidates, its own money and
an incumbent voters know. Cuellar’s evidence came from House returns,
which this data does not carry; the map’s drafter answered him with
statewide returns, which it does. You have been reading the drafter’s
kind of evidence, and should say so when you quote it.
The re-count also holds turnout fixed. It assumes the people who
voted between 2016 and 2024 are the people who will vote in 2026, in the
same numbers, with the same loyalties. Districts full of voters who
changed sides recently, like the ones along the Texas border, are
exactly where that assumption is weakest.
Nothing here reaches intent. The tables can say what the new lines do
to votes already cast, precisely and in every election available. They
cannot say why anyone drew them, which is the only question Florida’s
constitution actually asks.
What you should have learned
- In Texas and Florida alike, votes already cast elect more
Republicans under the new lines in every election available: 2–5 extra
seats across 17 Texas races, 1–4 across 13 Florida ones.
- A re-count holds the voters still and moves only the lines, which is
why it can attribute the difference to the map and to nothing else.
- Most districts do not move: of 646 Texas district-elections, the
ones that change hands are a small minority against a background of
seats safe before and after.
- A map drawn from recent returns is tuned to recent returns, so its
gains are largest in the years closest to the drawing and smallest in a
year that ran the other way.
- The district number is the key that joins one plan to another, and
it holds only while nobody renumbers. Texas did not; Florida renumbered
3 districts, and pairing its maps took a crosswalk built from census
blocks.
Extensions
None of these is answered above, and every one can be answered from
the tables the Sources section hands over.
seats_by_election.csv replays every election on both
maps and reports rep_gain. Count the elections where the
new map does worse for its authors. What does a column with only one
sign mean?district_summary.csv has mean_old beside
mean_new for each district. Sort by the difference. Which
districts absorbed the change, and which were left alone?tx_overlap_pop.csv gives the population shared between
each old and new district as a weight. Find the districts
that were rebuilt almost from scratch, and the ones that kept
everybody.- Pick one election in
district_shares.csv and compare
share_old against share_new across all
districts. Do the shares move together, or does the map take from some
and give to others?
- Stretch. Other states moved on their maps in the
same cycle. Find one state’s before-and-after plans and check whether
the one-sided pattern repeats.
Sources
Data
The data itself
The Texas figures above are drawn from these tables. They are plain
CSV, they open in a spreadsheet, and they are yours to take further.
district_shares.csv,
district_summary.csv,
seats_by_election.csv,
tx_overlap_pop.csv,
tx-districts-tween.json
One more file is not data about the subject – it is how this chapter
checked its own numbers: facts.csv, the individual
numbers the text above quotes.
Cases, constitutions and state law
- League of United Latin American Citizens v. Perry, 548 U.S.
399 (2006) — the challenge to Texas’s 2003 mid-decade map: no
presumption against mid-decade redistricting, and District 23 struck
under Section 2 of the Voting Rights Act. https://www.law.cornell.edu/supremecourt/text/05-204
- Rucho v. Common Cause, 588 U.S. 684 (2019). https://www.law.cornell.edu/supremecourt/text/18-422
- Fla. Const. art. III, §§ 20–21 (the Fair Districts amendments,
adopted by initiative in 2010), https://www.flsenate.gov/Laws/Constitution, and
League of Women Voters of Florida v. Detzner, 172 So. 3d 363
(Fla. 2015), holding the 2012 congressional plan drawn with intent to
favor a party and its incumbents; summarised with its filings at https://thearp.org/litigation/league-women-voters-florida-v-detzner/.
- National Conference of State Legislatures, Redistricting Law
2020 (Denver: NCSL, October 2019), ch. 10 for the states that bar
mid-decade redistricting and ch. 6 for what LULAC settled.
Frozen at 2019: a state could have added or repealed such a provision
since.
Press
Academic literature
Why this data cannot be fetched again
The data behind this chapter cannot be downloaded again by a script, and
it is worth knowing why before you try.
WHY NOT. The two spreadsheets were exported from Dave's Redistricting
(https://davesredistricting.org), which rebuilds election results at
the precinct level and re-sums them inside any district map you load.
Each export is a manual download from a browser session with a
particular map open; there is no address a script can call to get the
same file. The copies shipped with this chapter are the archive.
WHAT EXISTS INSTEAD. The site itself, free after registration. Open
the enacted 2021 Texas congressional map and the 2025 replacement,
turn on any statewide election from 2016 through 2024, and the site
shows the same district-by-district shares these files hold. Exporting
them one election at a time rebuilds the two spreadsheets, columns for
president, governor, senator, and the rest, rows for districts 1
through 38.
WHAT YOU CAN STILL CHECK. However you come by the numbers, these say
whether you are holding the same object:
- 38 rows per file, one per district; 17 election columns; the long
table assembled from them has 646 rows
- every cell is a Democratic share of the two-party vote, strictly
between 0 and 1
- in the 2024 presidential race, Democratic candidates lead in 11
districts under the old lines and 8 under the new
- across all 17 elections the new lines never elect fewer Republicans
than the old ones; the gain runs from 2 seats to 5
