How Runoff Rules Pick House Nominees,
2004–2022
The rule that picks the nominee, found without
reading a single statute
Democracy’s Data
Last updated September 15, 2026 at 4:42p EDT
In March 2014, Ralph Hall came first in the Republican primary — the
election in which the party’s voters choose its nominee — for Texas’s
fourth district. He had represented it for 34 years and was the oldest
person ever to serve in the House. He took 45.4 percent, and the man in
second, John Ratcliffe, took 28.8.
Two months later he was not the nominee. Texas requires a majority —
more than half of all the votes cast — so the top two went to a runoff,
a second election between the two of them alone, and Ratcliffe won it.
Nothing about the votes changed. The rule finished the count
differently. Had Hall’s district been in Pennsylvania, where a
plurality — simply the most votes, however far short of half — wins, he
would have been renominated on the night.
A nomination is usually described as a measurement of what a party’s
voters want. This brief asks how much of the answer is instead the rule
that finishes the count. Where does that rule live, how often does it
change the outcome, and how much of it can you work out without opening
a statute book?
The test
The data is the Federal Election Commission’s Federal
Elections series, the compilation the Commission publishes after
every two-year election, for 2004 through 2022, downloadable from the Commission.
For every federal candidate it reports the primary vote, the runoff vote
and the general-election vote, assembled from the official counts of the
state election officials. The ten workbooks are kept with the retirements
brief.
This source fits the question because it is the only federal record
that reports both rounds. Each state writes its own nominating rule into
its own election code, and there is no federal table of the rules. What
the Commission records is narrower and more useful: when a runoff
happened, it reports the runoff votes. That turns out to be enough to
recover the rule from its outcomes. This brief belongs to the cluster
the roster
chapter introduces — records about who ends up holding office, read
here for the step where a party settles on one name.
The test: for every state, squeeze the threshold from both sides of
its outcomes, and then count how often the second round reverses the
first.
The data
The ten workbooks hold 22,066 House candidate rows, of which 16,547
of 22,066 are on a Democratic or Republican line. The Commission has
changed its layout between editions — sheet names, column headings, and
party labels that are one letter in some years (D,
R) and three in others (DEM, REP)
— and a reader who takes only one form of the party label loses two
entire elections without a warning.
| Elections covered |
10 |
| Years |
2004 to 2022 |
| House candidate rows |
22,066 |
| Of those, on a major-party line |
16,547 of 22,066 |
| Contested primaries with usable vote totals |
3,705 |
| Of those, in all-party states, set aside |
576 |
| Runoffs |
173 |
The rows are grouped into contested primaries: one
state, one district, one party, one cycle, with at least two candidates
who received a vote total. That is the nomination that was actually at
stake. The 576 contests in California and Washington pool both parties
onto one ballot — a top-two primary, from which the two
leaders advance whatever their party — so a share there is not
comparable to a party primary’s, and they are set aside. Louisiana is
dropped because in most of these years it holds no party primary at
all.
Four tables come out. contests.csv is one row per
contested primary, with leader_share and a
runoff flag. runoffs.csv keeps the 173
contests that went to a second round, with leader_share and
leader_won_runoff. brackets.csv gives each
runoff state the highest leader share that still produced a runoff, the
lowest that did not, and the bracket_width between them.
by_year.csv counts contests, runoffs and overturns by
cycle.
Before you read on, write a number down. Across 2004
to 2022 there were 173 House primary runoffs. In each, the candidate who
came first in the primary faced the candidate who came second.
How many times did the person who came second go on to
win? Commit to a count out of 173 before you look at the
figures.
What it says about democracy
A runoff happens when the leading candidate falls short of some
threshold. You do not know the threshold. But you can see, for every
contest, what the leader took and whether a runoff followed.
So the outcomes squeeze the rule from both sides.
The highest leader share that still went to a runoff must sit
below the threshold. The lowest share that did not must sit
above it. The rule is somewhere in between.
Georgia shows the squeeze. Across 105 contested primaries it held 27
runoffs. The highest share a leader took and still had to face a second
round was 49.5 percent, so the threshold is above that. The lowest share
that settled a nomination outright was 50.2 percent, so the threshold is
below that. Georgia’s rule lives in a window 0.7 of a point wide, and 50
percent — a majority — sits inside it.
Figure 1. Where each state’s runoff threshold has to
be. One bar per state that held a House runoff, because the
finding is an interval, not a trend. The left edge of each bar is the
highest leader share that still produced a runoff, the right edge the
lowest share that did not; for a fixed rule, the threshold lies inside
the bar.
6 of the 8 states bracket cleanly, and every one of them brackets
half the vote. The tightest is GA, 49.5 to 50.2 percent, which pins the
rule to within 0.7 of a percentage point. Nobody looked up a statute to
get that.
Two states break the method, and they are the exceptions with a
story.
| TX |
80 |
306 |
49.6 |
47.6 |
| NC |
11 |
149 |
39.5 |
20.0 |
A bracket cannot run backwards if the rule is fixed.
A North Carolina contest at 39.5 percent went to a runoff, and one at
20.0 percent did not. No single threshold does both. So the rule was not
fixed. Split North Carolina by cycle and it appears:
| 2004 |
2 |
2 |
35.0 |
45.4 |
| 2010 |
3 |
3 |
39.5 |
40.2 |
| 2012 |
3 |
3 |
37.8 |
43.5 |
| 2014 |
2 |
2 |
34.3 |
40.9 |
| 2016 |
3 |
0 |
NA |
20.0 |
| 2020 |
1 |
1 |
22.7 |
46.3 |
| 2022 |
5 |
0 |
NA |
31.2 |
In 2010 the bracket closes at 40 percent, pinned to within 0.7 of a
point by two contests in one year. In 2016 three contests fell under 40
percent, the lowest at 20.0 percent, and none triggered a second round —
for that cycle there was effectively no threshold. Then it closes again,
lower: a 2020 contest at 22.7 percent went to a runoff, and by 2022 the
rule sits somewhere around 30. The threshold moved twice, and both
moves are visible in nothing but outcomes. Texas fails for a
smaller reason: its bracket runs backwards by 2.0 points because a
second-place finisher just under half can decline the runoff rather than
contest it.
How often does the second round matter? The record answers directly,
because it reports both rounds.
Figure 2. Every House primary runoff from 2004 to 2022, one
tick per contest. Each tick sits at the share the primary
leader took; the upper row is runoffs the leader went on to win, the
lower row runoffs the leader lost.
61 of 173. The candidate who came second in the
primary went on to win the nomination in more than a third of these
contests. Most readers guess a handful, because coming first sounds like
an advantage that carries. It carries, and it is not decisive. The
runoff is not a formality that ratifies the first round. It is a second
election, and it frequently disagrees with the first.
Most states, though, hold no runoff at all: a plurality wins, however
small.
| under half the vote |
593 of 2,956 (20.1%) |
| under 40 percent |
279 of 2,956 (9.4%) |
| under 35 percent |
157 of 2,956 (5.3%) |
593 nominations were settled outright by a candidate most
primary voters voted against. Every one of those would have
gone to a runoff in Georgia or Alabama. What would have happened there
is unknowable, but Figure 2 gives the only relevant guide: when contests
like these do go to a second round, the first-round leader loses about a
third of the time. The same distribution of votes yields one nominee
under one state’s rule and, often, a different nominee under another’s.
And the rule sits in fifty separate election codes, appears in no
federal table, and changes when a legislature feels like changing
it.
What this brief cannot tell you
It cannot tell you who would have won a runoff that never
happened. That is the central question and it is unanswerable.
A runoff is a different election with a different electorate, and the 61
overturns in Figure 2 are the nearest thing to an answer, not the
answer.
It cannot give you the rules themselves. Figure 1
brackets a threshold; it does not read a statute. Where a state held no
runoff in twenty years, this file says nothing about its rule, including
whether it has one. And a bracket is only as tight as the contests that
happened. Georgia’s is narrow because Georgia ran 27 runoffs across 105
contests; a state with two runoffs gets a wide bracket and no more
knowledge than that.
Turnout is missing, and it is not a detail. These
are shares of whoever voted, and a runoff electorate is usually smaller
than a primary’s. Two rounds can disagree because voters changed their
minds or because different people showed up, and nothing here separates
those.
And a rule is not only a threshold. Who may vote,
when the primary is held, how a candidate qualifies, and whether the
second-place finisher must request the runoff all vary by state. None of
that is in this file.
What you should have learned
- A nomination is a measurement plus a rule for finishing it, and the
rule does real work: 61 of 173 House primary runoffs from 2004 to 2022
reversed the first round.
- A state’s runoff threshold can be recovered from outcomes alone: 6
of 8 runoff states bracket cleanly around half the vote, the tightest to
within 0.7 of a point.
- A bracket that will not close is a finding too: North Carolina’s
threshold moved twice inside these ten elections, visibly, without the
file ever stating a rule.
- In plurality states, 593 nominations were settled outright by a
winner who took under half the vote — contests a majority-rule state
would have sent to a second round.
- The counting rule lives in fifty separate state codes and no federal
table, which is why the practical way to find it was to look at what
happened.
Extensions
None of these is answered above, and every one can be answered from
the tables the Sources section hands over.
brackets.csv finds each state’s threshold by bracketing
it. Sort by bracket_width. Which states can you pin down
exactly, and which are still ambiguous?by_year.csv counts contests and overturned
runoffs by cycle. Work out how often the rule changes the answer, year
by year. Is the overturn rate steady?runoffs.csv has leader_share and
leader_won_runoff. Find the leaders who lost. How far ahead
were they after the first round?nonclosing_by_year.csv holds the state-years where the
bracket does not close. Read North Carolina’s rows. At what cycle does
the change become visible?
Two stretch ideas point past the spreadsheet. Look up one runoff
state’s actual statute, Georgia’s or Texas’s, and check it against the
bracket Figure 1 found. Then find the North Carolina session law that
moved its threshold, and date it against the cycles above. And the same
workbooks carry Senate races: rebuild the brackets from Senate contests
and see whether they agree with the House ones.
Sources
Data
The data itself
Every figure above is drawn from these tables. They are plain CSV,
they open in a spreadsheet, and they are yours to take further.
brackets.csv, by_year.csv, contests.csv, nonclosing_by_year.csv,
runoffs.csv
One more file is not data about the subject – it is how this chapter
checked its own work: checks.csv, the
arithmetic verified before anything was drawn.
