The opinion
Opinion issued February 3, 2026.
In The
Court of Appeals
For The
First District of Texas
————————————
NO. 01-23-00921-CV
———————————
ERIN ELIZABETH LUNCEFORD, Appellant
V.
TAMIKA CRAFT, Appellee
On Appeal from the 164th District Court
Harris County, Texas
Trial Court Case No. 2022-79328
CONCURRING OPINION
Harris County’s administration of the 2022 general election was not the
pinnacle of excellence. Some polling places opened late, which led a court to order
they all stay open late. Some didn’t have enough ballots. Thousands of people, upon
being told of delays due to these problems, left polling places and may have given
up on voting altogether.
And those were just the problems apparent on election day. Subsequent
litigation showed well over a thousand individuals were allowed to vote despite
either not completing statutorily required forms or filling out residence forms in a
manner not consistent with being a Harris County resident.
Were the election a blowout, this may have gone unnoticed. But it was close.
Out of 1,107,390 votes cast, two countywide elections were decided by fewer than
a thousand votes, and five other races were decided by fewer than five thousand.
Twenty-one losing candidates filed election challenges. These cases were
assigned to the same visiting judge, the Honorable David Peeples. Judge Peeples
granted summary judgment for the contestees in fifteen cases, four contestants
dismissed their cases after discovery, and two cases went to bench trials.
Both of those trials—this case, and Jones v. Pierce—were appealed and I was
on the panel for both.1 I join the opinion of the court today because I believe Judge
1
In Jones v. Pierce, No. 01-24-00377-CV, the margin of victory in that case was 449
votes. Based on evidence that significantly overlapped with the evidence here, Judge
Peeples found the true winner of the election could not be determined and ordered
a new election. While the appeal was pending the contestee/appellant resigned his
bench and dismissed the appeal, leaving the trial court’s judgment in place. As best
I can tell, in terms of votes cast that is the largest election ever overturned in Texas
history.
There were a variety of legal issues on appeal in Jones, and nothing I say here is a
comment on the proper result of that case. I use the publicly-available information
2
Peeples was within the bounds of his discretion to conclude that Craft’s margin of
victory was too large to be undermined by the number of votes Lunceford proved
were affected by the various election administration mistakes.
With that said, having reviewed the records and arguments for both cases I
have noticed ways in which current Texas law makes election challenges from large
jurisdictions difficult, or perhaps impossible. Some of these are beyond the control
of the courts—it seemed awfully hard for the contestants to get basic information
from the county to verify its election results, such as 1) voter rolls for who was a
registered voter on election day, 2) an accurate list of who voted, 3) cancellation and
suspense lists that were accurate on election day—so I won’t dwell on them.
But one issue falls squarely upon the judiciary: What type of proof suffices to
prove a voter who cast a ballot actually voted in a particular down-ballot election?
To overturn an election Texas law requires not just that illegal ballots were cast, but
that illegal votes were cast in the specific race being contested. In previous cases,
contestants met this burden through testimony by the voters, but the numbers here
make that practically impossible. In future big-jurisdiction cases, I think courts
should be open to the sort of general statistical reasoning Judge Peeples found
persuasive in this case.
and arguments from Jones simply as a data point of a case that had some, but not
complete, factual overlap with this case on the single issue I address here.
3
I. Size Comparisons
With over a million votes, this case is the largest election contest in Texas
appellate law, and it’s not remotely close. The next largest I can find involved
roughly 4% as many votes. See Miller v. Hill, 698 S.W.2d 372 (Tex. App.—Houston
[14th Dist.] 1985), pet. dism’d, 714 S.W.2d 313 (Tex. 1986). Indeed, Texas election
cases typically involve small numbers of votes in small elections. See, e.g., Medlin
v. King, 705 S.W.3d 267, 278 (Tex. App.—El Paso 2024, pet. denied) (litigating
residency of 26 Loving County voters for election with around 80 total votes);
Woods v. Legg, 363 S.W.3d 710, 712 (Tex. App.—Houston [1st Dist.] 2011, no pet.)
(litigating residency of voters for Galveston city council runoff where winner got
209 votes and loser 200).
The large number of votes here obscures how small the margin actually is.
For instance, in Rodriguez v. Rangel, 679 S.W.3d 890 (Tex. App.—San Antonio
2023, pet. denied), the contest concerned a 7-vote margin in an election where 3,907
ballots were cast. A 7-vote margin sounds a lot smaller than the 2,743 margin here
or the 449 margin in Jones, but as a percentage of votes cast the 7 votes in Rodriguez
is 0.18%, which is only slightly smaller than the 0.26% margin here, and
considerably larger than the 0.04% margin in Jones.
I point this out to dispel any notion that the litigants here and in Jones
challenged an extraordinarily high portion of voters. It is simply the case that any
4
election challenge in a large Texas jurisdiction is going to involve a challenge to
thousands of votes. The judiciary must be able to resolve a claim like that in a fair
and practical manner.
II. Evidence Large and Small
In election cases with few votes it’s not a huge burden to put on live witnesses.
While having twenty-six challenged voters testify in Medlin was likely a big to-do
in Loving County, that number of witnesses falls within ordinary parameters for a
trial.
But testimony from the thousands of challenged voters here and in Jones does
not. And neither Lunceford nor Pierce attempted to produce such testimony. Instead,
they attacked the lawfulness of large categories of ballots, including ballots
supported by improper documentation and ballots cast after hours, among others.
With those sorts of claims, with these sorts of numbers, how should a court
determine which of the challenged voters actually voted in the relevant election?
Lunceford and Pierce argued the court should simply apply the undervote percentage
for the overall election to the challenged number of votes. In this case, 96.14% of all
ballots cast included a vote for the 189th District Court, so take the number of people
who voted illegally, multiply it by 0.9614, and that’s the number of illegal votes in
this election.
5
Judge Peeples called this approach “reasonable,” and he followed it both here
and in Jones. In both cases the contestees challenged this method, claiming the lack
of testimony from individual voters left this evidence insufficient.
III. The Alternative to Statistical Reasoning
The contestees’ observation that all previous case law involves individual
testimony is true, as far as it goes. Although Judge Peeples’s approach is novel, I
think it was a reasonable adaptation to the numbers involved in a Harris County
election contest. Requiring contestants to put on thousands of witnesses to challenge
even the slimmest election loss would make election challenges insurmountable in
large jurisdictions.
Judge Peeples’s approach was appropriate for the facts of these cases. First,
it’s important to remember the contestants were not seeking to be declared the
winner, which is what many election contestants seek. Here, the contestants’
requested relief was for the trial court to order a new election because it could not
“ascertain the true outcome of the election.” TEX. ELEC. CODE § 221.012(b). This
standard, on its face, embraces a fair amount of ambiguity.
Second, as the size of a sample goes up, the more likely it is to reflect the
whole. Had Judge Peeples found one ballot illegal, it would not have been reasonable
to infer whether that ballot undervoted a particular election. Had he found ten ballots
illegal, it still would have been unreasonable to infer much. But as the sample size
6
increases so does the reasonableness of inferring that it reflects the whole. With the
sample sizes Judge Peeples was dealing with—in Jones it was 1,483 affected votes,
here it was 2,891—in the absence of arguments or evidence to the contrary it would
be reasonable to infer these voters behaved more or less like the other voters.
Third, individual testimony in cases like this has its own problems. This is
shown by Green v. Reyes, 836 S.W.2d 203 (Tex. App.—Houston [14th Dist.] 1992,
no writ). Green involved a multi-race Democratic primary runoff election. There
were 32,338 total ballots cast, with 31,530 votes in Green’s race; Green won the
original count by 186 votes.2
The evidence showed 429 voters illegally voted in the Democratic runoff after
voting in the Republican primary. Id. The question before the trial court was to
determine how many of those voters voted in Green’s race, and, if possible, for
whom did they vote?
Green says that 313 of the challenged voters—0.97% of total voters—testified
(by live testimony, deposition, affidavit, or phone testimony) in an 11-day trial. Two
things stand out from this number. First, even with a concerted effort just a few
weeks after the election, only about three quarters of the challenged voters could be
2
Green has one of the largest vote totals of any Texas election challenge, and it is the
largest election where an appellate court has upheld a trial court’s decision to
overturn the results. The 186-vote margin there is much smaller in absolute terms
than the margin here, but as a percentage of the vote it’s more than twice the margin:
0.59% there compared to 0.26% here.
7
found. The trial court found it was “highly unlikely” that evidence could be obtained
from the other 116 “even with a protracted trial.” Id. at 206-07. If an election
contestant is obliged to put on witness testimony, but a quarter of challenged voters
will be unreachable, that’s a severe hindrance to large-jurisdiction election
challenges. And, intuitively, one would reckon the less legitimate the voter, the less
likely he or she is to respond conscientiously to service.
The second thing that stands out is that 313 witnesses in 11 days is a
remarkable pace, but it’s still insufficient for the numbers at issue in a large
jurisdiction election challenge. Had the trial court here heard testimony from 0.97%
of the voters—10,718—even at the pace of thirty witnesses per day, five days a
week, it would take almost a year and a half just to receive evidence. That would be
a stupendous outlay of judicial resources, and the cost to both parties would be a
degree of magnitude more than the government salary of the elected office at issue.
And how useful would this testimony be? The trial court in Green noted that
even though the case was tried within a few weeks of the election the testimony
“must be viewed with caution since many of the voters who disclosed their vote did
so with qualifications such as ‘I think’ or ‘I probably.’” Id. at 206. If the contestant
here or in Jones called thousands of witnesses to testify months after the election
about whether they voted in one of sixty down-ballot judicial races, any testimony
would need to be, at best, “viewed with caution.”
8
And then there’s the very obvious point that witnesses would be testifying
because they voted illegally or, at least, in a manner that did not demonstrate the
highest levels of conscientiousness. If a witness testified she illegally cast a ballot a
year earlier and she “thinks” she voted in a particular down-ballot race, I don’t know
what a factfinder is supposed to do with that.
Multiplying the marginal testimony of a marginal witness by several thousand
doesn’t create clear and convincing proof of anything. If 10,000 witnesses with
obvious credibility problems testified, on appeal any appellate court would defer to
the trial court’s credibility finding no matter what it was. If 10,000 witnesses
testified, “I illegally voted in the election, but I didn’t vote in the race for the 189th,”
or if 10,000 witnesses testified, “I illegally voted in the election, and I definitely
voted for the 189th,” as long as there were at least 10,000 votes and 10,000
undervotes, any finding by the trial court would be upheld. Thus, a trial court would
distill an 18-month trial into whatever result the judge wanted, and that would be
that.3
3
I’m not implying the result would be whimsical, necessarily. Forcing a judge to
make a decision based on overwhelming quantities of incredible testimony is like
forcing him to build a house on sand. The only question is which direction it will
fall.
9
IV. General Statistical Reasoning is Sufficient
I don’t think such an extravagant proceeding is necessary to produce clear and
convincing evidence that votes were cast when the number of challenged voters is
as large as it is here. Green is the only example in the case law of a court calculating
the number of illegal votes by multiplying the number of illegally cast ballots by the
undervote percentage for the contested race. Judge Peeples relied on Green for his
decision to use general inductive statistical reasoning here and in Jones. In both cases
the contestees argued that Green required more evidence than what the contestants
produced here.
The contestees’ basic point was correct: Green had more evidence than Judge
Peeples required. But under scrutiny it’s clear the additional evidence in Green was
unneeded or ignored.
The contestant in Green challenged 429 voters, of whom 313 testified. That
left the trial court with the problem of how to account for the other 116 illegal votes.
The chair of Rice University’s political science department testified about how to
calculate the undervote. Both here and in Jones the contestees pointed at this expert
testimony as distinguishing Green from these cases, where there was no such expert
testimony.
But the Rice professor’s testimony in Green is a great example of a highly
credentialed individual testifying to common sense. The extent of his testimony on
10
the undervote was that “[t]here is absolutely no reason to think that [the] two and a
half percent [undervote] that characterized the entire [contested race] would have
been anything other than characteristic of the [illegal voters].” Id. at 211.
In its findings, the trial court in Green calculated the undervote by applying
the 2.5% undervote to all 429 illegal voters. That is, it disregarded the voters’
testimony about whether they voted and instead used statistical reasoning.
The Fourteenth Court upheld this method based on the Rice professor’s
testimony, but I don’t think the professor’s testimony added anything of value. A
political science professor stating he has no reason to doubt a commonsense
inference about undervoting is like the director of the McDonald Observatory
testifying he had no reason to believe the daytime sky on a certain date wasn’t blue.
Calling an expert to testify he has no reason to doubt a commonsense inference is,
to the degree it is evidence, an affirmation that the commonsense inference is
reasonable. The commonsense inference would have been as reasonable and as
strong without the expert’s testimony.
The other difference between this case (and Jones) and Green is that the
contestant in Green called some, but not all, of the illegal voters, but here (and in
Jones) the contestant did not call any challenged voters. The argument goes that even
if calling thousands of witnesses would be too burdensome, the contestants should
have called some of the challenged voters.
11
This objection to statistical reasoning fails for at least two reasons. The first
is that once we realize any witnesses will have credibility problems and testify
something like “I think I did/didn’t,” it doesn’t matter how many there are. Zero, a
hundred, and a thousand are all less useful than basic statistical reasoning. That
seems to be the conclusion the trial court reached in Green when it applied the
overall undervote percentage rather than relying on witness testimony.
The second is that taking a convenience sample of a sample doesn’t prove
whether the original sample was biased. That is, if we’re trying to determine whether
the 2,891 affected voters in this case behaved like the total population of 1,107,390,
getting testimony from, e.g., 289 of the affected voters doesn’t answer that question.
Instead, it just raises the question of whether those 289 were representative of the
2,891.
Rather than sampling a sample, courts can take the original sample for what
it is. To be sure, if there was something distinctive about the sample of illegal voters
suggesting they were not representative—if they all had the same last name or had
addresses in the same precinct—that’s the sort of thing that could be pointed out to
the trial court as a reason not to make an inference. And contestees are always able
to put on witnesses explaining why the sample of illegal voters isn’t representative
of the whole. But here there was nothing rebutting the commonsense inference that
12
the challenged voters voted in the contested race at more or less the same rate as
voters at large.
V. Caveats
I’m not advocating a blanket rule for all election contests. If a trial court found
there were ten illegal voters and the contested race had an undervote of 50%, you
can’t infer anything of value from that. What makes this case such a good case for
statistical reasoning is 1) the fairly large number of affected votes and 2) the very
low undervote percentage. As the number of affected votes gets smaller and the
undervote gets larger, inferences becomes less useful. Perhaps more marginal cases
would need expert statistical testimony, but Judge Peeples’s inferences here were
justified based on ordinary reason and common sense.
I will make one suggestion for how inferences could be better. Judge Peeples
applied the overall undervote percentage to all affected votes. But we know from the
final results that the undervote varied for different types of votes. For instance, in
the election challenged here, mail-in ballots had an undervote of 8.3%, early voting
ballots had an undervote of 3.2%, and election day ballots had an undervote of 4.4%.
Rather than applying the overall undervote percentage to all ballots, it would be
better to apply the corresponding percentage to each subset. But this approach would
not have created a substantial difference in the final number of affected votes.
13
VI. The Big Picture
No statute or constitutional provision requires that courts use the “clear and
convincing” burden of proof for election cases. It’s purely a matter of common law,
and as such courts should keep an eye to ensuring its application is appropriate.
Anytime the undervote is relevant in an election challenge, it’s only because
the contestant has already proved the number of illegal ballots exceeds the margin
of victory in an election. Without putting too fine a point on it, that’s a bad situation.
If, in such bad situations, courts use unnecessarily high burdens of proof to make
election challenges practically impossible, the public may become skeptical of all
election results, and justifiably so. But if the judiciary allows contestants to use
general statistical reasoning in large election contests, as Judge Peeples did here, we
can render results capable of assuring the public that the true winner won, even if
the election administration was not the pinnacle of excellence.
Clint Morgan
Justice
Panel consists of Justices Rivas-Molloy, Gunn, and Morgan.
Morgan, J., concurring.
Publish.
14