Were the allegations of contestor’s petition supported by the required weight of the evidence, we would be required to set aside the election for Attorney General. However, our thorough examination of the evidence and arguments of the parties cause us to conclude that contestor has not met his burden of proof to overturn the election. Therefore, for the following reasons, we enter judgment for contestee.
I
Burden of Proof
We begin our analysis by restating established, fundamental principles applicable to the involvement of courts in election matters:
“This court is very much aware of the limitations on its power to intervene in the election of public officials. The right to vote has its source in and is guaranteed by the Constitutions of Ohio and of the United States. Since public elections belong to the political branch of the government, they are a matter of political regulation and questions arising in reference to elections are subject to judicial review or cognizance only in limited areas * * *. [Citations omitted.] As such, courts should be very reluctant to interfere with the election of public officials by the people, except to enforce rights or mandatory or ministerial duties as the statutes require. *105The survival of our system of government requires that proper respect be given to the will of the people as expressed at the ballot box. * * *” MacDonald v. Bernard (1982), 1 Ohio St. 3d 85, 86, 1 OBR 122, 122-123, 438 N.E. 2d 410, 411-412.
Moreover, we have long acknowledged that in an election contest courts exercise delegated political authority, not judicial authority:
“Elections belong to the political branch of the government, and the power conferred by the General Assembly of the state through authorization by Section 21 of Article II of the state constitution to ‘determine, by law, before what authority, and in what manner, the trial of contested election shall be conducted’ is not judicial power within the meaning of Section 1 of Article IV of the Constitution.” Foraker v. Perry Twp. Rural School Dist. Bd. of Edn. (1935), 130 Ohio St. 243, 4 O.O. 264, 199 N.E. 74, paragraph one of the syllabus. See, also, Williams v. O’Neill (1944), 142 Ohio St. 467, 27 O.O. 400, 52 N.E. 2d 858, paragraph one of the syllabus; McClintock v. Sweitzer (1941), 138 Ohio St. 324, 20 O.O. 383, 34 N.E. 2d 781; Thompson v. Redington (1915), 92 Ohio St. 101, 110 N.E. 652; and Link v. Karb (1914), 89 Ohio St. 326, 104 N.E. 632, paragraph four of the syllabus.
Accordingly, we have adopted stringent standards for granting relief in election contests. In Mehling v. Moorehead (1938), 133 Ohio St. 395, 408, 11 O.O. 55, 60, 14 N.E. 2d 15, 21, we stated that “* * * unless it is shown that the [election] result was contrary to the will of the electorate, it will not be disturbed.”
However, in Otworth v. Bays (1951), 155 Ohio St. 366, 44 O.O. 344, 98 N.E. 2d 812, at paragraph one of the syllabus, we stated:
“Where irregularities in an election are so great and so flagrant in character as to render it impossible to separate the illegal from the legal votes and raise a doubt as to how the election would have resulted had such irregularities not occurred, they must be deemed fatal to the validity of the election and warrant the rejection of the entire vote of the election district.”
Finally, in In re Election of Swanton Twp. (1982), 2 Ohio St. 3d 37, 39, 2 OBR 581, 582, 442 N.E. 2d 758, 759-760, we stated:
“In applying these standards [of Mehling and Otworth], this court has consistently held that there must be an affirmative showing that enough votes were affected by the alleged irregularities to change the result of the election.”
The message of the established law of Ohio is clear: our citizens must be confident that their vote, cast for a candidate or an issue, will not be disturbed except under extreme circumstances that clearly affect the integrity of the election.
The controlling cases require a contestor to prove two facts: (1) that one or more election irregularities occurred, and (2) that the irregularity or irregularities affected enough votes to change or make uncertain the result of the election.
We observe that previous election contest cases have nofi, clearly established the contestor’s burden of proof. For example, Swanton Twp.’s requirement of an “affirmative showing” could mean either a preponderance of the evidence or clear and convincing evidence. We adopt the clear and convincing evidence standard for these reasons. First, as we have stated, courts must be restrained in invalidating elections. Second, in McClintock v. Sweitzer, supra, at 327, 20 O.O. at 384, 34 N.E. 2d at 783, we approved the clear and convincing *106evidentiary burden for election contests in which the alleged irregularity involved fraud. Third, the relief sought — the rescission of an election — is equitable in nature. See R.C. 3515.11.
In Cross v. Ledford (1954), 161 Ohio St. 469, 53 O.O. 361, 120 N.E. 2d 118, at paragraph three of the syllabus, we defined “clear and convincing evidence” as:
“*** [«pjlmt measure or degree of proof which is more than a mere ‘preponderance of the evidence,’ but not to the extent of such certainty as is required ‘beyond a reasonable doubt’ in criminal cases, and which will produce in the mind of the trier of facts a firm belief or conviction as to the facts sought to be established.”
This is contestor’s burden of proof in this case. We now turn to the issues he raises.
II
Fraud
Contestor first argues that each of the alleged election violations constitutes fraud under R.C. 3599.42, which provides:
“A violation of any provision of Title XXXV of the Revised Code constitutes a prima-facie case of fraud within the purview of such title.”
For the purposes of this analysis, that characterization is not helpful. Not all the grounds raised by contestor are “violations” of R.C. Title 35. For that matter, some violations do not constitute the type of “fraud” that would warrant changing the election result.
For example, in this case, contestor cites the mechanical disrepair of certain optical scanning machines as one reason to invalidate the election. This apparently invokes no allegation of fraud or statutory violation. Similarly, some violations of R.C. Title 35 could hardly constitute fraud that would cause the invalidation of an election. Consider a violation of R.C. 3501.35, which prohibits loitering or congregating within the area between the small flags of the United States and the polling place. Would such activity constitute “fraud” under R.C. Title 35 to warrant overturning an election? We hold that it is immaterial which type of irregularity is alleged in an election contest. The inquiry remains the same: whether there is clear and convincing evidence that an irregularity occurred and, if so, whether there is clear and convincing evidence that it affected enough votes to change or make uncertain the result of the election.
Ill
Improper Ballot Rotation
A
Contestor’s strongest argument is that the Mahoning County Board of Elections improperly rotated the order of names on the ballots in Mahoning County in violation of Section 2a, Article V, Ohio Constitution, and R.C. 3505.03. Section 2a of Article V provides in part:
“The names of all candidates for an office at any election shall be arranged in a group under the title of that office. The general assembly shall provide by law the means by which ballots shall give each candidate’s name reasonably equal position by rotation or other comparable methods to the extent practical and appropriate to the voting procedure used. * * *”
R.C. 3505.03 provides in part:
“The names of all candidates for an office shall be arranged in a group under the title of that office, and, except for absentee ballots or when the number of candidates for a particular office is the same as the number of candidates to be elected for that office, shall be rotated from one precinct to another. * * *
“The method of printing the *107ballots to meet the rotation requirement of this section shall be as follows: The least common multiple of the number of names in each of the several groups of candidates shall be used and the number of changes made in the printer’s forms in printing such ballots shall correspond with such multiple. The board of elections shall number all precincts in regular serial sequence. In the first precinct, the names of the candidates in each group shall be listed in alphabetical order. In each succeeding precinct, the name in each group which is listed first in the preceding precinct shall be listed last and the name of each candidate shall be moved up one place. In each precinct using paper ballots, the printed ballots shall then be assembled in tablets.”
In early to mid-September, a bipartisan group from the Mahoning County Board of Elections met with Walter Labazon, a representative from the company that had the contract to print the ballots. Those attending included board member William C. Binning, director James Dellick, both Republicans, assistant director Mark Hanni, and employee James Tablack, both Democrats. Although Binning did not remember attending the meeting, he did not deny that it took place. Labazon informed the group that he could not print the forty-two different ballots required for proper rotation of candidates’ names in time for the November 6 election. The group then authorized printing seven rotations, knowing it violated Ohio’s election laws.
No evidence shows that this matter was ever formally placed before the full Mahoning County Board of Elections, or that any effort was made to publicize the informal decision. The printer, Youngstown Lithographing, delivered the seven-ballot rotation to the board “about a week” before the election.
Contestor agrees, and we find, that there is no evidence that the ballot rotation was designed or approved to benefit any candidate.
Contestee appears to admit that the improper rotation technically violated Section 2a, Article V of the Ohio Constitution, R.C. 3505.03, and Directive No. 90-41 of the Secretary of State issued September 7, 1990, but argues that the “imperfect” rotation was not legally significant because (1) there was no fraud; (2) the decision to use the seven-ballot rotation was made to permit the election to take place and not made with intent to favor or disfavor any candidate; and (3) the difference of only thirty extra precincts in which contestee’s name was wrongly placed first represents substantial compliance with the law, given that no other improper ballot rotation was alleged in the state’s other 13,624 precincts.
We find that contestor has established by clear and convincing evidence the fact of an illegal ballot rotation.
We reject the argument that the people or the General Assembly require only “substantial compliance” with precinct-by-precinct rotation. Moreover, as we have stated, we find it irrelevant in an election contest whether an alleged irregularity was motivated by evil or benign purposes. Our focus is on the irregularity. The evidence before us suggests that the group that authorized the seven-ballot rotation acted without improper motives in order to allow the election to go forward on November 6. However, we do not find that their decision, made six to eight weeks before the election on the printer’s statement alone that proper rotation could not be accomplished in time, and without any apparent attempt to meet the obligations imposed by R.C. 3505.03, excuses noncompliance with the statute.
Similarly, we disregard contestor’s *108argument that the board allowed the printer to post an insufficient bond and that a proper bond could have provided a fund for additional printing costs of the proper rotation. This evidence, if true, would provide only an additional reason why the seven-ballot rotation constituted a potentially actionable ballot irregularity. We find that there was such an irregularity on other grounds.
B
Having established that the seven-ballot rotation was a statutory violation, contestor must prove by clear and convincing evidence that it, alone, or in combination with other irregularities, affected enough votes to change or make uncertain the result of the election. Contestor argues that illegal rotation of his and contestee’s names on the Mahoning County ballot deprived him of enough votes to change the result of the election. In his petition, he claims that contestee’s name appeared in the first position on Mahoning County ballots approximately 13,553 more times than his. Contestor arrives at this figure by subtracting from the product of four-sevenths (the fraction of precincts in which contestee’s name appears first on Mahoning County ballots) of 94,869 (the total number of votes cast in the Attorney General’s race in Mahoning County), three-sevenths of that total vote. However, in his brief, he estimates the number of “illegal ballots” to be approximately fifty thousand, by counting the number of ballots cast in the precincts in which the wrong candidate’s name appeared first on the ballot. He then argues that whenever improper rotation occurred in a precinct, regardless of whether his or contestee’s name was placed first, the ballots are “illegal.” Moreover, he argues that this precinct-by-precinct analysis shows that contestee received 15,970 more first place ballot positions than he did, not the 13,553 stated in the petition. (The difference in these two numbers results from the fact that unequal numbers of votes are cast in different precincts; thus, counting the exact number of votes cast in precincts in which the wrong candidate’s name was placed first produces a total different from applying a fraction to the total vote cast.)
To support these claims, contestor first argues that this many “illegal” votes is sufficient to overturn the election as a matter of law. Further, he contends that expert testimony on the subject of ballot rotation is inadmissible because the people, in Section 2a, Article Y of the Ohio Constitution, and the General Assembly, in R.C. 3505.03, have determined that precise ballot rotation by precinct is required.
He appears to reason that the adoption of Section 2a of Article V in 1949 fixed the principle of substantially equal ballot rotation on a voter-by-voter basis; that our decision in State, ex rel. Roof, v. Bd. of Commrs. (1974), 39 Ohio St. 2d 130, 68 O.O. 2d 85, 314 N.E. 2d 172, precipitated the 1976 amendment of the section and delegated to the General Assembly the duty to provide for proper ballot rotation; that the General Assembly amended R.C. 3505.03 to require ballot rotation by precinct; and that any deviation from the precinct-by-precinct ballot rotation requirement of R.C. 3505.03 will invalidate an election, or at least invalidate any election where the number of “illegally” rotated ballots exceeds the winner’s margin of victory. He notes that both the Roof decision and the report of the Ohio Constitutional Revision Commission, which recommended the 1976 amendment to Section 2a of Article V, cited Note, California Ballot Position Statutes: An Unconstitutional Advantage to Incumbents (1972), 45 So. Cal. *109L. Rev. 365, which stated the following conclusion at 376:
“ * * * [0]ne can attribute at least a five percent increase in the first listed candidate’s vote total to a positional bias. Rather than an average bias, this figure represents a ‘safe minimum’ that will be exceeded in most elections Hi * * ft
Contestor implies that this study is now part of the law of Ohio.
First, we acknowledge that since 1949, when Section 2a of Article V was adopted, the people have determined that substantial equality of ballot position must be achieved, and that the law now requires precinct-by-precinct ballot rotation to accomplish this. We do not find, however, that any deviation from this requirement necessarily invalidates an election. Such a conclusion would ascribe a supervening preeminence to the principle of ballot position equality. Contestor’s view of improper ballot rotation suggests that every vote for every candidate named first in a precinct with improperly rotated ballots is tainted by position bias. We find little in law or social science to support this view.
Both former Chief Justice O’Neill, in Roof, and the 1977 report of the Ohio Constitutional Revision Commission cited the Southern California law review article as evidence that ballot position has an effect. That, presumably, was the evidence before them, or the evidence they chose to examine.
Similarly, in McLain v. Meier (C.A.8, 1980), 637 F. 2d 1159, cited by contestor, the court found that ballot position bias in North Dakota violated federal equal protection guarantees, citing the same law review note. The trial court had assumed a ballot position effect of five percent in that case, and the appellate court noted that “[n]o contrary evidence was introduced.” Id. at 1166. We conclude from these cases that improper ballot rotation could have an effect, but the extent of this effect depends on the record developed in each case.
Moreover, although the Ohio Constitutional Revision Commission also cited the above-quoted conclusion that a position bias of five percent or more may occur, we do not find a sufficient reason to believe that the people adopted this or any other percentage as a fixed value to be applied in all cases. The primary difference between the constitutional provision of 1949 and the provision after the 1976 amendment was that after 1976 the General Assembly was required to determine “the means by which ballots shall give each candidate’s name reasonably equal protection by rotation or other comparable methods to the extent practical and appropriate to the voting procedure used.” The former, self-executing constitutional provision had required that candidate’s names “shall be so alternated so that each name shall appear (in so far as may be reasonably possible) substantially an equal number of times at the beginning, at the end, and in each intermediate place, if any, of the group in which such name belongs.”
In R.C. 3505.03, the General Assembly fulfilled its duty by requiring precinct-by-precinct ballot rotation. Nothing the people or the General Assembly did in changing the Constitution and implementing the change suggests that they approved any fixed percentage that courts must use in evaluating the effects of position bias.
Accordingly, we hold that, in an election contest based upon a violation of R.C. 3505.03, the contestor must prove by clear and convincing evidence that the statute has been violated and that the effect, if any, of the positional bias created by such violation changed *110or made uncertain the result of the election.
In this case, the parties submitted evidence from three expert witnesses to show the effects of the improper ballot rotation on the race for Attorney-General in Mahoning County. Dr. Térry Buss of the University of Akron testified for contestor that in some cases “when they’re [voters] not sure what the names are, they will choose the first name that is presented to them on the ballot.” In addition to name identification, he also testified that other factors influencing voter choice are party identification and whether a particular race appears at the beginning, middle, or end of the ballot. He said that a proper evaluation of a particular race would include analysis of all these factors. “Once you do that and take out the effects of all those things, it’s possible that one might observe a 5 or even 10 percent change in the preferences of voters as a result of that. It’s a statistical probability after having done that.” Asked on what basis he formed that opinion, he replied that: “it’s something that’s standard in our curriculum.” He also testified that he knew of no particular factor in the race for Attorney General in Mahoning County that would have “rendered inapplicable * * * [his] standard range of 5 to 10 percent differential in nonrotation situations.”
However, on cross-examination, Buss also admitted that he had made no study of the significance of ballot rotation in the race for Attorney General in Mahoning County. He said that he had no opinion on the effect of positional bias in that race and that it was possible for ballot position to have no effect in some contests. Without fully evaluating the aforementioned factors, he said he could not express any opinion as to whether the significance of ballot position in this race would be a “small percentage, large percentage, or no percentage.” Dr. Buss referred almost exclusively to possibilities rather than probabilities. His testimony was thereby rendered virtually nonprobative on the effect of rotation bias.
Contestee called two expert witnesses. Dr. Jon A. Krosnick, Assistant Professor of Psychology and Political Science at the Ohio State University, testified that there are circumstances in which a candidate’s position on the ballot can affect the number of votes the candidate receives. He also testified that significant factors involved in determining the effect of ballot rotation are the number of persons competing for an office, political party identification, the degree to which voters are informed about candidates in a particular race, the position of a particular race on the ballot, and whether voting in every race is compulsory. He further testified that ballot position would tend to be more significant where there are more than two candidates. He added that party identification on the ballot should reduce the effect of ballot position because it provides useful information to the voter. He said that a high degree of publicity about a race would tend to decrease the effect of ballot position, and that the position of a race close to the front of the ballot would tend to decrease the magnitude of position effects. Further, he said that noncompulsory voting also decreases the significance of ballot position.
Unlike contestor’s expert witness, Krosnick made a study of the 1990 Attorney General’s race in Mahoning County and concluded that ballot position had no effect. He began by separating the county’s four hundred twenty precincts into two groups: the two hundred forty precincts using ballots with contestee’s name first, and *111the one hundred eighty precincts using ballots with contestor’s name first. He then examined “a series of characteristics of those precincts to determine whether the precincts assigned to those two groups by the assignment method were quite similar to one another, as would be expected through a random assignment procedure.”
Dr. Krosnick then looked at three characteristics: the number of voters registered in each precinct, the number of voters who actually cast ballots in each precinct, and the average percentage of turnout per precinct. He found the two groups of precincts similar.
He did not compare demographic factors of the two groups. On cross-examination, he agreed that “we really don’t know much about the demographic similarities of these precincts,” but he did not consider this information necessary because “widely accepted scientific methods dictates [sic] that controlling for these factors is not necessary if, in fact, random assignment of ballot orders to voters has taken place.” He concluded that the two groups of precincts so closely resembled each other in the characteristics he had examined that he had “great confidence that it is * * * highly probable” that random assignment had taken place.
He then computed the percentage of votes each candidate received in each group of precincts. Contestee received 63.788 percent of the vote in the precincts where his name came first, and 63.658 percent in the precincts where contestor’s name came first, a difference of only 0.13 percent. Thus, contestee “did slightly better when” his name was listed first.
He stated that factors other than ballot position could explain this difference. Although he considered the two groups of precincts similar, he did not expect them to be identical. For example, both groups should have similar, but not identical, proportions of male voters, college-educated voters, and supporters of contestee. Such differences between the two groups might explain the different results. Therefore, he performed a statistical significance test to see whether the 0.13 percent difference in results could be attributed to ballot position or chance. The test compared the size of the difference between contestee’s percentages in the two groups of precincts to the size of variation in voting within those precincts in order to determine whether the difference between the groups was likely to be due to chance alone or to the candidates’ different ballot positions. The test revealed “almost near certainty” that the difference was due to “chance alone.”
Krosnick thus concluded, to a reasonable degree of scientific certainty, that “the ordering of the names on the ballot in this race in Mahoning County had no impact on the outcome.”
Contestee also called Dr. Robert Darcy, Professor of Political Science and Statistics at Oklahoma State University. Dr. Darcy had previously performed studies of ballot position in Oklahoma and Colorado and had concluded in the Oklahoma study that there was no position effect, and in the Colorado study that ballot position had no statistically significant effect. When discussing prior research in this field, he stated “that there were no instances in the literature where an American style election that involves choosing between several candidates, making one choice between several candidates with party identification provided, there was no example in the literature of a significant difference produced by ballot position.”
Darcy’s analysis and conclusion *112relate to the 1990 Attorney General race in Mahoning County and were similar to Dr. Krosnick’s. Darcy concluded that ballot position had no effect on the outcome of the race.
Contestor argues that a ten percent ballot position bias applied to contestee’s 1,234 vote margin of victory will reverse the outcome of the election. However, as stated, we find little probative evidence supporting the existence of such a bias.
There is little doubt that in some elections ballot position can affect the results of the election. However, all three expert witnesses did agree that position effect could be reduced or eliminated by certain factors. Those factors are party identification on the ballot, position of the race toward the top of the ballot, significant preelection publicity, and the American practice of noncompulsory voting. Although there was no specific evidence offered to attempt to quantify the publicity this race received, we presume that there was a significant amount. All other factors were present in this election — party identification was provided; the race was second on the ballot, after the race for Governor; and voting in the race was, of course, not compulsory. Accordingly, we find evidence that these factors reduced the significance of ballot position in the Attorney General’s race.
Moreover, we find additional evidence of a minimal effect of ballot position in the analyses performed by Drs. Krosnick and Darcy. While we do not necessarily endorse their analyses as the definitive models for this type of case, we find them more persuasive than the testimony of Dr. Buss, who performed no analysis of this race and yet was willing to presume a five to ten percent ballot position effect, even though he admitted that the factors we have just mentioned could reduce the percentage to zero. Accordingly, although we cannot say precisely what percentage affects vote totals in this case, we find from the range of alternatives presented by the evidence — from zero percent to ten percent — that the correct percentage would be closer to zero than ten percent.
However, even a low percentage may reverse an election if it affects enough votes. Thus, we must determine to which votes the percentage should be applied. Identifying this presents a significant problem because expert testimony has not made clear whether the percentage should be applied to total votes or total votes received by the candidate whose name was wrongly placed first on the ballot. Accordingly, we have made both computations.
R.C. 3505.03 is clear:
“* * * The board of elections shall number all precincts in regular serial sequence. In the first precinct, the names of the candidates in each group shall be listed in alphabetical order. In each succeeding precinct, the name in each group which is listed first in the preceding precinct shall be listed last and the name of each candidate shall be moved up one place. * * *”
Contestor has shown that on a perfect model, when a seven-ballot rotation is used for four hundred twenty precincts instead of a forty-two-ballot rotation, the ballots in every other set of seven precincts are improperly rotated, i.e., the wrong name is listed first in every other set of seven. Then, by applying this model to the list of precincts actually used for rotation purposes in Mahoning County and the official returns from those precincts with improper rotation, contestor shows that he received 16,774 votes and that contestee received 26,422 votes, and he concludes that the difference of 9,648 votes represents a “net gain” to contestee.
We agree with contestor’s basic *113theory, but not his totals or conclusion. Our own analysis shows that the precinct list actually used was not a perfect seven-ballot rotation, the sequence having been interrupted in two places before the seventh precinct in the set of seven was reached, although, in total, contestee’s name was first in two hundred forty precincts, and contestor’s name was first in one hundred eighty. We observe that the wrong candidate’s name was placed first on the ballot in two hundred six precincts, one hundred eighteen with contestee’s name first and eighty-eight with contestor’s name placed first.
As suggested by contestor, this process allows us to identify the exact precincts in which' each candidate’s name wrongly appeared first and, by referring to the vote returns by precinct, Exhibit 58, we are able to ascertain that 24,726 votes were cast in precincts in which contestee’s name was wrongly placed first and 18,144 were cast in precincts in which contestor’s name was wrongly placed first.
If a position effect of ten percent, the highest percentage suggested by expert testimony and the percentage argued by contestor in his brief, is applied to these totals, the products are 2.473 votes and 1,814 votes, respectively (ten percent of 24,726 equals 2.473 and ten percent of 18,144 equals 1,814). The difference is 659 votes in favor of contestee, which is enough to change the result of the election, the evidence having shown that 618 votes (one half of contestee’s 1,234 vote margin of victory, plus one) would have to be shifted from contestee to contestor to change the result of the election. Actually, a ballot bias of only 9.39 percent would be sufficient (9.39 percent of 24,726 minus 9.39 percent of 18,144 equals 618). However, as we have stated, we find no evidence to support either of these percentages.
We reach the same conclusion if the computation is based on votes the candidates actually received in precincts in which their names were wrongly placed first. Using these numbers, we find that contestee received 15,288 votes in precincts where his name was wrongly placed first, and contestor received 7,298 in precincts where his name was wrongly placed first. To produce the necessary 618 vote shift would require applying a bias percentage of 7.73 percent to these totals (7.73 percent of 15,288 equals 1,182 and 7.73 percent of 7,298 equals 564), with the difference being 618 votes.
Even the totals suggested by contestor require finding a position bias of 6.405 percent to produce the required 618 vote shift (6.405 percent of 26,422 minus 6.405 percent of 16,774 equals 618).
Again, the evidence simply does not support either of the percentages.
Accordingly, we find that contestor has not sustained his burden of proof on this issue by clear and convincing evidence.
C
Contestee argues that contestor is estopped to contest the election because the improper ballot rotation was a matter of public record that could have been discovered before the election. We reject deciding the case on this basis.
We have held that equitable estoppel must be determined on the particular facts of each case. Hampshire Cty. Trust Co. v. Stevenson (1926), 114 Ohio St. 1, 11, 150 N.E. 726, 729. In cases in which this court has found estoppel in an election contest, including the cases cited by contestee in support of his argument, the irregularities were plain on the face of the ballot, and the contestors were well aware of the alleged defects before the *114election. Improper ballot rotation is not so evident. Even if the ballots were available before the election, a candidate would have to be forewarned of the legal requirements for rotation, informed as to how to calculate a proper rotation, and vigilant enough to check the ballots at all eighty-eight counties in a statewide race.
Accordingly, we hold that absent actual knowledge by the candidate or his or her agents of improper ballot rotation before an election, estoppel may not be invoked to dismiss a contest of that election based on improper rotation.
IV
Optical Scanners
Contestor asserts that optical scanning machines used by the Mahoning County Board of Elections produced unreliable results because of differences in the number of overvotes and undervotes they recorded. An overvote occurs when there appears to be a vote for both candidates in a race; an undervote occurs when there appears to be no vote for either candidate.
For the recount, the board had ordered only one program from the machine’s manufacturer, which would cause the machines to reject, as uncountable, overvotes and undervotes in the race for Attorney General and also a race for common pleas judge, which was also being recounted.
According to contestor, on December 10, 1990, approximately one hundred thousand ballots were run through the optical scanning machines, which “kicked out” about six thousand ballots as either overvotes or under-votes in either or both races. Teams of board employees, composed of one Republican and one Democrat, then sorted these six thousand ballots into three piles. The first pile contained ballots the employees believed were kicked out as overvotes or undervotes in the race for Attorney General, the second pile contained ballots the employees believed were kicked out as overvotes or undervotes in the judicial race, and the third pile contained ambiguous ballots requiring examination by the board. When this sorting was complete, employees ran piles one and two through the machines with the kick-out mechanisms deactivated to be counted again.
This process was then repeated on December 11, 1990 for the recount of the Attorney General’s race. The six thousand ballots originally kicked out were again separated into three piles by employees making the same assumptions. Then about five thousand seven hundred ballots (six thousand minus three hundred seventeen, which were deemed ambiguous) were run through the machines to be recounted.
Then, according to contestor, all ninety-four thousand ballots that had previously been counted on December 10 were run through the machines on December 11, and this produced more kicked-out overvotes and undervotes on December 11 than had been kicked out on December 10. When the board’s employees attempted to run these additional kicked-out ballots through the machines to see if they could be counted on subsequent attempts, contestor’s representative protested, and these ballots were then added to pile three to be examined by the board.
Subsequently, on January 9, 1991, using eight machines, contestor ran tests on ten ballots that had been counted on December 10, but kicked out on December 11. This test showed that as to seven of the ten ballots, all eight machines varied in whether they would count or kick out a ballot. Contestor ran another test on these ten ballots with the kick-out mechanisms *115deactivated. This test showed considerable variance as to whether a machine would count a vote as for a candidate or as an undervote. At the extremes, machines three and six counted five votes for contestee, two votes for contestor, and three under-votes; machines seven and nine counted one vote for contestee, one vote for contestor, and eight under-votes. No machine counted more votes for contestor than for contestee.
A
Contestor raises several objections to the recount process. First, he argues that R.C. 3515.04 mandates separate recounts and that the board violated the statute by having the machines kick out overvotes and undervotes in both contested races on the same day. R.C. 3515.04 provides in part:
“* * * [T]he board shall not permit the counting or tabulation of votes shown on the ballots * * * for election to any office or position, * * * other than the votes shown on such ballots for the * * * election * * * concerning which a recount of ballots was applied for.”
While it was probably a practical error to order only one program to conduct the recounts, we do not find a violation of R.C. 3515.04. The words “election,” “office,” “position,” and “recount” therein may be read as plural. See R.C. 1.43(A). Therefore, we find that R.C. 3515.04 requires separate action on ballots subject to recount from action on ballots not being recounted, not separate action on various recounts. Accordingly, we hold that contestor has failed on this issue to meet his threshold burden of showing an irregularity by clear and convincing evidence.
B
Contestor next argues that it was error for board employees to decide whether the kicked-out ballots were attributable to the race for Attorney General or the judicial race and then run these ballots through the machines again. This, he claims, affected at least five thousand seven hundred ballots, enough to reverse the result of the election.
Several witnesses testified that machines have different sensitivity levels. One scanner may read a ballot that another scanner would “kick out” because it was unable to detect a pencil mark. Therefore, we find no irregularity in attempting to cause a machine to read a ballot that another machine could not read. In a perfect election, all votes would be machine-counted as a vote for a candidate. Admittedly, in this case, the recount program caused employees to make an extra decision when a ballot was kicked out — namely, which race caused the overvote or undervote. However, we are not persuaded that this additional judgment significantly affected the process because, by contestor’s own account of events during the December 11 recount of the Attorney General’s race, all ballots thought to be rejected in either race were run through the machines again for recounting. Thus, it made no difference if a ballot was put in the first pile or the second pile. Both were run again in an attempt to have machines count them in a process we do not fault. Accordingly, we find that contestor has not sustained his burden of proof by clear and convincing evidence on this issue.
C
Contestor next argues that his tests prove how unreliable the board’s machines were. Contestee argues that *116since this type of machine has been approved for use in Ohio, as required by R.C. 3506.05, contestor’s evidence to the contrary is inadmissible, citing State v. Vega (1984), 12 Ohio St. 3d 185, 12 OBR 251, 465 N.E. 2d 1303. Vega holds that the reliability of intoxilyzers, having been legislatively determined, may not be generally attacked by an accused. Admittedly, some of the contestor’s statements appear to attack the general reliability of optical scanning machines. However, the tests he performed go to the reliability of the eight machines tested, not all machines. We therefore reject contestee’s assertion.
We find that contestor’s tests prove no more than was apparent from testimony — that different sensitivity levels of machines will produce greater or fewer numbers of overvotes and undervotes. Most significantly, contestor’s tests do not establish that any vote was changed from one candidate to another. Moreover, contestor could have verified his charges by asking for a court-supervised recount under R.C. 3515.13 as part of this election contest. He did not. Accordingly, we find that contestor has not sustained his burden of proof by clear and convincing evidence on this issue that an irregularity occurred.
V
Remake Ballots
Contestor asserts that discrepancies in the process of remaking torn and tattered ballots in Mahoning County cast doubt upon the validity of the election. First, he asserts that the number of “remake” ballots was not reconciled with the number of torn and tattered ballots at the December 11 recount, but was only later reconciled by board employees on January 3, 1991, out of the presence of witnesses.
Michael Slanker, a representative of contestor, testified that, at the recount on December 11, an attempt to reconcile the torn and tattered and remake ballots showed that there were four fewer remake ballots than torn and tattered ballots. Slanker also testified that James W. Dellick, Director of the Mahoning County Board of Elections, told him that he (Dellick) thought he knew where the four ballots were, but that finding them would be a lengthy process.
Dellick testified that when Slanker asked for the reconciliation, to the best of his (Dellick’s) knowledge there were thirty-seven torn and tattered ballots, but he could not say how many remake ballots were then counted. Dellick stated that he was not especially concerned about reconciliation at that time because from past experience he knew that remake ballots were sometimes put back with other ballots in the precincts from which they were drawn. He further stated that on January 3, 1991, he sent employees James Tablack and Charles Zamary to find the missing four ballots, which they retrieved from the containers in which the precinct ballots were kept.
Zamary testified that he was able to retrieve the four remake ballots from the precinct containers because the torn and tattered ballots were marked with a precinct number and the remake ballots filed with other ballots from that precinct were identified on the ballot as remake ballots.
Thus, contestor’s statement of events appears essentially accurate. However, he does not charge any specific irregularity to be deduced from these events alone. Instead, he offers further evidence that a handwriting expert, Dr. David A. Crown, could not authenticate the initials of Dellick and Carol Rimedio-Righetti on eight remake ballots and the stubs from which they were severed. He *117does not relate the four “missing” ballots to the eight with questionable initials except to state:
“The Mahoning County Board of Elections made no attempt to reconcile the TT and RM ballots during the recount. The ‘Infamous Eight’ clearly do not have genuine initials on them. Although Rimedio has sworn that each of her initials (C.R.) on each of the RM and TT ballots and stubs was made by her own hand, according to the world’s leading expert, they were not. No satisfactory explanation has been advanced. These eight ballots appear to be yet another symptom of a larger problem in Mahoning County. None of its results are reliable.”
Thus, contestor invites us to speculate on the depths of this problem and relate it to other issues. We decline to do so. We do not find a necessary relationship between the four “missing” and later retrieved remake ballots and the eight ballots on which the initials were questioned. Moreover, Dr. Crown’s testimony was met by that of Dellick and Rimedio-Righetti, who testified that they placed their own initials, or those of the other, on the torn and tattered ballots, the ballot stubs, and the remake ballots in the presence of witnesses. Thus, we do not find clear and convincing evidence that either the four “missing” remake ballots or the eight questioned ballots were improperly remade.
Regarding the remake ballots, the petition states:
“* * * Later during the recount process the Deputy Director of the Mahoning County Board of Elections was observed to be in the room where the remake ballots were stored in an open tray along with one of his employees, and the Deputy Director was observed to have a short, stubby, ‘sawed-off’ pencil concealed within the palm of his hand. Subsequently, a number of additional remake ballots were ‘discovered’ and the total of the remake ballots were [sic] then determined to be complete, although the total remake ballots never did match the total number of torn and tattered ballots.”
We note that the only evidence produced on this point was testimony by Michael Slanker, who stated that he observed Deputy Director Mark Hanni enter the vault where ballots were kept with a “very short pencil” (“about 2 inches”) in the “palm of his hand.” Slanker observed Hanni from a doorway across the hall from the entrance to the vault, at a distance of fifteen to twenty feet. According to Slanker, Hanni walked to a plastic container that Slanker believed contained remake ballots. Shortly thereafter, James Tablack entered the vault, and Hanni “dropped the pencil into his pocket.” Slanker estimated that Hanni and Tablack were in the vault for fifteen to thirty seconds. Slanker saw Hanni enter the vault, observed him the entire time he was in the vault, and saw him leave the vault. Slanker did not see Hanni touch a ballot, and there is no other evidence from which we can conclude that the “sawed-off pencil” was used to violate any election law.
We find no clear and convincing evidence to support the allegations of the petition on this issue.
VI
Overvote X Votes
Contestor asserts that the process of counting “overvote X” votes in Mahoning County casts doubt on the election. The overvote X phenomenon occurs when a voter blackens ovals beside both candidates’ names (indicating a vote for both candidates) and then places an “X” over one of the ovals. This raises the question whether *118the “X” means a vote “for this candidate” or “not for this candidate.”
Contestor states that the board decided this issue as a vote for the candidate whose oval was “X’d” at the judicial recount on December 10, and the opposite way at the recount of the Attorney General’s race on December 11, when twenty-six of twenty-seven ballots so marked had the “X” through contestor’s oval.
Evidence shows that during the recount of the judicial race between judicial candidates Krichbaum and Smith, the board was presented with two ballots with the overvote X phenomenon, that the Republican board members voted that the ballots were a vote for Krichbaum, and the Democratic board members voted that they were a vote for no one. Each ballot was referred to a representative of the Secretary of State, who was present, and he either voted or suggested — the record is not clear — that the ballot be counted as, a vote for Smith, over whose oval the “X” had been placed. The board apparently so voted.
On the next day, in the recount of the Attorney General’s race, the vote was consistently given to the candidate whose oval was not marked with an “X.” Of the twenty-seven ballots selected for review by contestor, eighteen were decided by the board, and nine were decided by the Secretary of State’s office in order to break a tie vote by the board. Contestor alleges an irregularity by treating this phenomenon differently on successive days. We agree that logic suggests consistency from day to day. However, the irregularity we perceive occurred at the judicial recount when the board apparently informally accepted the recommendation of the Secretary of State’s representative who was present. As to the Attorney General’s race, the board and the Secretary of State took consistent positions. Accordingly, we find no clear and convincing evidence of an irregularity with respect to the Attorney General’s race on this issue.
We note that with respect to this issue the petition states:
“* * * Each and every one of the ‘X’ marked overvote ballots appeared to be marked identically, and all but one of the forty to fifty overvote ballots in Mahoning County was cóunted as a vote for Candidate Fisher. The ‘X’ marks appeared to have been made by the same hand on affected ballots.”
Mark Hanni testified that he did not alter any ballot, nor did he know of anyone altering a ballot. Dr. Crown, the handwriting expert, could not “associate any of the Xs” and could not support the claim of the petition. Nor, examining Exhibit 25, photocopies of the ballots in question, can we. Accordingly, we find no clear and convincing evidence to support these allegations of the petition.
VII
“Walk-in” Ballots
Contestor alleges that the H.B. 237 “walk-in” ballots were not integrated with other absentee ballots as required by our decision in State, ex rel. Bennett, v. Bds. of Elections of the Counties of Ohio (1990), 56 Ohio St. 3d 1, 564 N.E. 2d 407. Judge R. Scott Krichbaum, the winner in the judicial race that was recounted on December 10, testified that he instructed his representatives at the board to try and keep the H.B. 237 ballots separate, but was not sure this was done. James Tablack flatly denied the allegation of the petition that the H.B. 237 ballots were stored separately. In his brief, contestor cites no contrary evidence in the record. Accordingly, we find no *119clear and convincing evidence supporting this allegation.
VIII
A
Contestor alleges that ballot security at the Mahoning County Board of Elections was so lax as to throw the results of the election into doubt. He first argues that ballots were stored in unsealed plastic containers at the board of elections, and that this violates a provision of R.C. 3505.31.
However, the language of the statute is applicable to polling places, not the board of elections. The final sentence of the paragraph that contains the provision cited by contestor states:
“The presiding judge shall thereupon forthwith deliver to the board such containers of ballots and the sealed sets of pollbooks, poll lists, and tally sheets, together with all other election reports, materials, and supplies required to be delivered to such board.”
Accordingly, we find no clear and convincing evidence of an irregularity on this issue.
B
Contestor next alleges that the vault where the ballots were stored in plastic containers “was left open and accessible, and there was evidence that people were observed alone in the vault with the ballots.” To support this assertion, he cites the incident where Slanker observed Mark Hanni and later Hanni and Tablack in the vault. Deputy Sheriff Dean F. Huntley, who was also present on that occasion and saw Hanni, testified that Hanni did not “have time to alter any paperwork.”
Huntley and other deputies provided security at the vault door after November 6, 1990. He stated that the deputies maintained around-the-clock surveillance at the vault door after that date. It is unclear on what date this surveillance ceased, although it may have been January 9, 1991. Huntley did testify that there was “an occasion when the vault was not in view for maybe just a moment or two, but not with someone present in it.” However, contestor does not relate his allegation of lax security on this point to anything except the time when Slanker observed Hanni in the vault. We do not find this to be clear and convincing evidence of an election irregularity. Therefore, contestor has not proved the threshold issue by clear and convincing evidence.
C
Contestor next alleges that “ballots were moved from one optical scanner to another during the recount with no procedure in place to ensure that a ballot was counted only once * * *.” For this proposition, he cites only the following exchange during the testimony of James Tablack:
“Q. What do you do to insure that precincts are not counted twice? What is the process, procedure to make sure that a precinct’s ballots are not counted twice?
“A. I’m not involved in that.”
We do not find from this interchange clear and convincing evidence of an election irregularity.
D
Contestor next alleges that “[e]ven within the vault ballots were not stored in sealed containers or organized in any logical fashion. Rather, some ballots were left in envelopes lying on the floor.” He neither cites testimony for this allegation nor does he relate it to any effect on the election. We do not find clear and convincing evidence of any irregularity-
*120E
Contestor next alleges that the Mahoning County Board of Elections has lost fifty-five ballots. He arrives at this conclusion by comparing the official count of ballots cast, which totaled 100,850, with the official recount, which totaled 100,795. Contestee explains this apparent discrepancy by reference to the board’s summary of its rulings on its disputed ballots, which shows forty-eight “no votes” decided by the board, and the Secretary of State’s opinion letter, which shows that the Secretary of State broke a tie in seven cases, which he called “overvotes.” These fifty-five votes were then eliminated from the official count by the recount.
We have examined the evidence and are inclined to agree with contestee’s interpretation. In any case, contestor’s allegation on this point is not supported by clear and convincing evidence.
IX
Contestor alleges prima-facie election fraud in Summit County because some precincts there recorded more votes than the number of persons signing the poll books, and some recorded fewer votes than the number of persons signing the poll books. He states that this must be due to “extremely sloppy election procedures or fraud,” and that “one must view the results in Summit County with skepticism since the procedures evidence a lack of reliability * * *.”
Alexander R. Arshinkoff, the Chairman of the Summit County Board of Elections, testified that in fourteen to sixteen precincts the number of ballots cast exceeded the number of persons who signed at the polls and that in about six precincts fewer votes were cast than voters who signed at the polls. When the board certified the election results on November 26, 1990, it stated:
“* * * [T]here were sixteen extra ballots counted above the number of voters. There were six ballots less than voters in certain precincts.”
Accordingly, we find clear and convincing evidence of an irregularity in Summit County. However, contestor does not relate this irregularity in a way that would affect the outcome of the Attorney General’s race. Accordingly, we find that he has not produced clear and convincing evidence to show how votes were affected.
X
Geauga County
Contestor alleges that a voting machine in a particular Geauga County township precinct showed a substantially different count in its vote totals than did the other voting machine in that precinct. To establish this, he called as a witness Dean J. Schanzel, a member of the Geauga County Board of Elections. Schanzel testified:
“* * * [T]he only conclusion that I was able to draw was that at some time during the course of the election the counter behind Paul Pfeifer’s name on that machine either began to slip or quit registering votes altogether.”
In Exhibit 94, contestor shows that only 58.9 percent of voters voting on the allegedly broken machine cast votes in the Attorney General’s race, whereas 94.9 percent of voters voting on the other machine in that precinct voted in that race. Exhibit 94 also establishes that on the allegedly broken machine contestor received only forty-three of one hundred eighty votes cast, whereas on the other machine contestor received one hundred one of one hundred seventy-five votes cast. Contestor did not have the machine or its counter examined. The board considered having the machine examined, but decided not to, reason*121ing “that this counter would end up in court in some level of discussion * * * 1}
Contestor’s evidence permits a strong inference of a malfunction of the machine in question. Exhibit 94 reveals that contestor received 40.6 percent of the votes cast in the race on the allegedly broken machine, a far lower percentage than on any of the other thirteen machines used in that township. Because the burden of proof in this case is clear and convincing evidence, we find that contestor has not proved the existence of an irregularity. Moreover, since one hundred six votes were cast for Attorney General out of the one hundred eighty votes cast on the machine, the most contestor could gain would be seventy-four votes. Even assuming contestor would have received those votes or some percentage of those votes, that fact would certainly not warrant an invalidation of the election for Attorney General.
XI
Having examined each issue raised, we find that contestor has proved by clear and convincing evidence the existence of irregularities with respect to ballot rotation in Mahoning County and the failure to reconcile voters and votes cast in Summit County. However, we find that he has not proved by clear and convincing evidence that enough votes were affected thereby to change the result of the election or make it uncertain. Therefore, pursuant to R.C. 3515.14, we enter judgment for contestee.
Judgment accordingly.
Sweeney, Holmes, Wright, H. Brown and Resnick, JJ., concur. Douglas, J., not participating.