Nov 09: Audit vs. Scanner Counts

A. Ballot Count Accuracy

Among our greatest concerns are the discrepancies in data where no thorough or reasonable explanation is provided by election officials. The table below, shows all districts with ballot count discrepancies.  In seven (7) of these districts, the scanner counted more ballots than were counted by hand.  In ten (10) of these districts, the scanner counted fewer ballots than were counted by hand.

Scanner Counted Ballots	Hand Counted Ballots	Difference	Percent Difference
919	904	15	1.6%
1315	1298	17	1.3%
771	762	9	1.2%
1164	1169	-5	0.4%
492	494	-2	0.4%
1046	1050	-4	0.4%
1007	1004	3	0.3%
677	679	-2	0.3%
1362	1366	-4	0.3%
449	450	-1	0.2%
1900	1904	-4	0.2%
961	963	-2	0.2%
762	763	-1	0.1%
861	860	1	0.1%
2046	2044	2	0.1%
1877	1876	1	0.1%
3357	3358	-1	0.0%

Table 1: Discrepancies in Numbers of Ballots Counted by
Hand vs. Counted by Scanner[1] in Ten Districts, November 2009 Audits

Based on observer reports, we do not believe that all of the hand counts are accurate because of the questionable counting methods observed.  On the other hand, because of these discrepancies, we also have no basis to conclude that the scanners counted all ballots accurately.

B. Vote Count Accuracy

Even considering confusion over ballots with questionable votes, an analysis of the district reports submitted to the Secretary of the State indicates that vote count discrepancies remain.

For example, the table below presents, by number and percentage, vote differences greater than 10 between hand-counted votes and machine-counted votes when all ballots with questionable votes are included[2]:

Col C Machine Totals (tape)	Col D Undisputed Vote Totals	Col E Questionable Vote Totals	Col F Overall Hand Count Totals (D+E)	Difference	Percent Difference
2042	2103	0	2103	-61	3.0%
612	541	11	552	60	-9.8%
1045	1088	0	1088	-43	4.1%
556	518	11	529	27	-4.9%
932	910	0	910	22	-2.4%
1488	1466	0	1466	22	-1.5%
1453	1435	0	1435	18	-1.2%
1279	1263	0	1263	16	-1.3%
1140	1124	0	1124	16	-1.4%
992	976	0	976	16	-1.6%
1323	1309	0	1309	14	-1.1%
1198	1184	0	1184	14	-1.2%
1420	1407	0	1407	13	-0.9%
588	567	8	575	13	-2.2%
267	253	2	255	12	-4.5%
1067	1055	0	1055	12	-1.1%
2083	2072	0	2072	11	-0.5%
496	480	5	485	11	-2.2%
775	760	4	764	11	-1.4%
465	439	16	455	10	-2.2%
387	373	4	377	10	-2.6%
973	963	0	963	10	-1.0%
552	532	10	542	10	-1.8%

Table 2: Candidate counts where Hand-Counted Votes and Machine-Counted Votes Show Discrepancies of 10 Or More Votes.

The following table shows the number of candidate counts with various levels of count differences between the optical scanners and the hand counts, considering ballots with questionable votes:[3]

Count Difference	% Of All Counts	Number of Candidate Counts
0	56.6%	427
1-3	30.9%	233
4-6	7.6%	57
7-9	1.9%	14
>10	3.1%	23
Total	100.00%	754
	Average Difference:	1.6 votes

Table 3: Distribution by Difference of Candidate Counts between
Hand-Counted Votes and Machine-Counted Votes.

Using the same data as the previous table, omitting small counts with small differences[4], the following table shows the number of candidate counts with various levels of percentages of differences between the optical scanners and the hand counts, considering ballots with questionable votes:

Range of % of Count Difference	% Of All Counts In Range	Number of Candidate Counts
0	55.3%	380
> 0 and < 0.5 %	19.5%	134
0.5 % and < 1.0 %	9.3%	64
1.0 % and < 2.0 %	11.4%	78
2.0 % and < 5.0 %	3.2%	22
5.0 % and < 10.0 %	1.2 %	8
10.0 % and greater	0.0 %	0
Total	100.0%	686
Average Difference %	0.37%

Table 4: Distribution by Difference of Significant Candidate Counts between Hand-Counted Votes and Machine-Counted Votes By Ranges Of Percent Of Differences.

We note that if we were to trust these counts as an accurate representation of the optical scanner’s counting:

• For over 25% of candidate vote counts, the machine count difference is greater that 0.5% which is the maximum level for requiring a close vote recanvass.  For statewide races the threshold is significantly lower since the maximum difference for an automatic recanvass is 2000 votes, which is normally less than 0.15%

We do not believe that all of these counts are accurate.  But we have no reason to believe that all the hand count differences can all be attributed to human counting error. For public confidence it would seem important that all unsatisfactorily explained discrepancies between machine counts and official, final, audit results should be significantly lower than the maximum threshold for automatic recanvasses.

We continue to support investigations and recounting in public of all unexplained differences over an agreed upon threshold per count.

C. “Questionable” Votes and “Undisputed” Ballots

Observations and comments from election officials indicate confusion about classifying “undisputed ballots” and about counting “questionable votes.”[5] An undisputed ballot is a ballot with no apparent problem or questionable votes on it. A questionable vote  is a mark on a ballot that may not have been read properly by the optical scanner.  Audits exhibited a variety of interpretations of what constitutes “undisputed” and “ballots with questionable votes”.  Audit statistics confirm these observations.

• Sixteen (12) districts were reported as having zero (0) ballots with questionable votes.
• On average, audits reported 2.0% of votes as questionable votes.
• The districts with the largest percentages of questionable votes reported for all candidates and races for the district was 10.8%, 8.9%, 7.9% and 5.7% questionable votes.  Yet the data show that in most cases the vast majority were counted accurately by the scanner.

The following table has some examples of candidate counts with the largest percentages of questionable votes. Note that, in general, the optical scanners seem to have counted accurately many of the votes classified by officials as questionable.

The table shows one count per municipality.  In several of these municipalities many candidate counts had similar questionable vote totals.

Col C Machine Totals (tape)	Col D Undisputed Vote Totals	Col E Questionable Vote Totals	Col F Overall Hand Count Totals (D+E)	Percent Questionable
68	58	9	67	13.2%
335	291	42	333	12.5%
975	857	118	975	12.1%
235	218	17	235	7.2%
629	589	42	631	6.7%
196	182	13	195	6.6%
647	605	41	646	6.3%
170	160	10	170	5.9%
1621	1557	70	1627	4.3%

Table 5.  Examples of Candidate Counts with the Largest
Percentage of Questionable Votes (one example per town)

The following table shows the number of questionable candidate counts with various levels of percentages of questionable votes for each candidate[6]:

Percent Difference	% Of All Counts	Number of Candidate Counts
0	29.8%	225
> 0 and   < 2 %	37.4%	282
2 % and  < 5 %	21.9%	165
5 % and <10 %	8.9%	67
10% and greater	2.0%	15
Total	100.0%	754
Average Difference:	1.9%

Table 6: Vote count differences by counts and percentages.

We note that these are huge numbers of ballots with questionable votes.  From our observations too many are classified as questionable.  The problem with too many being so classified is that it provides an opportunity to miss real discrepancies as machine undercounts can then incorrectly be attributed to “voter error”.

---

[1] The law and audit procedures often use the term “Tabulator” to refer to election machines.  We use the terms “Scanner” or “Optical Scanner” to make the report clearer.

[2] This is the most favorable interpretation of the audit reports, giving every benefit of the doubt to the accuracy of machine counts and the accuracy of hand counts.  When Total Hand Count Totals is less than or equal to the Machine Totals, then the Questionable Vote Totals are included.  When Undisputed Totals is greater than or equal to the Machine Totals then all Questionable Vote Totals are excluded.  In the remaining cases enough Question Vote Totals are included to make the difference zero.

[3] This table and the following table provide data similar to that provided by the University of Connecticut in analyzing the November 2008 post-election audit, available at: http://voter.engr.uconn.edu/voter/wp-content/uploads/2008-Nov-Hand-V10.pdf

[4] The table omits candidate counts with tape counts less than 50 votes that have differences less than 3 votes.

[5] Part of the confusion comes from as the terms “Undisputed Ballots” and “Questionable Votes”.  One term refers to ballots, the other to votes, where the process must focus at different times between classifying ballots and classifying votes.  Also the terms can add to the confusion between votes which might have been read two different ways by the scanner and votes that should have been read one way by the scanner, yet that reading would not accurately reflect the voter’s intent.

[6] Once again, This table provides data similar to that provided by the University of Connecticut in analyzing the November 2008 post-election audit, available at: http://voter.engr.uconn.edu/voter/wp-content/uploads/2008-Nov-Hand-V10.pdf