[EM] Report on the 2006 Burlington Mayoral election.

Sun Jan 4 20:22:46 PST 2009

Since Burlington has made ballot images available, and since we only 
have, from summary results, part of the story of this election, I 
decided to analyze the images. I'm writing this as I do the work, 
both as a way to record what I find and to report it. I will 
interpret the data elsewhere, so, unless I make mistakes, there 
shouldn't be any controversy about this.

There are instructions that Burlington gives for loading the data 
into a spreadsheet. The votes are contained in a series of .prm 
files. Each record begins with a precinct and ballot number, but, I 
noticed, in some cases these are duplicated, they could represent 
different counting batches or some other unexplained anomaly.

In the Excel file I compiled, there were 9865 records, which agrees 
with the total number of ballots as reported. Burlington reports 77 
invalid ballots.

I'm not going to report the IRV results, per se, those are available 
at 
http://www.burlingtonvotes.org/20060307/2006%20Burlington%20Mayor%20Round.htm

I find 77 ballots with no choice at all; as far as the ballot images 
are concerned, these are blank. (In an audit, it's not impossible for 
some of these ballots to be found valid, it depends on rules. For 
example, no ballots showed a blank first choice and then some later 
preference. That's unlikely, this is a known and reasonably common error.)

This leaves 9788 ballots. The software they used was general-purpose 
STV software, I believe it's open source, so the report mentions the 
Droop quota, which is a simple majority of the valid ballots, if we 
assume that even overvoted ballots are valid: so the software is 
seeking, until it's found or all but two candidates are left, 4895 
votes as a majority.

Under the Burlington rules, according to the instructions at 
http://www.burlingtonvotes.org/20060307/manualverification.php , a 
ballot with equal ranking (two or more candidates at the same rank) 
is "exhausted" if more than one of the candidates is not eliminated 
when that rank is reached. Quite a number of voters overvoted in 
first rank, which will result in immediate exhaustion. I'm going to 
list all these initially exhausted ballots

71 000071-00-0640 10006 1  C01=C02 C03
31 000031-00-0172 10002 1  C01=C04 C02 C05 C03
42 000042-00-0832 10003 1  C01=C04
71 000071-00-0646 10006 1  C01=C05 C02=C04 C03 C06
71 000071-00-0641 10006 1  C01=C06 C04
31 000031-00-0397 10002 1  C03=C04
41 000041-00-0233 10003 1  C03=C04
41 000041-00-0242 10003 1  C03=C04
21 000021-00-0013 10001 1  C03=C06 C05 C01 C02
41 000041-00-0379 10003 1  C04=C06 C02 C05

The candidate names, for reference, are this:
.CANDIDATE C01, "Louie The Cowman Beaudin"
.CANDIDATE C02, "Kevin J. Curley" (Republican)
.CANDIDATE C03, "Bob Kiss" (Progressive)
.CANDIDATE C04, "Hinda Miller" (Democrat)
.CANDIDATE C05, "Loyal Ploof"
.CANDIDATE C06, "Write-ins"

Overall usage of ranks:
5 candidates ranked: 2609.
	of these, a write-in was in:
		5th rank: 96 (plus one C05 = C06 vote)
		4th rank: 67 (plus two C05 = C06 votes)
		3rd rank: 49 (plus two C05 = C06 votes)
		2nd rank: 38
		1st rank: 25.
	excluding votes with a write-in, simply or as an overvote, 2330 
voters ranked all five candidates.

(There were five candidates on the ballot, and five ranks to be 
expressed, thus full ranking was allowed even if the voter ranks a write-in.)

4 candidates ranked: 348.
3 candidates ranked: 1770.
2 candidates ranked: 3256.
1 candidate ranked:  1806.

1st rank votes (overvotes are listed with the first candidate in sequence):
C01: 119 plus 5 overvotes,  C01=C02, C01=C04, C01=C04, C01=C05, C01=C06.
C02: 2609
C03: 3809 plus 4 overvotes,  C03=C04, C03=C04, C03=C04, C03=C06.
C04: 3106 plus 1 overvote,  C04=C06
C05: 57
C06: 78.

2nd rank votes:
C01: 608 plus 2 overvotes: C01=C03=C04=C05, C01=C04
C02: 1427 plus 3 overvotes: C02=C03, C02=C04, C02=C05
C03: 2444 plus 1 overvote: C03=C06
C04: 2877
C05: 456
C06: 163

Total first and second rank votes (neglecting overvotes):
C01: 727
C02: 4036
C03: 6253
C04: 5983
C05: 513
C06: 241

Were this a Bucklin election, and if the voters voted the same 
pattern in the first two ranks, two candidates would have gained a 
majority: C03 and C04, the Progressive and the Democrat. C03, as with 
IRV, would have prevailed. No claim is made, however, that voting 
patterns would remain the same. (This will be discussed in further analysis.)

Breakdown of votes by number of ranks expressed:

1 rank:
C01: 32 (26.9% of valid votes for C01) plus 1 overvote, C01=C04
C02: 767 (17.4%)
C03: 454 (11.9%) plus 3 overvotes, all C03=C04
C04: 536 (17.3%)
C05: 5 (1.0%)
C06: 8 (3.3%)

2 ranks:
C01>C02: 10
C01>C03: 6
C01>C04: 2
C01>C05: 1
C01>C06: 1
(C01=C02>C03: 1)
(C01=C06>C04: 1)

C02>C01: 89
C02>C01=C03=C04=C05: 1
C02>C03: 203
C02>C04: 296
C02>C05: 19
C02>C06: 28

C03>C01: 33
C03>C02: 187
C03>C04: 1095
C03>C05: 39
C03>C06: 35

C04>C01: 18
C04>C02: 269
C04>C02=C03: 1
C04>C03: 870
C04>C05: 19
C04>C06: 10

C05>C01: 0
C05>C02: 5
C05>C03: 2
C05>C04: 3
C05>C06: 0

C06>C01: 0
C06>C02: 5
C06>C03: 4
C06>C04: 3
C06>C05: 0

3 ranks (by first preference):
C01: 18
C02: 451
C03: 762
C04: 499 (plus one overvote C04=C06)
C05: 14
C06: 25

4 ranks (by first preference):
C01: 9 (plus two overvotes  C01=C04, C01=C05)
C02: 69
C03: 164 (plus one overvote C03=C06)
C04: 91
C05: 4
C06: 8

5 ranks (by first preference):
C01: 40
C02: 686
C03: 1040
C04: 793
C05: 24
C06: 25

Analysis of first preference overvotes by involved candidate, 
compared to valid first pref. vote for that candidate.:

C01: 5/119	4.2%
C02: 1/2609	0.0%
C03: 4/3809	0.1%		
C04: 6/3106	0.2%
C05: 1/57	1.8%
C06: 3/78	3.8%

5-rank voters by first ranked candidate, subsorted by last ranked candidate:

C01>>C02: 8
C01>>C03: 2
C01>>C04: 11
C01>>C05: 17
C01>>C06: 2 (bottom ranked write-in!)

C02>>C01: 189
C02>>C03: 95
C02>>C04: 185
C02>>C05: 183
C02>>C06: 34

C03>>C01: 342
C03>>C02: 330
C03>>C04: 151
C03>>C05: 193
C03>>C06: 24

C04>>C01: 277
C04>>C02: 227
C04>>C03: 63
C04>>C05: 192 (plus one overvote C04>>C05=C06)
C04>>C06: 33

C05>>C01: 6
C05>>C02: 8
C05>>C03: 2
C05>>C04: 7
C05>>C06: 1

C06>>C01: 2
C06>>C02: 4
C06>>C03: 2
C06>>C04: 6
C06>>C05: 11

One statistic I'd like to compile that I haven't: there were a huge 
number of "marks on the ballot," the large majority of which were not 
officially counted. Just for starters, we have, among the voters for 
the top two in first preference, 1833 voters who *fully ranked*, so 
at least four times that number of votes were not counted from them.

The total number of spoiled or blank ballots (blanks plus overvotes 
in first preference) was 77 apparently blank plus 10 overvotes in 
first rank (which totally spoils the ballot, even if there were lower 
preferences listed).

Corrections appreciated.