Print Friendly and PDF
only search openDemocracy.net
jon.bright

The de Borda Institute New Economics Foundation OurKingdom

Home ¦ About ¦ University Signup (closed) ¦ Public Signup (closed) ¦ Guidlines & Schedule ¦ Contact

The experiment:

Phase 1: "Ideas" forum (closed) ¦ Phase 2: "Debate" forum (closed) ¦ Phase 3: "Ballot" forum (closed) ¦ Vote (closed) ¦ The results

 

An experiment in consensus voting and e-democracy

The de Borda Institute, the New Economics Foundation (nef) and OurKingdom (openDemocracy's UK politics section) are collaborating in an innovative experiment in online consensus decision making. This is the project's homepage.

 

THE ANALYSIS

Peter Emerson

Abbreviations

AV = alternative vote (which is the same as STV)

BC = Borda count

MBC = Modified Borda Count (the technical name for a consensus vote).

NGO = non-governmental organisation

STV = single transferable vote (or AV)

Contents

The Results

The Turnouts

Further Comments and the Process

A Second Analysis

A More Mathematical Approachtin

A Composite?

A Comparison

Conclusions

Annex: The Options

Feedback

The Results

There were two constituencies, the first of university students, the second of members of the public. The results were as follows:

Option

Universities

Public

B

8th

8th

C/E1

7

4th

C/E2

4th

3rd

D/F

3rd

5th

G/H

2nd

2nd

H2

1st

6th

I/J

5th

1st

L

6th

7th

M

9th

9th

Both constituencies agreed that the options of M and B were the worst; and both accepted option L was also not good.

On the positive side, H2 was the students' favourite, I/J was the public's choice, and both constituencies gave option G/H their 2nd preference.

The Turnouts

Only 16 students voted, from a voters' list of 90. That's a 17.8% turnout. On the public side, 60 out of 170 persons voted, a turnout of 35.3%. In comparison with other votes taken over the internet, both of these figures are above average. It must nevertheless be said that a sample of 16 is far too small for any definite scientific conclusions, and even a vote of 60 persons is not the best of bases. It is still possible, however, to do an analysis.

Of the 16 students who voted, 12 submitted a fall ballot of 9 preferences, 1 a partial ballot of 6 preferences, 2 of 5 preferences, and 1 of 4 preferences.

In the public constituency of 60 voters, 34 submitted a full ballot, 2 of 8 preferences, 3 of 7, 4 of 6, 2 of 5, 2 of 4, 8 of 3, 2 of 2, 1 of 1, 1 of 0 [1] and 1 was garbled.

On the whole, therefore, even if the process was "just too complicated," to quote one participant, most people managed to state some if not all of their preferences. Furthermore, as another voter noted, "none of the options listed capture[d] my preferences. In selecting options I tried to pick those which, if implemented, would cause the least harm."

The level of comprehension, then, was actually quite high. One person made a mistake in that he/she cast two 3rd preferences and voted 9-4-3-2-3-7-6-1-8, obviously casting a 3 for a 5. The computer program reads a vote like this as a valid partial vote: 0-0-0-2-0-0-0-1-0. A second person forgot to say ‘6' and voted 3-8-7-5-1-4-2-8-9, so the computer said

3-0-0-5-1-4-2-0-0. And a third person voted 9-2-2-9-1-9-2-2-9 and said that the 1 was his/her 1st preference, a 2 was "ok" and a 9 was "those I oppose";[2] the computer read this as 0-0-0-0-1-0-0-0-0.[3]

Further Comments and the Process

Certainly, many participants found the process complex. And it was complicated. It would have been much easier, for example, to have had a multi-option debate, (admittedly two years ago), on the venue for the 2012 Olympics, for in that situation, there were just the five options ‘on the table'. In our debate on party funding, however, as everyone (now) knows, there were lots of factors to be considered, and umpteen permutations and combinations of these which could be combined into any one option.

The theory of consensus decision-making is as follows. We start with a problem: how do we fund the political process? The question, then, concerns funding, but of course the answer depends on what sort of political process is envisaged. And many comments bemoaned the fact that the UK still uses a most simplistic and adversarial electoral system, to name just one aspect. Nevertheless, we are where we are, and the question still has to be dealt with.

Now in theory, at the beginning of a decision-making process, one starts with as many opinions as there are participants. (If not, indeed, with a few more!) The purpose of a consensus debate, having accepted that there are many possibilities at the beginning (stage 1), is to gradually whittle these down to a more manageable number (stage 2), before coming to short list (stage 3), and finally, via a vote on this list, (stage 4), to arrive at the outcome of just one fully coherent option.

In stage 1 of the process, then, when participants were asked for their ideas and suggestions, many points were raised on a wide spectrum of thoughts. The consensors tried to put these into some sort of order, (stage 2), in what became a draft list of 13 options. Given the complexity of it all, we tried to simplify matters by categorising these options into groups or ‘families', and we also presented a summary. Alas, many participants still thought the process was far too convoluted, and one even resorted to the wet towel!

Some pointed out that the human mind is best able to deal with about half-a-dozen options. This was therefore our goal when we set about producing the draft ballot paper, (stage 3). Despite our best efforts, however, it was over 6 options long. Even then, some pointed out that it did not cover all the points raised and so, as you know, we finished up with a ballot paper of 9 options.

Now in some consensus debates, the consensors draw up the list of options and then, before producing the actual ballot paper, put these options into random order. Given the nature of our own debate, however, we decided to leave them in the order used during the earlier stages and, as you also know, the ballot paper was presented with the summary as shown below.

The consensors also endeavoured to ensure that each option was of about the same length. In our own example, the number of characters in each option was as follows:

 

Options

Option

B

C/E1

C/E2

D/F

G/H

H2

I/J

L

M

Length

360

372

344

313

424

354

441

330

357

Option I/J was the longest (and it is hoped the sheer size of it did not make it look more important). This longer than average volume was partly because, in the light of quite a few comments, it was considered necessary to explain just how the two vote system would work.

THE SUMMARY

 

 

Family

 

 

Option

INCOME

EXPENDITURE

State funding,

to be paid to national party (P)

or local constituency (C)

Non-state funding

Activities:

(what do parties do with their funds?)

are activities to be further restricted?

individual

corporate

A/B - state + corporate funding

B

state funds those parties/candidates with a minimum number of elected reps

P

yes

yes

no more than at present

C/D/E/F

state funding but no corporate funding

C/E1

as per number of local members

C

yes

no

yes, nationally

C/E2

as per number of local members

C

members' fees only

no

yes, both locally and nationally

D/F

all (above a fixed votes threshold) equally

C

yes

no

no more than at present

G/H

state funds democratic

process

G/H

state + corporate funding goes into a democracy fund

C

yes

yes, but pooled

yes, both locally and nationally

H2

state funding in kind only

C

yes

yes

yes, both locally and nationally

I/J - state funding

/votes

I/J

each voter votes twice: state funds parties

/candidates as per their share of second votes

C

yes

no

yes, nationally

K/L/M

state funding banned

L

no state or corporate funding

n/a

members' fees only

no

yes, nationally

M

no state funding

n/a

yes

yes

yes, nationally

A Second Analysis

In consensus decision-making, a preference vote can be used either as a straw poll and/or as the final decision-making process. In our own exercise, if it had been decided prior to the debate that the outcome had to achieve a minimum consensus coefficient (of which, see below) and if the outcome in the vote failed to pass that threshold, the vote could have been used as a straw poll. On this basis, the consensors could have declared that there was indeed a consensus that options M, B and L were definitely not the answer, and that the debate was now reduced to the following:

THE REFINED SUMMARY

 

 

Family

 

 

Option

INCOME

EXPENDITURE

State funding,

to be paid to local constituency

Non-state funding

Activities:

(what do parties do with their funds?)

are activities to be further restricted?

Individual

Corporate

C/D/E/F

state funding but no corporate funding

C/E1

as per number of local members

yes

no

yes, nationally

C/E2

as per number of local members

members' fees only

no

yes, both locally and nationally

D/F

all (above a fixed votes threshold) equally

yes

no

no more than at present

G/H

state funds democratic

process

G/H

state + corporate funding goes into a democracy fund

yes

yes, but pooled

yes, both locally and nationally

H2

state funding in kind only

yes

yes

yes, both locally and nationally

I/J - state funding

/votes

I/J

each voter votes twice: state funds parties/candidates as per their share of second votes

yes

no

yes, nationally

Because our own exercise consisted of only one debate-plus-vote, the answer (in the public constituency) remains as option I/J.

A More Mathematical Approach (analysing the public vote only)

In consensus voting, every option gets a certain overall score, and one can calculate the consensus coefficient of all the options according to the following formula:

If SA is the MBC score of option A, if V is the valid vote, and if n is the number of options to be voted on, the consensus coefficient CA is defined as SA/V.n In other words, an option's consensus coefficient is its MBC score divided by the maximum possible score; and the coefficient varies from good to bad, from 1 to zero.[4]

In the public debate, the social rankings were as shown:

Public

Option

MBC

rankings

B

112

8th

C/E1

224

4th

C/E2

231

3rd

D/F

212

5th

G/H

248

2nd

H2

195

6th

I/J

272

1st

L

183

7th

M

106

9th

The maximum possible score, for a consensus co-efficient of 1, would consist of a 1st preference on a full ballot from everybody, i.e., 9 points from everyone, which is 9 x 59 = 531. Accordingly, the consensus coefficients of all nine options are as shown below:

Public

Option

MBC

rankings

consensus coefficient

B

112

8th

0.21

C/E1

224

4th

0.42

C/E2

231

3rd

0.44

D/F

212

5th

0.40

G/H

248

2nd

0.48

H2

195

6th

0.37

I/J

272

1st

0.51

L

183

7th

0.34

M

106

9th

0.20

The winning option, then, I/J, has a consensus coefficient of 0.51. In a consensus democracy, such a level might well be regarded, constitutionally, as insufficient. In which case, the vote would indeed be regarded as a straw poll, and as already implied, the debate would be resumed on the basis of an agenda requiring further discussion on options C/E1, C/E2, D/F, G/H, H2 and I/J.

A Composite?

In a very simple debate on, say, dog licences, with let us say options £1, £2, £5, £10 and £20 on the ballot paper, if there is no clear winner but rather two most popular options, £10 and £20, the consensors may well decide to form a composite and declare the outcome to be £15.[5]

In our own debate, again in the public constituency, 17 voters considered the two leading options - I/J and G/H - to be, as it were, adjacent, and a further 7 voters regarded them as being only one step removed. So maybe the consensors could have considered the possibility of a composite.

They did not, but let us nevertheless consider the possibility. The two options were as follows and, of course, it must be remembered that I/J was the more popular of the two:

G/H State and corporate funding pooled into a "democracy fund" to finance innovative, democratic activities by individual political parties and/or ngos. Administered by Electoral Commission.

Donations/membership fees to parties/candidates to be from individuals only.

Political activities at both constituency and national level to be under tighter controls, both fiscal and functional (certain activities, ads etc, banned).

and

I/J Each voter has two votes, one elects a candidate, the other funds a political organisation. State funding to parties/candidates in proportion to the number of second votes cast.

Parties/candidates may also receive donations/membership fees but from individuals only.

Political activities in the constituency subject to current expenditure limits; nation-wide party activities to be under tighter fiscal controls during election campaigns.

In forming a composite, the consensors would take I/J as read, and add anything from G/H which they considered to be compatible. The combined outcome, then, could well have been as follows:

I/J/G/H Each voter has two votes, one elects a candidate, the other funds a political organisation. State funding to parties/candidates in proportion to the number of second votes cast. In addition, the Electoral Commission shall administer a "democracy fund", to finance innovative, democratic activities by individual political parties and/or ngos.

Parties/candidates may also receive donations/membership fees but from individuals only.

Political activities in the constituency subject to current expenditure limits; nation-wide party activities to be under tighter fiscal controls during election campaigns.

But all this is for the day when consensus decision-making is both more properly understood and more widely practiced.

A Comparison

Just out of interest, it was also decided to analyse the vote on the basis of one joint constituency: the full analysis, then, is shown here under ‘Both':

Options

Universities

Public

Both

B

8th

8th

8th

C/E1

7th

4th

5th =

C/E2

4th

3rd

3rd

D/F

3rd

5th

4th

G/H

2nd

2nd

2nd

H2

1st

6th

5th =

I/J

5th

1st

1st

L

6th

7th

7th

M

9th

9th

9th

Given the relative size of the public constituency, we should not be surprised to note that the social rankings in the ‘both' constituency are very similar to those achieved in the public one.

It is also interesting to compare the outcome under the MBC with what would have happened under other decision-making voting systems.

 

Option

VOTING SYSTEMS

Plurality

Vote

Two-round voting

AV/STV

BC

MBC

Condorcet

B

9th

9th

9th

8th

8th

8th

C/E1

7th =

7th =

7th =

5th

4th

4th

C/E2

5th

5th

4th

3rd

3rd

3rd

D/F

7th =

7th =

7th =

4th

5th

5th

G/H

2nd

2nd

2nd

2nd

2nd

2nd

H2

6th

6th

6th

6th

6th

7th

I/J

1st

1st

1st

1st

1st

1st

L

3rd

3rd

4th

7th

7th

6th

M

4th

4th

5th

9th

9th

9th

As was noted during the seminar in Oxford,[6] "There are two defensible procedures for aggregating votes: the Condorcet rule and the Borda rule. The Condorcet rule selects the option (if one exists) that beats each other option in exhaustive pairwise comparisons. The Borda rule selects the option that on average stands highest in the voters' rankings."[7] And here it is, in practice: in the table above, both the MBC and the Condorcet give very similar social rankings, the only differences between the two being in connection with options H2 and L, where instead of being 6th or 7th, they are the other way round.

There is also not a lot of difference between the BC and MBC, but that is because so many people cast a full set, or a nearly full set, of preferences.

While all the voting methodologies give the same 1st and 2nd social rankings, there is considerable difference in the overall rankings between the rest.

Nothing, of course, is perfect. The MBC has some weaknesses, a Cordorcet count has others. They are, nevertheless, the two best voting procedures, and many social choice scientists have devised methodologies which take both into account.[8] Furthermore, if the outcomes from both counts are the same or very similar, one can rest assured that the procedure has worked: that the participants have engaged in a broad-ranging debate; that the consensors have drawn up a balanced list of options; that the voters have taken the exercise seriously, expressing that which are, if but in their own minds, single-peaked preferences;[9] and that the outcome represents "the will of the people"!

Conclusions

It works.

Consensus voting can and does produce an outcome and, on many occasions, the outcome may be taken as a definite result. In other situations, as was said above, the consensus vote might act like a straw poll. But in nearly every circumstance, the information gleaned from a consensus vote will always be useful.

ps

An important part of any transparent process is an impartial and independent "electoral office". Our own teller reported that some voters used unregistered e-mail addresses, while others voted in the ‘wrong' constituency. In any future exercise, more effort must be made to ensure the security of the vote, all of which is perfectly feasible.

Peter Emerson, Belfast, 25.3.2008

Annex - THE OPTIONS

B State funding to be in direct proportion to the number of elected representatives, if that number is above a certain minimum.

Parties'/candidates' donations/membership/affiliation fees may be obtained from either corporate and/or individual sources.

Political activities not to be limited in scope, but only to current financial restraints in election campaigns.

C/E1 State funding to constituency parties to be dependent on number of local paid-up members.

Donations/membership fees to parties/candidates to be from individuals only.

Political activities in the constituency subject to current expenditure limits; nation-wide party activities to be under tighter controls, both fiscal and functional (certain activities, ads etc, banned).

C/E2 State funding to constituency parties to be dependent on local paid-up members.

Only membership fees (without any additional contributions) to parties/candidates, and only from individuals.

Political activities at both constituency and national level to be under tighter controls, both fiscal and functional (certain activities, ads etc, banned).

D/F State funding to constituency candidates (to those who pass a certain threshold of votes gained in the subsequent election).

Donations/membership fees to parties/candidates to be from individuals only.

Political activities not to be limited in scope, but only to current financial restraints in election campaigns.

G/H State and corporate funding pooled into a "democracy fund" to finance innovative, democratic activities by individual political parties and/or NGOs. Administered by Electoral Commission.

Donations/membership fees to parties/candidates to be from individuals only.

Political activities at both constituency and national level to be under tighter controls, both fiscal and functional (certain activities, ads etc, banned).

H2 State funding to be in kind only, as administered by Electoral Commission.

Parties'/candidates' donations/membership/affiliation fees may be obtained from either corporate and/or individual sources.

Political activities at both constituency and national level to be under tighter controls, both fiscal and functional (certain activities, ads etc, banned).

I/J Each voter has two votes, one elects a candidate, the other funds a political organisation. State funding to parties/candidates in proportion to the number of second votes cast.

Parties/candidates may also receive donations/membership fees but from individuals only.

Political activities in the constituency subject to current expenditure limits; nation-wide party activities to be under tighter fiscal controls during election campaigns.

L. State funding and corporate funding to be banned.

Only membership fees (without any additional contributions) to parties/candidates, and only from individuals.

Political activities in the constituency subject to current expenditure limits; nation-wide party activities to be under tighter fiscal controls during election campaigns.

M. State funding to be banned.

Parties/candidates may receive donations and/or membership/affiliation fees from either corporate and/or individual sources.

Political activities in the constituency subject to current expenditure limits; nation-wide party activities to be under tighter controls, both fiscal and functional (certain activities, ads etc, banned).

 


[1] The person who submitted this blank ballot participated at length in the debate and was "happy to support [this] intellectual exercise" by talking... but not by voting.

[2] It was just a pity, therefore, that he/she thought four options were all equally ok, and the other four all equally bad.

[3] A case could be made for reading this as 7½-3½-3½-7½-1-7½-3½-3½-7½.

[4] During the transition stage in South Africa, they used the term ‘sufficient consensus', but it was never defined. Long Walk to Freedom, Nelson Mandela, Abacus, 1994, p 714.

[5] In such a debate, voters may well be expected to have what are called ‘single peaked preferences'. If someone's 1st preference is £1, then presumably his 2nd preference is £2, his 3rd preference is £5, and so on. If another voter has a 1st preference of £5, then her 2nd preference will be either £2 or £10; and if she does vote 1st-2nd, £5-£10, then her 3rd preference will be either £2 or £20. It would be very odd if someone voted £1-£10-£2-£20-£5, i.e., with two ‘peaks'. And even odder if someone voted £5-£1-£20-£2-£10, with three ‘peaks'.

Well, if (nearly) everyone does vote with single-peaked preferences, then it is perfectly reasonable for the consensors to form a composite.

[6] Prior to this particular pilot, two seminars were held in Oct. 2007: one in Oxford to discuss the theory of voting, and a second in London to talk with the practitioners.

[7] A program to implement the Borda and Condorcet rules... Iain Mclean and Neil Shepherd, Nuffield College Politics Working Paper 2004 -W11, University of Oxford.

[8] Designing an All-inclusive Democracy, Emerson (Ed.), Springer-Verlag, 2007, p 19.

[9] Admittedly, some voters tended to do a bit of simple stuff, just going down the ballot, 6-7-8-9. This is another reason why, in ideal circumstances, the ballot paper should be in random order.

Feedback

We would welcome any feedback from participants on what they thought of the process (though many people have already fed back as it went on). If you'd like to add more thoughts, you can do so by leaving a comment below (you will need to be logged in to see the comment form).


We encourage anyone to comment, please consult the
oD commenting guidelines if you have any questions.