Thursday, 5 November 2009

Statistical Notes - Referendum and Votes

The vote on Montfort Tadier's proposition took place today, and unfortunately did not succeed.

This was to let Islanders decided in a referendum whether they want the Constables in or not.

Deputy Angela Jeune put in an amendment that the referendum should be binding -  on condition that the turnout is greater than 50%, no doubt to stop making the referendum a pointless exercise if the States decide to ignore the results anyway.

But even though this proposition failed - both Montfort's proposition and Angela's amendment - they raised some significant and instructive points on the statistical background involved in referenda, and in voting in general.

Any election or referendum is essentially the same as a self-selecting opinion poll, i.e., although conducted in a more formal manner to make sure that only legitimate voters vote and once, it is the same statistically as a JEP "text in what you would like poll". It is not a random sample, and statistically may not represent the views of the population as a whole. It represents the people who choose to vote.

People who do not vote may not vote for a variety of reasons, which is why there is a UK electoral reform group NOTA, which suggests that voting slips should have "none of the above" on any list of names, to allow a protest vote to be heard as there is no mechanism in our voting system for that. Obviously in the case of the constables staying in the States, where there is a yes or no answer, there undoubtedly would not be the equivalent of this - an undecided vote. The "don't cares" can be certainly assumed not to vote, and there is no easy mechanism (apart from compulsory voting) to count them.

Voting is about making decisions - who gets in, what is to happen etc, and is not a random sampling mechanism. It is judged that people can choose not to vote, but that is their choice, and only those who vote can have a say.

Obviously, in the case of the referendum, the larger the sample size, the more representative it will be, because the less exclusions there will be. Actually the same is true of any vote be it for a referendum, or a vote for any election, which is why a poor turn out gives an implicit message that those voted in have not quite a good a mandate as they should have. The lower the turnout, the more it represents "activists" rather than everyone, and politicians instinctively know this.

Parish Assemblies - with no postal voting - are often the very worst example of this, where 40 people can decide and "rubber stamp" on one evening what a Parish of thousands should do. The move to have both postal and day long polls at Parish halls for Procurers du Bien Public etc is a step in the right direction.

An example of voting size will demonstrate why - for a self-selecting population - a greater size is better. Suppose we conduct a survey or phone in about the Fort regent swimming pool, and 100 people phone in, of which 90 say keep the pool, and 10 say do not. And now suppose that 20,000 people phoned in, and 55% of those (11,000) say keep the pool, and 9,000 say do not. With such a difference, it is easy to see what is true statistically, that with a working population of say 40,000 or more, the second result is more representative of the whole even though it is self-selecting.

So Angela Jeune's 50% is more significant than 33% because it means - if we look at the vote and non-vote - that 50% of the total island population care one way or the other, and only 50% we are not sure of - while on perhaps 33% turnout (which is often the case), the missing 66% of the total island population may well just not care or have no opinion one way or another. The missing figures cannot be assumed to be representative of those who voted.

The alternatives are either random sampling or stratified sampling. Random takes a purely random and statistically large enough segment of the population, and polls them completely at random. Stratified sampling (often used by statisticians) tries to take a random sample but weights it by the same ranges within the population as the last census (same age brackets, income groups, male/female etc)

Interestingly there was a random sample which was very well conducted a few years back - when there was an idea to have a massive and extraordinary bridge to connect the waterfront to the rest of town - connection being a perennial idée fixe among certain politicians - a statistical sampling company was appointed to make a significant random phone sample of the population, which had quite the opposite result to the JEP's self selective survey, and came up with a clear "no" to the project, much to the disgust of the politicians!. The only weakness, but it was a small one, was that while it included all numbers - phone book and ex-directory, and different times of day - not to miss selected people - it of necessity polled only people with landline phones.

No comments: