You can ‘prove’ anything with statistics

                Mathematics does not lie.  A statistical formula or algorithm will give a correct result.  Statistics based on natural sciences such as biological data are altogether dependable.

But whenever sampling is based on human behaviour, such samples suffer from errors resulting from the whims of human behaviour, psychology, and indeed, wilful bias. This includes results of electoral samples meant to demonstrate the mood of voters at a point of time.  Electoral polls based on sampling suffer from the vicissitudes, inherent fears and caprices of human behaviour of selected respondents and may indeed be incorrect, sometimes significantly so. 

Even if the sample is structured to ensure a correct representation between electoral districts, sex, age, education and social classification to ensure a stratified random sample that is really representative of the entire statistical population from which it is drawn.

And this applies equally to electoral polls in our island and the potential inherent error of same, even if made very close to the electoral appointment.  The potential error is much more than the level of confidence often stated to be close to 95%, with a margin of error of 2.8%, presumably based on defined standard deviation levels from the sampling mean.

In this perspective, it is worth analysing electoral polls such as that published last Saturday, May 2. Others in the same vein are expected to follow.

I am convinced that the publication of these opinion polls by Marmarà are nothing other than a propaganda stratagem meant to demoralise Labour opponents.  The covert, but obvious, intention is that it will lead some voters to argue they might as well as not vote against the ruling regime.  That if you cannot beat them, you'd better join them. That there is no real personal incentive to be on the losing side of attempting to vote out an incumbent party dishing out favours, gifts, favours and pledges, irrespective of  their sustainability. 

By nature, people prefer to side with the winner. To side with the underdog, and to give truthful replies when quizzed in a poll, takes moral courage and fortitude, with nothing to gain at a personal level.

Many polls have been spectacularly wrong.  Most recent examples are those of the last electoral test of the European Parliamentary elections in June 2024. Many predicted Labour victories of anything between 20,000 and 30,000 votes.  Labour eventually won with an 8,500 vote majority.

The big limitation of polls also emanates from poll fatigue. There are so many polls conducted on various aspects of life that people are fed up with them.  Additionally, some pollsters – not electoral polls – offer sweeteners for replying such as some exchangeable voucher or a chance to win a prize – and this makes voters even less likely to want to reply to polls that do not give them something in return. Besides the resistance of many to give a truthful answer to an unidentified source about something that is so sensitive as electoral preference.

Up to around 2017, local electoral pollsters used to publish the number of people they had contacted and how many accepted, giving a clear picture of refusals. 

If among those refusing to reply there is a skew – that is they are not perfectly representative of the true picture of electoral popularity at the specific time – the replies of those who accept to reply will necessarily be biased.  When the refusal rate is high – the prevalent refusal rate in electoral polls may be above 50% – and when one factors the various behavioural motives mentioned earlier, it is reasonable to assume that PN voters are likely to be the lion’s share of reply refusals. When one needs to contact around 4,000 voters to attain a sample of 800 positive replies, the poll result is bound to be speculative. 

Assume one is starting with a population that is split 50-50.  Suppose one has to contact 4,000 from respondents, with a sample that should have a 2,000 – 2,000 split.  Since 3,200 will have refused to reply, one has to consider whether the 800 who replied are skewed in favour or against one party or the other.  I believe that the PN voter has every reason to abstain from replying, or to give an untruthful reply.  Even a slight skew will give unreliable polls.  A slight skew amongst the actual respondents in the poll may easily give an 8% gap for the likes of Marmarà to tell their audience that Labour will win by 28,000 votes from a population that may actually be much closer to being 50-50.

Pollsters need to be challenged to publish the number of refusals. When they do, we will be better able to gauge how reliable or unreliable they are.

The reported “scientific” method of machine learning to predict how undecided respondents will likely vote also gives far from reliable results.

Hence, for those who believe that May 30 is the date when they may possibly send this corrupt government regime home, as Hungarians did to Victor Orban a month ago, I say they should not be disheartened by polls. 

Marmarà in particular is a manifestly conflicted statistician.  He and his companies have received dozens of public contracts, most by direct order, from state entities over the past decade. That alone makes it legitimate to ask questions about the reliability of his polls.

Kuraġġ people of goodwill. The trust confided in Labour in 2013, by myself include, to govern according to the mantra of transparency, accountability and correct governance, has led to the most corrupt administration in Malta’s political history since independence.  The electorate still has a chance of getting this morally bankrupt government out of power at the forthcoming elections of May 30, a chance much more possible than Marmarà's polls may suggest.

Paul Bonello is a financial advisor.