Christopher Charles | Challenges with opinion poll interpretation

The media river is in a spate with misinformation, and, unfortunately, I am trying to swim upstream, an impossible task when ignorance abounds. The party leaders’ performance ratings should be compared, since these are the two choices the voters have.

Party leader A or B can only be high or low in comparison to each other, their previous ratings, or other leaders, but the first is most important. Only compared differences over several polls is useful. The same mistake is made in the reporting on homicides, for example, by comparing the number of homicides in September of this year with the number for September last year, when only a trend of several years is explanatorily useful.

The answer choices for a question in August-September Anderson poll on leader performance ratings were a five-point scale of 1-very poor, 2-poor, 3-average, 4-good, and 5-very good. One news report only tallied the good and very good answer choices to get the positive ratings. On the scale, average is the midpoint and must be added to the ‘good’ and ‘very good’ answer choices for the positive rating. Therefore, only the below average responses such as ‘very poor’ and ‘poor’ can be interpreted as negative.

The scientific interpretation of the August-September Don Anderson poll shows that Holness has a positive rating of 66.2 per cent and Golding, 52.7 per cent. Care should be taken in interpreting this difference when both leaders score over 50 per cent positive ratings and the prime minister’s role is larger and more important than the opposition leader’s role. The prime minister has the higher percentage because he sits as the emperor of governance and directs government outputs, compared to the opposition leader who only leads his party in parliament, sets the direction of his party, criticises the government, and meets with party supporters and other stakeholders.

The negative ratings of both leaders were below 50 per cent. One report gave the reasons for the negative ratings as if they are coming from the total sample and not a subset. This is a conflation of the numerator and the denominator which creates confusion. The best way to have approached this was to say, for example, that, out of the minority that rates the prime minister/opposition leader negatively, 20 per cent gave a particular reason. Also, the registered voters were able to choose more than one answer which is acceptable, but the tally was more than 100 per cent. Again, these proportions (the reasons for the negative evaluations) were reported as if they were out of the total sample. The context of letting the readers know one is talking about a subset of the sample rather than the sample is very helpful for readers.

TYPE OF QUESTIONS ASKED

Asking the right questions is very important. In one poll, respondents were asked who is best suited to lead the country. They were given the prime minister and other members of government and asked to rate them together, and then the opposition leader and other senior members of his party were rated, too. A question like this must mirror the choices that the voters will have on election day, which is two. What was interesting was that the opposition members pitted against Golding were very popular and they took more support away from him than the other government members took from the prime minster. The framing of this question, and the multiple answer choices with differential popularity weighting favoured the prime minister. Bias was introduced into the data collection, processing and interpretation.

THE WESTMINSTER CONTEXT

Westminster is leader-centric because, structurally, the party leader is the centre of voter attraction. In a recent poll, the party leaders had the same approval ratings but there was a difference of a few percentage points in the party standings, most of which was within the margin of error. If one doesn’t use the margin of error in this example, the leader approval rating is always more important than the party standing in assessing the likely election outcome, since the leaders drive the popularity of the parties and not vice versa.

The governor general appoints the leader of the majority party as prime minister, who hires and fires cabinet ministers, controls them through collective responsibility, nominates most senators, guides his party in parliament, and guides the party direction and slate of election candidates. The opposition leader does the same in his party, appoints shadow ministers, and nominates the minority of senators. Election mobilisation revolves around the party leaders. The party leaders’ approval rating is more important than the party standings. It is also very important to note that a poll can show a more popular leader and party, but this leader and their party can lose the election because of how this popularity was distributed across the 63 seats in terms of votes.

Some solutions are that media managers should employ statisticians to train reporters in statistical thinking and reasoning. Pollsters should publish a column on the interpretation of their polls with one of the major newspapers, which becomes a part of the public record. The universities that train journalists should create and teach a course on statistical thinking and reasoning for journalists.

Christopher Charles, PhD, is a professor of political and social psychology in the Department of Government at The University of the West Indies, Mona. Send feedback to columns@gleanerjm.com