‘No validity’ in increasing poll sample size
Don Anderson says 1,000 is representative enough for accuracy
Published:Tuesday | July 4, 2023 | 1:32 AM
Erica Virtue/Senior Gleaner Writer
POLITICAL POLLSTER Don Anderson has dismissed suggestions that the 1,000 poll sample size used for political polls should be increased for greater accuracy. The seasoned market researcher told The Gleaner that an increase to 1,500 would only reduce the margin of error by an insignificant percentage.
In addition, the change would make polling prohibitive for organisations, as it would increase the cost by 50 per cent.”
“No. Absolutely not. There is no validity in increasing the sample size of the polls. People talk about increasing sample size, a sample need to be large enough to give prominence to the smallest sub-unit you want to analyse. If we wanted to know how a man within a certain age group thinks in Kingston, then you need a large enough sample size of that which would then make the sample much larger to accommodate it. You don’t need to do that. You don’t need a sample size more than 1,000 persons,” said Anderson in an interview with The Gleaner.
His Market Research Services Limited typically uses 1,000, 1,003 and 1,012 sample size.
His recent People’s National Party (PNP) commissioned poll findings showed the party five points ahead of the ruling Jamaica Labour Party (JLP).
It was criticised by JLP, which called the findings ‘contrived’ and ‘fictitious’ in a release under the signature of its general secretary, Dr Horace Chang. It said it rejects as fictitious and contrived the results released by the PNP, which purports to show them ahead.
“The poll is sponsored by the PNP and its findings are designed to artificially influence public opinion. The release of (a) party-sponsored poll is part of the PNPs propaganda and misinformation campaign intended to shore up a weak leader, sow discontent and distract the country with political mischief...” it said, inter alia.
The findings at the heart of the political ‘bun’, however, was the following question.
“If you decide to vote, which of the two parties do you intend to vote for” in June 2023?
Findings showed 30.2 per cent said the PNP and 25 per cent said the ruling JLP, a clear five-point lead.
In February 2023, both parties were in a statistical dead heat with 28.1 per cent said they would vote for the PNP compared to 27.9 per cent for the JLP. One year ago, the JLP led by a massive 13 points over the PNP, with 31 per cent compared to 18 per cent for the PNP.
However, those who are undecided or are not voting at June 2023 is at 44.8 per cent, a one per cent increase from 44 per cent in February.
The JLP has yet to produce its own poll results, which it claims does not reflect its own findings.
Anderson said in the United States the best pollsters in the world hardly uses a sample of more than 1,000 to project the outcome of the presidential elections. That country’s population is more than 350 million, he said.
While accepting Anderson’s findings, political scientist and attorney-at-law Dr Paul Ashley suggested that an increase in the sampling size may become necessary given the increase in Jamaica’s population, and given that 1,000 persons have been used for years.
“Don Anderson’s market research has been used by the private sector, and he has been the most accurate of pollster operating in the current space. I don’t’ think one can question the findings of Anderson,” Ashley said in defence of Anderson’s reputation.
He said it was a time-worn practice to attack the messenger if the findings did not suit you. The dismissal by the JLP, he said, is a replica of what currently obtains in the USA involving the Republican Party and former President Donald Trump.
“It’s fake news if the findings or the report is unfavourable to Trump. It’s the same thing here. It’s Trumpian. Discredit the messenger to discredit the message,” Ashley stated.
Jamaica’s total population was 2.83 million in January 2023
Anderson said the larger the sample size, the more costly the poll will become.
“It is old school that the larger the sample size, the more accurate the findings. Rubbish, rubbish, rubbish. Sample size is not a fact in the validity of the data. It’s the representativeness of the sample. If the sample is representative with 1,000, and accommodates all the people within the population, within their known numerical width, representativeness is more important than sample size,” Anderson explained.
He said, however, that there are instances when sample size matters.
“The only time it matters is if you want to explore a small segment. For example, if you want to look at men in Kingston, in a normal 1,000 sample, men in Kingston would give you about 35 interviews. In a sample size of 1,000, you get about 215 interviews in Kingston…” according to Anderson.
He provided the following example.
An interview with 125 persons in Kingston who are asked if you voted in the last election, 75 responded in the affirmative. If asked further how did you vote, 50 said they voted for the JLP and 25 for the PNP. Those percentages, he said, would reflect 60-odd and 30-odd percentages.
“Can you imagine a report which said 60-odd per cent of men said they voted for the JLP and 30-odd per cent for the PNP. That is because the sample size was not large enough to allow you to do that the kind of disaggregation,” her explained.
Anderson said such polling would need about 700 men in Kingston, broken down and disaggregated.
“The larger sample size only becomes important if you need to do some serious amount of disaggregation of the data, then you reduce the dynamism of small numbers. The smaller the sample, the more dynamic the numbers and therefore the percentages would be staggering,” he told The Gleaner.