Polling pitfalls Mark Wilson Published: Sunday, December 7, 2014 Outside Track It’s polling season. Nigel Henry’s Solution by Simulation poll in last Sunday’s Express was a fine start to the crop—plenty interesting questions, and a few hints at an answer. Sensibly, the polling focused on the four most marginal seats, which are likely to swing the election. But first, the big pitfall: its sample size. Pollsters rightly argue for a good-sized sample. Their clients, with an eye to the budget, generally resist. Lord Ashcroft is a British businessman with big Caribbean investments. He is a former deputy chairman of Britain’s Conservative Party. He has deep pockets. Looking ahead to Britain’s general election next May, he published a poll last month covering 18 landmark seats. He paid for a sample of 1,000 voters in each—so 18,000 in all. That gave him a 3.1 per cent margin of error. The Sunday Express paid for a sample of 635, spread over four seats. Within each seat, that’s an average of 159—so an 8.1 per cent margin of error, as Henry said very clearly in his professional report. In St Joseph, the poll had each main party at 32 per cent. So that means real support between 24 per cent and 40 per cent. One part could be as much as 16 points ahead. It gets worse. There’s a one-in-20 chance of a rogue poll, with results outside the statistical margin of error. A Nacta (North American Caribbean Teachers’ Association) poll published in August looks iffy. It reported a sample size of 1,000—but broke this down to report on ten marginal seats. It looks like a sub-sample of around 100 in each marginal—so that would be a margin of error around ten per cent. The report claims a four per cent margin, but it is not clear where that number comes from. The Nacta poll puts PNM support in Arima at 45 per cent. Unless I have missed something, the true answer could be anywhere between 35 per cent and 55 per cent. In a marginal seat with a close result, any poll with a sample below 1,000 is pretty much a waste of time. Two British pollsters—YouGov and Populus—generally go to around 2,000 for a national poll. The statistical margin of error is not the only pitfall. If the sample is not truly random, we get polling bias on top. That may happen if there’s some quirk in the polling method that gives too many old people or too many young, too many Indo or Afro-Trinidadians, too many politically-aware voters or too few. Next problem: answers may not reflect true voting intention. I talked to a pollster years ago in Antigua, when the Birds were in office. Rightly or wrongly, in that small society, some voters felt they were being personally tracked. Were opposition supporters reluctant to state their true opinion? That said, last week’s poll had some interesting pointers. There was a huge difference between the four seats. A larger sample poll in each would be useful follow-up. In Moruga/Tableland, the poll shows a swing around ten per cent towards the Partnership. If this is accurate, which it probably isn’t—and repeated nationally, which it clearly won’t be—it would leave the PNM with just six seats in Parliament. The MP, Clifton de Coteau, says he’s “humbled.” And humble he might be, because apparently his personal ratings in the seat trailed far behind those of his party and his prime minister. But the poll opens up a couple of questions. Is the Partnership doing extra-well in rural southern seats—perhaps because they have been playing catch-up with infrastructural spending? Or are they making up for historic under-performance in a seat with a lot of Indo-Trinidadian voters where the PNM has previously done well? The Moruga/Tableland swing may look counter-intuitive. But then, most journalists and commentators probably don’t spend their long weekday evenings gossipping about politics in Barrackpore or Sixth Company. In Tunapuna, the poll shows a swing of around 17 per cent to the PNM. That would leave them far ahead of the 52 per cent they scored in 2007. Repeated nationally, that swing would leave the Partnership with just 12 heartland seats (assuming they win back Chaguanas West). So is that the corridor swinging the opposite way from the rural South? In San Fernando West, the poll shows a tiny swing of under three per cent to the PNM. In a general election, that would give Keith Rowley just two extra seats, far short of a majority. And St Joseph? The opinion poll’s tied result translates into a three per cent swing to the UNC since last year’s by-election—or an eight per cent swing to the PNM since 2010, which, repeated nationally, would leave them with 22 or 23 seats—and a long list of gains including Moruga/Tableland. What about Jack and the smaller parties? Between six and 12 per cent in each seat. That’s enough to make a difference if we go to two-round voting. So…time for some more polling. And next time, please, splash out on a good-sized sample. Columnist Source:: Trinidad Guardian The post Polling pitfalls appeared first on Trinidad & Tobago Online. #trinidad
Posted on: Sun, 07 Dec 2014 09:20:30 +0000
Trending Topics
Recently Viewed Topics
© 2015