Trying to explain why the Ashcroft National Poll appears to be so erratic
Turnout filtering might be the key
At the excellent PB gathering in Ilkley last night the big topic of conversation was the extraordinary Ashcroft national poll which saw the 2% CON lead of last week become a 7% LAB one.
At 4pm each Monday afternoon since the start of May Lord Ashcroft has been publishing his weekly national phone poll. This has been a great addition to the overall polling mix particularly because testing political opinion in this way has become rarer since Populus stopped being the regular pollster for the Times eighteen months ago. Its twice weekly Westminster VI surveys are online.
The initial Ashcroft one in May caused a big shock because it was the first national poll of any kind in 26 months to record a CON lead. That didn’t last long and over the next few polls, as can be seen in the chart.
YouGov’s Anthony Wells suggests that the Ashcroft volatility is down to the much smaller samples that we see in phone polls. Of 1,000 in a phone poll only about 500 of those asked often give a view which statistically increases the margin of error.
-
But if accuracy was down to sample size alone then how come YouGov’s 6,000 sample final online poll fared so badly in the GE2010 accuracy table behind five pollsters with 1,000 sample phone polls?
My suggestion to Lord Ashcroft is that he should look at the ICM approach to turnout filtering. The latter has a much less judgemental “what did you do last time” question. This generally results in a smaller proportion of the sample saying they voted in 2010. It then discounts the views of non-GE2010 voters by 50%.
As a general rule YouGov doesn’t do turnout filtering.