Two types of outlier
This morning TNS released a new poll showing figures of CON 28%, LAB 35%, LDEM 6%, UKIP 18%, GRN 7% (tabs). The seven point Labour lead is striking, and out of line from the broader trend. Usual caveats apply, but for once I haven't seen many people over-react to it. Most have sensibly enough seen it as the just the other side of the coin to the ICM poll earlier this week showing a solid Tory lead - two outliers in opposite directions. However, it is worth looking at the different reasons why these two polls went against the trend.
On average ICM produce figures that are slightly more Con/less Lab than average, but only by a tiny bit, generally they show Labour -v- Tory leads that are much the same as those from other companies. The reason that their poll on Monday showed a four point Tory lead when other companies are showing the race almost neck-and-neck is almost certainly just random sample error. Samples vary from day-to-day, month-to-month and sometimes you get one that's a bit Laboury or a bit Conservativety in ways that weighting does not correct. I expect ICM's next poll will be in line with those from other companies.
TNS is a different sort of outlier. The seven point Labour lead may be different from that shown by other pollsters, but its actually in line with TNS's previous polls. In their previous four polls they showed Labour leads of 6 points, 0 points, 7 points and 7 points. Over on my chart of house effects TNS's polls on average show a Labour lead three points larger than other companies (and if anything the gap is growing!) This isn't random sample error, this is a consistent methodological difference between TNS and other companies, and unless they switch methods I expect their future polls will continue to show bigger Labour leads than other polls.
Two outliers, but one is probably just a random blip from a pollster who normally shows the same as other companies, one looks like a typical poll from a pollster who regularly produces polls that show a bigger Labour lead than the pack.
