This halloween polls made scary reading for Labour, SNP taking more seats in Scotland*, one poll even putting the Conservatives one point ahead, but how does the model digest them? Well, it still predicts Labour as largest party, just with a mildly reduced chance of majority.
How can this be? Quite simple, the model includes a "Reversion to the Mean" assumption, that says there's a pull back towards the last polling result that needs to be overcome with momentum. The "Reversion to the Mean" has been quite a strong benefit to Conservative figures over just looking at a poll average. But it would be silly if we assumed that there would always be a "Reversion to the Mean" of this kind, because then governments would never change hands.
So there are three things that stand against "Reversion to the Mean". First is the momentum of the polling, which either offsets against it, or re-enforces it when the incumbent has the momentum. Second is the assumption that "Reversion to the Mean" is weaker when consumer confidence is low. And third is the constant progress of time, eroding the weight of "Reversion to the Mean"'s effect on the prediction.
The consumer confidence component also means that when GFK/NOP's consumer confidence index went from -1% to -2% today, Labour actually had a minor uptick in the prediction against one run yesterday. Remembering of course, that "Reversion to the Mean" is still against Labour at this point.
* Scottish specific Polling is monitored by the model to estimate seats that the SNP run in.