UPDATE: This post was written early yesterday (Monday), but I chickened out of posting it, because I thought it was too much of a ballsy call. However, Roy Morgan are themselves making the call, so I think it is worthy of discussion. In particular, because this is not a recent change, but part of a consistent pattern for them.
The Herald Digipoll, ONE News Colmar Brunton, Fairfax-Research International, and 3 News Reid Research are all showing support for National dropping, but Roy Morgan has National support steady, and comfortably above 50%.
Polls are invariably reported with their Margin of Error, an estimate of the precision for a party with exactly 50% support. The Margin of Error is not very useful for comparing whether there is a difference between two parties, and it certainly is no use at all for considering change over time. And change over time ought to be what we are interested in. Is support for Party X climbing or dropping?
One crude way of doing this is to add a line fitted through the datapoints, but one of the hidden aspects of variation, as I tried to illustrate on Friday is that different polls will use slightly different methods, which may mean that that poll produces consistently different results in a certain direction. As I noted then, Roy Morgan generally shows lower support for National. However, over the last month, the other polling companies have all shown a clear drop, while Roy Morgan have had National’s support climbing.
Roy Morgan sample over a week to a fortnight, which could explain them being slower to catch on to this most recent trend. Alternatively, it may be that some particular element of their survey method is producing a consistently different result for them. This is brave: if they are proved right on election night, it could be that their method is superior. If they are wrong, then they may want to reconsider how they are polling. Irrespective of the outcome, however, it brings me back to a recent summary of Daniel Kahneman’s work looking at the success of fund managers: whoever is most successful in a given year (or election?) may just be due to chance.
UPDATE 2: The downward trend exists for the other 3 polls, but because Fairfax Research International and Herald Digipoll have relatively few data points, it is hard to display in a tidy fashion. Also, Roy Morgan’s trend does not diverge in the same way for Labour and the Greens, only National.
The data is derived from wikipedia, and fit with a modified version of the graph there, using a LOESS fit, with a span value of 0.3.