Random thought, 18 May 2001
It is part of the political folklore in the Alliance and other places that Labour and the Alliance scrap over some of the same votes. Thus when one goes up, some of that is won at the other’s expense.
And intuitively, this folklore seems to hold merit.
Moreover, in the mid-1990s it did seem from the polls that Labour’s recovery leading up to the 1996 election was at least partly at the Alliance’s expense.
Broad poll comparisons seem to bear this out. Here are the numbers given by averaging the four main nationally published polls:
Labour six months Sep 95-Feb 96 22%
Average Alliance six months Sep 95-Feb 96 22%
Average Labour October (election month) 96 27%
Average Alliance October (election month) 96 14%
Labour up 5%, Alliance down 8%
That is evidence of transfer
Labour election 28%
Alliance election 10%
Labour up 6% on Sep95-Feb 96 figures, Alliance down 12%
That is evidence of transfer
And if you take election month polls against election result:
Labour up 1%, Alliance down 4%
That is maybe evidence of transfer
But if the swap assumption was true in the mid-1990s, it doesn’t seem to have held since. Take the 1996-99 Parliament and compare the middle year with the campaign month and the election.
Average Labour 1998 42.2%
Average Alliance 1998 7.9%
Average Labour Nov 99 (election month) 36.7%
Average Alliance Nov 99 (election month) 8.5%
Labour down 5.5%, Alliance up 0.6%
Maybe slight evidence of transfer but hardly counterbalancing and. . .
Labour election 38.7%
Alliance election 7.7%
Labour down 3.5% on 1999 average
Alliance down 0.2% on 1999 average
No evidence of transfer
Perhaps there is some tenuous evidence if you take the poll results for November 99 against the election:
Labour up 2.0%, Alliance down 0.8%.
But that says nothing about recovery from the middle of the term.
However, the Greens were emerging as a force and went from 0% in early 1998 (average 0.5% for that year) to 5% at the election. If anything that should have taken the Alliance vote down as Green votes formerly in the Alliance were siphoned off. The combined Alliance-Green vote in the election was 13.0%.
So you might say Labour went down 3.5% between 1998 average and election and
Alliance/Green went up from 8.4% total average 1998 to 13.0% in election and thus up 4.6%.
That looks like a transfer.
But probably at least half (I would say more but there is no hard evidence to guide us) of the Green vote were new votes or votes that did not go to the Alliance in the 1996 election or in 1997/98 polling, so the combined figure has to be treated with great care.
And it says nothing about what might happen between 2001 and the 2002 election since the Greens and the Alliance are separate parties.
So let’s compare the first quarter of this term (Dec 99-Sep00) with the second quarter (Oct 00-May01)
Average Labour first quarter 45.9%
Average Alliance first quarter 4.5%
Average Labour second quarter 43.7%
Average Alliance second quarter 4.3%
Labour down 2.7%
Alliance down 0.2%
There is no evidence of transfer
If you compare each single poll with the same preceding poll (eg, DigiPoll with the preceding DigiPoll) for the whole of the time since the 1999 election to May 2001, the Alliance vote moved in the opposite direction from Labour only 20 out of 38 times (about half). (If one takes the movement between each point on my smoothed three-poll moving average, the number is 5 out of 41 � but this is probably not a good comparison.) The 20/38 figure may mask movements between Labour and the Alliance when both move in the same direction but one moves more in that direction than the other, shedding to or gaining votes from the other partner as part of that greater movement. But even so, I don’t think one can draw a conclusion.
There is some slight evidence if one does the calculation another way, asking whether the Alliance’s share of the combined Labour/Alliance vote rises when Labour’s vote falls and falls when Labour’s vote rises. That happened 26 out of 38 times (or two-thirds of the time). But if this calculation was to give strong evidence of an exchange of votes, I would think a higher correlation would be needed.
Does the Alliance pick up going into an election, as Jim Anderton asserts? Maybe it will in 2002 compared with this year (as New Zealand First did in 1999 compared with 1998.
But if one takes a 3-poll rolling average, the Alliance got 8% through 1998, the same as its 1999 result. The Alliance average from November 1995 to May 1996 was 17.5%, well above its 10% in the 1996 election. (Taking individual polls gives a similar result).
That, of course, says nothing about what might happen next year.
Comment by Jack Vowles (19 May) on reference to this in the Herald:
I see you’re quoted in the Herald as saying there is no apparent relation between Labour rises and Alliance falls in polls since 1996. My work looking at individual-level movements in survey data suggests there is a relationship – I was surprised you don’t see one in polls.
Anyway, over the last 30 minutes or so I have regressed first differences (change in party vote intention for each party from one poll to the next) from One News-Colmar Brunton polls, assuming changes to Labour influence changes to Alliance. This is what we get –
Variable B SE B Beta T Sig T
LABCH -.124182 .056306 -.303334 -2.205 .0322
(Constant) -.027713 .193909 -.143 .8870
Which means as Labour vote intention rises by 1 per cent, Alliance intentions go down .125 of a per cent. Not as close as I had thought it might be, but of course Labour has more intentions than Alliance, so if Labour goes up 10 points the Alliance goes down 1.25 points, which is a bit more significant on their lower base of support. The relationship is also statistically significant, and consistent post pre- and post 99 elections.
This is a crude way of doing it, though – and a more sophisticated method should probably show more effects, because post 1999 election one would have to control for other factors which would influence support for the government parties together, and then partial out the Labour effects on Alliance – the same might apply in reverse when the two parties were in opposition, particularly as they moved closer together.
So it we control for change in both parties combined share, we get something much better – for every 1 per cent increase in Labour support net of the influence of changes in the combined vote of the two parties, Alliance support goes down nearly 0.9 of a per cent, and both estimates are highly significant.
Model B Std. Error Beta t Sig.
(Constant) 2.622E-0 .042 .626 .534
LABCH -.877 .027 -2.141 -32.610 .000
GOVCH .886 .028 2.060 31.375 .000
This greatly exaggerates the influence, though, as the government change variable is not independent of the Alliance change variable (which it should be according to the assumptions of regression). A good more truly independent variable would be disapproval of the government prior to the 1999 election, and approval of it since. But I don’t have it in my spreadsheet going back all the way, and anyway, this is enough work for a Saturday afternoon…