Do minor parties do better in the election campaign?

I noted in a NZ Herald discussion on how badly most minor parties were polling, it is taken as gospel that minor parties pick up support in an election campaign, and major parties lose support.

Now this is a pretty logical proposition, but what is the empirical evidence.  Let’s start with 1996.

In 1996 Labour averaged 17% in the CB poll before the campaign, and got 28%.  National averaged 39% and got 34%. NZ First averaged 22% and got 13%. Alliance averaged 12% and got 10%. ACT averaged 3% and got 6%.  Christian Coalition averaged 3% and got 4%.

Conclusion: Not true in 1996.  Labour got 11% more than it polled, and from the minors only ACT got significantly more than it polled.

In 1999 Labour averaged in all polls 42% and got 39%. National averaged 36% and got 31%. NZ First averaged 4% and got 4%.  Alliance averaged 7% and got 8%. ACT averaged 5% and got 7%.  Greens averaged 1% and got 5%.

Conclusion: Partly true in 1999.  Labour and National got a bit less than they had been polling, and some minors got a bit more.  Greens were the big improver but this is more because like ACT in 1996 they were a new party.

In 2002 National averaged 32% and got 21%. Labour averaged 51% and got 41%. ACT averaged 4% and got 7%. NZ First averaged 3% and got 10%. United averaged 0.4% and got 7%. Greens averaged 6% and got 7%.

Conclusion: True in 2002.  Labour and National each got around 10% less than they had been polling, and all minors got a lot more except for the Greens. 

In 2005 National averaged 36% and got 39%. Labour averaged 43% and got 41%. Maori averaged 2% and got 2%. ACT averaged 2% and got 1.5%.  NZ First averaged 7% and got
6%. United averaged 2% and got 3%. Greens averaged 5% and got 5%.

Conclusion: Not true in 2005.  Labour and National each got a higher vote than they had been polling. All minors were within 1% of their average polling level.
So the assertion that minor parties pick up support in the campaign period from major parties is far from a rule.  It has only conclusively been true once in 2002.  It was partly true in 1999 but false in 1996 and 2005.