The Opinion Polls and the 2005 Election.

Some material I prepared just before the election, which my son put on his Vorb website, where there are the graphs and some subsequent comments.

Keywords: Political Economy & History; Statistics;

I thought you might be interested in the following two charts. They combine four polls (NZH, Fairfax, TV1, TV3), interpolate between polls and project after them (out to election day – a very foolish thing I add).

Some material I prepared just before the election, which my son put on his Vorb website, where there is also commentary and the graphs

Keywords: Political Economy & History; Statistics;

Briefly they suggest
1. Labour hit its nadir on the 6th Sept and has been slowly recovering.
2. National peaked on 30th August, and has been falling steadily since.

3. Little gains or losses to NZF or Greens but, if you trust the polls, both are above 5 percent (but given what has been going on in Tauranga, who knows).

4. Not much movement in ACT or Maori Party.

5. Some growth in Progressive but this may be the effect of sampling error.

6. UF seems to be a gainer from the falling off from National (as I expected).

7. LPG has always been the dominant party, but was struggling at the end of August.

BUT it does not have enough votes to govern without support from UF, or NZF, or MP.

Couple of other things.
A. Take the last 5 observations with a grain of salt.
B. I don’t trust the polls to give good levels. It is still a close run thing.

As Jim Bolger famously said ‘Bugger the Polls’.

But as Nobel prize winning economist Bob Solow famously said of the addicted gambler’s response to the pokie machines in a far from salubrious joint, ‘I know they are crooked but they are the only game in town.’

Comment: with hindsight

As a statistician I have long had an interest is surveys and opinion polls, and did a lot of work on them in the early 1980s. For most of the time the outcome cannot be verified. Hence the interest in analysing them statistically just before the election.

Public opinion pollsters do not randomly survey (which is what they theory of sampling requires) but they ‘quota’ sample, that is they divided the population into groups, calculate the share in the overall sample that each group should have (the ‘quota’), and telephone around until they fill the quota. We have to take it on trust that the procedure gives an unbiased outcome. (‘Bias’ here is used in the statistical sense, not the political one. I am not saying particular pollsters favour particular political groups.)

The public polls were all over the place. Two taken at about the same time could be statistically inconsistent. They are great for commentators and as talking points. But did they contain anyinformation?

My assumption was that their levels were statistically biassed but the trends were not. (The trends could be biassed, if different quota groups were behaving differently.) So I pooled them. (Others pool them too, but they dont usually allow for timing differences.)

It’s all a bit rough, if less rough taking the polls on faith. You can see the graphs of the trends at Vorb. Note I projected the trends out to election day.

For the record, the election day projections and actual outcomes were as follows:

Party Share (%): Pre-election and Prediction and Post-Election Actuals

PARTY	Prediction	Outcome
Labour	40.6	41.1
National	38.8	39.1
NZF	7.0	5.7
Greens	5.3	5.3
Maori P	1.5	2.1
United F	3.9	2.7
Act	0.6	1.5
Progressive	1.5	1.2
Other	0.9	1.2

On the whole the fit is not bad, although not perfect. (I could make it look even more impressive by rounding to the percent.) It is certainly within the ‘margin of error’ (whatever that means, as any statistician must dryly add.

I found this fascinating. It suggests what while the levels in each poll are unreliable parties, the trends may not be, and the average (poll of poll) levels may have been about right. Apparently all the idiosyncratic quota cancel one another out.

A cheer for Gauss and his Central Limit Theorem?