There’s an aspect of the political polls that I suspect many people are unaware of.
As percentage support for each party is reported, most people probably assume that more people are supporting the parties that show an increase in percent support and fewer people are supporting those that show a decrease.
It’s not quite that straightforward.
As ‘swordfish’ has emphasised in some posts on the blog ‘Sub-Zero Politics‘ (in-between amazing photos of Norway and the Faroe Islands), if you ignore the undecideds you can get a very misleading picture of the state of the political mind of the electorate.
Looking back over the last two years of Fairfax/Ipsos polling with due consideration given to the undecided voters there’s some interesting insights to be had into that mind.
Those insights also raise questions about how poll results are being reported and why they’re being reported in the way they are.
Below is a graph (and corresponding table) of party support over the last two years of Fairfax/Ipsos political polls expressed as the actual number of people per poll who affirmed support for the four main political parties.
There’s a few interesting points to make about this graph. There’s a couple of things to remember, though, before making those points.
- Each poll sampled roughly 1,000 people (usually marginally over). Avowed support for each party can therefore be expressed as a percentage of (roughly) a random sample of one thousand people. The Green Party, for example, is supported by just under or just over 100 people per poll. Avowed support per 1,000 New Zealanders is therefore 10.0% of the adult population.
- For now, I’ll ignore all the debates over polling methodology (e.g., landlines versus cellphones, who is more likely to answer phones or agree to participate, etc.). I’ll assume, that is, that the (roughly) 1,000 people sampled in each poll is a truly random sample.
Given those two points, what does the graph show?
First – and perhaps most surprising for many people who’ve read the headlines about these polls – the number of people in a sample of 1,000 adult New Zealanders who express support for the National Party has hardly changed over the past two years.
The greatest number of people out of the polling samples over that time who expressed support for National was 442 in November 2012. The smallest number was 372 in February 2013. That’s a range of 70 people over the two years – or about 7% of the 1,000 voter sample.
If you look at the trend line for the National Party (the thinner, reddish line running through the blue National Party polling line) it shows a very small incline from left to right (for the statistically inclined, the R² value is 0.02723, very low). It’s about the equivalent of adding something less than 10 voters out of 1,000 – or 1% of the sample.
Further, the largest number of people avowing support for National was in the November 2012 poll at 442 (out of the roughly 1,000 polled). That’s 44.2% of adult New Zealanders. The percentage party support reported in the poll – based on only the preferences of the ‘decided’ voters who were likely to vote – was 46.3%.
The latest poll, however, has a reported percentage support for National of 56.5%. The number of actual people in that poll who avowed support for the National Party was 439 – three fewer people than in the November 2012 poll.
So, the reported support for National between November 2012 and June 2014 appears to show a 10% increase in the proportion of New Zealanders supporting the party. Yet, there are (marginally) fewer people declaring their support for National in June 2014.
Put bluntly, National has pretty much no more support amongst New Zealanders now than it did two years ago. National is not, that is, ‘running away’ in terms of popularity. Its avowed support has barely shifted, if at all. Any headlines that speak of National ‘surging’ in the Fairfax polls or ‘increased support’ for National are therefore quite misleading.
While National’s solid but stagnant level of support might be initially surprising it shouldn’t be. The National Party vote in the 2011 election was, in absolute votes, only marginally greater than it had been in 2008. It gained 1,058,636 party votes in 2011 and 1,053,398 in 2008. That’s a total increase of 5,238 votes over a period when the total number of registered voters increased by 80,088 electors.
Second – The Green Party support, as mentioned, has also been pretty steady. As with National Party support, those avowing to support the Green Party has varied within a reasonably narrow range (though, at lower overall support, that variation does matter perhaps more for the Green Party than for National).
The number of people expressing support for the Green Party has ranged between 106 (in July 2012) to 84/85 in February 2014. (The trend line has a R² of 0.11803). The range is the equivalent of 21 people or 2.1% of the 1,000 sample size.
For both National and the Green Party it seems that the voting base remains committed and entrenched. (It’s beyond the scope of this post, but the reason for the similarly steady polling may be that the demographics overlap or commitment is more ideological for these groups – and ideology is hard to shift.)
Putting aside questions about the proportion of ‘swapping’ between parties occurring in the population, the nett effect is a reasonably constant appeal.
Third – The stand-out ‘story’ in the graph (and table) is the decline in the number of decided in the polls. The trend line for the decided vote shows a decline over the two years of almost exactly 100 people (and an R² of 0.5474). Put another way, there has been a significant increase of about 100 people in the 1,000 voter sample who are undecided who to vote for or are unlikely to vote.
To an extent that trend line is affected by the very low number of decided voters in the latest Fairfax poll, but some noticeable decline exists irrespective of that poll.
The ‘story’ of the increased number of undecided voter leads to the fourth point.
Fourth – The number of people (in the Fairfax polls) who have avowed support for the Labour Party over the last two years has declined steadily. The trend line shows a drop of about 100 people over that time – pretty much identical to the increase in the undecided vote.
The variation in support is large, mainly due to the latest poll. Labour had a high of 329 people declaring their support in November 2012 and a low of 180 in June 2014. That’s a range of 149 voters. When the latest poll is not included the range is from 244 to 329, or 85 voters. That’s a slightly higher range than for the National Party voter support.
Even before the latest poll is included the Labour Party’s avowed support is more highly correlated with the proportion of the decided vote than is the National Party support. I’ve made a table of the correlation coefficients of each party’s support over the two years and the number of decided voters in each sample.
Over all the polls (the top half of the table) the Labour and NZ First support is highly correlated with the level of the decided vote. That is, the higher the decided vote the higher the number of people declaring support for these parties. Both National and the Green Party – while still positively correlated with the number of decided voters – show a lower correlation.
Basically, that means that for National and the Green Party supporters, as already noted, have a greater tendency to be ‘entrenched’ in their decision at least at the population level. Fewer of their supporters appear to switch from support to ‘undecided’.
For Labour and New Zealand First the story is different. Their support has a greater tendency to switch to ‘undecided’ (or to be less likely to vote).
When the latest poll is not included in the correlation, however, National’s level of support shows closer correlation to the level of the decided vote (the correlation coefficient goes from 0.325234216 to 0.700879977) but still less than the correlation for Labour and New Zealand First support. But it’s important to remember that, while correlated with the decided vote, the actual fluctuation is greater for Labour and New Zealand First than for National.
The major change in the correlation coefficient for National between inclusion and non-inclusion of the latest poll reflects the fact that the National Party support increased (from 393 to 439) while the number of decided voters decreased from 826 to 777.
Even without the latest poll included the Green Party, interestingly, still has a relatively low – though positive – correlation with the level of the decided vote.
There’s a further correlation that adds some insight into what’s going on. Below is a table of between party correlations.
All correlations are small. The ‘largest’ is a negative correlation between the National and Labour Party support. (Negative means that as one party’s support goes up the other party’s support goes down.) It’s also interesting that the relationship between support for Labour and the Green Party is practically non-existent in the Fairfax polls. (Interestingly, when the latest poll is removed from the calculation of the correlation between Labour and National there is a positive correlation between the support for the two with a correlation coefficient of 0.391170648. That is, when National’s support goes up, so does Labour’s.)
The main ‘story’, from the Fairfax polls at least, is a story of the relationship between Labour Party support and the level of ‘undecided’ vote. That is, when the avowed Labour Party vote goes down so does the level of decision in the electorate overall.
Polls, including the Fairfax polls, headline the percentages for the declared voter support. Without taking into account the level of the undecided vote this presents a misleading picture of how many voting age New Zealanders support each party.
Of course, by limiting reporting only to those who have declared party support and are likely to vote the reporting may very well reflect what could happen in an election.
With that point in mind, here’s perhaps the most revealing correlation between the ‘headline’, reported voter support and the level of the decided vote in the Fairfax polls:
[Without the June 2014 poll result, the correlations are: Labour 0.259225079; National -0.416534116. That is, still a positive correlation, though lower, for Labour and still a negative correlation, though slightly lower, for National.]
There’s a very clear story in these two correlations: Put simply, as the decided vote goes up so does the reported percentage vote for the Labour Party.
Conversely, as the decided vote goes up, the reported percentage vote for the National party tends to go down.
The closer the election draws the more likely it is that people will make a decision.
But then there’s one more step – getting people to put that decision into action and actually vote.