Wednesday, 29 July 2015

Party Vote Demographics: Part I

I was thinking about what the “average voter” for each political party looks like, and my marketing research training kicked in. Pull a bunch of demographic data from the census, try to match it to party vote data from the last election, and hey presto, we should be able to build some reasonably interesting profiles.

It turns out that it’s a bit harder than it seems, and as a result there are a lot of caveats. We can’t match individual demographics to individual votes, because if that data was available it would be a reasonably significant breach of privacy. So we have to decide what level of granularity is sufficient for our analysis – after digging through meshblocks and area units, eventually I settled on electorate demographic comparisons; in electorates where there are more people of x category, were there more or less votes for y party?

Note that I am being careful with my wording there (and I will do my best to be careful with my wording throughout these posts, but will inevitably slip up somewhere). Most importantly, these are correlations, which does not imply causation. This is important, so I will say it again – correlation does not imply causation. If you read this post and make unconditional declarative causative statements I will be upset at you. For many of the correlations I found, it is entirely plausible for it to be a coincidence or for there to be some other factor that explains the relationship. It is tempting to say that certain groups of people are more or less likely to vote for a particular party, but we must remind ourselves that these statistics do not imply causation; these statistics cannot prove that the fact that an individual belongs to a particular demographic group causes them to vote for a particular party. I should also note that all discussions in these posts about why the correlations exist are largely guesses/opinions and not scientific.

Just so that I don’t bore the general audience too much, other statisticsy things that I did to try and make things robust and fair are explained in an appendix post for those who are interested.

For each of the correlations, I’ll include the correlation coefficient, or r statistic. This is a measure of how strong the relationship between the two variables is, ranging between -1 and 1. If r is negative, then as one variable increases the other decreases, and if the number is positive then as one variable increases the other also increases. r=0 would indicate exactly no relationship. The larger the magnitude of the number, the stronger the relationship. For example, r=-0.25 would be a weakly negative relationship, r=-0.8 would be a very strongly negative relationship, r=+0.4 would be a reasonably strong positive relationship, and so on.

To give us some direction, I figured maybe what we should do is explore some commonly held stereotypes about the political parties, and see if they were reflected in the demographics and voting statistics. Let’s start with income and age.

“Richer people vote for ACT, the Conservatives, and National, poorer people vote for Labour”
In electorates where there are more people (aged 15 and over)…
… with zero income, more people voted for ACT (r=+0.358), fewer people voted for Democrats for Social Credit (DSC) (r=-0.472)
… in the $10,001-$35,000 income bracket, more people voted for NZ First (r≈+0.6), and slightly more people voted for DSC (r≈+0.25)
… in the $15,001-$25,000 income bracket, slightly more people voted for the Conservatives (r≈+0.23)
… in the $25,001-$30,000 income bracket, fewer people voted for ACT (r=-0.575)
… in the $25,001-$40,000 income bracket, more people voted for Ban1080 (r≈+0.23)
… earning $50,001 or more (per year), slightly more people voted for National (r≈+0.25)
… earning $50,001 or more (per year), more people voted for United Future (r≈+0.4)
… earning $60,001 or more (per year), more people voted for the Greens (r≈+0.5)
… earning $70,001 or more (per year), more people voted for ACT (r≈+0.38)

Discussion: The statistics would suggest that electorates with richer people do vote for ACT and National, but also vote for United Future and interestingly the Greens! The Green relationship in particular is possibly explained by the support for the Greens in central urban areas, especially in Wellington, that also happen to be areas with higher income individuals. Electorates with more low income individuals did not vote more for Labour. Surprisingly, electorates with more people with zero income also had more people vote for ACT, which could possibly be explained by those electorates having more stay-at-home housewives or young students with no income, dependent on the high(er) income of the main breadwinner in the household.

“Old people vote for NZ First”
In electorates where there are more people…
… aged 15-39 years old, fewer people voted for NZ First (r≈-0.4)
… aged 50-79 years old, significantly more people voted for NZ First (r≈+0.7)
… aged 50 years and older, more people voted for the Conservatives (r increasing from +0.209 to +0.65)
… aged 50-84 years old, more people voted for DSC (r≈+0.47)
… aged 60 years and older, fewer people voted for InternetMANA (r decreasing from -0.164 to -0.406)
… aged 45-54 years old, more people voted for the Māori Party (r≈+0.25)
… aged 60 years and older, fewer people voted for the Māori Party (r decreasing from -0.124 to -0.417)
… aged 55 years and older, more people voted for National (r increasing from +0.252 to +0.515)

Discussion: The stereotype largely holds up – electorates with more young people voted less for NZ First, and electorates with more old people voted more for NZ First (Bay of Plenty, Tauranga, Coromandel, and Whangarei). Other “old-friendly” parties included the Conservatives, DSC, and National, while InternetMANA and the Māori Party were less popular in electorates with more people aged 60 years and older.

“Young people are more left-wing”
In electorates where there are more people…
… aged 20-29 years old, slightly more people voted for the Greens (r≈+0.23) and the Civilian Party (r≈+0.16)

Discussion: The Greens do well at attracting the youth vote on the back of long-term sustainability policies, and they had a lot of party votes in student-heavy electorates like Christchurch Central, Dunedin North, Auckland Central, Rongotai, and Wellington Central. Younger people probably also take politics less seriously (or alternatively are more disenfranchised with the system), hence the Civilian Party. It’s a little odd that there is no statistical relationship between young people and Labour though.

The interesting thing (at least to me) about these statistics is how they reveal people’s biases. The statistics are hard cold truth, but how we choose to interpret the statistics is another matter. Whether we allow ourselves to question our biases or just selectively reinforce them is something I find fascinating. Coming up – more voter demographics!

No comments:

Post a comment