Election spinner

Bernard L. sent in this chart a while ago, and with the looming British elections, it's a good time to show it, and ask readers how to spin this election. (Via Guardian)

In particular, could someone help me understand the tri-color spinner? Given that the change in seats for the three parties combined should be zero, I don't get how this can fit into a concentric-circles presentation. If you click on the link to the original chart, you can move the black dot around the circle.

In addition, I'm mystified why the constituencies can be depicted on a graph paper, each one the same size as the other. This is not the first time I have seen the U.K. mapped in this way so there must be some reason behind this choice. (For reference, I have never seen the 50 states mapped in this fashion.)

Reference: "Election Map and Swingometer 2010", Guardian (UK), April 5 2010.

Comments

[kiwispoon]

I think the way the spinner works is that if the black dot is in the centre then the map shows the seats won by each party in the previous election. When you move the dot the 3 figures to add to zero. In the case above there it predicts the number of seats that the conservatives will win if the vote swings 3.91% to them (at the expense of labour 1.37% + Libs 2.54%). I think this assumes a uniform swing across the whole electorate. Sorry I am a kiwi so I can't answer your question about the size of electorates (I imagine they are roughly equal populations so I would expect city based electorates to be geographically smaller than rural seats.
Cheers
Stephen

[Chris Baker]

Hi, British reader here,

The circle makes sense to me, the further out from the origin you get, the more seats a party wins, if you move the point as far as it goes, in the middle of the segment, the party takes seats 50/50 from each of the others.

It actually controls two things:
How many seats a party takes, and who they take them from.

Your second point, the reason the constituencies can be plotted like that is that each constituency has the same voting power (Each elects one MP)

I understand our system can be a little complicated, but then so can the US one!

Thanks,
Chris Baker

[Ken]

I think the mapping works because population density in the UK is less variable than elsewhere and where the density is lower the electorates tend to have smaller population size.

All I can say about the swingometer is that it seems to work.

[Keith Henry]

It's due to the funny and broken way parliamentary seats work in the UK.

Basically each seat tends to be skewed to a particular party.

For instance a rich Southern countryside seat will tend to have a high density of Conservative voters.

Conversely, a poor Northern industrial town will have disproportionally many Labour voters.

The problem is that polls are national - changes in voter opinion are spread across seats that are already biased one way or another.

So the statisticians came up with 'swing'. They take the change in support at the national level and apply it as a ratio to each seat.

So, suppose the Conservatives have 5% more support nationally and Labour 5% less. To apply the 'swing' we add 5% Con votes in each seat, and take 5% Lab votes off, and then see who wins in each case.

Some seats have huge majorities and are called 'safe' they need a swing of 20% or more and that just doesn't happen.

Other seats (like mine: http://www.voterpower.org.uk/wirral-west) are marginal - a small % could swing it either way.

They've used swingonometers in UK elections for years, but this is the first in about 90 years where it's become a 3 horse race, hence the funny three way dial. I don't think they've quite figured out how to make swing work for 3 parties yet.

You don't get this in proper democracies (which we technically aren't by the UN's 2005 definition) because it only makes sense with our weird first-past-the-post system.

Also each seat has the same power, regardless of the relative population size or area covered - we don't have the weighting based on population that US states get. That's why the seats get shown as a grid.

[widdowquinn]

Each constituency elects one MP, as Chris says, and additionally should contain approximately the same number of people (see http://www.parliament.uk/about/how/elections/constituencies.cfm for more details).

[Meng Bomin]

The cartogram is clearly designed to show the result of voter swing on the seat count in Parliament. The concentric circles represent the net percentage of voters switching their votes in a given party's direction. The whole point of the exercise is to convert a predicted voter swing into the amount of seats in the House of Commons, because like all single member district plurality elections, the popular vote only matters insofar as it alters the number of seats, as those are what determine the actual balance of power in the next British government.

U.S. constituencies aren't typically mapped the same way because the elections for House of Representatives don't get nearly the attention that the elections for the House of Commons, which makes sense as the House of Commons elects the Prime Minister, whereas the President is elected separately in the United States. The U.S. also lacks a strong third party, which the UK has in the Liberal Democrats and strong third parties make SMDP votes more interesting and unpredictable than anything you would see in a House of Representatives election.

If the map were to show the constituencies at their geographical size, it would artificially inflate the perceived performance of the Conservative Party, as the Tories do well in more rural districts, whereas Labour has more urban strongholds. That is why they show constituencies as blocks on a grid. It gives a visual representation of the seat count in Parliament while maintaining some semblance of geographic locality.

Now, the main problem with this map is that it uses a very simplistic and likely inaccurate model that involves a uniform voter swing. Nate Silver has a good run down of some of the problems with that model.

That said, a similar feature is offered at the BBC website, though the set-up is better, in my opinion: users manipulate the percentage of the vote received by each party rather than the swing and the cartogram (which uses hexagons instead of squares) includes sub-national boundaries to better allow users to understand the locations involved.

Finally, the seat count change of the three major parties need not add up to zero precisely,because there are regional parties like the Scottish National Party that may gain or lose seats in the House of Commons in addition to the seat shifting between the big three.

[Barry Rowlingson]

There's some similar gridded maps of the USA here:

http://commons.wikimedia.org/wiki/Category:2008_U.S._presidential_election_maps

this one is scaled and shaded by number of electoral college votes:

http://commons.wikimedia.org/wiki/File:Cartogram-2008_Electoral_Vote.gif

[Wade Dorrell]

The "spinner" makes sense to me as a what-if control to adjust 3 numbers which must sum 0, but the concentric circles are mostly non-functional & confusing. They are supposed to help you set or maintain a particular swing value for one party (say 2%, the 2nd ring) while adjusting the other two, but for maintenance I think it's a rare person who can follow a curve with a computer mouse... the lines did not help me even stay within 1% as I tried to change only the other two. It was easier to move the mouse/black circle in a straight line from the center and watch the 3 labels; in that case the rings are not of much help, watching 2 labels is as hard as watching 3.

[derek]

I've seen US Congressional seats represented the same way as a cartogram, so I'm surprised you're surprised.

I think the circle is the small central part of a larger triangle, and the swings can be no more than 10% in any direction.

[Matthew]

"In particular, could someone help me understand the tri-color spinner? Given that the change in seats for the three parties combined should be zero, I don't get how this can fit into a concentric-circles presentation. If you click on the link to the original chart, you can move the black dot around the circle."

There are other parties, but the change does (in seats above) approximate to zero.

In addition, I'm mystified why the constituencies can be depicted on a graph paper, each one the same size as the other. This is not the first time I have seen the U.K. mapped in this way so there must be some reason behind this choice. (For reference, I have never seen the 50 states mapped in this fashion.)

It's to show population. Roughly the constiuencies are the same size in population terms, about 90,000 - there is a spread, with the Isle of Wight something like 120,00 as it is an island - and some inner-city ones about 50,000, but most are similar. So the idea of that chart is to show how the country really splits. If you show it geographically, then the Tories look completely dominant as the urban seats shrink. See http://www.visionofbritain.org.uk/images/statmaps//elections/election_1997-05-01.png

That's Labour's (RED) famous landslide, on about 390 seats to the Tories (BLUE) 190 or so.

[Kaiser]

Thanks for all the helpful comments.

I think the spinometer is a graphical failure (I understand it's been used for years, typically with only two parties.) I take it that there are really just two variables (Labor swing and Tory swing, assuming constituency swing is equal to national swing).

First problem with the Guardian graphic is the implied symmetry between the three parties. When you move the dot around, you'll notice that the swings are not symmetric; in other words, putting the dot on the outer edge of the blue sector does not give the same numbers as putting the dot on outer edge of the red sector. Of course, the asymmetry is reality. I'm just saying the chart conveys the wrong impression.

Second, it would be much simpler to dissociate two concepts: one party's projected positive swing, and the proportion taken from each of the other two parties (or just one). In this spinometer, the former is the distance from the center, and the latter is the angle. Would it be simpler if they just have a few scroll bars to set these variables?

The BBC graphic is definitely better. But it seems to allow certain extreme cases (completely unrealistic ones) which are nonsensical.

As for the cartogram, interesting to learn that each constituency has about the same population, and one MP. That's different from the U.S. Senate in which each state has a Senator but obviously very different populations.

[Matthew]

But not that dissimilar from the US House of Representatives, surely?

[Tomp]

Hi, I built the BBC's version of this app so thanks for the kind words.

We tried several different interface mechanisms before settling on the pie-chart control. First up we tried a ternary plot with a moveable point, this had the disadvantage of being limited to 3 parties and we wanted to show 'others' in order to closely link the graphic to opinion poll data (or more specific parties in the future, perhaps for Scottish parliament, Welsh assembly election etc.). Also the ternary plot is perhaps a little bit esoteric for a general audience.

A set of sliders was unsatisfactory as the UI for locking unlocking different party sliders and determining from which party votes should flow to which other party was a usability nightmare where more than 3 sliders are concerned, plus there are potential editorial policy issues (i.e. we might be open to accusations of bias if by default votes transfered to a particular party).

I think the pie control is intuitive (whatever we think about pie charts everyone understands how they work) and the only real draw back is, as Kaiser points out, that unrealistic scenarios are available to users. We hoped to guide people as to what 'realistic' scenarios would be with the links to current opinion polls and historical vote shares. We considered limiting the range of the pie sections but where do you draw the lines? and in all honesty playing with the full range is sort of fun and hopefully illuminating as to the mechanics and shortcomings of the model.

With regard to the proportional representation, we chose that not only because the traditional geographic map gives the physically larger traditional Conservative and Lib Dem seats disproportionate visual prominence but also in order to further emphasise the crude nature of the uniform national swing model. For the same reason we don't give information on which hexagon represents which constituency (though the identities of some are obvious and most can be worked out by an astute reader with time on their hands).

[Joel]

More of an odd map than a junk chart, but Ben Hennig at the University of Sheffield has created three charts mapping constituencies geographically, by equal weight, and by population distrbution. The last one looks slightly wacky at first glance. I'm not sure it's particularly readable but it's a different take on data mapping:

http://benhennig.postgrad.shef.ac.uk/?p=481

[Tic]

In my opinion ternary plot would be perfect. Yes, it allows only three parties, but there are only three parties in this situation! And it is impossible to move votes between parties with more that three parties whatever plot is used.
Moreover, with the pie control I was not able to move all 10% of the votes to a party to another party, that is, to make a party rate +10%, another party rate -10% and the third one unchanged. I think it is not possible with the current control settings. Am I wrong?

[Tomp]

(Most delayed response to a commebnt ever!)

I was not able to move all 10% of the votes to a party to another party, that is, to make a party rate +10%, another party rate -10% and the third one unchanged

That is possible Tic, but depending on what parties you choose it may be a two step process. Our ternary plot prototype showed it was really tricky to adjust accurately, moving the mouse in precise direction and increment over a 2d plane is much more chanenging than sliding it round a single dimension.

I recently wrote a blog post about the development of the app...
http://blog.pointlineplane.co.uk/post/2174827384/case-study-seat-calculator