Swingometer

[ciphergoth]

Paul's two dimensional three-way swingometer

This is still a draft version of the swingometer, posted here to get some feedback on how to make it more comprehensible and more useful. A three-way election is a rather challenging thing to illustrate on a diagram, so I hope you'll forgive this being a little harder to follow than a normal two-way swingometer!

• Point the mouse at the "2005" on the image and you'll see the result for 2005; the share of the vote for each party, and the number of seats they hold. Strictly speaking it's a projection of 2005's result onto the 2010 constituencies.
• Move the mouse around the diagram and you'll see projections for what sort of parliament you might see as the vote share changes. These projections are based on the very simple "Uniform National Swing" model, which has many failings but AFAICT more sophisticated models don't do much better in practice.
• The colour of each hexagon illustrates who has the most seats; a majority is shown in a darker colour. Larger majorities get darker colours. Where two parties are exactly equal, we use an in-between colour. Just compare the colour of the hexagons with the share of seats indicated for each one to understand the colour scheme.
• The straight lines across the diagram indicate the lead in points of the party with the largest vote share. Where the three lines meet is the point where all three parties get the same number of votes.
• A fixed proportion of votes are shared between the three main parties It would be better if you could change how many votes go to other parties, but it's much harder to do.

I have plans for updating this diagram during election night, but they're even more complicated, so I'll come to that later. Please let me know what you think - does it make sense?

Comments

[mooism.livejournal.com]

It makes sense to me, and looks good.

Possible improvements:

Sometimes the vote share / seats totals box disappears off the top of the screen. This seems to be because I'm using my netbook, so there isn't much vertical screen, and you have either the top or the bottom of the box aligned with the mouse cursor. Neither top- nor bottom-aligned leaves the whole box visible when the mouse is in the middle of my screen.

It would be nice if when it showed the UKPR average, it also showed the vote share and seat totals that it shows for other points on the map.

An explicit sense of scale would be nice. Are the lines spaced at 10% swings?

Maybe show the most recent polls from various pollsters as well, to provide a sense of how much uncertainty there is about the current position?

[ciphergoth.livejournal.com]

Sometimes the vote share / seats totals box disappears off the top of the screen. I'm using the jQuery Tooltip library to create this box; I'm not enough of a JS hacker to try to fix this I'm afraid.

An explicit sense of scale would be nice. Are the lines spaced at 10% swings? Yes. These need labels, but I'm not sure exactly where to put the labels. Also, do these labels need to say any more than just "10%"?

Maybe show the most recent polls from various pollsters as well, to provide a sense of how much uncertainty there is about the current position? Yes, that's definitely part of the plan.

It would be nice if when it showed the UKPR average, it also showed the vote share and seat totals that it shows for other points on the map. I want that. The trouble is that the polls vary in an extra variable - the number of non-big-3 votes - that my diagram can't chart. Probably I just need to calculate a special box for each poll; I worry that would cause confusion but I don't see a better way.

Thanks for your comments!

[josington.livejournal.com]

Would putting the stuff that's currently in tooltips in a separate box to one side be an option? I'm also having netbook tooltip issues.

[josington.livejournal.com]

Actually, the tooltips are displaying fine for me as long as I view it fullscreen.

[ciphergoth.livejournal.com]

That's an acceptable fix. Would be good to work on netbooks since that's what people will have on their laps while watching election, but the tooltip effect looks really nice...

[henry_the_cow]

I agree with much of the above. It's good. The main thing missing is some sort of labelling on the main map; currently it is like a topographical map in which the contours aren't labelled, so we don't know if we're looking at mountains or foothills.

Also, the pop-up bar chart doesn't always display the +/- from 2005. (When it does, the plus sign is too small for my old eyes to clearly distinguish from the minus sign).

As you note, polls have the problem of small parties and undecided votes. Could the size of uncertainty be indicated by circles on the main map? I suspect this would add a degree of reality to the impression that people take from these polls. You'd probably want an option to toggle it on/off though, otherwise the map would be completely swamped by opinion polls and their margins of error.

[ciphergoth.livejournal.com]

Hmm, need to clear that up - the +/- refers to the majority finishing line, not 2005. Is it worth displaying change in seats from 2005?

Gathering enough information to properly display poll uncertainty could be hard - hopefully the way the polls scatter would get that across.

[henry_the_cow]

Oh I see now. I didn't grok the meaning of the +/- at all, not the horizontal line. And I wondered why they disappeared sometimes (i.e. when they are all <-100).

This negates my comment about the size of the + sign. I was assuming, without looking too closely, or thinking, or anything, that if one number was a -, the other must be a + (because I was assuming a swing).

Ideally, these negative/positive numbers would somehow visually indicate the distance from the horizontal line, but I don't see how to achieve that, especially when the edge of the bar is close to the line.

Maybe a simpler approach would be to label the horizontal line with the number required for a majority. (I did

I agree with the suggestion of adding % signs to the upper graphic in the pop-up.

[ciphergoth.livejournal.com]

I'm labelling the line with the majority number now; I'm going to add text as drdoug suggests for the rest.

I tried adding the % signs to the upper graphic but that really doesn't look good; I've put a % in the centre of the graphic instead, which I hope does the same job but looks quite a bit nicer.

[pengshui-master.livejournal.com]

I wondered if you where going to do this, given your work on the US election. Good show.

BTW, You get slightly different results that this site here:-http://ukpollingreport.co.uk/blog/advanced-swingometer-map

Which is as far as I can tell all done in javascript so you can probably easily tell where your models are different - which is hopefully the reason.

Would it be possible to show the path of the polls, eg draw the line which represents the changes in the polls as the campaign progreses?

[ciphergoth.livejournal.com]

Weirdly, that page is blank for me - decoration but no payload.

[pengshui-master.livejournal.com]

Hmm,

Ahh. The page is JS + SVG are you using a browser without SVG support?

[ciphergoth.livejournal.com]

I'm using Firefox 3.5.9 under Ubuntu 9.10.

[pengshui-master.livejournal.com]

Curious. Works fine here under Google Chrome (4.0.266.0) and Iceweasel 3.0.6 both on a Debian lenny/squeeze mix.

[ciphergoth.livejournal.com]

Curiouser still - it's just started working.

[lizw.livejournal.com]

It makes sense to me, yes.

[lovingboth]

I am not sure if I kept it, but the Guardian did a paper one in 1992 or thereabouts, and it looked just like yours.

I suspect the main issue will be the model's accuracy. I think UNS did badly in 1997, for example, because it misses the effects of people going 'I will vote for whoever will beat X' (i.e. the Tories in 1997).

[ciphergoth.livejournal.com]

I remember a Guardian swingometer based on a hexagonal chart; they took two pages, and wrote the result into each hexagon. Can't remember if it was 1992.

[lovingboth]

The other possibility is 1987... I'll see if I kept it.

[ciphergoth.livejournal.com]

1987 is a better fit for the memory I have of it!

[lovingboth]

Found it - it was 1987. Do you want it?

[ciphergoth.livejournal.com]

If I get around to doing more with that swingometer, a picture of the Guardian one enough to give the flavour might be nice!

[drdoug.livejournal.com]

Cool! As I tweeted, I grokked it pretty quickly when I first saw it. But that's probably not a good guide to general comprehensibility. :-)

I don't think you need the horizontal lines (marking round fifties of majorities SFAICT), just the single 'finish line' of a majority of seats. They're only remotely useful in the far outer reaches of the graph - where we are talking about majorities over 100 - and this election is so not going there. A single line would also be clearer as a finish/win line.

More labels would probably help the uninitiated, but obviously, many more and it'll get unusably crowded. Maybe try % signs after the vote shares in the popup? Would make that instantly clear what that means, I think - vote shares are the main thing that are percentages in elections, apart from swings.

The vote-share black lines could be labelled near the edges (since they're least useful), and you could leave the centre line unmarked. Then label the segments of lines something like 'L 10% > LD', 'C 20% > L', or 'L vs C +10%'. The minimal solution is just labelling them '10%', '20%', which'd be neater. But I like the idea of labelling the ends of the lines separately to cue people in to what they mean. I doubt that most people would instantly understand that - for example - the black line in the bottom left-hand corner showed where Labour had 30% greater vote share than the Lib Dems in the upper segment but where it had a 30% greater vote share than the Tories in the bottom segment. And understanding that is key to understanding the whole chart, I think.

I wonder if you could lose the +/- majority bit from the top of the seat graph columns. The +/- strongly makes one think of a change from the last election (or at least it does me), so this is another confusion you could lose. It is handy information, though - perhaps you could add it in as a line of text underneath? Saying, e.g. 'Lab maj 25' or 'Hung, Lab 65 short, Con 63 short'. (Using the same algorithm for when you show both Lab and Con short of a majority rather than just one - which empirically seems to do the job beautifully but I couldn't instantly see how it worked.) Might take up a lot of space, though.

The 2005 point is cool, and to me made it clear that the boundary changes would have seriously eaten in to Labour's majority last time. But that's because I know that the Labour majority was way more than the 21. Can you add a specific point in that popup? Or add to a series of notes at the bottom of the page something like 'The 2005 general election vote share is shown on the graph for information. In 2005, Labour won a majority of 66, but this model suggests that with the boundary changes since then they would only have had a majority of 21 seats.'

The two purple hexes look more significant than they are (SFAICT they are where the vote share whole-number percentages happen to map on to a perfect Lab/Con tie in seats) - maybe change to a vertical red/blue split?

Another suggestion for an explanatory note: You can see that the red area of the graph (Labour majority) is much larger than the blue area (Conservative majority), and both are hugely larger than the yellow area (Lib Dem majority). This dramatically illustrates the difference between our current electoral system and a pure proportional representation system (where the areas would be equal).

I don't think you want a 'change from last election' in there (crowded enough). Maybe yet another note - perhaps just adding seat numbers to 2005 note. Ah - perhaps you could have it as an extra thing underneath the whole graph. Actually, it'd be good if as well as the popup you could have a panel underneath that would show more/fuller/more fully labelled information, updated by where you last mousedover. (Mouseovered?) Ooh - actually, it could fix underneath if you clicked on a hex. That'd be very cool because as a user you'd be able to click on a point of interest, then navigate away from the page to sth else (e.g. Twitter, other poll results, or the live feed, or your blog, or whatever) and the data from the point of interesting in the graph would still be there for comparison without you having to hunt down that specific pt again. But can imagine the JS being hairy.

Out of time & chars!

[ciphergoth.livejournal.com]

This is incredibly helpful!

Will eliminate all but one of the horizonal lines as you suggest.

Have added "10%" markers on lines, but could add party names; will try it. Also will add % marker to vote share donut.

Will lose +/- in favour of text description as you suggest. Algorithm no more than cutoff at -100.

Didn't realise impact of boundary changes was so great! Yes, will add to text.

Vertical red/blue split quite awkward to achieve but very desirable - will think about that one.

Will definitely comment on how unfairness of electoral system is illustrated in text.

Really like the "click" idea. All ideas on what should be presented in the oodles of space that would allow very welcome.

Many, many thanks!

[drdoug.livejournal.com]

Algorithm no more than cutoff at -100.

Yay for absurdly simple algorithms that do a job well!

Will definitely comment on how unfairness of electoral system is illustrated in text.

My suggestion isn't quite accurate anyway, I realise.

the "click" idea. All ideas on what should be presented in the oodles of space that would allow very welcome.

For hung parliaments, it would be cool to show what majorities were possible - specifically, whether or not Lab/LibDem and/or Con/LibDem alliances would command a majority. Prob not much point to show a Lab/Con alliance commanding a majority, since it almost always would - maybe invert that one in the shiny yellow sliver and show 'Lab/Con alliance no majority' for the unlikely scenarios where this applies, to keep LibDem dreamers happy.

Many, many thanks!

You're welcome. Many, many thanks for making this! It's a cool tool and I shall use it repeatedly between now and May 8th.

[drdoug.livejournal.com]

... another quick thought on hung parliaments: in many scenarios where a party is close to a majority, the minor parties become hugely important. A Labour/SNP/Plaid/SDLP coalition is not entirely impossible, for instance, and nor is a Con/UUP/DUP(!) one. So if you're enumerating possible majority Government alliances, it'll be painfully obvious that restricting to the three biggest parties isn't covering all possibilities.

But OTOH, when you get down to that level of a handful of seats, you're well beyond the likely accuracy of a model like this, I reckon. Especially given that your model has to assume a fixed number of seats to 'others'. And there's all sorts of obscure fencepost type issues that generally don't matter but might suddenly do so in close-to-majority situation, like how many non-voting seats Sinn Fein occupy, the Speaker+Deputies, seats vacant due to death of candidates (there's at least one), etc. And of course the entirely non-zero probability that actual Independent MPs are highly significant.

So maybe just another point for the Notes.

[ciphergoth.livejournal.com]

Yes, I think trying to track the non-Big-3 parties is going to be out of reach; best that can be done is to say how many seats are held by "Other" parties, how far each Big 3 is from majority, and how far Lib-Con and Lib-Lab are.

One annoyance: if you've got 5 out of 9 seats, you have a "majority of one", but if you have 5 out of 10, you're "one short of a majority" - like BC/AD, there's no zero.

[ciphergoth.livejournal.com]

You're welcome. Many, many thanks for making this! It's a cool tool and I shall use it repeatedly between now and May 8th.

thanks! Will soon discuss plans for how to use it during election night - broadly, diagram will show makeup of Parliament assuming that *remaining* seats show a particular swing.

[fizzyboot.livejournal.com]

I think you should make the 3 sections where there is a hung parliament a lighter colour, to indicate this more clearly. Also, you might want to explicitly label the different sections.

[juudes.livejournal.com]

I get a general idea of what it does but I'll need to think about it some more!

When you say "Where the three lines meet is the point where all three parties get the same number of votes" do you mean "same number of seats"?

Can you have a button that toggles extra labels on and off, so that when you understand it you can get them out of the way?

[ciphergoth.livejournal.com]

Where the three regions of colour meet is where they have the same number of seats. Where the black lines meet is where they have the same number of votes.