Tuesday, 21 June 2016

Breferendum Branalysis

It's less than 24 hours until voting in the UK referendum on leaving the European Union begins, so another brief UK-focused post (or "Brpost") must be due.

Unfortunately, referenda in Britainia lack the nuance and complexity of our sophisticated Australian ways; they've opted for an odd system whereby they count the votes, and decide the winner on shear numerical volume. This austere simplicity deprives amateur analysts like myself (known as "Bramateur Branalysts" in the UK) of most of the fun number-juggling, and leaves us with a sense of disgust that can only be properly expressed by the great Clint Eastwood.

I mean, all any competent person needs to do is apply the most recent polling data and, allowing for the poll's margin of error, they have a result. Anyone looking to delve any deeper than that must be a truly sad and obsessed individual.

The Deeper Delvings of a Truly Sad and Obsessed Individual

While the polling for this referendum necessarily accounts for all the latest swings in popular opinion, it's history so far has been erratic and volatile. That means today's polling is little use tomorrow, but also that extracting meaningful trends or predictive information is (at first glance) problematic.

Data sourced from the link above.
I know. I know.

Things only slightly improve visually if we look at these polls as snapshots that capture some volatile trend, ignoring the fact that there is no reason to suspect the variation in the data represents any pattern of rises and falls.

Now this looks like the early race had the remain vote ahead, and that things have tightened up over the last few months. To help clarify this data, I've conveniently shaded this chart with the colours leading votes at any given point in time:

On this data, it looks as though the remain vote truly did dominate the early contest, but its moments in the lead have become shorter and its interruptions more frequent. On the other hand, this might just be the result of more regular polling as the referendum draws closer - particularly online polling.

Online polling is cheap and easy, but also problematic for a number of reasons - even ignoring the risk of poll hacking. For one thing, online polling attracts a certain demographic of people and does not represent society as a whole (though this is a far larger problem in compulsory voting systems)

For another, online polling is often (I haven't checked in these cases) un-corrected to ensure the demographics captured are adjusted to reflect the likely voter demographics (this is harder to do in non-compulsory voting systems, though, as the voter demographic is harder to establish). Where corrections do occur, they often rely on very small sample sizes for key demographics, in particular the elderly.

Another common issue is that online voting is almost always opt-in, which might not seem like a huge issue in a non-compulsory voting system, but it tends to disproportionately attract vocal minorities.

So, just for the sake of curiosity, what happens if we only use the phone-polling - a slower and more expensive process, but an opt-out one often targeted to ensure decent demographic sampling and corrected for any demographic shortfalls:

Son of a ballot...

While the trend for exit votes is still more evident in the recent months, this data makes these look more like outliers or sampling errors than anything predictive of the referendum. Now, yes,phone poll data has its issues too. Phone polling tends to favour older voters, with fewer young households maintaining a landline phone - although phone polling is addressing this through dialing mobile numbers as well for many national campaigns. And the elderly are statistically more willing to have a talk with pollers, telemarketers or anyone who calls to interrupt the monotony of their day and distract them from the fast approaching shadow of the grim reaper.

Then again, older voters are stereotypically more conservative and less likely to vote to remain in the EU. And, as I said, targeted polling and data correction make phone polling more reliable than its malformed online sibling.

I don't watch Game of Thrones but I believe it is topical, and also maybe this is the most generous way to describe online polling?

In the interest of fairness, though, I took a look to see if the online data on it's own was neat and informative. Let me preface these results with a simple:

Don't say Clint didn't warn you.
My problem with this isn't that there's a lot more chopping and changing. It's that there's no obvious pattern, beyond the tentative slide towards more pro-exit voters obvious in our earlier data.

Part of the problem could be the mix of different polls. Different polls on different sites asking different questions of different visitors lured into the poll by different means. So here are the repeat polls (those with more than one data point) individually:








This allows any possible trends captured by individual polls to emerge, but to be honest the data seems all over the shop. We could aggregate these by overlaying the colours...

... which makes things as clear as mud. But if we then divide these various shades into stark categories of exit/remain/undecided...

We can, at least, see once again that support for an exit is growing.

In Conclusion...

From all the data we have, it seems that support for leaving the European union is on the rise as a general trend. Whether this is enough to carry the day, however, is another question. The latest combined polling and phone polling results both have the most recent polls as votes to remain in the EU, with an upwards trend. The less reliable online polling has the most recent data as pro-exit, but on a downwards incline.

While I wouldn't read too much into the slope of these graphs, this does suggest in the short term that the referendum is likely to support remaining in the EU. Additionally, looking at the duration of pro-remaining leads (and assuming there are actual cycles and trends to be observed), it seems as though the current remain lead in the more reliable polling should last for more than enough days to carry the referendum.

On these numbers the Brexit referendum, like the Scottish Independence vote, has been labelled "to close to call" by much of the media, but once again I feel confident to predict no change and the perseverance of an established union.


  1. I very much enjoyed reading these deeper delvings of a truly sad and obsessed individual. Nice work Dan.

  2. Slight issue. You've used red for Remain and blue for Brexit. These colours are swapped the exact of opposite of the way that all the locals for it.
    Apart from that, we had some good giggles here. Thanks!

    1. BBC's gone Blue for Brexit and Yellow for Bremain.

      Maybe we need a vote on what the colours should be.