Archive for the 'Commons seat predictions' Category

h1

Combination therapy. An occasional reminder that using seat predictors on current polling is stupid

Sunday, August 11th, 2019

You see it happen regularly, on here and on twitter. A new opinion poll comes out, showing dramatic news. Immediately, we decamp to Electoral Calculus and Flavible, to discover that such a poll, if replicated at a general election, would produce a hung Parliament with Plaid Cymru the largest party, able to form a coalition with the Greens and Lady Sylvia Hermon. The oracles have spoken. “Cor blimey”, we expostulate.

Why do we do it? At a time when the polls are in ferment, we all intuitively know that the polls aren’t an accurate reflection of what is going to happen at the next election, even if that next election might be only three months away (disclosure: I’m still firmly betting against a 2019 election, but that’s by the by). Numerous commentators, including me, have pointed out that seat predictors just aren’t equipped to deal with the types of poll movements that we have seen since the last election.  

So in another attempt to try to scotch this practice, let me illustrate the futility of this from a different direction.  As at the time of writing, the most recent poll is from YouGov. It showed the Conservatives on 31%, Labour on 22%, the Lib Dems on 21% and the Brexit party on 14%. Plugged into Electoral Calculus this gives: Con: 348, Lab: 191, Lib Dems 53 and Brexit: 1 (Clwyd South, if you’re curious). The poll before that, from ComRes, showed Labour on 30%, the Conservatives on 29%, the Lib Dems on 16% and the Brexit party on 15%.  Electoral Calculus turns that into Labour 290, Cons 265, Lib Dems 38 and Brexit 3 (Boston & Skegness, Montgomeryshire and Carmarthen West and Pembrokeshire South this time).

Hmm. Let’s look at that YouGov poll from a different angle. It shows the Conservatives and Labour on a combined vote share of 53%. Electoral Calculus show them winning 541 of the 632 British seats. In 2017 they shared 83%.  In 2017 they shared 579 of those 632 seats. The predictor is suggesting that their votes are going to be very efficiently distributed to withstand mislaying 30% of the electorate between them.

As it happens, we can test the likelihood of this against reality. In 2017, there were just 24 seats in Britain (excluding the Speaker’s seat, which is a special case) in which the Conservatives and Labour shared less than 53% of the votes.  Now bear in mind that if the Conservatives and Labour tally 53% of the votes between them, we can expect them to get less than that mark in roughly half of the seats. Of those 24 seats, between them they won precisely one.

Now if I tell you that all bar seven of those seats were in Scotland, you might triumphantly tell me that invalidates my observation, because the SNP are dominant there and it’s a gigantic special case (I don’t agree, because in any constituency in which they are recording low combined levels of votes, there are going to be other parties in the mix). But Labour and the Conservatives didn’t win any of the other seven either.

There are seats, such as Orkney & Shetland and Ceredigion, where neither Labour nor the Conservatives are in contention. There are two horse races, such as Eastbourne or Glasgow Central, where one of the two main parties is essentially absent. But only in Gordon did the Conservatives take the constituency.

The tipping point seems to be 60%. Above the combined Labour/Conservative 60% mark, only five constituencies were won by a different party (the SNP in all five cases). Below that level, another party winning was the norm rather than the exception. Between them, they took just 14 of the 61 constituencies in that category in 2017. Bear in mind that once they record two thirds of the vote between them (as they did in all bar 88 constituencies in Britain), one of them is bound to take the seat and you realise the idea of the major parties being super-efficient with their votes on low vote shares is hard to reconcile with such data as we have.

What all this suggests is that if the two major parties’ support declines as far as polls currently suggest, their seat count is going to be very dependent on the distribution of their opponents’ vote. There is no particular reason to assume that their new opponents’ vote is going to be especially inefficient: unlike the Alliance in the 1980s, both the Brexit party and the Lib Dems seem to have particular hot spots (in the case of the Brexit party, the more run-down bits of the country, and in the case of the Lib Dems, the more affluent bits of the country). The lumpier that support for each of them, the more likely they are to convert that into seats.

What does this mean in practice? If Labour and the Conservatives achieve a combined 60% of the vote at the next election, as currently seems an entirely reasonable starting point, they will poll above that level in roughly half the constituencies and below that level in the other half.  We can expect them to take substantially all those where their combined polling is above 60%, but they will be doing very well (and way beyond how they did in 2017) if they take 50% of the seats where their combined polling is below 60%.

If you do the maths, you find on such polling that a combined total of 475 seats between Labour and the Conservatives would be very good going for them.  That’s 80 less than Electoral Calculus predict on the ComRes poll (where they have a combined poll share of 59%).

To be clear, this is not to knock Electoral Calculus, who do a great job. I’m trying to show just how hard their job is. And I’m trying to persuade you that when the polls shift in the way that they have, past certainties break down. At times like this, the only wise course of action is to have some honest doubt.

Alastair Meeks




h1

Going back to your constituencies. Alastair Meeks on not taking seat predictors too seriously in times of change

Wednesday, June 5th, 2019

Obviously, you should not treat opinion polls very seriously at all, especially when no general election is on hand. Respondents are being asked an artificial question (there is no general election tomorrow) with no real-world consequences hanging on their answer. 

You might be getting a general expression of enthusiasm for a particular credo, a message to the voter’s normal choice or simply a casual choice made without much thought, and that’s before you get into the methodological adjustments that the pollsters use to produce their confection. Headline voting polls at this stage in an electoral cycle should be treated as an expression of mood rather than a forecast.

For all that, they are pretty well the only useful data points that we have for Westminster elections. So they are pored over endlessly by political geeks, deconstructed by subsamples and slotted into seat predictors.

The latest YouGov poll was particularly sensational, putting Labour and the Conservatives in joint third on 19% behind the Lib Dems on 24% and the Brexit party on 22%. Not content with this, Nigel Farage complained that YouGov’s failure to prompt for the Brexit party nefariously suppressed his creation’s ratings (regrettably, he did not give a detailed psephological explanation of how YouGov should correct for their past overstatement of Brexit party support: no doubt they would have welcomed his new-found expertise).

These findings were then plugged into the seat predictors. What was most striking about these was how Labour in particular, but also the Conservatives, held onto far more seats than one might expect. Electoral Calculus comes up with Labour 202, Brexit 141, Lib Dems 119, Conservatives 110.

Forgive me if I have my doubts about the accuracy of such predictions. Seat predictors are made for a world where polls move up or down a few percentage points, like boats bobbing on the swell of the ocean. If this poll were to be anything like accurate, Labour and the Conservatives are about to be hit by a rogue wave.

The traditional method of calculating seat movements at elections is by uniform national swing. That implies that the percentage swing between parties is exactly the same in every seat. So if the Conservatives start on 48% in a seat and Labour on 30%, and there’s a 10% national swing from the Conservatives to Labour, Labour would be expected, all things being equal, to take the seat 40% to 38%. Historically it has worked better than the more intuitive proportional swing (where each parties’ share of the vote would move proportionately in each seat).

The problem with uniform national swing where the movements are so wild is that it breaks down completely. If, for example, the Conservatives were to drop from 43% to 19% at successive elections, then uniform national swing would imply that the Conservatives got negative votes in every seat in which they tallied 24% or less of the vote in 2017. There were 98 such constituencies; this is not a theoretical problem.

Electoral Calculus is more sophisticated than this. It operates what is known as a strong transition model. As the link explains, the model divides party supporters into strong and weak supporters. The model allocates them to each seat based on vote share in each seat, and then assumes that the weak supporters fall away first when a party’s polling droops.

There are two difficulties with this approach in turbulent times, particularly so far as both Labour and the Conservatives are concerned. First, it assumes that their votes would drop off particularly quickly in seats where they start from the position of a low vote share. This might well be true for minor parties or parties with particular regional strength.

But both main parties have substantial and to date enduring strength across the nation: Labour did not lose a deposit in 2017. It is not at all obvious that beyond a certain point in any given constituency that their voters are weak.

Secondly, and more fundamentally, the model copes poorly with a fundamental realignment. Many lifelong former Labour supporters voted for other parties in the EU elections. An absolute majority of Conservative party members appear to have voted for the Brexit party in the EU elections. The concept of a strong voter breaks down when new binary divides spring up.

None of this is to disparage electoral models and Electoral Calculus is excellent. They have proved terrifically useful in the past in understanding the dynamics of close battles (and you should always remember that the predictions are only ever going to be as good as the inputted data). But where the public shifts allegiance dramatically, they are going to be much less useful. It would be like predicting the outcome of a duck race in a whirlpool.

At the vote shares indicated by that sensational YouGov poll, my expectation is that both Labour and the Conservatives would have far more wasted votes than any electoral model currently suggests and as a result lose considerably more seats than any model put forward.

We’ve since had an Opinium poll showing the Brexit party in the lead and the Lib Dems in fourth. Electoral Calculus showed the Brexit party with 306 seats and Labour with 202, but with the Conservatives on a mere 26 seats.  I have to say that feels a more believable outcome from such a result.

In summary, supporters of the main parties, particularly the Conservatives, should not assume that incumbency and traditional strength is going to save them if the polling continues to show them gurgling round the plughole, even if the seat calculators suggest otherwise. This is no time for complacency.

Alastair Meeks




h1

Marginal improvements. Looking at the reliability of seat predictions from polls

Monday, February 5th, 2018

Last week, Mike Smithson noted the Conservatives seem to have an in-built advantage in the electoral system over Labour – if they got an equal number of votes, the Conservatives could expect about 15 seats more than Labour even if Labour had a 0.5% lead in the polls, if Electoral Calculus is to be believed. That begs the question whether seats are likely to move consistently at the next election in the way that seat predictors assume. Let’s have a look at the possibilities.

Electoral Calculus uses what is called a strong transition model. For major parties in seats in which they are in contention, this is close to uniform national swing. For present purposes, I’m going to stick with looking at uniform national swing, which has the great merit of simplicity.

First things first, what is uniform national swing? It assumes that as votes move from one party to another, they will do so in precisely the same percentages in every constituency. So if the Conservatives hold a seat with 51% of the vote and Labour have 46% of the vote, a 5% swing would see those percentages reversed. If UNS applies, then if the Conservatives had just 20% in a seat and Labour had 70%, we should expect the Conservatives to drop to 15% and Labour to get 75%. The point to note is that you look at the absolute percentages, you don’t try to adjust for the proportion of voters in each constituency.

Logically, UNS can’t predict seat counts in all contingencies – it breaks down completely mathematically in more extreme cases (this is what Electoral Calculus’s adjustments for a strong transition model are designed to overcome). Obviously, seats do vary according to local circumstances and the quality of candidates on offer, and few seats will swing by exactly the national swing (Mike’s home seat, Bedford, is quite unusual in that it has tracked the national swing closely at every election since its creation in its modern form in 1997). In practice, UNS has done pretty well in normal circumstances in coming up with broadbrush predictions, particularly in two party systems.

Why? Two different points are worth considering. First, when you’re dealing with a large number of swing seats, you’re likely to produce a bell curve of distributions of swing, clustered particularly around the median swing. All other things being equal if the swing isn’t outlandishly big, those large numbers of seats are likely to be almost randomly distributed by size of majority across the bell curve. A seat might fall to an unexpectedly big swing but it is likely to be counterbalanced by a seat that didn’t fall owing to an unexpectedly small swing.

Secondly, seats of a similar type are likely to behave similarly (and probably did so in the past too, assuming they haven’t changed). Lincoln, Bolton West and Keighley, for example, have oscillated between the two main parties for decades. So the ordering of seats at a previous election is likely to be quite indicative of the ordering of seats at the next election, though the colour of the rosette of the winner might change.

Do these considerations hold next time? Well, the second one looks very suspect. Here are the marginals that the Conservatives are defending and here are the marginals that Labour are defending (including all seats vulnerable to a 10% swing). Focus on the final column, which shows you what the swing was last year. Nationally, the swing was about 2.2% from the Conservatives to Labour.

But when you look at the most marginal seats, the swings vary wildly and no pattern is easy to discern. Stroud and Bishop Auckland nestle side by side in the Labour defence list at numbers 12 and 13, but Stroud secured an outsized swing to Labour while Bishop Auckland saw a solid swing to the Conservatives. On the Conservative defence list at numbers 12 and 13, Broxtowe and Stoke-on-Trent South show similar and similarly contrasting swings.

The last election saw relatively few seats change hands but both main parties’ coalitions seem to have changed substantially, whether because of Brexit or because of the very different direction that Jeremy Corbyn has led Labour. It seems unlikely that seats that responded very differently to those stimuli will respond similarly to either the continuation or the withdrawal of those stimuli. Broxtowe and Stroud might swing in similar ways next time, and Bishop Auckland and Stoke-on-Trent South might swing in similar ways next time, but all four swinging in harmony does not look particularly likely.

There is a betting consideration to this. Next time the constituency markets will not be proxies for the national result. You won’t be able to deduce from a 3% swing in the polls from the Conservatives to Labour that Chingford & Woodford Green, North East Derbyshire and Carlisle all look in more or less the same amount of peril for the blue team. Constituencies will need to be considered either individually or in relatively small groups.

But what of the bell curve? I’ve had a look at the 2017 results and plotted the seats by size of swing, grouped into bands, here. As you can see, this looks pretty much like a bell curve centred around the 2.2% swing to Labour I mentioned earlier. Not only that, the seats in each segment divide roughly evenly between the blue and the red team in each band (with the exception of those seats which swung up to 5% to the Conservatives, which are twice as likely to be Conservative-held as Labour-held).

So UNS, and similar methods, look as though they still have life in them. What you shouldn’t do next time is look at a given seat and extrapolate even indicatively how it is likely to perform next time from an opinion poll. Local considerations look set to be far more important than national trends when looking at individual seats.

Alastair Meeks




h1

Latest Electoral Calculus projection has CON 15 seats ahead even on a 0.5% lower average vote share

Thursday, February 1st, 2018

This is, of course, on the current old boundaries

I have made this point before but the latest projection from Martin Baxter’s Electoral Calculus is an excellent example of how under current boundaries the system works in favour of the Conservatives.

As can be seen in the January projection above LAB had a 0.5% average poll lead but when fed into the Martin Baxter computation Corbyn’s party ends up with 15 fewer MPs.

The boundary changes, if agreed, make this even more pronounced. If GE17 had taken place under the proposed new electoral map the Tories on June 8th would have secured a majority of 4.

Bedford, where I live, is one of the changes, moving, according to Baxter from a LAB majority of 789 to a CON one of just 9 votes. making it just about the tightest marginal in the country. It is one of just two seats where the gap is down to single figures.

All this means that LAB has to be securing polling leads which are considerably bigger than that which they have been enjoying recently something that doesn’t seem to dampen the red team’s rhetoric.

Mike Smithson




h1

Seat projection from today’s ICM poll has CON ahead on MPs even though behind on votes

Tuesday, January 16th, 2018

When’s the Corbyn Clique going to work out that the system now works against LAB?

The latest ICM Guardian poll out and the figures – C40/LB41/LD7 – are included above in the seat projection from Martin Baxter’s Electoral Calculus.

As can be seen that although the Conservatives are one point behind on votes this, according to the projection, will put them one ahead in terms of seats.

This reflects a big trend that first was noticed at GE2015 when the Lib Dems were hammered after the years in Coalition. The Tories who had suffered most in relation to LAB on the votes/seats in the previous four elections found themselves benefiting disproportionately from the sharp decline of what was then Clegg’s party.

    The effect is that now on the same vote share the Tories will probably get 20-30 more seats than Labour. If the system is biased then it is to the blue team though they’ll continue, no doubt, to whinge quoting 2005.

LAB, for want of a better term, “wastes” more of it votes chalking up big shares in its heartlands than the Tories who are more vulnerable to the LDs.

Interesting in the latest projection above that the LDs move up two seats even though their share is down on GE2017. The reason, of course, the that in this poll the Tories are down even more.

Projections on the proposed new boundaries have the system biased even more to the blue team.

All this means is that Mr. Corbyn’s LAB needs vote share leads far in excess of anything it has managed to achieve since June to be sure of him becoming PM.

Mike Smithson




h1

In late April the Tory data chief, Jim Messina, told senior Tories that his modelling pointed to a CON majority of 290

Tuesday, December 5th, 2017

Put this on your “Christmas” list

I’m just back in the UK after my holiday on the West Coast of the US visiting my son, Robert, and his family who have moved to LA from London in July.

Part of my holiday reading was Betting The House: The Inside Story of the 2017 Election by Tim Ross and Tom McTague which was published last month.

It is an absorbing read giving a detailed account of GE2017 based on conversations with many of the key players and provides interesting revelations that look remarkable given what we know now.

The top point for me that put all this into context is the one highlighted in the heading – what Tory chiefs were being told ten days into the campaign. This is an extract.

“.. Ten days into the campaign, Jim Messina, the American data consultant working with the Tories, told Stephen Gilbert, Lynton Crosby, Mark Textor and other senior figures that his modelling suggested the Conservatives would win 470 seats – enough for a staggering majority of 290, more than double Margaret Thatcher’s 1983 landslide – and an exponential improvement on David Cameron’s winning margin of twelve.

It was an extraordinary moment and one that caused serious concern among those at the top of the campaign who already feared expectations were spiralling out of control. Messina’s forecast was the high-water mark for the tide of Tory optimism, but right up to election day the most senior campaign officials thought they would make strong advances into Labour territory… “

Although this inevitably got ratcheted down as the campaign progressed the view throughout the seven weeks that an increased majority was a certainty had a totally adverse impact on Conservative thinking. Quite simply it skewed the party’s whole management of the election and approach to seat targeting.

This is how the authors describe what happened when in late May a poll had the lead narrowing sharply. They were so convinced of the outcome that it was dismissed.

“.. On May 25th YouGov ran a poll in the Times, cutting the Tory lead over Labour to just five points. It was the first clear sign that a real change could be happening but was widely dismissed by commentators and analysts as unrealistic. Five days later, YouGov produced something even more dramatic: a seat projection model that said the Tories were on course to lose their majority in a hung parliament. Jim Messina and Mark Textor did not believe it. Sitting inside CCHQ, Messina composed a message on Twitter: ‘Spent the day laughing at yet another stupid poll from .@yougov. Hey .@benleet do you want to bet for charity? I’ll take the over.’ Messina showed it to colleagues and asked if he could tweet it, before doing so…”

But it wasn’t just the Tory campaign that was getting a distorted view of what was happening. Labour’s private pollsters were also giving a gloomy picture for their client. This from just before polling day:

“.. The picture from Labour’s own pollsters BMG was pessimistic. For most of the campaign, BMG had been forecasting a Tory majority of 150. On election day, they thought May was on course for a majority of 80…”

In a telling point on LAB targeting the authors report that activists believe they would have come a lot closer to the CON seat totals if they’d known what was really happening.

If you are being pressed to suggest ideas of Christmas presents for yourself then mention this book. It is a must read for all who follow polls and election forecasting.

Mike Smithson




h1

June 8th 2017 is a day that the election predictor/modellers will want to forget

Wednesday, June 14th, 2017


Wikipedia

It is little comfort to the election predictors/modellers that Wikipedia has now decided to record for posterity how successful they were in predicting the party seat outcome of GE17. The chart is above.

As can be seen only the YouGov model based on 50k+ of its own interviews came out of this well.

Throughout the campaign the forecaster/modellers aimed to produce projections of the party seat totals which, are course, based on the outcome of 650 separate first past the post elections in the different constituencies. This is sharp contrast to the BREXIT referendum or, say, the final round of the French election where it is a binary choice of two based on national totals.

The reason the GE17 modellers got it so terribly wrong was that their main data sources, the opinion polls, had, with one very honourable exception, a huge polling fail. Never was the saying garbage in garbage out so appropriate.

Virtually all of the models were following standard swing theory in their approach to seat predictions which meant a big error on top of everything with their LD seat projections. The party increased its MP total by 50% with a reduced national vote share. The exception was YouGov with its own exclusive and large polling data source.

A problem was the overwhelming CON landslide narrative which dominated everything and was reinforced by the adjustments almost all the pollsters had made following the GE15 polling fail. Any pollster that produced numbers that didn’t fit the general perception were attacked and their findings ignored.

This meant that sharp move away from the Tories after the manifesto saga was much less noticeable at the time and things like the consequential drop in CON 65+ support are only now being observed.

Lots of lessons from GE17 then which no doubt they will try to avoid next time.

The problem, of course, is that there is a huge appetite during a campaign for information on how the battle is going in terms of seats. Gamblers in particular are a key audience. The forecaster/modellers satisfy that need.

No doubt the Wikipedia table will be rolled out next time as a reminder to treat projections with caution.

Mike Smithson




h1

The CON GE2015 target seat over-spending issue throws into question the mathematics of GE2017

Monday, May 1st, 2017

Keeping within the limits could make a difference

Becasuse so much of the effort to predict and analyse the next election is based on what happened in each seat at GE2016 we are in something of a quandary because of what we know now about the Conservative approach to constituency expense limits.

It is entirely possible that the Tories will be defending seats on June 8th that would not have been won if spending had been kept with the constituency limits.

Before GE2015 political scientists such as Prof John Curtice were predicting that the Conservatives would need a GB vote margin over Labour of 11% in order to achieve a majority. That they reached that target with a gap of just 6% was a remarkable feature of the election and how many punters got it so wrong.

Now thanks to the work of Channel 4’s Michael Crick and others we have a greater awareness of what went on in the election and this has been closely studied by the Electoral Commission which was very critical of the party.

    Thus we cannot now say with certainty that seat X requires a swing of Y% because the base figure, the GE2015 result, could have been different if spending limits had been kept to

What would have happened in that seat if expense limits had been adhered to and that is very much in the air.

This is all going to make betting on some single constituencies much more difficult and also raises questions the standard poll shares to seats calculations

If we do get news of action by the CPS before polling day that could also have an impact.

Mike Smithson