Libdemgeddon? You don’t want to miss a thing.

It is fiendishly difficult to predict the performance of the Liberal Democrats. Their historical seat performance has been surprisingly uncorrelated to their national vote share in comparison to the other mainstream parties.

At the 2010 General Election, following a surge in support during the debates, the consensus amongst almost all punters (including myself) was for the Lib Dems to pick up at least 15-20 seats. In the event, they lost five, including some of their safest; Winchester, Harrogate and Knaresborough, Montgomeryshire and Oxford West and Abingdon. They also came close to losing St. Ives too. All had majorities of over 7,000 votes.

If that happened when the Lib Dems enjoyed a surge in support, then how will they fair when the current UK polling average gives them just 7%, just under a third of the 23% they polled in the 2010 General Election?

No-one knows. The Liberal Democrats draw much of their strength from local grassroots, and the personal votes for their candidates; they are seen as more apolitical and community grounded than MPs from other parties. However, because the result this time could hinge on just a handful of seats, we must at least *try* to model them.

Prepare for some heroic assumptions.

In my last post, I focussed on predicting Scotland as a country. In this post, I will focus exclusively on the Liberal Democrats in England and Wales alone. But, put the two together, and.. hey presto: you have my forecast for the performance of the Liberal Democrats in the 2015 General Election.

We must first understand where the votes have gone. Peter Kellner did a very interesting analysis in the Guardian last month. Adjusting for England and Wales only, the 6.37 million votes cast for the Liberal Democrats in the 2010GE has collapsed to just over 2.1 million now. But I’ve assumed that some of the floaters will return to the Lib Dems and arrived at a forecast figure of 2.41 million. That’s 38% of the 2010 General Election tally, and equates to a national vote share of 9% assuming turnout is similar to last time. So, a little higher than the polls are currently showing.

That’s a bit too neat for a rule-of-thumb forecast. So, I’ve looked at aggregated data within the sub-samples of the opinion polls. I’ve then forecasted the national poll shares forward (how these might move) to come up for a forecast for Election Day. One (important) health warning: I didn’t do anything to adjust for turnout, new voters, voters who’ve died, moved abroad etc. because (a) I didn’t think it would have a huge bearing on the results, and (b) I have better things to do with my time.

This has allowed me to come up with a “standard split” for where the 2010 Lib Dem votes might go. I’ve used a 35% retention as a guide:

2015 standard split (as % of 2010 Liberal Democrat vote)

• Liberal Democrat (loyal) 35%
• Conservative 15%
• Labour 25%
• UKIP 14%
• Green 8%
• Others 3%

If we apply these shares to all 46 Liberal Democrat English and Welsh seats (making no adjustments for turnout) we get (drumroll) .. no Liberal Democrat holds.

Model 1 – Uniform Standard Split: Conservatives pick up 33 seats, Labour 12 seats and Plaid Cymru 1 seat. No Lib Dem holds.

This obviously isn’t going to happen. Why? Because the Liberal Democrats vote is disproportionately concentrated in their existing seats. In the 2010GE they polled 16.4% of their national vote in just these 46 seats, when proportionately to the number of English and Welsh seats in total it should have been 8%. They score better where they have a local council and parliamentary presence.

Secondly, the Lib Dems are concentrating almost all of their resources in defending their existing seats, so any reduction in national vote share should be “bunching up” where they are actively fighting and have that local presence. And, finally, and most importantly, we have polling evidence from Lord Ashcroft to tell us this uniform standard split isn’t so.

The Ashcroft polling is interesting. He has polled all but eight of these 46 seats over the last year. A simple average of his findings gives a vote retention of 68.2% of the 2010 vote share, and a mode of 63.8%. Almost an exact flip of the 35/65 national retention.

So, for my second model, I’ve simplified to a “LD constituency split” for where the 2010 Lib Dem votes might go within each constituency.

2015 LD constituency split (as % of 2010 Liberal Democrat vote)

• Liberal Democrat (loyal) 65%
• Conservative 9%
• Labour 15%
• UKIP 7%
• Green 3%
• Others 1%

I then tweaked the UKIP vote in each constituency (doubling it) as in most cases it was starting from a relatively low base. I also adjusted both the Tory vote and Labour vote to 95% of their 2010 score (assuming the votes would be down, but the differences would broadly level out) and applied that model uniformly across the same Liberal Democrat constituencies. Once again, nothing fancy with turnout was done.

Model 2 – LD Constituency Model: Conservatives pick up 27 seats, and Labour 11 seats. Lib Dems hold 8 seats.

However, that still didn’t look right. We know from Lord Ashcroft that some Lib Dem incumbents are doing much better than that. They have large personal votes, encouraging local results, and a lot of local tactical voting in their favour. Furthermore, the Conservative and Labour votes are not holding up that anything like that uniformly across the country. Not to mention the regional rises in the UKIP vote.

So I tried one more time. This time, I applied a constituency weighting to each of the Liberal Democrats, Conservatives, Labour, UKIP and Green 2010GE vote tallies in each constituency. But I maintained the standard 2015 LD constituency split.

Statisticians will hate this. Yes, to some extent, the two different weightings will compensate and/or overlap with each other (take your pick) and it’s ugly and imprecise. But I reviewed each constituency in turn, adjusted the weightings, and think they look broadly correct.

My assumptions:

• In no constituency will the Liberal Democrats retain more than 85% of their 2010 vote. Yes, there are large personal votes. But when you are down 2/3rds on your last election this will tell (to some extent) everywhere. Some of the Ashcroft polling showed a 97%+ retention. I could be wrong, but not all votes are personal. It seemed fantastical to do anything else.
• The flipside is that nowhere have I assumed the Liberal Democrats will drop below 50% of their 2010 vote. Incumbency and the local profile will have some effect, and this should help mitigate their voter losses.
• Liberal Democrats *will* drop further where there is no incumbent re-standing. I have them dropping to an average of 60% of their 2010 vote where they are not re-standing. It’s closer to 75% where they are re-standing. Liberal Democrats really will be relying heavily on personal votes this time.
• Whether it’s a Tory 40:40 target, or not, makes only a nominal difference in this analysis. Except it may make the 2010 Tory vote share a little more resilient.
• The Tories drop their 2010 vote almost everywhere. I have them ranging from 55% retention in Bradford East to 95% in Bath. In most cases between 80-90%.
• Labour increases their 2010 vote share almost everywhere too. I have them ranging from a ‘freeze’ on 100% in Brecon to 150% in Twickenham. In most cases between 110-125%.
• UKIP increases vary. A lot. From no change in Bristol West to 800% in Portsmouth South, the variations are highly regional and demographic-based. In most cases they are between 350-450% up on their 2010GE vote share.
• Greens increases vary too. From no change in Mid Dorset and Poole North, to 500% in Bristol West. Mostly in the 150-250% bracket.

And here are the results:

Model 3 – Casino Royale adjusted LD Constituency Model (England & Wales): Conservatives pick up 13 seats, and Labour 10 seats. Lib Dems hold 23 seats.

You can see a detailed breakdown here.

Adding this to the Scottish seats forecast gives this grand UK total forecast:

UK Wide Forecast	2010 GE	2015  GE	Changes
LD Seats:	57	25	-32
Con Seats:	0	14	+14
Lab Seats:	0	10	+10
SNP Seats:	0	8	+8

However, another health warning: the most interesting thing I found is that 16 of the seat results were very marginal – with a majority of less than 5% – and 7 ultra-marginal, with a majority of less than 1%.

For the Liberal Democrats, it’s very much a case of their seat spread being a mile wide but an inch deep. They could go as high as 33 seats or as low as 17 seats. But them retaining more than 35 seats is for the birds.

For the Tories only 7 seats look absolutely nailed on gains: Portsmouth South, Wells, Mid Dorset and Poole North, Chippenham, Solihull, Taunton Deane and Somerton and Frome. However, on current polling, they are very likely to do better: some of the remaining Lib Dem marginals will fall to them. It’s a question of which.

For Labour there appear to be 8 seat gains in the bank: Hornsey and Wood Green, Norwich South, Redcar, Bradford East, Cardiff Central, Burnley, Manchester Withington and Brent Central. Their extra two (Birmingham Yardley and Bermondsey and Old Southwark) are forecast to go to them but you couldn’t be certain of either. Particularly not Bermondsey, where I have them winning by 226 votes.

Finally, I overlaid the betting odds on each of these constituencies to see where value might lie.

My conclusions are below. If I’m right, there’s both money to be made on the Conservative ‘dead-certs’ as well as value in the more marginal possibilities. I am not betting on the Liberal Democrats anywhere (you may wish to do so but I don’t) because the combination of odds offered, and uncertainty of black swan losses this time round, doesn’t make me confident enough to back them anywhere.

• Cheltenham – 11/5 on the Conservatives with Bet365. I had the Liberal Democrats holding this by just 119 votes. However, the Ashcroft poll showed the Conservatives 15% ahead (40% Con to 25% LD) before the consistency specific question on voting intention was asked. That’s a huge gap to close on personal/tactical votes alone, and it looks like it could be very close to me.
• Bath – 7/2 on the Conservatives with Ladbrokes. Like Winchester and Harrogate in 2010, what looks on the face of it like a safe Lib-Dem hold may not be. Particularly with the incumbent of 23 years (Don Foster) standing down. I have a Lib Dem hold by just 442 votes. But I think it could go either way, and the current odds represent value.
• Brecon and Radnorshire – 6/4 on the Conservatives with Ladbrokes. I think this could be like Montgomeryshire in 2010. The polling is very close and I have the Conservatives picking it up by 598 votes. Worth a punt.
• Chippenham – 8/15 on the Conservatives with Ladbrokes. I have the Conservative taking this by a majority of over 7,000. Why the odds are still this long, I don’t know, but I think it’s excellent 50% return on your money inside 8 weeks. I’m on.
• Cardiff Central – 1/2 on Labour with Bet365. Similar reasons to Chippenham – I have Labour taking this by almost 5,000 votes. Seems a good price to me.

However, usual warnings apply: please do not rely on my model for your own betting, in fact I’d strongly recommend you don’t, and please do your own research.

Final note: I did a “back check” to see how the patterns of the votes for the Liberal Democrats might fall, nationwide, this year. Counting up the votes forecast from within these 46 seats (using my model) and assuming the Liberal Democrats poll 45% of their 2010 general election score in their next 47 target seats (those they don’t hold, but are the next best prospects for them) means that they’ll get 45% of their votes this year in just 93 seats.

That leaves 55% of their vote in the remaining 480 England and Welsh seats. That would give an average vote share in those remaining seats of just under 3,000 votes.

Given that most English and Welsh constituencies have over 60,000 electors that puts the Lib Dems at risk of losing an awful lot of deposits.

How this translates to the primary Con-Lab marginal battleground will, of course, be crucial to the election. And it’s that battleground that will form the subject of my next post.