Full tabs for the YouGov/Sun on Sunday poll are now up here. The slightly larger sample than usual was to make sure they had a good sub-sample of Sun readers, which the Sun used in yesterday’s analysis of the poll to look at what their own readers thought. The Express, however, has decided to report the Sun reader crossbreak as a national poll – obviously it wins the coveted UKPR crap media reporting of polls award. Just to be crystal clear UKIP are not in second place in this poll. The headline figures for this poll were CON 33%, LAB 34%, LDEM 8%, UKIP 15%. The figures quoted in the Express relate only to respondents who read the Sun.

160 Responses to “No, UKIP aren’t in second place”

1 2 3 4
  1. Well that’s three in a row. The traditional media did their worst to try to convince that the Tories losing another by election was clearly worse for Labour, but the response from voters seems quite clear – a significant hit from three companies on Tory VI.

    Meanwhile, the media seems to be picking up on Labour’s policy announcement on cutting business rate relief on private schools. From what we know, the policy seems reasonable. Additional requirements are being made to ensure private schools as charities are actually doing something charitable, in exchange for significant relief on business rates.

    Sections of the press have already resorted to labeling this ‘class war’, In the old days of New Labour triangulation, this would have been something the spin doctors would have been anxious to avoid, but I do wonder whether voters may be more receptive to such messages today.

    It’s difficult, as in the past, when Labour identified those who would lose under tax changes, the general impression of Labour as a high tax party allowed their opponents to paint them as anti aspirational, so those well down the scale and unlikely ever to be affected by the policy itself still felt threatened.

    ‘All in this together’ resonates now for many, but for the wrong reasons. Maybe a small and relatively insignificant policy such as this could be well received by sufficient voters to help shore up Labour support.

  2. @AW


  3. @RAF / AW

    Fascinating stuff, thanks. I had always assumed that the speaker would not vote at all, and that in the event of a tie, there would be much wrangling, whipping and bribery.

    Too cynical? Moi?

  4. @CMJ

    “I’ve looked at such data some months ago, and found it delivered poor r squared values.”

    Yep .. The r-square values are nothing to shout about: 0.34 for Labour, 0.29 for the LibDems and even lower for the other two. So, these fits don’t provide a wonderful account of polling sequences.

    The question is whether a noisy forecast is better than no forecast at all. Do you have better ways of predicting what the pattern will be next May?

    As an alternative, the Fisher model looks seriously flawed. Both parties of government are supposed to have benefitted from swingback over the period covered. But in both cases the best linear fits have a (reliable) negative slope. So the data are showing a tiny amount of ‘swingaway’ from the coalition parties. In short, there is good evidence that the patterns that emerged in previous elections are not repeating themselves.

  5. I’m a bit surprised that the Sun has so many Labour-voting readers. I guess this confirms that love of tits has nothing to do with one’s political inclinations.

  6. @Unicorn

    If I knew a good way to accurately predict the next GE, The bets would already be on, and the house purchase to be made with the winnings be planned.

    Sadly, I all I have is a bit of data and good guess. :-)

  7. I agree the next GE will rewrite the rules.

    All previous patterns will be rendered meaningless.

  8. Factcheck: Ukip is NOT ahead of Labour, despite what the Express says!


  9. Possible implications of moving away from the Uniform Swing Projection model

    This is a long post and not immediately linked to recent exchanges. Apologies for both. If you are not interested in how VIs turn into seat tallies then please click on to the next post.

    Summary (for those of you who don’t have the time or inclination to read the detailed arguments set out below).

    Given the evidence of recent marginal polling data, the Uniform Swing assumption is demonstrably wrong. A question that arises is whether this imperfection results in systematic distortions when the model is fed with nationwide polling data and used to project the distribution of seats in a putative parliament based on current data. In the notes below, I argue that use of the UNS Model may lead to a modest (5-10 seat) overestimate of the number of Labour seats likely to be returned in a future parliament.

    Based on current averaged polling data, the USP model projects a notional parliament comprising 274 Conservative seats, 325 for Labour, 23 for the LibDems with the remaining 28 seats going to various other parties. The number of Tory seats drops because the current average VI figure (32%) is four points lower than the 36.05% share they secured in the 2010 GE. Similarly, seat projections for Labour increase because the current VI (c. 33%) is about four point *higher* than their 2010 share (which stood at 28.99%). The model transfers from Labour all the Tory seats in which the majority (over Labour) was less than 8%. This includes all marginals up to but not including Stevenage (2010 Majority: 8.1%). Similarly, both Labour and the Tories gain at the expense of the LibDems.

    The question I am going to tackle is whether these transfer patterns are realistic, and I want to argue that they are not. From churn analyses (including AW’s post on November 7th) we know that the VIs for both Labour and Conservatives are being boosted by the changed expressed voting intentions of 2010 LibDem voters. To date, all analyses have shown that these changes have boosted Labour more than the Tories. (AW’s estimates were for a 4.6% gain by Labour compared to 1.8% by the Conservatives). What this implies is that the Labour VI is currently enjoying slightly higher boost from the LibDem influx. Part of the modelled shift in seats from Tory to Labour will have been produced in this way.

    The problems arise because Uniform Swing assumes that the national average swings are reproduced in each and every seat. However, we know for certain that the LibDem swing has not been uniform. The drop in their VI has varied systematically across seats to produce a national *average* of about 15%. In about 120 seats the LD vote was less than 15% and in these cases the drop must obviously have been *less* than 15%. To yield an average fall-off of 15%, the decrease in other seats must have been higher than 15%. Moreover, actual seat by seat polling data (from Ashcroft) show far from uniform drops. In fact, instead of being fixed across all seats the drop is close to being proportional to the 2010 recorded vote (see details below).

    If Labour’s VI is benefitting by picking up a good share of the 2010 LD votes it follows that Labour’s boost relative to the Tories must be smaller where there were fewer LD votes to share out in the first place.

    What happens if we take this into account and model a seat-by-seat transfer of votes rather than retaining the assumption that there is always a uniform swing across all constituencies? The exact calculations depend on the detailed transfer assumptions you make. (I favour using Ashcroft constituency polling data to regress LibDem polling drop against LibDem 2010 vote%, and using the slope and intercept of the resulting regression equation to predict LibDem drop in all unpolled seats. At present this gives the following equation: current LD VI = 0.47 x (2010 LD % vote) – 3%. Applied to the average national LD vote in 2010 (23%) this gives a ”prediction” for the current LD VI of 0.47 x 23% – 3% = 10.8%-3% = 7.8%, which is pretty accurate. It also gives a rather good account of the marginal VI data (R-squared = .88). Armed with this information, churn data (e.g., from AW Nov 7) can be used to reallocate these “lost” LibDem votes on a seat-by-seat basis.)

    However without getting caught up in too much detail, it is possible to give a rough idea of the changes that would be produced by any such adjustment. The basic idea is that the Tory->Labour swing will be smaller in any and all constituencies where there were a lower-than-average percentage of LibDem 2010 voters available to boost the Labour VI. If the most marginal Tory/Labour constituencies are seats that happen to have a low LibDem presence, then the Tory->Labour swing will be lower than that suggested by the national averages (and also by the UNS model). On the other hand if the swing region is populated by seats with a high LibDem presence we can then expect a larger-than-uniform swing.

    As it happens, in 18 of the 20 (currently) most marginal seats, the LibDem 2010 vote was lower than the national average (and less than 15% in some cases). It follows that a few of these seats are not likely to transfer to Labour in the way they are projected to do under uniform swing assumptions. To take an example, Harrow East had a 7.1% Tory-over-Labour majority in 2010, and so falls to Labour under current Uniform Swing assumptions. However, with a 2010 LD vote of only 14.27%. the prediction of the “modified vote-reallocation” version of the model is that the local drop will be about 60% of the national attrition, reducing Labour’s flight-from-LD windfall from 4.6 – 1.8 = 2.8 nationally (AW churn figures) to 1.68 locally. The 1.12% shortfall prevents Harrow East from falling to Labour under these adjusted assumptions. The same thing happens for Pendle (Labour’s margin to overcome being 7.96%) and Swindon South (with its 2010 margin of 7.52%).

    Obviously one has to take the individual seat projections with a pinch of salt. But the take-home message is that with what I would argue are reasonable assumptions about VI-reallocation (churn) Labour turns out to do a little worse in its seat-conversion than it does under the demonstrably false uniform swing assumptions. The next obvious question is whether this is just an accidental property of seats with majorities near the current (8% gap reduction; 4% swing) setting of the Swingometer. The answer seems to be that it isn’t. For the majority of seats where the Tories and Labour were close in 2010 the LD presence was below average. (Where the LD vote was above 23% it was likely to have been the seat winner or the runner-up, forcing one of the other parties out of contention.) So the same bias is likely to be in place even if swingback takes place over the next few months (or, indeed, if Ukip defections eat away at the Tory VI).

    Further note: A corresponding analysis might suggest that there are systematic biases in the relative damage Ukip defections cause to Labour and Conservatives respectively. However, it is not at all obvious that there are significant seat-by-seat differences of this kind. Regression data suggest that ‘uniform swing’ is much closer to providing an accurate description of the way the Ukip vote has built up since 2010. (The equation predicting Ukip VI from Ukip 2010 vote has a high intercept (12% – 13%) and relatively low slope (beta = 1.7), indicating that 2010 vote levels are not accounting for all that much of the subsequent change). This needs further work, but my impression is that Ukip churn is unlikely to alter the picture painted by uniform swing models.

  10. Alec,
    Small and relatively insignificant it may be,but it draws a clear line in the
    sand between labour and Conservative in educational policy.I have long felt
    that we will never have a strong educational system in this country whilst a
    private education system runs alongside the state system.I believe that in Finland ,the reform of education began with abolishing the private schools and
    they are now widely admired for their educational system.Whatever,it is good to see a policy that is not Tory lite.

1 2 3 4