{"id":556,"date":"2017-05-30T22:59:27","date_gmt":"2017-05-30T21:59:27","guid":{"rendered":"https:\/\/marriott-stats.com\/nigels-blog\/?p=556"},"modified":"2017-05-30T23:22:43","modified_gmt":"2017-05-30T22:22:43","slug":"uk-2017-general-election-forecast-3-a-description-of-my-final-model","status":"publish","type":"post","link":"https:\/\/marriott-stats.com\/nigels-blog\/uk-2017-general-election-forecast-3-a-description-of-my-final-model\/","title":{"rendered":"UK 2017 General Election Forecast #3 – A Description of my Final Model"},"content":{"rendered":"

For the last 6 weeks, I have been making forecasts of the number of seats that each party will get in the 2017 General Elect<\/a>ion.\u00a0 If you have been following my forecasts, you will know that I have developed a variety\u00a0of prediction models which all predict something different.\u00a0 With 10 days to go, I decided it was high time to settle on a single Final\u00a0Model which is described in\u00a0this post.<\/p>\n

<\/p>\n

First a reminder of the 5 models\u00a0I developed.<\/p>\n

    \n
  1. URS – Uniform Regional Swing, now superseded by URS+S.<\/em><\/li>\n
  2. URS+S – Uniform Regional Swing\u00a0+ Standown Adjustment<\/span><\/li>\n
  3. URS+T – Uniform Regional Swing + Tactical Voting<\/span><\/li>\n
  4. EU16R – Brexit Realignment<\/li>\n
  5. nURS – Non-Uniform Regional Swing<\/li>\n<\/ol>\n

    For more details on the methodology of all\u00a05 models, please read my prediction for the seat of Bat<\/a>h .\u00a0 The same process described for Bath is then repeated for all seats to arrive at my forecasts.<\/p>\n

    Until now, my official forecast was given by my URS+S model whilst my URS+T model was an alternative forecast.\u00a0 Both my Brexit Realignment & non-URS\u00a0models were intended as a sense check for the other two models but increasingly I have hinted that these models may become more important.\u00a0 I have now made the decision to retire my URS+S and URS+T models, bring my EU16R and nURS models up to date and use an average of these two models as my Final Model prediction.<\/p>\n

    \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 FINAL MODEL = Average of (revamped) Brexit Realignment Model & (revamped) non-Uniform Regional Swing Model.<\/strong><\/span><\/p>\n

    Let me explain how I came to this conclusion.<\/p>\n

    Is 2017 a normal election or a realignment election?<\/span><\/strong><\/h3>\n

    In 2015, there was no question that Scotland saw a fundamental realignment of voter behaviour that allowed the SNP to sweep the board and become the dominant party in Scotland for the first time.\u00a0 The rest of Britain saw a more normal election with what amounted to\u00a0a realignment among the smaller parties (Lib Dems,\u00a0UKIP & Greens) but the fundamental Conservative\/Labour battle\u00a0largely followed traditional patterns.<\/p>\n

    Then came the EU referendum in 2016 which revealed a new fault line in British politics which had always been there but was now in the open.\u00a0 I estimated that 400 out of 650 seats voted Leave\"\" <\/a>but what threw everything up in the air was that the likelihood of a seat voting Leave bore almost no relation with the 2015 General Election results (with the exception of the Nationalist seats which were overwhelmingly Remain seats).\u00a0 The tables to the left show that outside of London (EngXL), 75% of Labour & Conservative seats voted Leave.<\/p>\n

    My favourite chart\u00a0which shows how the referendum appeared to change the rules is the one below.\u00a0 This is for the 573 seats of England & Wales and the seats have sorted into deciles based on a demographic variable which is the % of households in each constituency that do not own a car.\u00a0\u00a0The reason for choosing this variable is\u00a0that my analysis showed it was the greatest differentiator between the Conservatives & Labour in 2015.\u00a0 You can see in the lowest decile on the left (where\u00a0on average only 12% of households do not own a car), the Conservatives\u00a0had a 43% lead over Labour.\u00a0 Conversely\u00a0in the highest decile on the right (where on average\u00a051% of households do not\u00a0own a car), Labour had a 33% lead over the Conservatives.\u00a0\u00a0The effect of these extreme differences is that in the lowest deciles, the Conservatives took all seats\u00a0whilst Labour took practically\u00a0all the seats in the\u00a0highest deciles.<\/p>\n

    \"\"<\/p>\n

    When you stop and think about it, the NoCar Household rate makes a lot of sense.\u00a0 After all, what kind of households are least likely to own a car?\u00a0 The answer is the young, the poor and those living in city centres.\u00a0 All are demographics known to vote Labour and the opposite demographics elderly, rich, rural dwellers are known to vote Tory.\u00a0 For the\u00a0smaller parties, the relationship is less strong though the Lib Dems are closest to the Conservatives whilst the Greens do best and\u00a0UKIP do worst in areas with few car owning households.<\/p>\n

    The black dashed line in the left hand chart is the Leave vote share\u00a0for each decile.\u00a0 I still find it astonishing that there is no relationship with\u00a0the NoCarHousehold rate until you get into the highest decile which is the only decile that voted Remain on average.\u00a0\u00a0So the\u00a0strongest differentiator of\u00a0voting intentions in 2015 turned out to be almost useless in 2016.<\/p>\n

    Instead,\u00a0the referendum vote was driven by Class, Education and Occupation.\u00a0 The chart below\u00a0is for something I call the Occupational Differential (for England\u00a0&\u00a0Wales again) which is the %people working\u00a0in Managerial\/Professional roles minus the % people\u00a0working\u00a0in Lower Routine roles.\u00a0 A positive differential indicates that managers\/professionals outnumber routine workers, a negative differential indicates that routine workers outnumber managers\/professionals.\u00a0 Occupation is a good proxy for\u00a0Class which is why I like this variable.<\/p>\n

    \"\"<\/p>\n

    This time the relationship between Occupation and Leavers is strong.\u00a0 It is also strong with areas with high numbers of UKIP and Labour voters.<\/p>\n

    So to come back to the question of whether the 2017 election is going to be a normal one or a voter realignment, looking at voter intention from the polls in terms of Class and Referendum voting will give us a clue.\u00a0 Charts B4 & B5 below will be familiar to you if you have been following my Opinion Poll Tracker<\/a>.<\/p>\n

    \"\"<\/p>\n

    B4 shows that among Remain voters, Labour has gained over 10% since the election was called and are approaching 50% of all Remain voters.\u00a0 B5 shows that the Conservatives are up almost 10% among Leave voters and were approaching 2\/3 of all Leave voters though they have slipped back a bit recently.\u00a0\u00a0 So Remainers are becoming more Labour whilst Leavers are becoming more Tory.\u00a0 But my earlier chart for the Occupational differential shows that in middle class areas, the Conservatives were ahead of Labour and those areas tended to vote Remain whilst in working class areas which tended to vote Leave, Labour was ahead of the Conservatives.<\/p>\n

    The next chart is a new one for my Opinion Poll tra<\/a>cker and for me is the most striking of all that I have seen in 2017.\u00a0 D1 & D2 show voting intention by class in 2017, based on pollsters that segment their respondents using the standard ABC1 (middle class) and C2DE (working class) definitions, and how this compares with all elections since 1974.\u00a0 The earlier data for 1974-2015\u00a0comes from Ipsos-Mori who have carried out post election polls over that timeframe<\/a> to identify voting patterns by demographics.<\/p>\n

    \"\"<\/p>\n

    For the first time ever in British political history, the Conservatives are expected to be the largest party among the working class whilst Labour could achieve their highest ever vote share among the middle class.\u00a0 It is well known that prior to 1974, class based voting was even stronger than shown in the charts here so 2017 could be the election where Labour which founded to represent the interests of the working class find themselves supplanted in that role by the Conservatives.\u00a0 In effect, if you consider charts D1 & D2 together, class ceases to be differentiator between the two main parties.\u00a0 If that is not a sign that 2017 is a realignment election then I don’t what is.<\/p>\n

    Which of my models is the best in a Realignment Election?<\/strong><\/span><\/h3>\n

    Until now, I have used my URS+S (Uniform Regional Swing + Party Stand Down Adjustment) model as my official forecast with a Tactical Voting variant of this as an alternative forecast.\u00a0 At the heart of this model is the assumption that whilst swings between the parties vary between the regions, within each region a similar swing can be expected across all seats in the region.\u00a0 This was certainly the case in Scotland in 2015 when it behaved differently to the rest of the UK but within Scotland, a largely similar swing from Labour & Lib Dems to the SNP was observed.\u00a0 But in a realignment election, I do not think it is tenable any longer to assert that a uniform swing will be observed within each region.\u00a0 Instead the swing will be dependent on the Referendum vote and the Class make up within each seat.\u00a0 Accordingly, I am retiring my URS+S and URS+T models.\u00a0 It is still possible that these models will be accurate at the national level but given that there are significantly more Leave than Remain seats, I would surprised if this was the case.<\/p>\n

    This leaves me with my EU16R (Brexit Realignment) & my nURS (non-Uniform Regional Swing) models.\u00a0 Unlike 2015, very few constituency level polls have been published in this election.\u00a0 Only 5 have been published but I have used these to see which of my 4 models were closest to the seat polls.\u00a0 In all 5, my URS+S & URS+T were not close and it was either my Brexit Realignment model or my non-Uniform Regional Models that were closest.\u00a0 The split was as follows:<\/p>\n