On the 25th Mike Smithson of Political Betting went to London for the annual local elections briefing organised by the Political Studies Association. Profs Colin Rawlings and Michael Thrasher concentrated on their by-election prediction model. Smithson showed these two slides.

This is an increase on their forecasts in the Times earlier in the month, which gave Conservative +50, Labour -50, Lib Dem +100, UKIP -100. However, the seat change are is still well below my own, much more basic, forecasts. I have summarized how the seat changes stack up against both the 2013 and the previous 2009 round.

I should note that my seat changes make no allowance for boundary changes, or even changes in the councils. On this basis, there is no real difference between the forecasts for the Liberal Democrats or UKIP. The difference lies with the two main parties, where I predict nearly three times the gain for the Conservatives and four times the loses for Labour. In so doing I am basically saying that the seats will roughly revert back to the seat numbers of 2009. Figure 2 (from my post on the earlier Rawlings and Thrasher forecasts) shows this very volatility.

Note the 2013 figures included Ynys Môn council in Wales, along with 33 English Councils.

At the moment we are living in turbulent times in terms of shift in opinion. I think it is not unreasonable to expect similar swings in seat numbers to those that occurred in these two previous elections, especially

Using local by-elections as a means of predicting local elections seems intuitively appealing. Every Friday is published the data on the previous days by-elections at both Politicalbetting and ConservativeHome blogs. The availability of this data is in no small part due to work done over many years by Rawlings and Thrasher. Yet these results are for the whole of England, whilst these elections are for only a seventh of the total. So local issues may be different, especially when these elections are in predominantly Conservative areas. The divergence is shown in the unchanged modeled National Equivalent Vote share from 2013 to 2017 for the Labour Party. It seems to me unlikely that this 29% share, (compared to actual 2013 English vote share of 21.1%) should be unchanged when the UK General Election opinion polls have dropped around 12% from around 38% to 26%. The current relative position of the Conservatives to Labour is about the same as in 2009, with maybe the Conservatives in a slightly stronger position currently. For this reason, I think the benchmark for forecasts should be 2009 in England.

However, if Rawlings and Thrasher, by their modelling can produce something nearer to the actual results this coming Thursday then they will have greatly added to election forecasting by managing to go beyond the rough ratios of swings in proportion to movements in national opinion polls.  The key will be in the Labour party losses, where the R&T forecasts appear most divergent. That is, without evidence of a large move in the opinion polls (and I am writing this before the opinion polls for the Sunday papers are published), if Labour seat losses are less than 200, then the explanation for this divergence will be at least in part be from predictions based on by-election results. The model will have been a predictive success. Even if Labour loses 300 seats, the fact of developing a rigorous model that is falsified by the data, could develop our understanding of council by-elections as a predictor relation to the full results. This previously unavailable information could still be useful. In less than a week we shall know.

Kevin Marshall