Hung Parliament looks likely in UK elections
The biggest difference between opinion polls in India and in the UK, which goes to polls on Thursday, is in what they report. Indian opinion polls primarily focus on predicting the range of the number of seats each party is going to get in the house, and sometimes don’t even bother reporting estimated vote shares. Polls in the UK, on the other hand, almost exclusively report the estimated vote shares, with the seat distribution almost being an afterthought.
The other big difference is in volume. Since 18 April, when Prime Minister Theresa May dissolved Parliament and called for fresh elections, there have been nearly 60 opinion polls published by 15 different agencies—that’s an average of more than one poll a day. In India, opinion polls are much sparser. There are far fewer polling agencies, with each conducting a maximum of three or four polls over the period of a few months leading up to the elections.
The third difference is in the ownership of the polls—in India, agencies mostly tie up with news television channels, with channels making a big spectacle of the broadcast of the results of each poll.
In the UK, on the other hand, barring one instance (ICM, which has tied up with ITV), pollsters have preferred to partner with the print media.
That said, what are the UK opinion polls saying? As one would expect from polls that forecast vote share rather than seat share, the polls are largely consistent (in India, a large source of deviation between opinion polls is the formula used to convert vote shares to seat share).
Figure 1 summarizes the predictions of all the polls for the four biggest parties since the time the elections were called. As we can see, opinion polls have been consistent in predicting that the ruling Conservative Party will win the highest vote share, though they are predicted to do far worse now compared to when the elections were called. Labour, the principal opposition party, started off far behind the Conservatives, but has bridged the gap significantly over the last six weeks, drawing votes from all remaining major parties.
The Liberal Democrats (LibDems), who formed the government in coalition with the Conservative Party between 2010 and 2015, and the anti-immigration UK Independence Party (Ukip) started off weak, and have since lost significant support, going by the opinion polls—with Labour being the biggest beneficiary of this decline.
So, assuming that the polls are accurate, what does it mean for the next UK Parliament (this is a big assumption, since opinion polls were widely off the mark in both last year’s Brexit referendum and the 2015 general elections)? With the UK and India having similar election systems, can we apply our learning from Indian elections to forecast what might happen in terms of seats in the UK?
For starters, the opinion polls make it clear that it is going to be a straight fight, a “two-cornered contest” between Conservatives and Labour.
Having analysed such contests in the Indian context in the past , we had seen that even a small lead in terms of vote share can translate to a landslide victory. That should suggest that the Conservatives should do well, even if their vote share lead over Labour has come down in the last few weeks.
How the votes will translate into seats, however, is governed by the relative strength of the parties in different seats, and in order to measure that, we can use two variations of the “swing model” that is commonly used to convert votes to seats in the Indian context.
In both variations, we take the forecast vote shares (from opinion polls), and look at what “swing” it represents from each party’s actual vote shares in the preceding election.
Then, in each constituency, we adjust the vote share of the party in the preceding election by this swing. The two methods vary by the way this adjustment is made, which we will describe using an example.
Let us assume a party had got 30% vote share in the previous elections, and opinion polls suggest the party will get 40% votes in the forthcoming elections. According to the first method, this represents a “swing” of 10 percentage points. Then, in each constituency, we increase the party’s vote share by 10 percentage points. We repeat this exercise for all parties, and then look at which party has the maximum votes in each constituency, summing up which we can project the seat shares in the forthcoming elections.
While this method works well in most cases, it can come unstuck in the presence of regional parties (for example, this is why Election Metrics’s predictions for the performance of the Janata Dal (Secular) in the Karnataka elections in 2013 turned out incorrect.
Regional parties have strong influence in only one restricted region, and any swing in favour or against them should be applied in that region alone.
A generalized swing model like the above underestimates any swing in their favour in their region of influence. In order to adjust for this, we can use ratios rather than differences.
Again going to our example, if a party got 30% of the votes last time round, and is expected to get 40% of the votes in the forthcoming elections, we will adjust the number of votes won by the party in each constituency by a factor of 4/3 (40% divided by 30%). Doing this for all parties, picking winners and summing up will help us make an effective prediction.
Using ratios means a party gets impacted only in its areas of influence, and can thus better account for regional parties. Applying these two models using data from the 2015 general elections and the recent opinion polls, we can generate forecasts of what the UK Parliament might look like. It must be mentioned that we are stepping into dangerous territory here since even renowned election forecasters such as Nate Silver have shied away from making seat predictions for the UK elections.
Figure 2 shows the seat distribution across parties in case we were to use the first model (where we look at the swing as a difference in vote share), while Figure 3 shows the seat distribution across major parties when we look at the swing as a ratio of vote share.
The trend is again similar to what the vote share projections suggested—the Conservatives started with a big lead, which has come down significantly.
In fact, going by the ratio swing model, which might be more accurate considering the presence of a significant regional party—the Scottish National Party, a hung Parliament looks increasingly likely.
The UK is headed for interesting times.