This article is part of a small series on voter migration at the 2019 election. See also Voter Migration by Party, Voter Migration by Group, and Voter Migration by Seat.
Tactical voting was much discussed at the 2019 election. The "first past the post" electoral system is hostile to electoral blocs which are split between parties, and rewards groups which can unite behind a single party banner in each seat. For this reason, both Remain and Leave blocs put effort into mobilising their voters to vote tactically to maximise their seats. On the Remain side, this meant convincing Lib Dem supporters to vote Labour in seats where Labour was more likely to win and to persuade pro-Remain Labour supporters to vote Lib Dem in seats where the Lib Dems are likely to be stronger. On the Leave side, this usually meant persuading Brexit party supporters to vote for the Conservatives.
Electoral Calculus was involved in this activity since we were commissioned by Remain United (a leading pro-Remain campaigning group) before the election to estimate what the likely result in each seat would be, and identify which seats could be vulnerable to pro-Remain tactical voting. Our final estimate was that there were about 20 marginal seats which could be influenced by pro-Remain tactical voting. There were additionally other marginal seats which could be influenced by pro-Leave tactical voting.
Since the Conservatives won the election with a handsome majority, there was little discussion of tactical voting in the immediate aftermath. But did tactical voting happen or not? And if it did, how many seats were affected? This article addresses these questions using a detailed analysis of the constituency results.
The main way to identify seats where tactical voting happened, is to look at seats where there was a significant difference between the predicted result and the actual result. This difference might be either that the actual winner of the seat was different from the predicted winner, or that the winning majority was larger or smaller than expected. If tactical voting took place in a particular seat, then the political trends (changes in vote share) in that seat should be different from national trends, which will lead to a difference between the predicted and actual results.
Of course, these differences can be caused by other factors. Differences can also be caused by model error (if the predicted vote shares are poor predictors), strong or weak local candidates (who get more or less votes than an average candidate would have received), idiosyncratic local factors (where local sentiment is driven by non-national issues), or other causes. An analysis of our model error (such as underestimating Plaid Cymru) and other factors can be seen in our track record of 2019. Nevertheless, we can try to identify tactical trends through the noise generated by these other causes.
The final campaign Electoral Calculus estimate was that 18 seats were vulnerable to pro-Remain tactical voting. This was based on the final poll-of-polls which showed a Conservative lead of 9.4pc. The polls were fairly accurate, but not exact, and the actual poll lead was 11.7pc, which was more than 2pc higher. If we had known the true Conservative lead at the time, it would not have changed the number of tactical seats by much (it would have been 21 seats rather than 18), but it would have changed which particular seats were identified as tactically marginal.
In order to identify the tactically-voting seats, it makes sense to take the model prediction based on the true poll lead of 11.7pc rather than the estimated poll lead of 9.4pc. This is to remove the noise generated by the moderate poll error so we can see the tactical effects more clearly.
The tactical voting model was based on two assumptions. The first was that people had a propensity to vote tactically which could be estimated with opinion polling. With our partners Savanta:ComRes we asked over 12,000 people across the country how likely they were to vote tactically. On average, around 40pc-45pc of Labour and Lib Dem supporters in England said they might vote tactically for the other party. The second assumption was that there would not be any pro-Leave tactical voting. This was not fully realistic but it was used to provide an illustration of how much impact tactical voting could have in one direction.
Using the true poll lead, there were Conservative 21 seats which were identified as vulnerable to pro-Remain tactical voting. There were a further 15 seats which were predicted to be very marginal Conservative victories, but could easily go either way if there were tactical voting. In total 36 seats were likely or possible to be pro-Remain tactical victories. This was based on the two assumptions above: that pro-Remain voters were 40pc-45pc likely to vote tactically, and ignoring any pro-Leave tactical effects.
Analysing these 36 seats, we count 10 seats where pro-Remain parties won the seat against the non-tactical prediction. This gives an upper bound on the likely effect of pro-Remain tactical voting. Compared with the likely target of 21 tactical seats, it is about half as much. What happened in the other seats? The table below shows the five types of seat, and how many of each type there were among the 36 vulnerable seats.
|Seat Outcome||Full Description||Num Seats|
|Pro-Remain victory||Seat won unexpectedly by pro-Remain party, possibly due to tactical voting||10|
|Unchanged||Seat won by predicted winner, with no big evidence of tactical voting||9|
|Pro-CON swing||Conservatives received more votes than expected, suggesting pro-Leave voting by Labour supporters or tactical voting by Brexit supporters||9|
|Hybrid swing||Combination of pro-Remain tactical voting and pro-Conservative swing, resulting in stalemate||5|
|Not enough||There was evidence of some pro-Remain tactical voting, but not enough to change the result||3|
In the ten seats which were won tactically, the share of third-party voters who defected was around 50pc (and higher in some seats). This was in line with our first assumption about how many voters might vote tactically. These seats were particularly common in Anglia, West Midlands and Yorks/Humber.
But in other seats (the "Unchanged" and "Not enough" groups) the share of defecting voters was only around 20pc, and tactical voting was less common. This happened in 12 seats in London, East Midlands and the North West.
And there were 14 seats where there was a large or medium amount of pro-Conservative voting. The nine seats with the clearest effect voted 66pc to Leave in the EU Referendum on average, indicating those seats were fairly committed to leaving. So this could be either pro-Leave voting by Labour supporters, or tactical voting by Brexit party supporters. These seats were focused in pro-Leave areas of the North and Midlands.
The other side of the tactical voting coin consists of seats which the Conservatives were not predicted to win, but which were vulnerable to pro-Leave or pro-Conservative tactical voting. Analysis of these seats was not in the scope of the original Remain United brief, but we can look at them now.
Performing a similar analysis, we can count 16 seats which the Conservatives won against the national trends. These break down into five categories:
|Seat Outcome||Full Description||Num Seats|
|LAB to CON||Classic swing directly from Labour to Conservative||7|
|Marginal||Very marginal seats which could have gone either way||3|
|Pro-CON||Swing to Conservative from both Labour and Lib Dems (High Peak, Wolverhampton SW)||2|
|Pro-Union||Possible anti-SNP tactical voting in Scotland (Banff and Buchan, Dumfries and Galloway)||2|
|Special||Special circumstances, such as retiring popular incumbent (Norfolk North) or poor Welsh modelling (Delyn)||2|
Additionally there were 8 other seats which were not won as expected, these break down as follows:
|Seat Outcome||Full Description||Num Seats|
|Special||Special circumstances, such as popular incumbent (Dagenham and Rainham); SNP candidate suspended for anti-semitism who won anyway (Kirkcaldy and Cowdenbeath); or a lingering dislike of Lib Dems (Sheffield Hallam)||3|
|Plaid Error||Model error underestimating Plaid Cymru support in Wales (Arfon, Ceredigion)||2|
|Surprise||Some seats had a very surprising result which went against the trend (Putney, Dunbartonshire East)||2|
|Pro-Union||Possible anti-SNP tactical voting in Scotland (Fife North East)||1|
So far we have seen 10 seats where pro-Remain tactical voting could have made a difference. There were also 9 seats where pro-Conservative swings were important (plus 7 marginal, pro-Union or special seats). Additionally there were 8 other seats which were not won as expected. The total of 34 seats which were not won as predicted can be broken down into these groups:
|Seat Outcome||Full Description||Num Seats|
|Pro-Remain||Pro-Remain tactical voting against the Conservatives||10|
|Pro-CON||Swing to Conservatives from Labour and other parties||9|
|Pro-Union||Anti-SNP tactical voting in Scotland||3|
|Marginal||Very marginal seats won by Conservatives||3|
|Plaid Error||Model error underestimating Plaid Cymru support in Wales||2|
|Special/Surprise||Special circumstances or surprising result||7|
Tactical voting did take place at the 2019 election, and it was enough to change the result in several seats. However, tactical voting was directed both against and towards the Conservatives. The two effects mostly cancelled each other out and the net effect was fairly small. Of course, that does not mean that each side's efforts were in vain, as each side gained over ten seats from those efforts. But overall tactical voting campaigns resembled an arms race where both sides made progress without achieving much relative gain.
The differences between the predicted and actual results can be seen seat-by-seat in graphical form. The interactive graphic below is a "political triangle" which maps Conservative, Labour and Lib Dem support into a triangle, where each party has a corner of strength and is strong in points near that corner, and weaker for points further away. Each seat is represented by a line which travels from its predicted political position to the actual 2019 result (with a dot coloured to represent the actual seat winner). The colour of line indicates the type of trend which affected the seat. A party should win a seat if it falls inside the kite shape indicated by the internal dotted lines. (See seat migration for more details.)
Interactive Usage: You can hover over any dot to see the name of that seat. You can also press the left-hand "Cycle" button to cycle through four seat filters: all seats; only seats with a large difference between predicted and actual results; only seats with a small difference between predicted and actual results; only seats with a difference between the predicted and actual party winner. You can also individually toggle the other five buttons to show or hide the five trend types.
(Note that the triangle only shows those 557 seats where the Conservatives, Labour and Lib Dem were predicted to be in the first three places. This excludes most Scottish and Welsh seats, where different dynamics apply. To see all seats in this format, please visit the 2019 track record page.)
There are five trend types which describe the type of difference between the predicted and actual result and its probable cause. These are:
Using the triangle we can get some interesting insights about tactical voting and other changes.
Remain. It is noticeable how much pro-Remain tactical voting there was in many seats. By toggling all types off except for 'Remain', we see there are many seats with long yellow tails. Some of these are heading in a direction parallel to the big orange arrow (marked 'TV swing from Lab to Lib') indicating tactical voting by Labour supporters in favour of the Lib Dems. Other go in the opposite direction showing tactical voting by Lib Dem supporters in favour of Labour. Generally seats which are predicted to be more Lib Dem than Labour (those which are nearer the Lib Dem corner than they are to the Labour corner) have swings from Labour to Lib Dem and vice versa. Although many seats move a long distance, very few (only five) cross the vital dotted lines which change the winner of the seat. Indeed many seats remain in the blue-shaded triangle which represents a Conservative vote share of more than 50pc. In those seats, anti-Conservative tactical voting will not be able to change the winner of the seat. But many other Conservative seats have become more marginal and these could be vulnerable to anti-Conservative swings at future elections.
Leave. There are also many seats where the Conservatives do better than predicted, which could indicate pro-Leave tactical voting by Leave supporters of other parties. In a few cases, these bring seats across the dotted lines to convert them into Conservative gains. In many other cases, the Conservatives pile up larger majorities which make those seats safer.
Hybrid. By combining both pro-Remain and pro-Leave effects, these seats tend not to change winner very much. There is a general trend to making seats into two-party contests by moving away from the centre of the triangle.
Anti-Con. There were a large number of seats where the Conservatives did worse than predicted, usually because Conservative supporters switched to Labour. This included four seats which Labour gained unexpectedly. But most of them were already safe Labour seats in Northern cities and London. So these gained voters did not bring a large benefit to Labour.
Tiny. By definition, these seats did not move very much. But there were six very marginal seats which moved enough to change the seat winner. You can see these by toggling 'Tiny' on, and pressing the 'Cycle' button until 'Gains' is selected.
There was actually quite large amounts of tactical voting at the 2019 election. Pro-Remain voters were enthusiastic tactical voters in many seats. However, they were able to change the seat winner in far fewer seats while only reducing the Conservative majority in many more seats. There was also a significant amount of pro-Leave tactically voting as Leave supporters voted tactically too.
Each side gained around 10 seats through tactical effects. But the net effect of these two types of tactical voting was broadly neutral as the efforts on both sides effectively cancelled each other out.
It was probably in each side's interests to maximise tactical voting by their natural supporters, since the other side was doing it, but the overall effect was more like a stalemate than a decisive advantage for either side.