By Michael Gallagher
Election campaigns feature extensive, some would say excessive, discussion of the horse-race aspects: in short, how many seats will the parties win? Until the votes are cast and counted, all we have to go on are the findings from opinion polls, and the challenge is to make accurate seat predictions from these.
One issue is the accuracy of the polls themselves, an issue highlighted by shortcomings shown up in the British election last May. This is an interesting subject in itself, but for the moment the question is how confident we could be, even if we knew for certain exactly how many votes each of the parties will win on 27 February, about being able to predict seat numbers.
Basically, there are two broad approaches to this:
(i) make predictions for every constituency and aggregate the totals;
(ii) make national-level seat predictions from national-level vote shares.
The obvious approach. Go through each of the 40 constituencies in turn and make predictions / best guesses, then add these, which produces national totals for each grouping. For whatever reason, this seems to be the more popular approach among pundits, but there are two downsides.
1a. That way madness lies. With 500+ candidates, producing a realistic estimate of the prospects of each one requires a massive amount of study. Most people have some sense of what’s going on in the constituency where they live or work, and some have contacts elsewhere that give them some insight into a few other constituencies. However, with all due respect to those who engage in the prediction of every constituency, no-one knows about the prospects of every candidate in the country, and indeed often those very closely in touch with individual constituencies, such as local journalists, freely acknowledge that they really don’t know what’s going to happen there.
1b. The model is not probabilistic. It considers each constituency contest in isolation and runs the risk that errors are cumulative. For example, suppose a party has an estimated 40 per cent chance of winning a seat in each of 5 constituencies. Examining each constituency separately leads to the conclusion that it probably won’t win a seat in any of them – this is the approach that leads to predictions that Labour might win fewer than 5 seats on the figures generated by some opinion polls. Taking a probabilistic approach, though, we can say that if a party has a 40 per cent chance of winning a seat in each of 5 constituencies, the most likely outcome is that it will win 2 seats there, even though we can’t say which two constituencies those will be. It might win zero, it might win 5, but 2 should be our best estimate.
A model that takes account of probabilities is much better than one based on the accumulation of discrete outcomes, because in real life parties tend to be over-represented in some constituencies and under-represented in others, rather than winning out, or losing out, everywhere. Thus, in six of the last nine elections, Labour won between 9 and 11 per cent of the votes, and it received 15, 16, 15, 17, 21 and 20 seats in these elections. The 32nd Dáil will be smaller, but not by much: even if these numbers were multiplied by 0.952 (the 158 seats in the next Dáil divided by the 166 in the previous 10 Dála), it still seems drastic to predict that Labour could win 10 per cent of the votes, or even a few points less than this, and yet be able to drive its TDs to Leinster House in the back of a single taxi.
This is much less time-consuming. It’s not necessary to know anything about any of the constituencies; all that’s needed is to have some idea as to how a given level of national support is likely to translate into overall seats.
This is easier said than done, though. In 2011 FG won 46.1% of the seats with 36.1% of the votes, a seats-to-votes conversion ratio of 1.28. If we could rely on this ratio holding true in 2016 as well, then with its current poll support level of 30% of the votes, FG would win 60 of the 157 contested seats as well as that of Seán Barrett, returned uncontested as Ceann Comhairle. However, it’s not that simple, because in 2002 FG won 18.8% of the seats with 21.5% of the votes, a ratio of 0.87. Applying that ratio to 30% of the votes on 26 February would give FG a mere 42 seats in all. Which is more likely to be right?
This seats-to-votes conversion ratio depends mainly on at least five factors: the vagaries caused by small constituency size, the fragmentation of the rest of the votes, the evenness of the party’s vote across the country, transfer patterns, and the degree of concentration of the party’s own votes. (For more on the ratio of seat shares to vote shares see Michael Marsh’s analysis on the RTE Election site.)
2a. The vagaries caused by small constituency size. The unusual feature of the Irish electoral system is not, as some people think, the intra-party electoral competition created by PR-STV, which is a common feature of European electoral systems, but the small constituency size: just under 4 TDs per constituency on average, which is exceptionally low for a PR system. This means that luck, the hop of the ball, plays a greater role than if Ireland used constituencies of 5, 6 and 7 seats, in which case the relationship of seats to votes would be likely to be fairer and more consistent. Usually small constituency size works to the advantage of the larger parties, but in any given case it might be a small party or an independent who benefits.
2b. The fragmentation of the rest of the votes. If a party wins 36% of the votes in a 5-seat constituency, how many seats will it win? If it’s a 2-party system and the other party wins 64% of the votes, then clearly the larger party will win 3 seats and the smaller just 2. But if the other 64% of the votes are dispersed among many other parties and independents, then 36% of the votes might suffice for 3 seats, as it did for FG in 2011 in Dublin South. With a multitude of small parties and independent candidates this time, the larger parties have a reasonable chance of being significantly over-represented.
2c. The evenness of the party’s vote across the country. For a small party, a ‘lumpy’ distribution is better than an even one: strong in some areas and pathetically weak in others is better than just plain weak everywhere. For example, if a party wins 5% of the votes nationally, then if it wins exactly 5% in each constituency it will probably win no seats at all, whereas if it wins 20% in a quarter of the constituencies and pretty much zero everywhere else, it should end up with 10 seats given that there are 40 constituencies. In 2011, FF won only 1 seat in Dublin because of its evenly-spread support here, whereas SF, with fewer votes in the capital, took 4 seats because of its lumpier vote pattern (discussed more fully in How Ireland Voted 2011). On the other hand, a stronger party might fare better if its support is fairly even, giving it a chance of winning a seat nearly everywhere. With the same national vote total, a party might be significantly over-represented or significantly under-represented in seat terms depending on whether its vote distribution is optimal (many seats just barely won) or sub-optimal (more votes won than needed to win its seats in each constituency but not quite enough to win an additional one). The impact of this factor can be assessed after the election but it’s near to impossible to predict in advance.
2d. Transfer patterns. The median party on the main dimension of competition stands to benefit from receiving lower preferences from all quarters while a pariah party is likely to receive a smaller share of seats than of first preference votes. Very often Labour has held the median position, being preferred by FG voters to any party further to the left and preferred by those on its left to either FG or FF. After its five years in government there does not seem to be much affection for it among voters further to the left, and SF (which used to be a pariah party, and still is for some, though it has by now shed this status in the eyes of many voters) is more likely to receive transfers from this quarter. Labour will still receive some transfers when the only alternative is FF or FG, though, and it will still receive a reasonable proportion of transfers from FG. In 2011 FF was the pariah party, one reason why it won just 11.5% of the contested seats with 17.4% of the votes, and it’s been estimated that it lost out on nine seats in 2011 due to other parties’ supporters using their lower preferences against it. The name ‘Fianna Fáil’ no longer produces quite the same apoplexy among the electorate that we saw in 2011, and while the party is hardly a transfer magnet it should at least win a seat share pretty close to its vote share.
2e. The degree of concentration of the party’s own votes. If a party’s vote is optimally distributed among the ideal number of candidates, its seats-to-votes conversion ratio will be higher than if its votes are being won by a larger number of candidates with a poor balance among these. For example, if it runs just one candidate in a constituency and she receives votes equivalent to 0.8 of a quota, it is likely to win a seat. If it runs two candidates and they win 0.94 of a quota, fairly evenly divided between the two of them, then it probably won’t win a seat (eg FF in Cork SW in 2011). If it runs three candidates and these win 1.1 quotas, fairly evenly divided among the three of them, then its chances of winning a seat are lower still (eg FG in Limerick W in 1987). This is where party head offices can earn their corn: picking the right number of candidates in each constituency, and balancing the votes optimally between these, can make a difference to the number of seats won. FF, for example, clearly nominated too many candidates in 2011. However, if things go wrong this is not necessarily head office’s fault: sometimes it becomes obvious only with the benefit of hindsight what the optimal number of candidates was, or head office might see clearly what needs to be done but be unable to get others to cooperate in implementing the strategy.
Predicting seats from votes remains an art as much as a science, but some methods hold more promise than others.
One thought on “How to predict the result of an Irish election”
I think predictions by constituency are best given the nature of Irish politics tending to be local (parish pump) based.