A model in progress

In April when this site launched I outlined what a model for Irish elections could look like. The reaction to the post was overwhelming and I want to thank everyone for their feedback and support.

Over the summer I’ve put my ideas to work and I’m happy to report some progress. But progress only, not a final model. Below I’ve shared some initial output and some issues are worth clearing up first:

  • The output below is derived from recent polls. Specifically, it takes polls from the past two months and averages them with proportions towards more recent polls. It assumes the holding of an election tomorrow and is not a projection towards 2016. No one knows what will happen then, although when polls begin to move, the output below will change too.
  • The model uses a list of candidates which includes announced and presumptive candidates. Many of my presumptions will be wrong (some candidates will retire/not stand) and other candidates will announce to run. For example in 2011 the late candidacies of Peter Mathews, Mick Wallace and Shane Ross provided a late shift in the state-of-play for their constituencies.
  • The model is still full of bugs and I have yet to sit down with other R-fluent people (people who can understand the code I’ve used) to iron out these issues and assess the model’s accuracy. For example, candidates are weighted on their past election results. Where such data are not available, a score is added subjectively. The code has also had trouble running multiple replications. In an attempt to run the code 1,000 times last night, only 419 replications were made before the code halted. One problem may relate to the handing of vote ties, something my code is not yet able to handle.

Initial output

Below are output graphs for 11 Irish political parties and groupings. These grouping are likely to change ahead of the election. The number of seats is listed on the x-axis (bottom) with the probability of a party obtaining this number of seats on the y-axis (left). Probabilities are very wide at this point. For example, knowing that Sinn Féin is 14% likely to win 32 seats means that it is 86% likely to win another total number of seats.

 

Fine Gael

Of all parties, the output for Fine Gael shows the widest range of possible seats. This is a shame because ideally the model should form a pyramid shape which shows a more distinct seat number as being the one most likely. Nonetheless, the strong grouping in the 43-48 seat range provides some clarity on the Fine Gael position ahead of the election.

FGseatprobabilities

 

Labour

Labour’s graph is closer to the pyramid shape we are looking for, showing the party with a c. 19% probability of winning 3 seats at the next election.

LABseatprobabilities

 

Fianna Fáil

Despite polling near its 2011 support levels, Fianna Fáil looks poised to make significant gains at the next election. There are a number of factors at play here, including its relative position to the larger Fine Gael party (now improved by FG’s decline), better selection of candidates at constituency level (i.e running the right number of candidates) and boundary changes which are largely favorable to the party. Seats in the range of 35-38 appear likely if an election were held tomorrow.

FFseatprobabilities

 

Sinn Féin

Despite polling at a similar level to Fianna Fáil above, transfers are less forthcoming for Sinn Féin, meaning the party is likely to win less seats than its opposition counterparts. Nonetheless, with 14 outgoing TDs at present, a doubling of seats for Sinn Féin would represent an historic result for the party on the year of the 1916 centenary.

SFseatprobabilities

 

Renua Ireland

With three outgoing TDs, Renua appears likely at this point to retain its existing seats, but no more. The competitive constituencies of Wicklow and Dublin Bay North present problems for Billy Timmins and Terence Flanagan, which explains the high probability for a return of just two seats. The potential for a 4th, 5th and 6th seat comes from Carlow-Kilkenny candidate Patrick McKee, Councillor John Leahy in Offaly and Senator Paul Bradford in Cork East.

RNseatprobabilities

 

Anti Austerity Alliance/ Socialist Party

With talks ongoing for a broad-left coalition in advance of the election, this graph may become obsolete in the future. Nonetheless, the model appears to have accurately captured the return of Paul Murphy and Ruth Coppinger to the Dáil.

AAAseatprobabilities

 

People Before Profit

Richard Boyd Barrett appears safe for a seat in Dún Laoghaire with either John Lyons (Dublin Bay North) and Gino Kenny (Dublin Mid West) likely to return a second (if not third) seat for the party. Once again, any formation of a broad left coalition could change the candidates and outcomes for constituencies. It is also worth nothing that in the case of Gino Kenny, the lack of other “Independent/Other” candidates means he takes nearly all available votes (of which there are a lot). This situation would likely change with the emergence of other candidates.


PBPseatprobabilities

 

Social Democrats

With only three candidates at present (TDs Roisin Shortall, Stephen Donnelly and Catherine Murphy) no extra seats can be derived for the newly formed party. If anything, the competitiveness of the Wicklow constituency puts the seat of Stephen Donnelly at risk, although this is a very rare occurrence in the model.

SDseatprobabilities

 

United Left

“United Left” is a broad term to describe other left-wing candidates that are not in either AAA/SP or PBP. Three seats is the most likely outcome for this grouping which comprises TDs Joan Collins, Clare Daly and Semaus Healy (WUAG). Once again, these groupings are preliminary and will change if “the left” comes together in some form.

ULseatprobabilities

 

Green Party

After the loss of all its TDs in 2011, the Green Party appears to be struggling to regain support although a few points are worth bearing in mind here. The first is that the Greens sometimes fail to register in national polls. The party often returns support of c. 2% nationally, meaning a poll average of 1% only captures half the likely Green vote. The second is that while its support has previously been highest in a number of Dublin constituencies (Fingal, Mid-West, Rathdown etc.) it is likely to concentrate at the next election in one key constituency – Dublin Bay South. Leader Eamon Ryan is contesting here and is likely to have a vote level far above all other Green candidates and above levels previously seen in other elections. The model says zero seats, but intuitively, one seat is more likely.

GPseatprobabilities

 

Independents and Others

While “Independents and Others” usually includes the above smaller parties (AAA, PBP, RN, SD, GP, UL) in this case it only covers a number of small parties such as the Workers’ Party and Direct Democracy Ireland. Since none of these micro-parties are poised to win seats, you can interpret this table as just being for “Independent” candidates. At 28 seats, an election held tomorrow would see an unprecedented number of non-party TDs in the Dáil.

IND-Otherseatprobabilities

 

The output above represents a model in progress. Its numbers are only a reflection of the state-of-play on today’s poll numbers, and not what may happen in 2016. Even if the polls were to stay static during the election, new candidates will emerge and disrupt the political landscape. Much more work is needed on this model, but I am content with the progress so far and will post when I have more updates to share. As always your feedback is most encouraged and welcome.

 

You can follow Irish Election Stats on Facebook and Twitter.

David Higgins tweets at @higginsdavidw

 

One thought on “A model in progress

  1. Hi David, interesting post.

    Do you plan to produce output for the model at constituency level? i.e. something like candidate A has x% change of winning a seat etc.

    Or do you intend for the model to give a national picture only, where errors at constituency level are evened out over 40 constituencies.

    I understand your model is a work in progress & all candidates are not yet declared at this stage. It would be interesting to see this constituency level output in the final few weeks leading up to the election, when all candidates are declared.

    Thanks for your work!

Leave a Reply