How accurate are these projections?
I don’t know. Making projections under a multi-seat PR-STV system is a nightmare. It’s probably better not to think of this as a forecast in and of itself, but rather an experiment in finding a functional methodology that can make seat projections in Ireland with at least some accuracy.
What are the challenges for these projections?
There’s quite a few, but to give some prominent examples:
Localism, for better or for ill, has a huge impact in Irish elections. An average ballot from Kilgarvan will be very different from one from Corca Dhuibhne. And its not just the rural issue it’s often stereotyped as – an average ballot from Pembroke East E will be very different to one from Royal Exchange. There isn’t really data to indicate how the geography of candidates and turnout will impact transfers, though Ian Richardson has done stellar work mapping the first preference results here.
There are problems with accurately gauging the Social Democrats, because they ran in so few constituencies in 2020 that the underlying data is deficient, and a lot of assumptions have to be made.
PBP/Solidarity have an issue whereby they have a handful of wildly popular and successful candidates, but also a lot of candidates who are going to struggle to keep their deposits; their vote is more volatile depending on candidate profile than any other major party, and a model is going to have difficulty factoring this in.
Independents are also difficult, because they vary so widely even within a constituency in terms of popularity, voter base, transfer friendliness/sources etc., but because of how polling is measured, they have to be treated as a cohesive unit, with minor parties lumped into the same measurement, and vary widely on transfers and swing.
Also, although a concern with polling rather than the model itself, the model might be seriously underestimating Aontú. Ireland Thinks has them much, much higher than anybody else does, but also doesn’t publish consistent provincial breakdowns, so the model can’t really factor it in. If Ireland Thinks are spotting something no other pollster is, as opposed to just having a really weird house effect, then the model will be off in this regard.
Finally, and most importantly, the most accurate version of the model won’t exist until we know for definite how many candidates each party is running in each constituency. This matters for four reasons: (1) the division of votes of larger parties (2) the relative performance of individual candidates running for the same party in the same constituency (3) parties or high-profile independents not running, and where their votes go (4) new high profile independents running, or parties contesting a constituency they didn’t contest in 2020, and who they will take votes from.
You can find out more about the modelling here.
What methodology did you use?
You can see the methodology, with explanations of the models involved, their strengths and weaknesses and notes on how I used them to create a final model, here.
Is the methodology static?
I hope not! Ideally I’ll be able to iterate as time goes on to refine and improve things, particularly looking to better calculate transfers and accommodate for local nuances, as well as move towards being able to articulate probabilities versus set numbers.
The website is currently very barebones; I plan on evolving it to be better over time.
What data source did you use?
For baseline numbers and transfer numbers, http://irelandelection.com/, with manual corrections to some data that’s miscalculated on the site (negative transfer numbers etc.).
Polling numbers are currently from Red C, Behaviour and Attitudes, IPSOS/MRBI and Ireland Thinks. If other pollsters (Amárach, Kantar Millward Brown, Panelbase, Survation etc.) start polling Ireland again, they will be included.
The seat diagrams are made using https://parliamentdiagram.toolforge.org/parlitest.php.
All maps used are digitally hand-drawn; you can reuse them under CC-BY-SA-4.0.
All modelling was built in Excel.
Who runs this website?
You can find this out on the about page.
How often will this projection be updated?
It won’t be with every poll. Ireland Thinks tend not to publish provincial breakdowns, and IPSOS/MRBI tend to lump PBP and Aontú in with “other”, both of which are not helpful (though I think the latter is due to a political choice by the Irish Times in how they present their data, rather than an issue with the pollster itself).
Also, because any polling shift that results in a change in elimination order means that entire constituencies will have to be re-calculated, the process is currently quite time-consuming. I don’t know if I will have bandwidth to update for every poll, though your support makes this easier. At this stage, updates will mostly depend on how frequent polls are, or if a radical shift in polling patterns emerges.
You’re wrong about constituency X, I live there and you haven’t considered local factors such as A, B or C.
Absolutely! This is a flaw with a mathematical model that cannot be overcome with current polling information. No model will ever fully capture the idiosyncrasies of Irish constituencies. The model has to assume that the swing in Cork City will be the same as the swing in Limerick County, because it simply doesn’t have more granular data to go on than provincial-level.
What on earth is going on with constituency X?
You can have a look at each constituency in detail; the full list with links to each page is here.
Why does your model hate Labour/PBP/SDs?
They’re not polling well. The model is going to miss some constituency nuances, but the polling numbers are what’s reflected in the outcome. Ultimately, left-wing parties were already squeezed by Sinn Féin in 2020, but SF didn’t run enough candidates to take advantage. If they do, a lot of those seats go to SF instead, even without SF’s current bump in polling.
Then why does it have the Green Party winning seats?
Because the methodology uses a rolling polling average, and the Green Party’s decline in Dublin hasn’t properly filtered through that yet; more recent bad polls for them are offset by more positive older ones. If polling continues on a downward trend, the rolling average will move down and the projections will change accordingly.
I love this, is there anything I can do to support this work?
This is all done in my spare time and for my own interest, but if you want to toss me a couple of bob to make my life easier and get some better kit to improve future work on this, you can do so here.
I hate this and I hate you, who do I complain to?
Please direct all negative feedback to your local Fine Gael branch.