British polling company YouGov published their first MRP report for the 2025 election cycle, becoming the second company to publish an MRP during this election cycle. For this post I wanted to run a comparison of the two most recent MRPs from two different companies. The two polls produced similar top-level figures, both in terms of seats and votes, but at the individual seat level there are some pretty significant differences.
First up: a brief explanation about MRPs. The acronym stands for multi-level regression with post-stratification. Basically they are polls with very large sample sizes which use data science to determine connections between various characteristics that predict how people are voting, and then use this to make predictions about the poll for each electorate, not just the national or state totals.
Since swings are rarely uniform, this sort of analysis is really useful for understanding which parts of the country might be swinging more or less than others, and this can sometimes change how many seats would be expected to change hands, depending on how efficient swings are. If swings are concentrated in safe seats, they can produce fewer changes in results than an election with a similar result where the swings were concentrated in more marginal seats.
Redbridge and Accent Research produced three MRPs throughout 2024. YouGov has now published their first MRP today. YouGov previously produced an MRP right before the last federal election. Shaun Ratcliff, who now runs Accent Research and was the guest on my most recent podcast episode, previously produced YouGov’s MRPs in Australia. YouGov has also produced MRPs in the UK and Spain. In the UK, a number of companies published MRPs in the lead up to the 2024 general election.
I don’t normally spend a lot of effort dwelling on the top-line polls – I recommend William Bowe’s Bludgertrack for national and state polling averages. Both William and Kevin Bonham do more on the polls. But I’m more interested in the seat by seat contest and these MRPs do offer opinions about the local races.
The latest Redbridge/Accent MRP was published in early December and had a sample size of 4,909. The poll gave a range of possible outcomes on a seat count, but it appears that the most likely outcome in each seat adds up to a total of Coalition 69, Labor 64, Others 9, Greens 4 and four ties (all Labor vs Coalition). The poll did not publish a national 2PP estimate, but it seems to be a small Coalition lead on 2PP (say 51%).
The new YouGov MRP was published this morning, and has a sample of over 40,000 interviews. It is based on a national two-party-preferred vote of 51.1% to the Coalition, and also gives a range of seat outcomes. On the website they make a prediction that the most likely outcome is 73 Coalition, 66 Labor, ten others and one Greens. If you drill into the individual seat data, it produces most likely outcomes of 70 Coalition, 66 Labor, ten others, one Greens and three ties.
It’s worth noting that a simple reading of the pendulum suggests fewer Coalition gains. A 2PP of 51.1% would be a swing of 3.2%. There are just seven Labor seats and two Crossbench seats on margins of less than 3.2% against the Coalition. This would take the Coalition from 58 seats to 67, and take Labor from 77 to 70. The recent MRPs suggest the Coalition could gain more like 15 seats, rather than nine.
I should note that from here on in I will describe their models and may use words like “winning” or “will be”, but they simply refer to the model of how things look right now. Neither pollster is predicting this is how things will end up at the time of the election.
Both of these polls produce remarkably similar overall results, both giving the Coalition a slight 2PP lead (consistent with most polls in early 2025 showing a slight Coalition lead). Both would produce a result of the Coalition winning more seats than Labor, falling a few seats short of a majority.
But if you drill down below the national level, they produce quite different outcomes at the seat level, between regions and between some states.
In net terms, the Coalition is leading in one more seat in the YouGov poll, and Labor is leading in two more. The Greens are leading in three less (4 in Accent, 1 in YouGov), the independents are leading in one more, and there is one less tie (three in YouGov, four in Accent).
Yet while those differences could be explained by just four differences, actually there are 19 seats that are flipped between the two polls.
About a third of these seats are explained by ties. Macquarie, McEwen and Shortland are all ties in YouGov, and all have a 51-49 lead in Accent (2 to Labor, one to the Coalition). Likewise Chisholm, Hawke, Reid and Sturt are tied in Accent but have 51-53% leads in YouGov.
But some others have bigger gaps. YouGov expects Dai Le to drop out of the top two and lose Fowler to Labor, and expect the Greens to lose all three of their Brisbane seats. Their model has the LNP falling into third in Griffith, while the Greens fall into third in Brisbane and Ryan, with Labor winning all three. Interestingly YouGov has Labor’s victory margins in Wills and Cooper over the Greens as significantly smaller than Accent.
Labor wins Solomon much more comfortably in Accent than in YouGov, while YouGov expects Lingiari to actually swing towards Labor while Accent expects the Coalition to win.
Interestingly YouGov expects a swing of only 1% to the Coalition in the large rural WA seat of Durack, compared to a 7% swing in Accent.
When I plot the expected swings onto the pendulum, YouGov generally has similar swings to seats on similar margins, while there is a bit more variety in Accent.
There are 124 seats where both polls agree that the final count is Labor vs Coalition, plus a few more where YouGov expects such a count. YouGov published a 2PP swing for all 150 seats, but Accent only did so where the 2PP matched the 2CP. If I compare average two-party-preferred swings amongst those seats, there are some interesting trends.
Both polls agree that swings are expected to be bigger in Labor seats than Coalition seats, but YouGov is expecting the swing to be bigger – 5.6% in Labor seats for YouGov, and 4.1% for Accent.
Amongst Labor seats on margins of 0-5%, Accent is expecting a slightly higher swing – 4.1% vs 3.6% for YouGov. But YouGov is expecting bigger swings in safer seats – 5.8% for seats on margins of 5-10% (compared to 3.6% for Accent) and 6.3% for those over 10% (compared to 4.5% for Accent).
Unsurprisingly, since the two polls seem to produce similar national figures, these bigger swings in Labor seats must be offset in Coalition seats. There is not much difference in the marginal seats. YouGov expects the swing in 5-10% Coalition seats to be a bit less (1.2% rather than 1.6%), but the big difference is amongst the safest Coalition seats, where they are expecting a 1.5% swing towards Labor rather than the 0.3% swing to the Coalition in Accent’s work.
This is very obvious on the pendulum. For the eight safest Coalition seats, held by margins of 13-23%, every seat is expected to suffer a swing of at least 3% to Labor, with Gippsland swinging to Labor by 7%. Accent also expects the Coalition swing to be muted here, but it’s much less dramatic – Gippsland still has a 3% swing to the Coalition and other seats range from 1% Coalition swing to a 2% Labor swing.
I find this a bit hard to swallow, but I haven’t done the analysis myself, so I’m presenting it here.
These swings are consistent with the idea that the Coalition is currently producing a very efficient vote, not wasting votes in seats they already hold, but YouGov’s model also suggests the Coalition is racking up big gains in safer Labor seats.
Finally it’s worth just checking in on four seats I’ve been paying close attention to. These outer suburban seats in New South Wales and Victoria have always been Labor-held but have been trending marginal. None of them are super-marginal but in all four cases both MRPs show them as being very close. Werriwa is within 1% on both models. Shortland is a tossup on YouGov and within 1% on Accent. Hawke is a tossup on Accent and within 1% on YouGov. Holt is a bit safer on YouGov (54% ALP) but remains very much in play.
Overall while there is a fair amount of variation, both polls have a common message. The Coalition is in the lead, they are on track to be the largest party with a chance of winning a majority, and they are gaining swings that produce more seat gains than the pendulum would suggest.
If I am reading the figures correctly, it seemed to me that there was a tendency towards the centre for all seats at the extreme. For example
SA Labors safest seat of Kingstown had a swing against labor of 10% but Barker had a swing towards Labor.
In victoria Fraser 6% against Labor whilst Gippsland 7% to Labor.
WA all swung against Labor but 10% in Brand and 1% in Durack.
Oxley -6%, Moncrieff +2%,
Newcastle -7% Farrer and Parkes +4%
Yeh agree with David, they don’t really pick up the factors that allow a seat to become ultra safe. YouGov in particular. Take the safe seat of New England, YouGov has Labor having their best result this century, Labor doing better in Gippsland than they did in 2013, Labor getting their best result in Mallee this century etc.
I don’t think there is a single traditional ALP vs Lib/Nat seat that has a margin over 15% in YouGov. There was 17 seats with a >15% margin after the 2022, vs zero in YouGov.
MRP is an interesting technique.
Had MRP been used in the polls prior to the 2019 federal election, it might have avoided all of the hand-wringing about the polls “getting it wrong” or succumbing to “herding”
The determining variable in the polls that year was whether or not One Nation was running in the electorate of interest. The pollsters assumed that One Nation would be running in 150 electorates, but it ran in only 50 electorates. This resulted in an error of 4.4% in the national projected One Nation vote. The ON voters then had to vote for their second choice. This might have split 60:40 Coalition:ALP
There was, however, one poll that got it right – the final giant Newspoll run a few days before the election. In essence, Newspoll did not ask voters in “Electorate X” whether they would vote for One Nation, if there were no ON candidate in that electorate.
It wasn’t MRP, but is was a simple equivalent that worked
What I cannot fathom from this MRP data is how it has been decided that there would be swings to the ALP in almost every inland electorate plus the 7% to Labor in Gippsland and 4% in Mallee. And that goes across SA, Victoria and NSW. Redbridge in some of their Victorian polls have had very odd results for regional areas when they have split the state into regions – perhaps too much slicing and dicing so that the margins of error get too high. They might have acknowledged this by changing the regional split each time. The whole thing is all very odd.
Thank you for this great analysis. Hard work. There is no surprise about inconsistencies between polls for sure. It will drive you crazy if you try to reconcile these polls locally. No criticism of any poll intended here.
Focusing on this YouGov poll – it relies entirely on paid (not much) volunteers. From 40,000+ odd they use census data & other data to weed down to a representative national sample of 8,000+. An immediate problem here is this technique is not totally random (what poll is) & will not capture certain demographics at all to weed down. For example, non tech device users, reluctant tech device users & informal voters. The 8000+ representative mince is then squeezed into electoral sausages which will become even less reliable because of the dilution of cross section.
Another problem is the electorate analysis model is not always accurate/up to date. Consider Calare everybody knows that Gee (an independent) will be re-elected as an independent against a strong Nationals’ challenge, but the poll gives it to Nationals v Independent. (& probably wrong independent). & Monash it’s a 4-way contest with Broadbent & it would take a preference re-alignment of Marne-like proportions for the Liberals to win that seat, but the poll says that it’s safe. And then in Groom it misses an independent altogether. Other seats like McPherson and Fairfax are distributed LNP v Labor but an independent will be in the final 2 – recently Albo hinted so & LNP are complaining that some retiring MPs aren’t doing much. It misses local demographic about retiring members like Hinkler & Leichardt. Although I can’t put my finger on it, but the results in Ryan & Griffith & Brisbane stink too. Ryan is a traditional LNP seat & if it goes (which I doubt) – well it goes there. How can Labor finish ahead of Greens in Griffith against Mr Charisma & LNP will certainly finish second. So although it’s a Labor seat if it goes, it goes LNP. & finally look at the margin in Brisbane of Greens over Labor. (Readers of my comments know that I do not hold a candle for the Greens.)
Finally, a non-uniform 2 party swing of about 2% in primaries to the coalition & 3.4% against Labor is about right. Think Werribee? But everybody on this site knows, two party national preferred analysis are pointless because to be relevant to a national result then no independent/green vote must actually be counted in the last 2 in any contest. But it is possible to identify at least 25 seats (ie 16% of all seats) where the vote will be counted & it is in those seats that the independent/green vote will also be massed. It is not unreasonable to estimate that of the total 24% vote estimated in this poll 12% will not be distributed between Labor/LNP.
Finally, finally on national two party preferred vote, although the One Nation vote is an overestimate (supporters are apparently encouraged to enrol for YouGov) we know that every one of those votes is distributed & ultimately at about 35 -36% to the left. Great news for the coalition? Not really because, in addition, another 15% of the One Nation vote will usually pass through independent/green (yes look at the 3 Brisbane seats) on its way to the LNP and in those crucial 16% of all seats it will irrigate the independent/green vote against the LNP.
Sorry to be so long. I have enrolled for YouGov & hope to get my $20 coles cards! We all should if we are retired and have the time?
With a sample size of 40,000+ across 150 electorates would mean that there’s roughly 300 per seat.
Caleb: but only 8000+ make the cut off for consideration and forvalid reasons. That’s 53 per seat?
271 and change per seat Caleb.
Roger I can’t see broadbent winning Monash the liberals have a great candidate who comprehensively won preselection in the branch so why would liberal voters do different. Nothing against him other then his age people will be looking to the coalition for Col relief and the fact labor party are in trouble in vic. Also atm I’m saying nats retain but gee could mount a seious challenge. If the nats primary drops below 35 they got problems though. Even under 40 he could likely win. Over 40 the nats should hold easily.
Also Labor could challenge for Griffith but it won’t be against the greens the lnp will make the 2cp.
@ john
Dutton has said latest version
There are no sugar hits and has opposed any alp attempts for cost of living relief
So he is relying on the anti Labor vibe to win.
I have no knowledge of the local liberal party in this area and the loss of preselection can be “you” guess… recruitment it’s time his age etc
I have been trying to tell people here that Mr Broadbent is more substantial than he appears at first sight.
We’ll see on the night… don’t think he can win but the seat is on a approx 3% margin.. may help Labor here
John: the problem in Monash is 4 way contest – Broadbent + Teal + ALP V Liberal. This is the preference re-alignment of Marne-like proportions that I referred to for the Liberals to win that seat. My bet is Teal. I pop up to Mudgee & Orange occasionally – Gee will be re-elected at a canter with ALP tactical preferences & then the other independent’s preferences. Agree LNP come second in Griffith but Charisma comes first & is elected on ALP preferences?
Mick labor are only going backwards in vic they won’t win Monash. If labor do well the teal can’t make the 2cp. If they couldn’t win in 2922 they won’t win now
The left right split was about 55right. 35 left and 10 in the middle which we can assume went to labor based on the 2pp.
@roger. The only way libs can lose is to broadbent taking enough of a personal vote from the libs to win on preferences from labor and the greens unless they can get onp and liberatarian preferences depending on who runs. The teal will likely stick to the vote 1 and you decide prefernces. Broadbent do the same or the libs because if he prefernces labor or anyone else it will just make it so much easier for the libs. And that’s assuming labor and the greens prefernce him to begin with. My guess status quo of lib vs lab as the grn will prefernce labor over libs and broadbent and prolly give labor enough vote to make the required 33.3% unless the grns get the teal over labor. Either way neither can beat the lib here. The centre right vot e will simply be too strong.
On Calare it’s the same situation labor and the teal cannot win as the centre right vote will be too strong only gee can win against the nats. But same problem teal votes will splinter as they don’t indicate prefernces other then 1. So on that I’ve got Nat retain. But I give gee wildcard status
@Roger Roughead – “And then in Groom it misses an independent altogether” – in Groom 2022, the teal independent, Susie Holt, polled 8.3%, in fourth place. Preferences from the UAP and Greens pushed her ahead of 3rd placed One Nation (who polled 9.6%). One Nation preferences flowed strongly to Holt, allowing her to overtake 2nd placed Labor (who polled 18.7%) and make the 2pp. She reached the 2pp as a result of very specific preference flows, which won’t happen again. It’ll go back to an LNP vs ALP seat.
On the Brisbane seats, anything can happen in all three. If the Greens lose Griffith, it will go to Labor. There is no way the LNP will ever be competitive in that seat. I think there’s a chance that MCM’s vote decreases as a correction of 2022’s large swing, Labor can sneak in the 2pp and win on LNP preferences. Brisbane is a toss up between The Greens and Labor, they’re extremely close on first preferences and Bates has been a low-profile local member. Ryan is a complete toss up, I expect the Greens’ vote to go down as a correction from 2022 (EWB has been low profile too), but I don’t know whether it’ll flow to the LNP or Labor. The LNP are throwing the kitchen sink into advertising, but the electorate doesn’t like Dutton and nuclear. The Labor candidate has been very active in the community so I wouldn’t rule out a decent swing back to them.
The yougov mrp also has a much more likely result for MacArthur at 53-47. Whereas the other had the libs winning which I think is highly unlikely
A.A.: Labor can sneak in the 2pp and win on LNP preferences.
****************************************************
Liberal/LNP have never been out of the 2CP in Griffith and 2022 was a low water mark for them, their PV dropped 8,121 from 2019 though 6,693 extra valid votes were cast in 2022. Same candidate both times for the LNP, MCM has had his turn, it’s been 3 years on stunting with no achievements. LNP protest vote to return to LNP, Labor to scrape in on Greens preferences, LNP to top the PV again.
John: Thanks re Monash. On your analysis it is Broadbent if he can come through the ruck. I am assuming that Green & Labor will preference him ahead of Liberals and each other. Re Calare: it is heroic to think that the strong teal independent’s vote will go anywhere but to Gee. He walked out of the Nationals on the Voice. I’m guessing Labor will try to throw Gee over the line against the Teals. Cynical: well I lived for a few years in Queensland & Labor/Beattie used to use Labor’s third place position on Gold Coast to throw a National over the line one term & then a Liberal the next term – like a revolving door. Eventually he won some of those seats because all his opponents were so dizzy.
AA: Yes Susie Holt finished second and the YouGov model pretends that it never happened? That’s my point. If she runs of course, she will get more than 8.3% this time, you agree? & she will finish in the top 2? You agree?
AA: I will defer to your local knowledge on Brisbane but personally I think that it is wishful thinking that Mr Charisma will not increase his vote in Griffith? His real problem is ALP voters want to cut off his preferences & his b..ls too?
Gympie: see AA. I do think that your dislike (who doesn’t) of Charisma is blinding you & AA to his impact.
I have updated my identification of seats where the vote of the independent/green vote will be counted in the final two to 30 (that’s 20% of all seats) & it is in those seats that will also be massed. I’ll drop Groome in respect to AA but they are Curtin, Mayo, Grey, Clarke, Goldstein, Kooyong, Monash, Indi, Wannon, Nicholls, Melbourne, Canberra, Calare, Mackellar, Bradfield, Wentworth, Hume, Farrer, Warringah, Sydney, Grayndler, Cowper, Fowler, Kennedy, Brisbane, Griffith, Ryan, McPherson and Fairfax
Some of the Yougov swings look pretty rogue – if the Liberals win Macquarie and Eden-Monaro Dutton will have the lodge to himself
Mrp is not intended to be a el cheap individual seat poll. I would rate the liberal party ‘s chances of winning Dobell Macquarie or Eden Monaro as slightly less than zero.
The non classic seats become complicated and may well depend on exclusion order. They are seats where Labor cannot win under normal circumstances so the choice if for people who wish to vote against the coalition
A put x the liberal candidate last would be helpful there
@Gympie while I agree that there will be a swing back to the LNP in Griffith, I don’t think it’ll be very much. Griffith’ LNP voters are mostly “small-l” and don’t like Dutton or his nuclear scheme. Having him as leader will probably turn off a lot of typical LNP voters.
@Roger Roughead
“If she runs of course, she will get more than 8.3% this time, you agree? & she will finish in the top 2? You agree?”
I don’t think Suzie Holt will get more than 10% of the vote. I also think she won’t make the 2pp, and will likely stay 4th, behind One Nation and Labor. The absence of the UAP means that One Nation will likely get a boost to their primary vote, helping them stay in 3rd. And I think that the absence of the morrison factor will shift votes from Holt to the LNP. Groom just isn’t a tealish electorate.
I’m not underestimating MCM’s influence on Griffith. I know he has a strong support base in Griffith. I said that there’s a chance of Labor winning. But there’s also a decent chance of him retaining.
I just don’t think he’ll increase his vote, as the anti-Morrison protest vote is gone, he’s been a controversial figure in regards to housing and The Greens have failed to win overlapping seats at the council and state level.
@Harwinian I’ve deleted your personal comment about another commenter for violating the comments policy. You’re warned.
Roger in Monash it’s highly likely the greens and labor will preference both the other and the teal over both broadbent and gee. While broadbent is and Ind he’s still of the right and former lib member. Who they preference out of broadbent and the lib candidate is anyone’s guess but yes they will preference gee over the Nat. The teals don’t indicate preferences so it’s not certain where their preferences will go so that will be up to the voters and it’s too early to be certain where those preferences will fall. I’m not too certain on those seats you named specifically Monash and all the qld seats bar Kennedy. Also I think both Bean and Franklin are a good chance for the inds. And Whitlam and possibly Moore as a maybe.