Talk:Opinion polling for the 2026 Portuguese presidential election

Second round polling

@Tuesp1985 About second round polling, I think the best division is between polling with confirmed candidates (which includes all polls that include two candidates that will actually run) and Hypothetical polling (with any matchup who has at least one candidate that will not run). If candidates like Paulo Portas or Santos Silva announce their candidacy, I'd add their polls to the confirmed candidates table, while if Ventura withraws his candidacy all polls that include his name should go to Hypothetical polling. And of course, if after the election there is actually a second round, the two candidates should have their own table and the other matchups should be added to the hidden section (maybe with two sections, one for actual candidates and one for non candidates).

For me, it makes the most sense that first round polling is divided by date, while Second round polling is divided by candidates. What do you think? I think we should settle this before the campaign starts and most polls start coming out. H3nrique Bregie (talk) 16:05, 20 June 2025 (UTC)

Hi there. You make a good point. Second round polls that don't have any of the confirmed candidate are, at the end of the day, hypothetical polling. In other pages, like the French 2027 election, there are separate second round poll tables. For me that's fine and confirmed candidates should have separate tables, and the rest of dropouts and/or people who rejected running go to Hypothetical polling. Cheers Tuesp1985 (talk) 22:25, 20 June 2025 (UTC)

Aximage Poll

I don’t think removing the Aximage/Folha Nacional poll is justified on the basis given. Wikipedia does not require ERC deposit for polls. Our inclusion criteria are verifiability in reliable sources, not whether a poll has already gone through every national legal formality. WP:V and WP:RS talk about sources, not ERC. ERC is a Portuguese regulator, not a requirement in enwiki policy. The poll is verifiable and attributed to a recognised pollster. Aximage is a long-standing, widely used Portuguese polling firm whose results are already included multiple times in this very article and in the main 2026 presidential election article (e.g. DN/Aximage barometers).

Furthermore, the presidential poll in question is published on Folha Nacional and mirrored on the Chega website, both explicitly describing it as an Aximage poll with a full set of figures (19.4% Ventura, 18.7% Gouveia e Melo, 15.9% Marques Mendes, 14.8% Seguro, 9.5% Cotrim). There is no source saying this particular poll is fabricated.

“Could be fake” is speculation and original research. Saying that a poll “could be fake or fabricated” because it hasn’t yet appeared in ERC is an editor’s own inference, not something supported by any cited source. That falls under WP:OR and also conflicts with WP:AGF. Unless a reliable third-party source questions the existence or authenticity of the poll, we can’t invent that allegation ourselves.

Past ERC issues don’t equal “fake polls”. There have been controversies about Aximage polls promoted by Chega/Folha Nacional without proper ERC deposit or technical data, and they were criticised in the press as “illegal” or non-compliant.

But even in those cases, the media and ERC treated them as real polls with legal/technical problems, not as imaginary numbers. That supports treating such polls with care, not pretending they simply don’t exist.

Partisan outlets is different than automatic exclusion. Folha Nacional and the Chega site are obviously partisan, but WP:RS and WP:PARTISAN don’t forbid using partisan sources altogether. They say we must: identify them as partisan where relevant; and avoid using them for contentious interpretation about rivals.

Here they are just publishing the numeric results of a poll by an external, established pollster. That is a straightforward, easily checkable factual claim (“Aximage found X, Y, Z%”), not an opinion about an opponent. Consistency with other articles. The same Aximage presidential poll (11–17 Nov, sample 816, with Ventura on 19.4%) is already included in the polling table at the main article on the 2026 Portuguese presidential election.

If there is a consensus to exclude it until the ERC record appears, then for consistency it should be removed from all relevant pages, not just this one. As long as it stays in the main election article, it makes little sense to single out this sub-article.

A more policy-consistent approach would be: keep the poll, clearly attribute it (“Aximage poll commissioned by Folha Nacional (Chega’s party newspaper)”), and if editors are worried about ERC, add a note once/if the ERC entry appears, or discuss in a short paragraph about legal controversies around party-commissioned polls. Completely excluding it on the basis of “ERC hasn’t published it yet, so maybe it’s fake” gives one regulator’s database more weight than Wikipedia’s own sourcing policies and relies on speculation that is not supported by any cited source Games30Top (talk) 18:26, 27 November 2025 (UTC)

Hi there Games30Top. The question isn't if the poll is fake or not, it's if the poll has "treated" the undecided, like Aximage normally does. For example, like I wrote in the view history section, the Aximage poll from October 2025 had GeM at 20.8%, Ventura at 18% and MM at 15.9%, raw figures these. But then, the polling company asked follow up questions to see where the undecided were leaning, and after this, the results became 24.4% for GeM and 19.2% for both Ventura and MM. We're not sure if the poll has this info also, and we'll only know when ERC publishes the full details, which will probably happen next week, as there's a 15-day embargo between the poll entry and it becoming public. Because Folha Nacional is Chega's newspaper, they could be publishing the "good numbers" only, and not the final ones, so, because of Aximage's past of treating the undecided numbers, we should proceed with caution. Tuesp1985 (talk) 12:19, 28 November 2025 (UTC)

Hi there @Tuesp1985, thanks for the explanation.

I fully agree that Aximage’s usual treatment of undecideds is relevant context and that the ERC deposit will clarify whether this specific poll has “raw” and “treated” figures, as in the October 2025 barometer. However, under Wikipedia e content policies, I don’t think that justifies removing the poll from the table at this stage.

Under WP:V and WP:RS, what matters is whether the numbers we are using are published in a reliable source. In this case, ECO (and Folha Nacional as commissioning outlet) publish a clear set of figures, with Aximage identified as the pollster. Those numbers are verifiable and not disputed by any secondary source.

Speculating that Folha Nacional “could be publishing only the good numbers” because it is Chega’s newspaper is, by definition, WP:OR / WP:SYNTH unless another reliable source explicitly questions the published figures or methodology. As editors, we shouldn’t pre-emptively discard data purely on the basis of who the client is, if the pollster itself (Aximage) is considered reputable and the results are properly reported.

The fact that Aximage may later release treated figures via ERC does not invalidate the currently published ones. It simply means the entry might need to be updated once those additional details become public. That’s normal for polling tables: we reflect the best reliably sourced figures available now and refine them when further information is released.

A reasonable compromise, in my view, is:

1. Keep the poll in the table with the currently published numbers and Aximage as pollster, properly cited to ECO/Folha Nacional.

2. Optionally add a footnote noting that Aximage sometimes publishes both raw and “treated” results after reallocating undecideds, and that full methodological details will be confirmed when ERC publishes the deposit.

3. When ERC releases the complete data, we can adjust the figures (e.g. to the treated numbers) and expand the note accordingly.

This way, we avoid original research and speculation, we respect WP:V / WP:RS, and we don’t create an artificial gap in the polling series while we wait for ERC, especially when no reliable source is currently contesting the published numbers. Games30Top (talk) 13:04, 3 December 2025 (UTC)

Semi-protected edit request on 27 November 2025

13 DG 13 (talk) 22:38, 27 November 2025 (UTC)

 Done. NotJamestack (✉️|📝) 23:54, 27 November 2025 (UTC)

The Supposely Aximage Pool that happened between the 11th adn the 17th of November is fabricated and not true... Ther eis no proof that that pool really exists and the link that you kepe puting in here that is shows as "reference" to said poll is from "FOlha Nacional" Which is a Far Right website known for posting fake news. Besides this post on this website there is no other refrence anywhere else to this poll ever existing. This pool is fake and fabricated so please remove it from the Wikipedia since it is spreading fake adn fabricated missinformation.

Semi-protected edit request on 2 December 2025

13 DG 13 (talk) 16:30, 2 December 2025 (UTC)

 Not done: it's not clear what changes you want to be made. Please mention the specific changes in a "change X to Y" format and provide a reliable source if appropriate. NotJamestack (✉️|📝) 16:56, 2 December 2025 (UTC)

I don't knnow if you are here representing this media that spreads Missinformation or if you just don't understand what Missinformation is... But I have showed enough times, evidence that the supposed Aximage pool from 10-17 of November 2025, IS NOT REAL. THere is no paper, reference or link taht leads to this pool, everything that exists about this "pool" is an article written by a Far-Right website that always spreads fake news and fake pools coming up with fake numbers and calling them a pool. Aximage has no record of this pool actually happening... And the only "proof" of it's existence is this article the website wrote pretending this poll is real, when in fact it is not, Aximage itself, the comapny has done no poll that have shown these resulst and you can't find this pool anywhere on Aximage's records... If you don't want to spread missinformation please remove this pool from the list...

Missing ECO poll and repeated reverts

I would like to note that the article is still missing the most recent poll published by ECO on 27 November 2025 (“Sondagem: Ventura e Gouveia e Melo em vantagem para passar à segunda volta na corrida a Belém”). Source: https://eco.sapo.pt/2025/11/27/sondagem-ventura-e-gouveia-e-melo-em-vantagem-para-passar-a-segunda-volta-na-corrida-a-belem/

In addition, the same group of editors repeatedly reverting this addition here have also been undoing equivalent updates in other language editions. This pattern raises concerns under WP:NPOV, WP:RS, and potentially WP:3RR.

If feasible, temporary page protection across all languages could help maintain article stability while the matter is discussed and consensus is established Games30Top (talk) 13:00, 3 December 2025 (UTC)

Hi there @Games30Top, that's the ISCTE poll, already on the table. The link you posted only talks of the data with the undecided. The link on the table has the SIC website that has the numbers after the undecided are distributed. A note could be added, like there is for the Intercampus poll on the next Portuguese election page, to show the date with undecided. Cheers Tuesp1985 (talk) 13:19, 3 December 2025 (UTC)

Hi there @Tuesp1985, thanks for the explanation.

But again, I still think the current approach risks introducing a bias in how this poll is presented. In this specific case we are choosing to use only the version after undecideds are redistributed (which clearly favours Gouveia e Melo over Ventura) while for other polls we have not systematically privileged “treated” results over the raw numbers.

From a methodological point of view:

1. The raw figures (with undecided included) are the direct outcome of the sample.

2. The version “after undecided are distributed” is already a modelled transformation of those numbers, which necessarily adds assumptions and increases uncertainty. It is not more precise; it is simply more processed.

If we only show the redistributed numbers for this poll, we are effectively giving editorial weight to one particular modelling choice that benefits one candidate, which goes against the spirit of WP:NPOV and consistency across polls.

A more neutral solution would be:

1. either to use the raw figures in the main table and mention the redistributed ones in a footnote (similar to what is already done for Intercampus on the other page),

2. or to show both, clearly labelled, so readers can see the original sample and the pollster’s reallocation.

What seems problematic to me is treating the redistributed version as the sole “official” result in the table when we haven’t done that systematically for other polls and when that transformation is, by nature, less directly tied to the observed data. Games30Top (talk) 13:30, 3 December 2025 (UTC)

@Games30Top, we could do the same as in the next Portuguese legislative election page: Add a simple rule of three introduction, and for those polls which don't distribute the undecided, we use the rule of three adding a note with the raw numbers. The polls who already distribute the undecided, are added without notes. This way it will be also easier to compare the election results with polling. What do you think? Tuesp1985 (talk) 14:54, 3 December 2025 (UTC)

I do get the goal of using a rule of three tbh: it makes the percentages easier to compare with the final election result by putting everything on a “decided voters only” basis.

My concern is that, from a statistical point of view, the raw figures are actually the most precise description of what was measured. The moment we rescale or redistribute undecideds, we are applying an extra layer of modelling. That is useful for some analyses, but it is less directly tied to the observed data.

The rule-of-three adjustment is neutral in the sense that it multiplies everyone by the same factor, but it still changes the information we show first to readers. It effectively assumes that what matters most is the distribution after we remove undecideds, whereas for many readers (and for anyone doing independent analysis), the original sample proportions (including undecided/abstention) are the key data point.

For that reason, I would actually support the reverse priority:

1. Show the raw figures in the main table, exactly as reported by the pollster;

2. Add the rule-of-three version in a note, clearly labelled as a derived calculation for comparison with the election result and between polls;

3. When a pollster already publishes a redistributed/“treated” version, we can also include that in a note, again clearly labelled as a modelled result.

That way the article remains fully comparable across polls, but we preserve the highest-precision information (the raw sample) as the primary data, instead of putting a transformed version front and centre. Games30Top (talk) 16:04, 3 December 2025 (UTC)

The issue with raw data is that they are, sometimes, misleading. I think that the simpler solution is the one used in the next Portuguese election page, but I will open a topic to try to resolve the matter. Tuesp1985 (talk) 16:56, 3 December 2025 (UTC)

Agreed that raw figures can mislead a casual reader if they are taken as “final” vote shares, but the same problem exists in the opposite direction: once we only show transformed data, we are no longer presenting what was actually observed, but a specific modelling choice that can also distort perception if not clearly framed.

I’ll keep this brief here and elaborate further in the new topic you’ve opened, so we can properly discuss how to balance readability with transparency about what is raw data and what is derived. Games30Top (talk) 17:09, 3 December 2025 (UTC)

Poll posting: Raw data or with the undecided removed?

@H3nrique Bregie, @Impru20, @NotJamestack, @Games30Top, @Mendes082013, @Danopt, @13 DG 13, @Number 57, @Quinnnnnby, @CoaxAndBotany, @Cmartim Hi there guys, to resolve once and for all the issue of how polls should be posted, and after a talk with @Games30Top, I decided to open a discussion topic on how to resolve the matter quickly, as election day is closer and closer and this kind of posting should have wide support. Games30Top defends that the raw data should have prevalence, while I support using a rule of three solution to match the polls with the future election results. Maintaining the numbers in the table as there are currently is also a possibility. So, discussion is now open. Tuesp1985 (talk) 17:06, 3 December 2025 (UTC)

First, thanks for opening the discussion and summarising our positions.

From my point of view, the key question is: what should Wikipedia show first? The numbers that are closest to what was actually measured in the poll, or numbers that already include an extra layer of modelling?

The raw figures (including indecisos/abstenção) are a direct description of what respondents actually answered. As soon as we apply either a rule of three or a pollster’s redistribution of undecideds, we are no longer showing the observation itself, but a model built on top of it. Both types of transform can be useful, but they are one step further away from the empirical data. For that reason, I think the most neutral and “Wikipedia-like” option is to give prevalence to the raw data in the main table and clearly mark any transformed numbers as derived.

I also agree that raw figures can mislead a casual reader if they are interpreted as “final” vote shares without noticing that undecideds are still in the denominator. But the same issue exists in the opposite direction: if we only show transformed data, a reader may not realise that those percentages are not what people actually said, but either (a) a pollster’s assumption about how undecideds will break, or (b) a rule-of-three rescaling that implicitly treats undecideds as if they simply did not exist. In both cases, the risk of misinterpretation is real if we do not make the nature of the numbers explicit. So it is not a simple “raw = misleading, transformed = clarifying” situation; each representation can mislead in a different way if presented without context.

The good news is that we do not have to choose only one and discard the other. We can preserve both accuracy and comparability. My suggestion would then be:

1. In the main table, always show the raw percentages, including undecided and abstention, exactly as reported by the pollster.

2. In a note or separate field, show the “decided-only” percentages that are useful for comparison with the election result: if the pollster already publishes redistributed numbers, we cite those; if not, we can optionally add a simple rule-of-three version, clearly marked as a derived calculation for comparability.

This way, readers who care about methodology and precision see the real sample first, and readers who mainly want to compare polls with the final result still have a consistent, “decididos apenas” number available. It also avoids forcing external analyses and models that use these tables (including my own) to try to reverse hidden transforms, which can create very distorted predictions.

So, in short: I defend giving prevalence to the raw data in the main table, and placing any transformed numbers (whether pollster-redistributed or via rule of three) in clearly labelled notes or secondary fields. I think this is the most "transparente" solution long term and it remains compatible with the goal of making polls comparable to each other and to the final election outcome. Games30Top (talk) 17:16, 3 December 2025 (UTC)

• Comment. Well! This is quite the topic, and I am not sure whether it can be resolved "quickly".

Short answer to the question: it depends, and should be determined in a case-by-case basis.

Long answer: opinion polls in some countries (I can think of Greece, Slovenia) typically use raw data directly as vote estimations (barring very few exceptions at most), so in those cases leaving or removing undecideds leaves essentially the same data (just using a simple rule of three to remove these. This would pose no problem as per WP:CALC). However, in some other countries, raw data does not equal to vote estimations, as pollsters typically use additional calculations for undecided distribution (i.e. Spain). For the first case, listing just one of these (either the raw data or the recalculation) would do the trick, as it's essentially the same data (though the use of footnotes is encouraged and, particularly, the fact that such a calculation is made should be made clear to readers in prose text). For the second case, use the official headline figures given by the polls and, if anything, add an additional table for raw data (without any recalculation). What is the best approach? I would say that would depend on what sources for that country's opinion polling typically do, so that is why I think this should be made in a case-by-case basis (I am not sure what approach is taken by Portuguese polls). I am, however, in favour of not using raw data as the main table. Why? Simple: as it has been pointed out, raw figures can mislead a casual reader if they are interpreted as “final” vote shares without noticing that undecideds are still in the denominator. Indeed, casual readers will read poll figures as being comparable to previous election results. But raw data cannot be directly compared with election results (except maybe if you consider also those abstaining, not just valid votes). And this is also a particular problem if some polls do not publish raw data, as this also prevents a direct comparation between different polls (which would mean they could not be shown in the same table). In those cases, it is best to add notes explaining what is being done. Impru20^talk 13:56, 4 December 2025 (UTC)

I appreciate both of your replies (this is exactly the level of nuance this deserves) but I still think there is a strong case for giving raw data primacy, and treating all transformations (whether rule-of-three or pollster redistribution) as clearly marked derivatives.

To @Impru20 point: I completely agree that cross-country practice matters. In systems where pollsters already publish a single “headline” vote estimate that is understood by everyone as the relevant number (for instance, Spain with standardised undecided redistribution), it is natural for us to mirror that. But even in those cases, there is an epistemic hierarchy in the data:

The raw figures are the closest we ever get to the underlying observation. They answer “what did people actually say at the time of the interview?”, including the proportion who are undecided or abstention-prone.

Any subsequent transformation (whether a simple rule-of-three or a more elaborate redistribution based on follow-up questions) is a model layered on top of that observation. It may be useful and visually cleaner but it is no longer a direct description of the sample or closer to what will happen on election day.

Because of that, the two main arguments advanced against using raw data as primary actually cut both ways.

First, “casual readers will misread raw data as final vote shares”. That is true if nothing else is explained. But the same casual reader will also tend to read redistributed or rule-of-three numbers as if they were somehow more “real”, without noticing that they embody assumptions (about turnout, about how undecideds will break, etc.). In one case, the risk is that readers ignore the undecided line; in the other, the risk is that readers ignore that they are looking at modelled, not observed, data. Neither representation is intrinsically immune to misinterpretation. The only cure here is clarity: we need to say explicitly which numbers are observations and which numbers are transforms.

Second, “raw figures cannot be directly compared with election results”. That is true if what we want to compare is “valid votes for each candidate” versus “percentage of all respondents, including non-voters”. But an opinion poll does not directly measure “valid votes”; it measures “stated intention at that moment, under that questionnaire, in that sample”. The undecided share is itself informative, especially as we approach election day. Suppressing it from the primary line of sight, and forcing it into a footnote, arguably distorts the story just as much as leaving only the raw data and pretending that they are final. For comparison with election results we can, and should, provide a standardised “decided-only” line; that does not require that we hide the original measurement behind it.

On @H3nrique Bregie suggestion to simply import the legislative-election method: I understand the attraction of symmetry, but I am not sure the legislative page provides a normative model we must follow. That solution was optimised for “one headline number per party, comparable to the final result”, at the cost of putting the transformed figures in the most prominent place and demoting the raw numbers to footnotes. It works, but it is a particular editorial choice: it privileges ease of visual comparison over strict fidelity to what was observed. For presidential polling (especially in a context where undecided levels and turnout are themselves politically and analytically relevant) there is a strong argument for inverting that hierarchy.

Nothing in WP:CALC says we must show our calculations first and the data they are based on second. It only says that straightforward arithmetic on reliable sources is allowed. Given that, my proposal tries to reconcile all concerns by:

1. We use the raw percentages in the main table, exactly as the pollster reports them, including undecided and abstention if given. That is the cleanest way to satisfy WP:V and avoid any doubt about what is being summarised.

2. We also provide, in a clearly labelled field or note, the “decided-only” figures needed for comparisons with previous elections and between pollsters. Where a pollster publishes its own redistributed headline, we cite that. Where it does not, we may add a rule-of-three rescaling, flagged as a Wikipedia-side normalisation for comparability.

This preserves what both of you are rightly trying to protect: for you, the comparability with election results and with other polls; for me, the transparency about what is observation and what is interpretation. It also avoids privileging those pollsters that already apply heavy undecided modelling over those that report their raw distributions more transparently.

In other words, I would invert the current customary hierarchy: instead of “headline/redistributed in the table, raw in the footnote”, use “raw in the table, headline/redistributed or rule-of-three in the note”. It is a small presentational change, but conceptually it puts the observed data back at the centre, and lets every reader (from the casual to the very technical) immediately see how much of what they are looking at is measurement and how much is modelling. Games30Top (talk) 18:49, 4 December 2025 (UTC)

the same casual reader will also tend to read redistributed or rule-of-three numbers as if they were somehow more “real” Rule-of-three numbers are the same as raw data, just excluding undecideds/abstentionists. Proportionally they are the same (that's what rule of three means). Unless a country's pollsters typically use different standardized methods to calculate headline figures, there is no reason to argue that these numbers are not "real"; this is what is done for legislative election articles in Portugal, as @H3nrique Bregie correctly pointed out below.

And election results are typically shown over votes, not over the whole electorate. So raw data is not useful for comparing with election results. It is that simple, really (and do not take this personally, but I do think you are depicting the situation as way more complex than it is). Impru20^talk 12:01, 5 December 2025 (UTC)

@Impru20" To claim the numbers are "proportionally the same" is a mathematical tautology that misses the statistical reality. When you exclude undecideds using a simple rule of three, you are not just cleaning the data; you are imputing data. You are treating every "Undecided" respondent as if they have already voted in exact proportion to the "Decided" respondents.

​This creates a false sense of certainty. A poll showing 40% A / 40% B / 20% Undecided describes a fundamentally different (and more volatile) race than a poll showing 50% A / 50% B / 0% Undecided, even if the "rule of three" numbers are identical. Presenting the former as the latter hides the volatility from the casual reader.

Furthermore, there is a strict epistemological difference between "raw data" and "recalculated data." Raw data represents the actual empirical evidence gathered by the pollster-measurement. Recalculated numbers represent a projection-prediction.

​While election results are indeed calculated over valid votes, polls are a snapshot of the electorate's current state of mind, which includes indecision. By stripping out the "Undecided" column, you are removing a critical data point: the degree of uncertainty in the electorate. Therefore, the recalculated numbers are arguably less "real" because they erase a measured component of reality (indecision) to fit a future format (election day results).

Even worst, the danger for the casual reader is precisely that they cannot distinguish between a candidate polling at 45% with high certainty and a candidate polling at 45% (recalculated from 30%) with high uncertainty.

If Candidate X has 30% support and 40% of the country is undecided, reporting Candidate X at "50%" (after removing undecideds) deceives the reader into thinking the candidate has secured a majority. It implies a "likely outcome" that the raw data does not actually support. Raw data prevents the reader from mistaking an open race for a settled one Games30Top (talk) 13:27, 6 December 2025 (UTC)+

First, try to summarize because your posts are turning into a WP:TLDR issue.

Second, I know what the numbers mean. When you say that When you exclude undecideds using a simple rule of three, you are not just cleaning the data; you are imputing data. You are treating every "Undecided" respondent as if they have already voted in exact proportion to the "Decided" respondents, this is true, but that is why I said that we should analyze it on a case-by-case basis. Are there precedents of Portuguese pollsters using different methods to treat "undecided" respondents and impute them to party vote results? If yes, please be specific and state which ones do and which ones do not so that we can move forward; if not, then I do not see how using a rule of three (ROT) is wrong, as we are essentially showing the same data as raw data, just without undecideds. In your replies above you have been specifically supportive of showing "decided-only" results; your divergence with other opinions shown in this discussion seem to come on which table should come first rather than on whether to use a ROT or not.

I think I have made my position fairly clear, and these lengthy posts will do little to sway me, frankly. Impru20^talk 13:34, 6 December 2025 (UTC)

From what I'm aware, Consulmark2 and Pitagórica use a simple rule of three, while UCP-CESOP and Aximage use follow up questions (the latter one depends on the poll itself, as in some cases they publish just raw data). Intercampus I believe also used rule of three in the past, but since 2021 it has stopped treating the undecided and publishes just raw numbers. ISCTE polls use a mix of voter characteristics, from gender to ideological position, to treat the undecided. Tuesp1985 (talk) 14:05, 6 December 2025 (UTC)

In that case, I would say that Consulmark2, Pitagórica and Intercampus would pose no issue with regards to using ROT (though that should be noted nonetheless, but that is already done for legislative elections). UCP-CESOP, ISCTE and Aximage would pose issues if they do publish raw data only, as from past precedent we would not be able to assume ROT would work for undecided treatment for these. Impru20^talk 14:10, 6 December 2025 (UTC)

UCP-CESOP and ISCTE rarely publish polls without the redistribution of undecided voters. Regarding Aximage, the polls that are published by the media, they always are without the undecided, but those which are commissioned by parties, mainly CHEGA, you a have a mix with some without the undecided and others just with the raw numbers. Tuesp1985 (talk) 15:48, 6 December 2025 (UTC)

@Impru20@Tuesp1985 I will keep this brief to avoid TL;DR.

We seem to agree that sophisticated pollsters (ISCTE, UCP) use complex modeling, not a simple Rule of Three (ROT). This creates a hard blocker for your proposal:

1. First, if we apply a simple ROT to ISCTE/UCP raw data, we are generating a number that contradicts the pollster's own scientific projection. We would be replacing their expert demographic imputation with our own calculator.

2. You argue ROT is "proportionally the same", but it mathematically amplifies noise.

Example: Candidate A has 18.01%, Candidate B has 18.00%. In raw data, this is clearly a statistical tie.

Scenario: If there is high uncertainty (e.g., 50% undecided), applying ROT doubles the gap and inflates the numbers. By shrinking the denominator, ROT mechanically scales up the margin of error, turning insignificant "noise" (0.01%) into what looks like a divergence. In highly volatile Portuguese elections, presenting a clear "36%" hides the reality that half the country hasn't chosen yet. Notice that even though ratios are preserved, absolute percentages carry psychological weight in political contexts (!)

We cannot mix "Official Imputations" and "Wikipedian ROT calculations" in the same column. The only standardized, neutral, and mathematically honest common denominator is raw data Games30Top (talk) 19:37, 6 December 2025 (UTC)

if we apply a simple ROT to ISCTE/UCP raw data [...] So, it has been said that UCP-CESOP and ISCTE rarely publish polls without the redistribution of undecided voters, so that pretty much solves the issue. For those, there is no raw data to redistribute.

You argue ROT is "proportionally the same", but it mathematically amplifies noise. It is still proportionally the same. That is what ROT means. 2 / 4 equals to 0.5. 4 / 8 equals to 0.5. 20 / 40 equals to 0.5. Different numbers, same ratios, it is not "noise". The issue is whether ROT can be used to "translate" some raw data into valid votes: it seems we can do that for some pollsters without any issue (because these have used/use ROT for that), not for others (but these others seem not to publish raw data anyway).

I do not think there is any complication, really. Impru20^talk 19:55, 6 December 2025 (UTC)

@Impru20 I think we’re talking past each other a bit, so let me be very clear about what I’m saying.

I’m not disputing the arithmetic of ROT. If A is twice B in the raw data, it stays twice B after ROT; on that we fully agree. My concern is how and where we use ROT in the tables.

1. When a poll has a very large share of undecideds/abstentions (e.g. ~50%), tiny differences like 18.01% vs 18.00% are basically noise. Applying ROT there multiplies everything by 1/(1–undecided), so the same noise becomes a larger gap in percentage points. But we don’t also enlarge the margin of error in the tables, so we end up visually amplifying noise in a context where half the sample hasn’t chosen yet.

2. For institutes like UCP or ISCTE, the published "headline" numbers are often not a bare ROT of raw figures; they include weighting, turnout modelling, etc. If Wikipedia takes whatever raw data exist and then applies a simple ROT, the result is neither the raw share of the full sample nor the pollster’s own modelled figure, but a third, unsourced number. That’s where the neutrality / no-original-research concern arises.

3. Because different pollsters use different methods (some model, some don’t, some change over time), a single column that mixes official imputations, our own ROT and occasional raw data is silently presenting incompatible estimators as if they were the same thing. That’s why I’m arguing the only truly neutral common denominator we can standardize on is the raw share of the full sample (with undecideds explicit). If the community prefers redistributed/modelled figures, then at minimum they should be clearly separated from numbers we ourselves derive by ROT, rather than blended into one column. Games30Top (talk) 15:02, 7 December 2025 (UTC)

At this point you are basically overreacting over an issue that is not as major as you are depicting it. Most pollsters do not show raw data at all, or show it alongside headline vote results. In those cases where pollsters only show raw data, you have some pollsters that are known for using ROT anyway, so these are also unproblematic. The actual issue only comes down to the few polls showing only raw data from pollsters that do not/have not used ROT as a method of undecided distribution. On this, you basically say we should prioritize a table with raw data (despite knowing—and having acknowledged it yourself earlier on—that casual readers will likely get confused with comparisons between raw data and actual election results). Excuse me if I do not share your concerns, nor your position. Thank you very much for your comprehension. Impru20^talk 17:41, 7 December 2025 (UTC)

@Impru20 To make it clear, I’d actually be fine with your case‑by‑case approach in general, just not for this specific presidential election, which is exactly the kind of race that is most vulnerable to the issues I’ve been raising (very high and shifting undecideds, mixed methodologies, and a lot of volatility).

So my concrete proposal for this election to final agree on a solution would be:

1. Main table always show raw data when available, including undecided/abstention
2. If only pollster "decided‑only" figures are published: show those, marked with '*' and a note like "* Pollster’s redistributed/decided‑only figures; undecideds not shown"
3. If we have to apply rule of three ourselves: show those, marked with '**' and a note like "** Decided‑only figures obtained by simple rule‑of‑three from published raw data".

That way we keep one table, protect casual readers from misreading a very volatile election as "settled", and still make clear where each number comes from.

Does that work for you? We could swap 2 with 3 if you want. Games30Top (talk) 22:13, 8 December 2025 (UTC)

But why are you so focused on me? I was asked for an opinion, I gave mine. I do not agree with you, and cannot see anyone else here agreeing with you so far. Thank you for your input and your comments, but I have little more to add in this discussion. Impru20^talk 22:19, 8 December 2025 (UTC)

I think we should use the same method as the legislative election pages: the polls that already have the redistribution of undecided voters should be added as they are, for the others we should use a rule of three and add a footnote about the raw data. I think that'd be the best and most simple option to show that data, and it would avoid more than one table for the same polls. H3nrique Bregie (talk) 16:33, 4 December 2025 (UTC)

I agree with this approach, which easily solves all aforementioned issues. Impru20^talk 12:01, 5 December 2025 (UTC)

What can't be done is what's happening now, because Aximage's polls have the undecided distributed and the others don't Ngfsmg (talk) 19:23, 18 December 2025 (UTC)

Opinion polling graphs

@Games30Top @Tuesp1985 @Mendes082013 @Petnog I think it's better for us to settle the graph situation in here than enter edit wars. I propose this to be the graph for this page, which aligns better with the graphs for the other elections:

• 
  My Proposal for the 2026 Portuguese presidential election
• 
  Games30Top Proposal for the 2026 Portuguese presidential election
• 
  Next Greek legislative election
• 
  Next South Korean legislative election

As you can see, both of the other versions (which are the same for almost all opinion polling articles) have the dots representing the polling results, with the lines being an interpretation of the evolution of polling numbers, being mostly an average. I don't really see how @Games30Top can say that my graph is misleading when it's in almost the same format as the other opinion polling graphs. Also, just like it shows in the graph for Greece, the lines don't extend indefinitely just like the Flourish made graph. That can mislead the reader into thinking that a candidate has had the same support since January, while they only started being included in polling since September.

I'd also like to state that I strongly disagree with the charts having projected results, as that breaks the purpose of Wikipedia. H3nrique Bregie (talk) 15:41, 20 December 2025 (UTC)

Like I wrote on the view history page, just copy the Greek or Spanish next election graphics, or for Presidential elections, the French 2022 one. There's no big discussion here. Cheers. Tuesp1985 (talk) 16:02, 20 December 2025 (UTC)

Those graphs are made using coding, with LOESS projections for the lines. I personally can't replicate them, so I've sent a request, but in case there's no avaliability to make the charts a provisional graph should be settled. H3nrique Bregie (talk) 16:07, 20 December 2025 (UTC)

Right. I actually also made a reply to @CoaxAndBotany, early this month, about making graphs for the next Portuguese election and the 2026 Presidential election, but had no response. Made a similar reply to @Quinnnnnby, nearly 1 year ago, about graphs for other Portuguese election, and, again, no response. In the case of no response, I would suggest your graphic, but, if possible, with the lines more smooth and with the date axis until election day. What do you think? Tuesp1985 (talk) 16:16, 20 December 2025 (UTC)

@H3nrique Bregie @Tuesp1985 I've just updated the graph to include the dots representing individual polling results, in line with the other opinion polling articles. With that addressed, I think the remaining concerns have been resolved. Can we close this discussion now? Games30Top (talk) 18:09, 20 December 2025 (UTC)

There’s still the question of the prolongued lines coming all the way from the start and the projections on the end. That’s my main concern with the graph rn. H3nrique Bregie (talk) 20:08, 20 December 2025 (UTC)

I agree with H3nrique. The prolonged lines since the start and the projected trend should be removed. But the date axis should remain until election day. The colors and candidate's name could be put on the right side. Tuesp1985 (talk) 20:20, 20 December 2025 (UTC)

@H3nrique Bregie@Tuesp1985@Petnog That concern has now been addressed. The updated graph removes the prolonged lines before candidates appear, removes any projected trends, keeps the date axis until election day, smooths the lines, includes the polling dots and places candidate names on the right.

In my opinion, it is also one of the cleanest and best-looking versions so far, while meeting all the criteria discussed. This brings it fully in line with established opinion polling graphics used elsewhere on Wikipedia, so I think this version should be considered final and the discussion can be closed. Games30Top (talk) 21:35, 20 December 2025 (UTC)

I like all of those except the portuguese one, because it's only one with those sharp edges that don't make a lot of sense and also make it look ugly. Petnog (talk) 21:34, 20 December 2025 (UTC)

I agree with this. All the referenced examples are consistent and visually coherent, except for the Portuguese one, which stands out for the wrong reasons. The sharp, angular edges don’t reflect how polling data evolves, make the graph harder to read and ultimately look unjustified.

More importantly, this isn’t just an aesthetic issue: the graph is also misleading and not methodologically sound. There are clear problems in how the trends are constructed and there are even critical mistakes in that exact graph (the Gouveia e Melo line is a good example if you look closely).

For these reasons, that version is neither visually nor analytically comparable to the other polling graphs cited. Games30Top (talk) 21:39, 20 December 2025 (UTC)

I saw the latest graph, it's much better now. Just one last thing: could you fix the colors for António Filipe and Jorge Pinto? It only makes sense for them to have the PCP (#FF0000) and Livre (#C2D216) colors, respectivelly. H3nrique Bregie (talk) 22:27, 20 December 2025 (UTC)

Thanks for pointing that out. Done! Games30Top (talk) 23:05, 20 December 2025 (UTC)

This version is ok. Thanks @Games30Top :) Cheers Tuesp1985 (talk) 00:53, 21 December 2025 (UTC)