What we don't know yet: limitations of the Italian forecast
Abstract. Every known weakness of the Italian forecast in one place: the assumed line-up, a correction that passes narrowly and rests mostly on the M5S, four elections of early final polls, independent errors, untested proportional swing, Futuro Nazionale's even spread, seats held at 2022, pollster ratings that cannot separate pollsters, second-hand polls, and the Senate. For each, how much it matters and what would change our mind.
Summary
The Italian forecast makes more assumptions than any other on the site. It corrects the polls, on the strength of a test that passes narrowly on four elections. It assumes a line-up of coalitions that the parties have not declared. It moves every collegio by proportional swing from 2022, spreads a new party evenly, and shares each coalition's collegi among its lists by a rule of ours. Its polls mostly reach us through Wikipedia, and its pollster ratings cannot tell Italian pollsters apart. Its seat calculation, by contrast, reproduces 2022 exactly.
This paper sets out each weakness, how much we think it matters, and what evidence would change our mind. The British counterpart is What we don't know yet: limitations and open questions in our UK forecast; the method is in Reading Italian polls: the correction and its test.
What has changed since this paper was written
Updated 9 November 2026. The rest of the paper is as first published, except where a sentence below says it was corrected.
The new electoral law is in force (8 October). The Chamber gave the new law its final approval on 8 October 2026, by 227 votes to 164, and the Italian forecast has used it everywhere since. As the forecast applies it to national shares:
Seats are shared in proportion to the vote, by Hare quota and largest remainder, among coalitions on 10% or more and lone lists on 3% or more. Inside a coalition, lists on 3% or more, plus the best list below 3%, share its seats; every list's votes count towards the coalition's total.
The bonus. The coalition or list that comes first with at least 42% of the valid votes gets 70 extra deputies and 35 extra senators, up to caps of 220 deputies and 113 senators.
The Senate is worked out the same way on national shares (the law allocates its seats by region). Seats abroad, Valle d'Aosta and Trentino-Alto Adige are shared in proportion to the national vote, an approximation.
The collegi no longer elect anyone. The maps and collegio pages show who leads each former collegio.
Where this paper describes the Rosatellum, it describes the law under which the 2022 Parliament was elected: it remains the basis of the check against 2022 and of the by-elections in this Parliament. /it/new-law shows the seats under both laws side by side.
- Only elections from 2000 shape the forecast (28 September). Earlier elections stay in the long view and on the history pages but no longer feed any test, correction or error size. In Italy the tests were already on elections since 2001, so nothing changed.
- Pollster grades rescaled (13 October) (Rating Italian pollsters: accuracy, the lead and openness).
The line-up
The largest uncertainty about the next Camera is not in the polls. Under the Rosatellum a coalition's candidate gets the votes of all its lists, so the same national votes give very different Camere under different line-ups:
| Line-up | Centre-right | Centre-left | Others |
|---|---|---|---|
| Centre-left 'campo largo' with the M5S (our projection) | 143 | 224 | 33 |
| Campo largo; Futuro Nazionale with the centre-right | 224 | 165 | 11 |
| The 2022 line-up | 223 | 96 | 81 |
Our default, the "campo largo", follows the centre-left's practice in regional elections since 2024; the other two rows show Futuro Nazionale inside the centre-right and the line-up of 2022 (Who stands with whom: the coalition line-up assumption). The simulation, the replays and the polling-day forecast all hold the default fixed, so none of their chances includes the chance that the default is wrong. We do not put a probability on the parties' decisions.
What would change our mind. The parties' declarations, about six weeks before polling day. Until then, a reader who thinks another line-up likely should read that row.
A correction that passes narrowly
The correction takes off half each list's average past miss because, with each election predicted from the others, it was closer on average:
| Election | Polls' average miss | Corrected, leave one out | Verdict |
|---|---|---|---|
| 2001 | 1.19 | 1.42 | further out |
| 2006 | 1.19 | 0.83 | closer |
| 2008 | 1.52 | 2.03 | further out |
| 2013 | 3.14 | 2.70 | closer |
| 2018 | 2.95 | 2.42 | closer |
| 2022 | 1.55 | 1.52 | closer |
| Average | 1.92 | 1.82 |
It passes by less than a tenth of a point per list, on four elections, and it wins on 2013 and 2018 while losing on 2008 and 2022. Most of its effect is one list: the M5S, understated at all three elections it contested, gains +2.9 points today. The M5S's miss has shrunk from 10.4 points to 4.8 and then 2.0. If it shrank again, or changed sign, the correction would make the next forecast worse.
Under the default line-up the correction moves points from the PD to the M5S inside one coalition, so its effect on the coalition totals is small (If we're wrong: replaying past Italian polling misses on today's polls). Under a line-up with the M5S alone it would matter more.
What would change our mind. The next election. The rule is checked every time the forecast is built, and if the correction stopped beating the polls on average it would switch off without any change to the method.
Four elections, and a ban
Everything the forecast knows about Italian polling error comes from 82 final polls at four elections. Before 2008 the final polls we could find give coalition totals only. The final polls are also early: the law forbids publishing polls in the last fifteen days, so the typical final poll finished fieldwork eighteen days before the vote. The record therefore mixes polling error with late movement, and the two cannot be separated. For the forecast that is less of a problem than it sounds, because the polls a reader sees before the next election will be subject to the same ban; but it means the record says little about how well Italian pollsters measure opinion on the day they poll.
What would change our mind. More elections; and party-level final polls from before 2008, if any can be sourced.
Independent errors, and the M5S's spread
The simulation draws each list's error independently, with a spread equal to its typical past miss. The M5S's is more than six points. At every election both contested, the PD and the M5S were missed in opposite directions, so their combined vote was missed by much less than either. Drawn independently, their sum is about twice as uncertain as the record suggests. Under the default line-up both are in the centre-left, so the centre-left's range is wider than the evidence supports (Simulating Italian polling error: where the chances come from). The M5S's spread is also not reduced by the correction, which takes off part of the miss the spread is built from.
What would change our mind. Drawing linked errors from past elections' patterns, as the British simulation does, once there are enough elections to make the patterns more than anecdotes.
Proportional swing and the collegi
Each list's 2022 vote in each collegio is scaled by its national change. Italian swings have differed between North and South, and proportional swing cannot follow a party that grows where it was weak: applied to 2018, it would have multiplied FdI's vote by six in every collegio, however small or large that vote had been. The projection has not been tested; a backtest from the 2018 result rebuilt on the 2022 collegi is possible and has not been run.
Three further simplifications sit on top:
- Futuro Nazionale has no 2022 vote and is given the same share of every collegio. A new party's vote is never that flat. It matters only when Futuro Nazionale is in a coalition.
- Candidates are not modelled. In 2022 a local list, Sud chiama Nord, won a collegio in Sicily against both coalitions.
- A list's collegi are our split of its coalition's wins, by the lists' national shares. The real split is negotiated between the parties.
What would change our mind. The backtest from 2018 to 2022; a model with a regional layer would have to beat proportional swing on it.
The seats abroad, the Valle d'Aosta and the SVP
The eight seats abroad are held at their 2022 result: polls do not cover Italians abroad. The Valle d'Aosta, which has no lists, moves with the coalitions its 2022 candidates represented. The SVP is not polled and its vote is held at 2022. Each is a handful of seats, and in a close Camera a handful can decide a majority.
The pollster ratings
Every Italian pollster we reviewed is graded E. The British scale gives no points for a miss as large as a poor Italian final poll's, and the register where pollsters file their full documents could not be read, so the openness we could credit is partial (Rating Italian pollsters: accuracy, the lead and openness). The ratings order Italian pollsters, a little, but their grades say almost nothing, and they cannot be compared with other countries'.
What would change our mind. Reading the register, which would let us check every pollster's method, sample and dates; and an Italian scale with its own end points.
Second-hand polls
Most polls since 2022 come from Wikipedia's tables (grade C). Where we could read the pollster's or the outlet's page, the figures were checked and upgraded; the rest were not. The official register, which would give every poll's full document, reset every connection.
The date
The election must be held by 22 December 2027, and can be called earlier. The polling-day forecast widens the ranges by a drift fitted on three elections at distances up to a year (From today to polling day in Italy); the next election may be further off than that, where the drift is an extrapolation.
The Senate and the law
The site forecasts the Camera only. A government needs the confidence of the Senate as well, elected the same day by the same system with different constituencies, and a majority in one chamber need not mean a majority in the other. The forecast applied the Rosatellum until the new electoral law was approved on 8 October 2026; it now applies the new law, as the note at the top of this paper explains.
Summary
| Limitation | How much it matters | What would address it |
|---|---|---|
| The line-up | More than any polling error | The parties' declarations |
| A narrow correction | Most of its effect is the M5S, whose miss is shrinking | The next election; the test reruns itself |
| Four elections, early final polls | Error and late movement mixed | More elections |
| Independent errors | The centre-left's range too wide under the default line-up | Linked errors from past patterns |
| Proportional swing untested | Collegio projections may be biased | Backtest from 2018 |
| Futuro Nazionale spread evenly | Matters when it joins a coalition | Regional polls or results |
| Seats abroad, SVP, Valle d'Aosta held | A handful of seats fixed | Polls abroad, which do not exist |
| Pollster ratings all E | The grades carry little information | The register; an Italian scale |
| Second-hand polls | Transcription errors would pass into the average | The register |
| The Senate not forecast | A Camera majority may not be a government | A Senate model |
Table: Known limitations of the Italian forecast.
The forecast's claim is narrow: Italian polls, corrected by a rule that has earned its place so far, carried through a seat calculation that reproduces the last election exactly, under a line-up we state. The next Camera election will be the first real test, and we will publish how it went.
References
DPR 30 marzo 1957, n. 361, as amended by legge 3 novembre 2017, n. 165.
Jennings, W. and Wlezien, C. (2018). Election polling errors across time and space. Nature Human Behaviour, 2(4), 276–283.
Legge 22 febbraio 2000, n. 28, article 8.
How to cite
Polling Forecast (2026). What we don't know yet: limitations of the Italian forecast. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/italy-limitations. Live figures change as new polls arrive; the date above is when the text was last revised.



