Method
The ballot changed twice, the polls have not
Unity withdrew on 4 September and the Reservists merged with the New Economic Party on 6 September. No poll has been fielded since either, so all 18 polls the forecast reads still ask about both. One new poll moved the printed table to 53 for the Netanyahu bloc, 55 for the opposition and 12 for the Arab parties.
The front of this page reads Opposition bloc and Netanyahu bloc are close, at about 55 and 53 seats, and Arab parties with 12, from 18 polls. Two days ago it read 54 and 54 and 12. One poll arrived in between, and the ballot changed twice with no poll taken after either change.
The poll
Lazar for Maariv, fielded 3–4 September, sample 600: Yashar 23, Likud 20, Together 15, the Democrats 10, Yisrael Beiteinu 9, Otzma Yehudit 8, the Joint List 7, Shas 7, United Torah Judaism 7, Ra’am 5, Religious Zionism 5, Amcha Israel 4. Blocs 51, 57 and 12.
An outside archive of Maariv’s own polls carries the same twelve figures and the same three bloc totals, Amcha Israel counted on the Netanyahu side in both.
It is one poll and it moved one median, Likud’s, from 23 to 22. That is enough to change the whole printed table, for the same reason it was two days ago. The chamber here is a rule: each list takes its median, a median below the 4 seats the threshold allows becomes nothing, and any seats left over go to the largest lists one each until the chamber is full. With Likud at 22 the surviving medians add to 119, one seat is left over, and it goes to Yashar.
- Netanyahu bloc: −1, from 54 to 53.
- Opposition: +1, from 54 to 55.
- Arab parties: 12, unchanged.
Underneath the table the simulations barely moved. The Netanyahu bloc’s chance of 61 went from 4.6% to 4.2%, the opposition’s from 8.3% to 8.8%. Across 20,000 simulations the opposition finishes ahead of the Netanyahu bloc in 57%, behind in 38% and level in 5% — counting a tie as half, the larger bloc in 59% against 57% two days ago. The opposition and the Arab parties reach 61 together in 93% of simulations, from 92%. That is addition and nothing more — six lists that do not act as one, and nothing here says they would.
What changed after the last poll
Candidate lists are filed with the Central Elections Committee on 8 September. In the two days before this round, two of the lists in the table above stopped being what the polls asked about.
- 4 September. Unity, the list Gilad Erdan and Yuli Edelstein formed on 6 August, withdrew from the election.
- 6 September. Hili Tropper announced he is running with Yashar, dissolving his alliance with the Reservists. The Reservists then merged with the New Economic Party.
Wikipedia has already opened a fresh table for what follows: a new header with no Unity column at all, and the old Zionist Home column widened to “Reserv.-NEP”. It carries the three dated events above and no poll. The newest poll in either source is still the Lazar of 3–4 September, and every one of the 18 the forecast reads was fielded on or before that day. All 18 ask about Unity and Zionist Home as lists that are running.
So the forecast is answering a question about a ballot that no longer exists, and the useful thing is to say by how much.
Unity takes 4 seats or more — the floor the threshold allows — in 0.5% of 20,000 simulations, and Zionist Home in 3.7%. Their mean contributions are 0.40 seats to the Netanyahu bloc and 0.32 to the opposition. Subtract each from the bloc it sits in and the Netanyahu bloc’s mean falls from 52.9 to 52.5 and its chance of 61 from 4.2% to 3.6%; the opposition’s median falls from 55 to 54, its mean from 54.6 to 54.3, and its chance of 61 from 8.8% to 8.1%.
That subtraction is not a re-run. In the draws where those lists do take seats the seats are real, and a model that never saw them would have given them to somebody. It is the size of the question, not its answer.
It is also the failure this project has already named. The ballot has not closed 51 days before the election, and the two ways to be wrong are to forecast a list that wins nothing and to forecast nothing for a list that wins. Both are on the page today, and both resolve on 8 September rather than by anything done here.
Where that leaves the order
Yashar is the list most likely to finish largest, at 55% against 53% two days ago, with Likud at 45% against 47%. Head to head, Yashar is ahead in 51% of simulations, Likud in 41%, and they are level in 8%. The printed table has Yashar with 24 and Likud with 22, but that is the leftover seat doing the work: the medians are 23 and 22, and one seat of the gap is a rounding rule rather than a finding.
The two lists on the threshold both firmed slightly. Religious Zionism, running with Zehut, takes 4 seats or more in 60% of simulations against 58%, and nothing in 24% against 26%. Amcha Israel is unchanged at 55%, and takes nothing in 23% against 25%. Both still print 4. Amcha Israel is counted in the Netanyahu bloc because Ofer Winter declared on 27 August that his list would recommend Netanyahu and no one else; nothing about a bloc is read from the polls themselves.
An eighth poll dated wrong
The 2026 compilation’s dating problem, reported here on 4 September, has produced one more. It has filed a Lazar poll under 4 August 2026 that is the 3–4 September Lazar for Maariv with the wrong month on it. Three things say so, and none of them depends on the others.
It reports lists that did not exist on 4 August: Amcha Israel, which no source read before 6 August; the reunified Joint List, which re-formed on 19 August; and the Religious Zionism–Zehut technical bloc, agreed on 1 September. Its seats sum to 122, not 120. And Lazar polls for Maariv on a weekly cadence that an outside archive of Maariv’s polls lists as 14, 21 and 28 August and 4 September, with no wave on 4 August at all — while the copy matches the 4 September wave on eleven of twelve lists, differing only on Yashar, where it carries a stale 25 against that poll’s 23. The extra two seats are the whole of the 122.
It is out of every aggregate here, and — as with the seven of last week — by luck rather than by design: it labels the Arab list “Joint List”, which before 19 August marks a hypothetical ballot rather than a real one. That is the same coincidence, holding for the same reason and no longer than the compilation keeps that column’s name. The date is not corrected here. Choosing between two dates from inside one source is a guess, and deleting the copy deletes a poll; the reading stands where the source put it, out of the aggregate, and the finding is written down.