⇩ Download the poll database (CSV) - every headline poll in one file, one row per poll × party, rebuilt daily. Left out: where a pollster published more than one lineup on the same date - an alternative scenario (say, a hypothetical merger), or a variant row we could not parse cleanly - only the headline poll is included, which is the same rule every chart on this site uses. So the file matches the poll count in the masthead rather than exceeding it. Each figure is an individual poll by the named pollster - every row records the pollster, who commissioned it, the source, and our verification level. Compiled by Ron Levi (University of Toronto) from those published polls; free to reuse with attribution (CC BY 4.0). The Download button under the table below instead saves only the rows you’ve filtered.
two_source_confirmed Confirmed against two independent sources - the strongest level (the N12/Midgam figures, cross-checked with Wikipedia).
primary_verified Independently checked, seat-by-seat, against the pollster’s own published release or a primary news report.
primary_confirmed Read directly from the pollster’s release.
aggregator_only Taken from Wikipedia’s poll-aggregation table, not independently re-checked.
conflict Sources disagreed; the value judged most reliable was kept and flagged here.
primary_manual Hand-entered by us from the pollster’s own report because it is newer than Wikipedia’s table (which lags the Hebrew press by a few days). Shown with an “awaiting confirmation” tag; it clears to two_source_confirmed automatically once Wikipedia carries the poll.
unconfirmed Not yet checked.
Each poll is recorded exactly as its source published it, even where the seats don’t total 120 - for instance one Direct Polls survey (16 Jun) sums to 127 and one Lazar survey (11 Jun) to 117 (each is flagged in its notes column below). Nothing in this raw table is changed. But before any average is taken, every poll is rescaled to the 120-seat house, so a source’s over- or under-count cannot tilt the aggregates on the other tabs.
A single survey occasionally enters the source record more than once - the same Channel 13 poll was logged first under one polling firm (Maagar Mochot) and later another (Midgam) as the attribution was corrected, and some polls appear under two nearby dates (a date later revised, or a fieldwork-end date shown next to a publication date). After a full audit of the record in July 2026 we adopted a consistent method for these, so no single poll is weighted twice in the averages:
How we identify a duplicate. Two entries are the same survey when they carry an identical seat breakdown from the same commissioning outlet. A poll’s identity is keyed on the outlet that ordered it plus its numbers - not on the polling firm’s name or the exact date, since both can differ across sources for one real survey.
Which copy we keep. The best-sourced one - the copy with a real sample size and an independent, seat-by-seat check, over a copy taken only from an aggregator. The duplicate is dropped.
When we do not merge. Identical numbers that sit far apart in time (weeks or more) are left as separate polls, because a race with little change can genuinely produce two matching results; we flag those for a source check rather than assume. We merge only where the evidence is unambiguous - and say so here when we don’t.
An independent tracker of opinion polls for Israel’s 2026 Knesset election. It gathers the seat projections published by the major Israeli pollsters into one database, and gives you tools to explore them - trends, blocs, a coalition builder, and an “if the vote were held today” simulation.
Israeli pollsters publish their results (how many of the 120 seats each party would win) but not their raw surveys, and no official body collects them - so this site aggregates published results (a “poll of polls”) rather than re-analyzing raw survey data - a true meta-analysis would need the underlying surveys, which pollsters do not release. The backbone is the Wikipedia poll-aggregation page (this build reads one specific revision), compiled together with pollsters’ own releases; each figure is cross-checked against the pollster’s own release or the Hebrew press where possible, and every number is labelled by how well it was confirmed. Some of them carry a second source beside that page. Most of that checking happens in Hebrew - 70 of 117 sources behind these figures are Israeli outlets (Zman Yisrael, N12, Kan, Channel 13, Maariv, Arutz Sheva). That is not incidental: the Israeli press carries a poll days before the aggregators do, and it is where disagreements get settled. Three this month were resolved by reading the pollster’s own Hebrew report - Kan’s seat graphic gave Shas 7 and United Torah Judaism 8 where the English table had them swapped; Channel 14’s write-up settled two more transposed pairs; and Ynet established that Ra’am’s primaries precede its reserved slots, not the reverse. Coverage is ~100 polls from 8 pollsters, checked for new polls automatically once a day. If a poll later disappears from Wikipedia, we keep our archived copy of it (flagged “archived”), so nothing we’ve captured is ever lost.
The Nowcast is a snapshot of where things stand now if the polls are right - not a forecast of the October result. We deliberately do not: predict the eventual outcome (that needs a history of how Israeli polls have missed past results), grade or weight pollsters by track record (same reason), or break results down by demographics (that needs the raw surveys, which we don’t have). We can show how the published numbers shift, how much they disagree, and what coalitions they imply. When we simulate, we let a whole bloc’s polls miss together, because real polling errors are correlated rather than independent, so the coalition odds carry a realistic spread rather than a falsely tight one; the full method (including this “correlated-error” term) is in the method note on the Nowcast tab.
Two. Likud held a members’ primary on 17 August 2026, and The Democrats held one on 20 July 2026 - the first fully online primary in Israeli politics. Every other party sets its list order another way: the leader decides in Yashar, Together, Yisrael Beiteinu and Otzma Yehudit; a party body votes in Religious Zionism, Hadash and Balad; and a rabbinical council approves the list in Shas, United Torah Judaism and Ra’am. Final slates were filed with the Central Elections Committee by 8 September 2026, when 38 lists were submitted. Each party’s slate and method is on the Candidate lists tab.
The election is on 27 October 2026, the end of the Knesset’s four-year term. The Knesset passed a dissolution bill and dissolved on 17 July 2026.
61 of the 120 Knesset seats. No single party wins a majority, so governing means assembling a coalition of parties that together hold at least 61.
The tracker pools every major poll into an “if the vote were held today” nowcast. The pro- and anti-Netanyahu blocs are close, and in most polling neither reaches 61 on its own - see the Nowcast and Coalition tracker tabs for the current picture.
Rarely in the polling: most surveys put the anti-Netanyahu bloc short of 61 without the two Arab-majority parties (Hadash-Ta’al and Ra’am). Use the Coalition builder to test any combination against 61.
All major Israeli pollsters (Midgam, Lazar, Kantar, Maagar Mochot, Direct Polls, Filber, Yossi Tatika and others), taken from each firm’s own release and cross-checked against Wikipedia.
An estimate of the current state from the latest polls: what the seats would be if the vote were held today, rather than a forecast of the eventual October result.
The tracker doesn’t call a winner, but the PM suitability tab tracks head-to-head “who is more suited to be prime minister” polling among the main contenders (Netanyahu, Eisenkot, Bennett), and the PM tab also carries a live prediction market on who forms the next government.
Compiled by Ron Levi · מאת רון לוי, אוניברסיטת טורונטו · University of Toronto (ron.levi@utoronto.ca), built, analyzed, and annotated with Claude Code (Anthropic). It is a compilation of publicly available polls - not original polling, and not affiliated with any pollster or news outlet. To avoid double-counting, it lists only individual published polls (each pollster’s own survey) and never other aggregators’ or models’ averages. Bloc alignments (pro- vs anti-Netanyahu) reflect the current moment, not a fixed feature of the parties. It is a work in progress, and corrections, suggestions and advice are very welcome - please write.
Israel’s electoral threshold is 3.25% of the vote, worth roughly four of the 120 seats. A party below it wins nothing at all. This matters for an average of polls because the threshold has already been applied before we see a number: pollsters publish seat projections, not vote shares, so a party a pollster leaves out (or shows at zero) is one that pollster has already judged to fall short. We are never applying the threshold ourselves; we are deciding what to do when pollsters disagree about whether a party clears it.
A party is counted in the two-week average if either test passes: it is seated in at least half the window’s polls, or in at least half of the six most recent. A counted party’s average uses only the polls that actually seat it, so a firm leaving it out lowers its chance of being counted rather than dragging its seat number toward zero.
The second test was added on 30 July 2026, and the reason shows what is at stake. Yoaz Hendel’s list (now Zionist Home) crossed the threshold partway through the window: it appeared in 7 of 17 polls, under half, so it was dropped from every average, yet three of the last four pollsters seated it at four or five. The first test alone counts how many polls seat a party and ignores which, so a rising party stays invisible for up to two weeks, exactly when it is newsworthy. The older test is kept alongside because the second alone would admit a single fresh outlier. Blue & White fails both, seated in 1 of 17 polls and by no pollster since 16 July. That is the point: one stale reading must not put a party on the board.
Two consequences worth knowing. Adding a party changes every other party’s seats, because the projection is rescaled to total exactly 120 - counting Hendel’s four seats shifted the blocs from 54/55/11 to 53/58/9, without changing the finding that neither bloc reaches 61. And a party sitting near the line can enter or leave the board between refreshes; when that happens the shift is arithmetic, not a surge in support.
The same rule now governs the nowcast simulation, which had a related flaw: it decided membership after evening out house effects, so max(0, 0 − a negative lean) could invent a seat share for a party a pollster had placed below the threshold. That was enough to put Balad - which no pollster seats at all - and Blue & White into the simulation. Membership is now judged on the raw published seats in every panel: a house effect may shift a party’s seat number, it may not decide whether the party is running.
Because this site pools published results, its central discipline is that each real survey is counted once. That is harder than it sounds: the same poll can enter the record under different labels, and counting it twice tilts every average without showing up anywhere. A full audit of the record in July 2026 turned up three ways it happens:
The rule we settled on: two entries are the same survey when they carry an identical seat breakdown from the same commissioning outlet - a poll’s identity is the outlet that ordered it plus its numbers, not the polling firm’s name or the exact date, since both of those drift across sources. We keep the better-sourced copy (a sample size, an independent check) and drop the duplicate; where identical numbers sit far apart in time and the case is ambiguous, we keep both and flag it rather than guess. The point is not housekeeping: before this pass, a strongly anti-Netanyahu poll was being counted twice, which made the opposition’s lead look steadier than the once-counted data supports.
A nowcast estimates where things stand now from current data, as opposed to a forecast of the eventual result. We treat the recent polls as a snapshot of opinion and turn their disagreement into probabilities:
The percentages are the share of the 10,000 runs in which each outcome occurred. Why 10,000? It is enough that the figures are stable to within about half a percentage point (the Monte-Carlo error is roughly √(p(1−p)/N) ≈ 0.3–0.5 pp here); more runs would not shift the headline, because the model’s own assumptions dominate the uncertainty. A fixed random seed keeps the numbers reproducible.
Why allied parties rise together - the “correlated-error” term. Polls are never exact, so in each simulated election we nudge every party off its polled seats a little. Real misses, though, come in groups: if ultra-Orthodox turnout runs higher than the pollsters expected, both Shas and United Torah Judaism land above their polls together - they draw on the same voters. Had we nudged each party on its own, Shas’s gain would often be cancelled by a chance dip in United Torah Judaism, so the pro-Netanyahu bloc’s total would barely shift between runs - making it look more locked-in than it really is. So we let a bloc’s parties drift partly together - a “within-bloc correlation,” set to 0.35. That 0.35 is our judgment, not a figure measured from past Israeli elections.
This is a deliberately simple version of the Monte-Carlo approach professional election forecasters use (e.g. FiveThirtyEight, who popularised the now-cast-vs-forecast distinction) - not a replica of it. It leaves out what makes those models elaborate: weighting pollsters by track record, fitting the full party-by-party error-correlation matrix from past elections, calibrating the spread to how far off polls have historically been this far from election day, and blending in non-poll “fundamentals.” (Evening out each pollster’s house effect is the toggle at the top of this tab; it is on by default, so the headline already evens out each firm’s house effect - and even then only relative to the other firms.)
What it assumes, and can’t do. The uncertainty reflects disagreement between published polls - we don’t hold the pollsters’ raw data, so we can’t model each survey’s own sampling and weighting. It is pure seat arithmetic: it counts who could reach 61, not who would actually agree to govern together (hence the separate columns for with and without the Arab-majority parties). And it is a nowcast, not an October forecast - months out, with new lists still forming, the real result can differ. Even with the shared bloc shock above, the spread is built from how far this cycle’s polls sit apart - usually narrower than how far final polls miss on the day - so the model can still run a little over-confident, and the figures above are best read as broad signals, not exact odds.
Parties can pair up to share leftover votes, and the count then awards remainder seats to the pair rather than to each alone - usually a seat or two. The numbers here are pollsters’ own seat figures, averaged and rounded to 120; that rounding is arithmetic, not the election’s allocation. A bloc shown a seat short could reach 61 on the night without any poll having been wrong. Each party may pair with one other, and a partner who finishes below the threshold contributes nothing. Four pairs have signed, none of them across the two blocs: Likud with Religious Zionism-Zehut on 8 September (Israel Hayom), Yashar with The Democrats and Yisrael Beiteinu with Together on 10 September (Times of Israel), and the Joint List with Ra’am on 12 September (Times of Israel), the deal Ra’am’s Mansour Abbas had said he wanted. That leaves Shas, United Torah Judaism, Otzma Yehudit and Amcha Israel unpaired, all four of them in the pro-Netanyahu bloc. The deadline is 16 October, the day after the deadline for changing or merging a filed list, so the rest can still sign. If any pair spans the two blocs, it will be said beside the bloc numbers rather than only in this note.
Two rules decide what never reaches an average. The first is that this site lists individual published polls only: other aggregators’ and models’ averages are excluded on sight, because pooling an average of polls with the polls it already contains counts the same surveys twice. The second is that a poll has to be sourced well enough to check. Where it is not, we keep the figures and say so here rather than dropping them without a word, so the record shows its own edges instead of implying it is complete.
Neither of these is a judgment that the numbers are wrong. Each is recorded in data/held_off.csv with its reason, so any of them can join the counted data the moment the missing source appears.
The seats-over-time line pools every pollster’s published polls into one curve with a Gaussian kernel (nearby polls count more), and a model-based (Kalman) option smooths the same series the way election forecasters do. The shaded 95% band is computed from how much the polls near each date disagree, so it widens where coverage is thin. A house effect is each firm’s persistent gap from the all-pollster consensus; “evened out” subtracts that gap before pooling, and the band is deliberately kept from the raw spread so de-biasing never fakes precision. We measure the gap, not its cause. The live figures - each firm’s current house effect per party - appear beside the chart on The parties.
Co-movement & factor structure (Dive into the data) - the matrix shows the correlation between two parties’ seat series across polls; the factor read applies principal-component analysis to the standardized (and compositionally transformed) series, and reports a range for the share of shared movement because that share is sensitive to method choices. Challenger parties - a fuzzy-set Qualitative Comparative Analysis (Ragin): each party’s membership in “large” is calibrated from its recent seats, and the panel reports which combinations of traits (newly formed, centrist) sit together with size, with consistency and coverage given in its footnote. Bargaining power (Coalition builder) - the Banzhaf index: across every possible coalition - with no regard to political compatibility - how often a party’s seats turn a losing combination into a winning one. Each of these keeps its live figures and fuller notes beside the analysis itself.
polls (one row per poll × party). Easiest way to start: click one of the grey example buttons below to run it, then edit it. (For yes/no fields, write scenario = false.)polls - one row per poll × party. Booleans: scenario = false.On the two-week polling average through 10 Sep 2026, neither side reaches the 61 seats a government needs. Netanyahu's coalition sits around 56, the anti-Netanyahu parties around 52, and the two Arab-majority lists (Joint List (Hadash-Ta'al-Balad) and Ra'am) near 12, currently aligned with neither camp. That leaves the opposition short of a majority unless those lists join it - the pattern in most polls since the spring.