Israel's exit polls missed by three seats. The early count missed by ten

The final count is in. The Central Elections Committee published the results of Israel's election on 4 November: Netanyahu's bloc 53 seats, the bloc led by Gadi Eisenkot's Yashar 48, and the two Arab lists 19. On election night the three main television channels had put Netanyahu's bloc at 50. Using their record since 2003, we said before a single vote was counted that it would land between 46 and 54. It did, near the top of the range.

The score

We made five calls on election night, and four held. Netanyahu's bloc landed inside its range; by the measure fixed in advance, a result this far above the exit polls comes about one time in twelve. Neither bloc reached 61, as we gave 99.8% odds it would not. Likud stayed the largest list, at 87% odds: it won 27 seats to Yashar's 21. The revised projections published after midnight did better still. They put the bloc at 51 on average, inside their range of 48 to 55, and placed Likud's 27 seats exactly. The call that missed was the other bloc. We gave it 49 to 56, and it won 48.

The exit polls under-counted Netanyahu's bloc by three seats, almost all of them Likud's. The three channels gave it 24 or 25 seats at ten o'clock; it won 27. Channel 14, which supports the prime minister and was left out of our average, gave his bloc 60 and missed by seven, in the other direction.

The bigger miss was the early count. Votes are counted by polling station, and the first stations to report are not a sample of the country. With a fifth of the vote counted the bloc stood at 63 seats, a majority; with a third, at 59. We said at the time that if it finished near 59 our forecast had failed. It finished at 53.

Does 2026 revive the folklore that the exit polls under-count the right? It adds one more election to the under column: six of ten, now, with an average miss of 0.8 seats against the right. That is still a lean the data cannot tell from chance (p = 0.24). Three seats is the second-largest miss in the record, after 2015's 3.3. The exit polls stayed noisy, not crooked. What follows is the article as published on election night, with the final count added to the charts.

On election night

At ten o'clock on Tuesday night Israel's three main television channels did what they do at every election: they named a number for every list before a single ballot box had been opened. Kan 11, Channel 12 and Channel 13 each put Binyamin Netanyahu's bloc at 50 seats of the Knesset's 120, the bloc led by Gadi Eisenkot's Yashar at 52 or 53, and the two Arab lists at 17 or 18. Channel 14, which supports the prime minister, gave his bloc 60. By one in the morning all three main channels had revised their numbers with the full sample: Likud rose to 27 seats on each of them, Kan and Channel 13 moved to a 51-51 tie, and Channel 12 put Netanyahu's bloc ahead, 52 to 49.

The exit polls are the story for two days, until the count is done. And every Israeli election night comes with the same piece of folklore: the exit polls under-count the right. In 2015 they gave Likud 27 or 28 seats and it won 30. In 2022 they gave Netanyahu's bloc 61 or 62 and it won 64. If that is a rule, 50 means something more like 53.

We checked. We collected each channel's first projection at the close of polls for every election since 2003, seat by seat, and compared it with the Central Elections Committee's final count. That is nine elections and 25 full projections from the three main channels, plus Channel 14's one previous attempt, in 2022. The blocs are counted as the channels themselves counted them on each night. The method, and the rule for what counts as a lean, were fixed before any number was computed.

Five seats, give or take

A projection of 120 seats across a dozen lists is never exactly right. The useful measure is how many seats it places with the wrong list: half the sum of the misses across lists. Over 25 projections the three main channels averaged five. The best nights were 2020 and September 2019, when two or three seats moved. The worst was 2003, when the channels still used telephone polls as well as sample ballot boxes, and nine seats moved. 2021 and 2022 were poor too, at six and seven.

No channel is reliably better. Kan and Channel 12 both average five seats misplaced, Channel 13 just over five and a half. The largest list is called right almost every time: 26 projections in 29 named the right winner, and the three misses were all in 2019, when Likud and Blue and White finished within a seat of each other twice. But the largest list's size can be well off. In 2003 Channel 2 gave Likud 33 seats; it won 38. In 2015 all three gave it 27 or 28; it won 30. In 2021 Channel 13 gave it 33; it won 30.

The misses cluster at the threshold. A list polling near the 3.25% needed to enter the Knesset is the hardest thing for an exit poll to call, and when the call is wrong every one of its seats goes to the wrong place. In 2021 all three channels put Ra'am below the threshold; it won four seats, and with them decided who formed the government. In 2022 all three gave Meretz four or five seats; it won none. In April 2019 two channels put Ra'am-Balad out; it won four.

No lean

Now the folklore. The chart shows Netanyahu's bloc, as the channels counted it on each night, in the three channels' average projection and in the final count.

The folklore has its evidence: 2015 and 2022 are real, and in 2003 and April 2019 the bloc also did better than the polls said. But in 2020 and 2021 the polls gave the right two seats more than it won, and the stories of those nights were revisions downward, not up. Over nine elections the average miss is half a seat against the right, with a standard error of 0.65 seats: a lean that the data cannot tell from zero (p = 0.43). The polls were under in five elections and over in four. The usual explanation, that religious voters are under-sampled, fares no better: Shas and United Torah Judaism were under-projected by 0.17 seats on average, against 0.10 for other lists of their size.

What is real is the spread. The three channels' average has missed the bloc by 1.7 seats in a typical election, with a standard deviation of 1.9, and by 3.3 at worst, in 2015. A single channel can do worse: Channel 12 was five seats under in April 2019. The spread is the number to carry into tonight.

Tonight's numbers

Take the three channels' average and add the past misses, as a distribution rather than a correction, and Netanyahu's bloc lands between 46 and 54 seats, with 90% confidence. The chance that it reaches 61 is three in ten thousand. The other bloc, Yashar, Together, the Democrats and Yisrael Beiteinu, averages 52 seats on the three channels and lands between 49 and 56; its chance of 61 is two in a thousand. The chance that neither has a majority without the Arab lists is 99.8%. Likud is 2.7 seats ahead of Yashar in the channels' average; the past misses on the gap between the two largest lists have a spread of 2.2 seats, so Likud stays the largest party with 87% probability.

Before publishing a forecast we test it on the past. For each election we fitted the lean and the spread on the other eight and scored the left-out result. Correcting for the lean made the forecasts worse, not better, so the forecast above uses none. The 90% ranges built without it covered the final result in eight of the nine elections; 2015 fell outside.

The forecast is built on the first projections, as the method fixed in advance requires. The revised projections published after midnight move Netanyahu's bloc to 51 on average, and the same method applied to them gives 48 to 55, with the other bloc at 50 and a range of 47 to 54. The chance of a majority stays below one in a thousand for either, and the Likud lead over Yashar widens to 5.7 seats.

Channel 14's 60 for Netanyahu's bloc is not in the average. Its one previous exit poll, in 2022, gave the bloc 61 and it won 64, a miss of three seats and in the right direction. One election is no record, and tonight it is ten seats from the other three channels.

What the revisions add

Two or three hours after the first projection the channels publish a revised one, with the full sample and the late voters. We have both versions for nine channel-nights, in 2019, 2021 and 2022. The revision placed fewer seats with the wrong list in seven of them, by about one seat on average, and made things worse once, for Channel 13 in September 2019. The revised numbers are better; they are not a different picture. Tuesday's revisions fit the record: two or three seats for Likud and one or two for the bloc.

So the exit polls are noisy, not crooked. On Tuesday night they said Netanyahu's bloc had 50 seats, and the record says the count will put it somewhere from 46 to 54. Counting began overnight; by early Wednesday the Central Elections Committee had tallied about a third of the vote, and a partial count is not a sample of the whole country. The final results are expected within three days. We will add them to this article and score the forecast, whichever way it goes.