Why Israeli Polls Are So Hard to Trust

The short version

Israeli polls are a decent guide to the shape of the race. They are a poor guide to the exact result.

Three things make them shaky. The system has a 3.25% threshold, so a party at 3% is either four seats or zero. Samples are small, usually 500 to 600 people. And some groups are hard to reach: Haredi voters, Arab voters, older voters and new immigrants.

The polls also disagree with each other right now. In late September, Likud ranged from 18 seats in one poll to 32 in another. The coalition ranged from 49 to 63. That gap is bigger than any normal margin of error. It tells you the pollsters are making different choices, not just drawing different random samples.

So use polls for direction and for the size of the blocs. Do not use them to call a single seat count. And do not treat any one poll as the truth.

The track record

On average, the final polls miss each party by one to two seats. That sounds small. It is not small when a bloc wins or loses 61 by a single seat.

The averages hide the big misses. A few of them:

• 2015. Final polls averaged about 21 seats for Likud. It won 30. The race looked like a tie and was not.
• April 2019. Likud polled around 27 and won 35. Blue and White polled around 29 and won 35. New Right and Zehut polled at 5 to 6 seats each and won none.
• 2021. Blue and White polled around 4 and won 8.
• 2022. The Netanyahu bloc won 64. Polls had it at 60 to 62. Shas polled about 8 and won 11. Meretz (3.16%) and Balad (2.91%) fell under the threshold, and about 300,000 votes were wasted.

There is a pattern in the errors. One analysis by Project 61 found that pollsters have historically overstated Likud and Labor, and understated Arab and Haredi parties. That is a historical tendency, not a law. Pollsters adjust after each miss, and the 2020 to 2022 errors were smaller than the early ones.

Why they miss

The threshold cliff. A party at 3.25% gets about four seats. A party at 3.2% gets none, and its votes vanish. Polls measure a smooth number. The result is a cliff. Several parties sit on that cliff this year, including Blue and White, the Reservists and Amcha Yisrael.

Small samples. A typical Israeli poll asks 500 to 600 people. The margin of error on that is about 4.4 points on a 50% question. For a party near 5%, a miss of a point or so is normal. A point is about four seats.

Hard-to-reach groups. Haredi voters are often under-sampled, especially online. Arab voters answer less, and Arab turnout has surprised pollsters in 2019 and 2022. Older voters and immigrants from the former Soviet Union are also hard to sample well. Pollsters weight to fix this, but weighting only works if they know the true mix.

Online panels versus phone. Some pollsters use online panels, some use phone calls, many mix both. Each method over-represents some people and misses others.

Refusals. Some voters will not answer, and the ones who refuse are not random. In 2015, one pollster reported about 15% refusals.

Undecideds. A large share of voters say they are undecided. Pollsters have to assign them to parties. Each house has its own method, and that choice moves the numbers.

Turnout. A poll asks everyone. The election counts only those who vote. Modeling who will turn out is guesswork, and it matters most for Arab and Haredi parties, where turnout swings are large.

Seat conversion. Pollsters turn percentages into seats with their own rounding and thresholds. Two polls with the same vote share can print different seat counts.

Late moves. Many Israelis decide in the last days. Parties merge, drop out or sign pacts late. Tactical voting grows as the threshold gets closer, with voters leaving a weak party to save their bloc. Polls taken before those moves miss them.

Outlets, pollsters and house effects

In Israel, a TV channel or newspaper commissions a polling firm. The firm does the fieldwork. Here is who works with whom this cycle:

• Channel 12: Midgam (R&C)
• Channel 13: Midgam (Project)
• Kan and Israel Hayom: Kantar
• Maariv and Walla: Lazar
• i24: Direct Polls
• Channel 14: Filber (Next Data)
• Zman Yisrael: Yossi Tatika
• Channel 16: Maagar Mochot

Pollsters have house effects. That means a firm tends to lean the same way poll after poll, whatever the outlet wants. A 2022 Times of Israel analysis found Direct Polls gave the Netanyahu bloc 61 or more seats in 7 of 9 polls. The other major pollsters did it in 4 of 45. The same analysis called the gap more methodological than ideological. In June 2024, one analyst noted Channel 14 had the coalition at 59 while Channel 12 had it at 51.

What we can and cannot say about outlet bias. Only two outlets have a clearly sourced alignment. Channel 14 is widely described as pro-Netanyahu. Israel Hayom is described as right-leaning and pro-Netanyahu. For the rest, we found no solid source for a political camp, so we do not assign one. Some sources describe Haaretz and Ynet as left-leaning, but the evidence we found is thin and dated.

This matters for how you read the polls page. We group polls by outlet, not because the outlet rigs the poll, but because the outlet's choice of pollster and the pollster's method both show up in the numbers. A gap between groups is a fact. The reason for the gap is a judgment, and the data cannot settle it.

This cycle the gap is large. Filber (for Channel 14) and Direct Polls (for i24) put the Netanyahu bloc around 60 or more. Midgam and Lazar put it near 50. Both cannot be right.

How to read a poll this month

1. Look at blocs, not parties. Bloc totals are steadier than single-party numbers. Even so, the coalition range this month is 49 to 63. Treat it as a range.
2. Average across pollsters. One poll is one data point. Ten polls from seven firms tell you more. The polls page does this and gives every firm equal weight, so a firm that publishes often does not dominate.
3. Check who ran it. Compare a poll to the same firm's earlier polls. A move inside one firm is more meaningful than a gap between two firms.
4. Watch the threshold parties. Blue and White, the Reservists and Amcha Yisrael sit near the line. Their seats are all or nothing. A change of one seat in the total can hide a party falling out.
5. Ignore small moves. One or two seats is noise. Four or more across several firms is a trend.
6. Expect a blackout. Israeli law bars publishing new polls in the final days before the vote. Expect the last public polls around Friday 23 October. Check the exact cutoff, because private polls keep running and parties act on them.
7. Remember the surplus-vote pacts. They can shift a seat or two between partners after the count. No poll prices that in well.

The honest summary: this race has two blocs close to 61 and a few parties on the threshold line. A real chance exists that the polls are off by enough to flip the result.

Sources

This article draws on published poll tables (including the Wikipedia compilation of 2026 polls), a Project 61 analysis of historical poll accuracy, a 2022 Times of Israel analysis of pollster house effects, a June 2024 analyst comparison of Channel 14 and Channel 12 polls, and election result tables for 2015 to 2022. Exact seat misses are rounded. Where sources disagreed, we used the more conservative figure or left the claim out.