How Wrong Were Israeli Polls? Six Elections, 2015 to 2022

The short version

Israeli polls have never nailed an election. But they have not been wrong in the same way every time, and that matters for how you read them now.

We took the final-week polls and the actual results for six elections. Three things stand out.

1. The big misses came early. In 2015 and April 2019, the biggest party finished 7 to 8 seats above the poll average.
2. The last three elections were closer. Most parties landed within one or two seats of the poll average.
3. Small misses still decide who governs. In 2022 the polls missed the winning bloc by about four seats, and that was the whole election.

If you do not follow politics: a poll is a snapshot of a few hundred people, not a forecast. If you do follow politics: the misses cluster on the same kinds of parties, and they are worth knowing before you read the next poll.

How we measured this

For each election we used the polls published in the last days before the legal publication ban (roughly the final week). We took the seat numbers each poll printed, averaged them, and compared the average with the real result.

• When a poll showed a party under the threshold, we counted it as zero seats.
• We left out one April 2019 poll whose seat figures added up to 127.
• "Miss" means the real result minus the poll average, rounded to whole seats.
• "Average miss" is the average size of the miss across every party that appeared in the polls.

The tables show the biggest misses, not every party. The full data is in the sources section.

The scoreboard

An average miss of one seat per party sounds fine. But 120 seats are split across a dozen parties, and the war for a majority is fought over a handful of them. A one-seat miss on each of four allied parties is a four-seat miss on the bloc.

2015: the race that looked like a tie

Every final poll had Likud far behind or level with Zionist Union, the Labor-led list. The polls had Likud at 20 to 23 seats and Zionist Union at 24 to 26. Likud won 30. Zionist Union won 24.

The polls did not catch the last days. Likud gained eight seats on the poll average, while the smaller right-wing lists lost ground. Yachad polled 4 to 5 seats in every poll and did not pass the threshold. Jewish Home finished three below its poll average.

Every one of the seven final polls had Likud at 23 or lower. None came within seven seats of the result.

April 2019: the tie that was not a tie

Final polls showed Likud and Blue and White level at 26 to 32 seats each. Both won 35.

This is the clearest case of the pattern. The two big lists finished six to seven seats above the poll average, and the smaller lists finished below it. Analysts often describe this as tactical voting: people who liked a small party backed the biggest one on their side. The numbers show the pattern, not the reason. Labor finished four below its poll average. New Right and Zehut were polled at about six seats each and ended with none.

September 2019: close, except for the edges

The polls had Blue and White at 31 to 33 and Likud at 31 to 36. The result was 33 and 32.

The two big parties were close. The smaller ones were not. Shas beat every final poll. The Joint List, Arab-led, also beat the poll average of 11 with 13. Otzma Yehudit was polled at 4 seats by several firms and won none.

2020: the most accurate of the six

This is the election where the polls did best. The average miss was 0.8 seats per party.

Even here, individual polls strayed. Final-week Likud numbers ran from 33 to 37.

2021: the small parties scramble the picture

Likud was right on the mark: polls at 29 to 32, result 30. The errors sat in the middle of the table.

Blue and White sat near the threshold in the polls, with most firms at 4 seats. It won 8. New Hope, Yamina and the Joint List all finished below their poll averages. No government formed until June.

2022: close on parties, wrong on the winner

In the last election, most party numbers were within one or two seats.

Look at the bloc. Likud, Shas, United Torah Judaism and Religious Zionism, the parties that formed the 2022 government, were polled at 58 to 62 seats, with an average of 60. They won 64.

Sixty is not a majority. Sixty-four is. That four-seat gap turned a coin-flip into a clear win.

Two things stand out. Shas beat every final poll by at least two seats. And three lists finished under the 3.25% threshold: Meretz (3.16%), Balad (2.90%) and Jewish Home (1.19%). Meretz was polled at 4 to 5 seats in every final poll. Balad and Jewish Home were polled below the line and stayed there. Votes for those lists counted for nothing, which gave every other party a larger share.

The patterns

Across the six elections:

• The biggest lists often finish above the poll average. Likud did in 2015, April 2019 and 2020. Blue and White did in April 2019 and 2021. The gap was largest in 2015 and April 2019.
• Small parties near the threshold get overrated. Yachad (2015), New Right and Zehut (April 2019), Otzma Yehudit (September 2019) and Meretz (2022) were all polled at seats and won none.
• Shas tends to beat the polls. It finished above its poll average in all five elections from April 2019 to 2022, by about three seats in April 2019 and 2022.
• The error is not always one-sided. In 2021 several parties finished below their polls. Do not assume the polls always lean the same way.

These are patterns in six elections, not rules. The polling firms change their methods after each miss.

Why the 2019 to 2021 polls were closer

The three most accurate elections, September 2019, 2020 and (to a lesser degree) 2021, came back to back, with the same parties and mostly the same voters. A plausible reason is that pollsters had fresh results to calibrate against. They could see how each group actually voted a few months earlier, and weight their samples to match. When four years pass, as it did before April 2019, that benchmark goes stale. The fixed-bloc politics of those years also helped: see The Netanyahu Divide: Five Elections, Two Blocs. This is a reasonable explanation, not something the data proves. 2022 was also accurate on most parties after a gap of about 19 months, and the pollsters still missed the bloc by four seats.

It matters for this year. The last election was in November 2022, nearly four years ago. Several of the biggest lists, Yashar and Together among them, did not exist then, so pollsters have little recent voting data for them.

Why this keeps happening

The article Why Israeli Polls Are So Hard to Trust covers the causes in detail. In short:

• The threshold. A party at 3.25% gets four seats. A party at 3.2% gets none. A poll can say 4 seats and the result can be zero.
• Tactical voting. Voters move to the party most likely to help their side, and they decide late.
• Small samples. A poll asks about 500 to 600 people.
• Hard-to-reach voters. Haredi and Arab voters, and older voters, are harder to sample well.
• Polls stop before voting does. The legal ban on publishing polls starts days before the vote, so late moves go unmeasured.

What this means for 27 October

1. Treat the bloc number as a range. In 2022 the bloc was off by about four seats. A four-seat gap on 61 is the whole election.
2. Treat threshold parties as coin flips. Several lists sit near the line this year. A poll can show 4 seats for a party that wins none.
3. Expect the large parties to gain at the expense of small ones. That happened in 2015 and April 2019. It may or may not happen again.
4. Do not rely on one poll. Even in the most accurate election, 2020, the final-week range for Likud was four seats wide.
5. The polls are a map, not a forecast. They tell you roughly where the race is. They do not tell you the result.

Sources

Poll figures and results come from the Wikipedia opinion-polling pages for each election (2015, April 2019, September 2019, 2020, 2021 and 2022), which compile the polls published by Israeli firms and media outlets, and from the official results shown on those pages. Poll averages and misses are our own calculations from those tables, rounded to whole seats. We used the polls published in the last week before each publication ban and excluded one April 2019 poll whose figures did not add up. Bloc totals for 2022 are calculated from the party figures.

• Opinion polling for the 2015 Israeli legislative election
• Opinion polling for the April 2019 Israeli legislative election
• Opinion polling for the September 2019 Israeli legislative election
• Opinion polling for the 2020 Israeli legislative election
• Opinion polling for the 2021 Israeli legislative election
• Opinion polling for the 2022 Israeli legislative election