Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

Tuesday’s real winners: Kulanu and the Joint Arab List

March 18th, 2015, 2:09am by Sam Wang

Kahlon's support is critical for Likud's coalition. Credit: Tsahir Abayov, AP

Moshe Kahlon of Kulanu, kingmaker

The votes are counted. Likud surged in the home stretch, making them the largest party in the new Knesset. The fifth-largest party, Kulanu, is likely to play an outsized role in determining who the next Prime Minister will be. This means that Benjamin Netanyahu’s return to office is not quite a done deal.

First, the results of Tuesday’s election:

Party Old 3/13 polls New
Likud 31 (+YB) 21 30
Jewish Home 12 11 8
Torah Judaism (UTJ) 7 6 6
Yisrael Beitenu (YB) see above 5 6
Yachad & Otzma 0 4 0
Total coalition 69 47 50
Kulanu 0 9 10
Yesh Atid 19 12 11
Shas 11 7 7
Meretz 6 5 5
Labor & Hatnuah 21 25 24
Hadash/UAL/Balad 11 13 13

The near-certain Likud coalition is indicated in red. Definite anti-Likud parties are “Labor-plus” in blue and Arab in green. That leaves┬áthree parties whose loyalties are not yet determined: Kulanu, Yesh Atid, and Shas.

Scoring the polls. Pre-election polls did fairly well, aggregated by voting bloc. However, they didn’t capture the massive surge for Likud, which is likely to have come from Netanyahu’s bald appeals to right-wing sentiment, which he made after the final polls were released. Unlike U.S. voters, Israelis have opportunites for strategic voting, a fact that Netanyahu exploited in the home stretch. Some of Likud’s surge came at the expense of Jewish Home, which fell short by 3 seats, and Yachad/Otzma, which fell below the 3.25% minimum. In the end, Likud overperformed opinion polls by 9 seats (equivalent to over 7% of the popular vote), but the “Likud-plus” coalition only overperformed by 3 seats, equivalent to only 2.5% of the popular vote. Since the average error in 2013 was 1.5 seats, we would have expected an error in the vicinity of 1.5*sqrt(5)=3.4 seats. So the error was about 1 standard deviation…not a massive shock, at least at the level of the whole coalition.

The second pre-election finding that came true was that the Joint Arab List is now the third-largest party in Israel, with 13 seats. Which blows the mind. (Here’s a very good argument that in the long term, this is bad for Israeli Arabs because it marginalizes them.)

This brings us to the importance of Moshe Kahlon, who I think most Americans have not heard of.

The importance of Kulanu. Assuming that President Rivlin gives Likud the first shot at forming a coalition, the task for Netanyahu is to recruit two out of the three key undecided parties. The combination of Yesh Atid+Shas probably won’t work because Yesh Atid is strongly secularist, and Shas is an ultra-Orthodox party. The other two combinations remaining both involve Kulanu.

Kulanu is a new party. It formed when its head, Moshe Kahlon, split with Netanyahu. On Election Day, Kahlon blamed Likud for making the current housing crisis worse. Betting markets and most analysts think Kulanu will fall into line with Likud. Before the election, Kahlon made clear his dislike for Likud. However, obviously he is under new pressure with Likud’s convincing performance on Tuesday. I think Netanyahu is still at risk. It will be interesting to watch.

There are some remaining minor hurdles to specific combinations of parties. Kulanu+Yesh Atid is iffy because their party heads both want to be Finance Minister. Regarding Kulanu+Shas, Shas might still be sore at Likud about the recently-passed law that mandated military service for haredi, members of an ultra sect. Maybe Shas will want that repealed.

Plan B, and Plan C. If Likud-plus can’t get it together within 42 days (read the rules here), then Labor-plus gets a shot for 28 days. Their most winning combination might be Kulanu+Yesh Atid (though see above). If that fails, then the last ditch is a “national unity” Labor-Likud government (though both parties say they have ruled that out). After that…there would have to be another election.

Finally, for a fairly acerbic take on how similar the possible outcomes were, especially for Palestinians, see this LRB piece by Yonatan Mendel.

Tags: Politics

16 Comments so far ↓

  • Amitabh Lath

    Wow, the polls were crap. What’s going on? Either the pollsters are hitting a heavily biased sample (so much so that weighing is not able to fix it) or people are flat out lying to pollsters. (If lying sounds too strong you could say “making up their minds in the last minute…”).

    • Sam Wang

      Polls have the same problem in U.S. primary elections, which is another case of high-information voters in a changing situation. The New Hampshire 2008 Democratic primary was one such case. Here, voters on the right had multiple choices. The error was 3 seats, which corresponds to 2.5% of the vote, not exactly a huge miss.

    • Sam Wang

      Also, see this interesting piece in the London Review of Books about the small substantive difference between the possible outcomes.

  • A New Jersey Farmer

    Could be that Netanyahu’s last minute announcents about a Palestinian state and Arab voters motivated his base to come out.

    • Sam Wang

      That, and other related statements. See this take in Haaretz.

    • Amitabh Lath

      The prediction as of March 13 was between 20 and 22 seats for Likud. The measured value was 31. Yes, perhaps there was a surge in the last five days, or Bradley effect or whatever. But the fact remains that from the US midterms to India and now Israel, the concept of cold-calling people on the phone and asking detailed questions about personal political opinion seems to be failing.

    • Sam Wang

      I don’t think that is the right metric. Instead, add Likud+JewishHome+YisraelBeitenu+TorahJudaism+YachadOtzma. This was 47 seats on 3/13, 50 seats actual. The difficulty is in strategic voting, as opposed to a two-party choice. I do agree that if one wants to capture those within-coalition shifts, some more dynamic sampling method like Internet polling or Xbox polling could be useful.

    • Amitabh Lath

      Yes, integration hides many ills.

  • Leor

    Perhaps if we had a model that combined fundamental variables (level of personal security, unemployment rate) with polls, it would have predicted that the incumbent prime minister (and his party) would stay in power.

    As for the within-coalition shifts, the story is quite simple:

    Right wing: Likud+ Jewish Home+ Yachad: predicted 36, actual 37

    All other parties were accurately predicted to within one mandate (Knesset seat)

    And still, many Israelis just love lying to pollsters.

    • Amitabh Lath

      This may be just my ignorance, but I have never understood how “fundamental variables” (GDP, unemployment, consumer confidence, phase of moon, whatever) is supposed to “fix” bad polling.

      The assumption is that polls are incorrect because a) sampling is wrong and since we do not have a good likely voter model we cannot weight the sample properly, or b) people lie to pollsters (Bradley effect).

      The “fundamentals” proposition is that errors (a) and (b) can be fixed by adding things like GDP etc. But how? Historic correlation? I simply do not understand this idea.

    • Sam Wang

      I actually do not think that such indirect predictors would have been of much help here, since we were in the home stretch. The best information was polling data – which was cut off the previous Friday. Press reports indicate that Labor and Likud camps were seeing movement toward Likud in their internal polls. That plus strategic voting can easily account for a 2.5% shift.

  • Leor

    Combining several sources of information might be a reasonable idea when each source contains a bias with unknown magnitude (such as the Bradley effect). There seem to be working examples for this (e.g., PollyVote).

    How to combine information sources?
    – Weighting, based on past accuracy and variability
    – Boosting – e.g., using index models as a specific predictor for parties/populations that deceived pollsters in the past

    • Amitabh Lath

      To correct the polling data, one would have to (at the very least) understand why it was off, and how the “fundamentals” fix the residuals.

      I can understand how something like say, unemployment rate or GDP can affect the vote. What I do not understand is how it affects the difference between what people tell pollsters and how they actually vote (or stay home).

      After all, the first order effects of these macroscopic fundamental variables are baked into the polling already.

      And without that understanding, any correction we concoct will be a just-so story. Given that there are a finite number of prior elections to tune on, and a nearly infinite number of fundamental variables to choose from, if your only metric is getting the past right, then how do you decide between one that uses GDP and another that uses the win/loss record of the Toledo Mudhens? If the latter has better a-posteriori predictive power?

      We were told as young’uns never to make ad-hoc corrections to data. The cautionary tale I remember is the 1968 experiment that saw evidence for the charm quark but “corrected” it away because it looked anomalous. Then in the 1974 “November revolution”, boom, Nobel Prize.

  • Amitabh Lath

    Is there any data (besides bloviating TV personalities) that points to large number of Israeli voters changing their minds between March 13 and the election? I know this has become the go-to story: Netanyahu’s scare tactics brought the right wing voters back to Likud, etc etc.

    But maybe the polls were wrong all along? Maybe Likud had 30 seats all along, and all the fluctuation including the last minute flutter towards Labor was just systematic noise and bad polling? It’s not as attractive a story, but let’s face it, the more mundane explanation is usually the correct one.

Leave a Comment