Israeli pollsters and news agencies churned out six polls in the 3 days after US President Donald Trump announced his Deal of the Century. Most of those polls were released in less than 24 hours, likely at great expense to the news agencies trying to generate headlines from the data. Those headlines varied from one saying nothing changed to another predicting Likud will surpass Blue and White. The truth is probably somewhere in the middle, but in order to find it, the various polls need to be combined to show the best information available.
That is what is shown here. Below is a discussion of the nature of polls, their strengths, their weaknesses, the problems of averaging them, and much more. But first, what’s the answer to the headline?
Did the Deal of the Century Affect the Election that’s a Month Away?
Before the deal was announced, the Right/Religious bloc was polling at an average of 54.9 seats. Those who nominated Gantz after the last election were polling at 54.4 on average. Both blocs needed 6-7 more seats to make a coalition (61+ out of 120 seats). The last poll before the deal was released on January 24th, before a host of different events including Bibi dropping his request for immunity, International Holocaust Remembrance Day, and the release of Naama Issachar. After all of those events, the Right/Religious bloc gained an average of 1.3 seats from the Left/Center/Arab Bloc. At the party level, Bibi’s Likud gained 1.6 seats in polls, partially at the expense of his allies on the right.
Below are charts of how everything breaks down including several different ways a coalition of 61 seats could be attempted, averting a fourth election.
Whether this change was the result of the Deal of the Century alone or a combination of all the events in the past week, this is definitely a win for Netanyahu. But unless this is a prelude to bigger growth from the Right/Religious bloc, this is not enough for an end to the endless elections. In fact, from a wider perspective, these polls speak more and more to how little is changing. The chart below with the results from September’s Knesset election is nearly identical to both of those above.
Many despair that the different blocs seem set in stone. This may be because the blocs are a good representation of the Israeli public. After all, the possible coalitions are only shifting by a single seat after one of the biggest political weeks for Israel in recent memory. Whatever the reason, if this deadlock continues, then only significant compromises by the parties themselves to make a coalition will result in an end to the constant election cycles.
Notes on Polls: What is Wrong with Them?
Before we go further, let’s reiterate why individual polls are only so reliable. We can see it when every new poll differs from others released just days or hours prior. If you looked into the polls, you’d notice that the margin for error is around 4%, bigger than the 3.25% threshold that is the do-or-die mark for small parties. Look even deeper and you’d see that polls are all predicting the size of parties as small as 4-6 seats by surveying around 500 total voters. This translates to estimating those 6 seats based on responses from about 40 people. Any survey that, by pure chance, polled just 20 more supporters of that party would report wildly different results. Yet these polls are not without merit. There is a reason they always seem to be in the right ballpark. If they could be combined, then many of these problems would shrink.
Notes on Polls: Three General Points
First, polls can become fairly accurate by averaging many polls together. The theory is simple and intuitive. Each survey will have some error, like missing some Haredi votes or over-representing Arab turnout. If each pollster makes a different mistake then combining their work can mitigate the error and randomness of any one survey. This can give significant precision, but it can also hide the truth behind deceptive “conventional wisdom.” A stark example is the fact that no American pollster predicted Trump would win because they were all making similar mistakes in gathering and analyzing their data. In Israel, this kind of error would miss factors like last minute spikes or drops in Arab and Haredi turnout, which missed the Shas surge to 9 seats in September when most predicted only 7.
A simple average seems like it would not be the best way to combine multiple polls. There should be some more intelligent way to take the data from each poll and use statistical analysis to give a more accurate result. The answer is that there are such methods. However, several problems prevent such an analysis including the lack of publicly available data, the complexity of factoring in samples taken at different times, and the issues arising from the different ways similar questions are asked. That said, a more simplified average does give good results and has been shown by some to reflect the eventual election outcomes with some accuracy.
Second, polls are good at predicting bloc sizes rather than party sizes. A survey of any one party will almost always miss the mark. A poll last election that said Labor had 4 seats instead of 6 was off by 33%. The same poll saying Left/Center/Arab parties have 54 instead of 57 seats was only be off by 5% on the bloc size. While those numbers are very simplified, they tell the same story that combining errors across the likely coalition blocs can give more accuracy to pundits and casual readers alike.
Third, despite accuracy problems for analyzing just one party at a time, polls still give a good idea of who is near the threshold. Falling below the 3.25% of votes cast means that a party gets nothing. You don’t have to be Naftali Bennett to know that being on a combinedlist with a party you disagree with can save your political skin. Knowing who may not make the cut can also help with figuring out who will be the next Prime Minister. Just be careful not to overestimate the consequences of one party dropping too low. Those “lost” seats go somewhere and odds are that half of them get distributed back to the party’s coalition allies. This election, the only relevant party is Otzma, which is averaging only 1.6% in polls since the January 29th and has only made threshold (3.25%) in one poll since the January 16th. For reference from September’s election, Otzma was polling around 2.8%, but only ended up with 1.8% of votes.
The notes below explain how numbers were produced for the charts above
Notes on Methodology: Averaging in this Article
In the chart above on post-deal polls, the six surveys were combined as a simple average (mean) because all arrived so close together. The same could not be done for the polls before Trump announced the deal because they spanned a much larger time period and they included several repeats from the same pollsters. A weighted average was used for these pre-deal polls. Those released more recently were given a higher significance so that they would be more reflective of the election if it happened on January 27th. (The weighting was linear based on the date of release and calculated so polls would have zero weight after 30 days.) In cases when a pollster issued more than one poll, the older of the two was thrown out so as not to give too much weight to the biases of that pollster.
Notes on Methodology: How to Divide Lists into Parties
Balad is one of the four parties that makes up The Joint (Arab) List. However, in the aftermath of the September 2019 election, Balad acted independently from the other parties in the list when it withdrew its 3 votes to nominate Benny Gantz for Prime Minister. Although it appears less likely that the Joint List will follow the same nomination path this round, it is worth differentiating between Balad and the rest of the list. The charts above separate Balad from the rest of the Joint List.
When averaging polls, most lists end up with a partial seat. If we are separately analyzing parties that are running in the election together, then that partial seat needs to be apportioned to the correct party. On the candidate slate for the Joint List, Balad has spots 2, 8, and 13. If the list gets 12 seats, only 2 will go to Balad. If the list gets 13, Balad gets 3. Because every vote between 12 and 13 goes toward getting the candidate at spot 13 into the Knesset, that fraction goes to Balad. So if the list is averaging 12.5 seats in polls, Balad averages 2.5 seats. Since the list is averaging above 13, Balad is averaging 3 in the above charts while the rest of the list gets the partial seat above 13. The makeup of the various party lists can be seen here: https://blogs.timesofisrael.com/23rd-knesset-election-lists-in-english/
Eight Polls between Party List Submittal (1/16) and the Deal of the Century (1/28): Ch11 Kantar, Ch13 Project HaMidgam, 103FM Maagar Mochot, Ch12 Midgam, Ch13 Direct Polls, Ch11 Kantar, Makor Rishon Panels, Walla Midgam.