Abstract

While election forecasting has arrived in some two dozen electoral democracies, hardly any effort has been made to forecast Israeli elections, due to several challenges. I discuss those challenges and offer a “reward-and-punishment” theory of the voting behavior of Israelis. I formalize the theory into a small model of election forecasting for Israeli elections using unemployment and incumbent approval approximately three months prior to the election and estimate the model using Israeli elections and voting intention data from 1973-2021. My forecast for 2022, given the sparseness of the model, was relatively accurate.

Introduction

Forecasting elections has proceeded apace in many democratically elected countries around the world [1-3]. The interest, of course, is high in trying to figure out the result of an election ahead of time, for policymakers, political players, and the public alike. Far from being a parlor game, having a good sense of the likely result will help direct the way in which campaign resources are allocated, amply candidate and party messaging, and create party platforms that conform more to the electoral environment.

One country notable for lacking a forecasting model is Israel. This could well be due to the complications of the multiparty Israeli system, where some two dozen parties regularly hold seats in the Knesset parliamentary chamber, and governments are always coalitions in some form. How can a forecasting model account for the kaleidoscope of Israeli partisanship? It may well be too tough to create such a model, in the end.

However, forecasting should still work, to an extent, even in complex environments such as in Israel. What we learn from these forecasting attempts may well be useful in other, similar electoral environments. (The Netherlands comes to mind as a polity with an enormous number of parties that contest for seats, and with a relatively low electoral threshold, allowing even tiny parties to win seats). In the spirit of learning, if nothing else, I will attempt such a forecasting model for Israel, using tried-and-true explanatory factors to hazard a guess as to the direction of the next government. The direction may be surprising.

I begin by offering a brief description of the complexity of Israeli government. Next, I delve into the literature to put forth an election forecasting model for Israeli parliamentary elections. I then estimate the model and assess it for its ability to forecast elections. I conclude by making the forecast and comparing it to the actual election results, from the 1 November 2022 elections.

Structure of Israeli Government

Since its founding in 1948, Israel has operated as a single-district multiparty parliamentary democracy. It is a remarkably open system, with a small electoral threshold of 3.25 percent voteshare for a party to hold seats in the Knesset. The openness has several origins: (1) the spirit of fair play and inclusion in the creation of a new democracy, (2) the founders of Israel were members of smaller parties, who favored proportional representation as a way to ensure their participation in politics, (3) the intention of including Palestine in the Israeli state required proportionality, and (4) the perception that proportional representation is simply more democratic [4]. Often, very small parties that tend to lie on the ideological extremes can hold the balance of power for a government. Israelis have generally supported their political system, though there is a high level of distrust in the Knesset and political parties in recent times [5].

One consequence of this open system is the extreme number of political parties. The most recent elections resulted in ten political parties receiving the 3.25% threshold¹. The largest party, Likud, received slightly more than a quarter of the seats, and is able to now govern in coalition with other right-wing parties in an eight-seat majority. This election was the fifth election in the past four years, as previous elections have resulted in collapsed governments.

Also contributing to the lack of trust and confidence in the Knesset is the turnover in political parties and shifting alliances within them. The mainstream parties, Likud on the right and Labor on the left, have undergone significant leadership and organizational changes, though the winner of the most recent elections, Benjamin Netanyahu, is the great survivor of these upheavals, becoming Prime Minister for the third time and the 15^th year. In fact, the Labor party descended to a historically poor performance in the 1 November election, dropping to four seats in the Knesset. This instability will affect forecasting efforts, which tend to rely on stable configurations of political parties.

Election Forecasting

Beginning with Sigelman [6] and Lewis-Beck and Rice [7], political scientists have endeavored to forecast, months in advance, election results for presidencies and legislatures. The general methodology is to model the election using a single equation, where the dependent variable is the voting result in some fashion (percentage voteshare for the incumbent party, seats won by the governing party, Electoral College results, for instance), and the independent variables are some form of voter evaluation of the government (approval of the government, state of the economy, absence or presence of war, for instance). The data collected are repeated across elections, and the model is applied to these data through multiple regression analysis, either ordinary least squares or a variant of it to capture the dynamic effect of the time-series data.

These forecasting efforts have become popular in the media, such that figures like Nate Silver and Nate Cohn have become data journalists and polling analysts, aggregating public opinion polls through U.S. election campaigns to hazard forecasts [8]. They have also translated to many countries outside of the United States [3].

Yet Israel has remained outside of all of these forecasting models, and perhaps for good reason. The open system makes it difficult to imagine the exercise of capturing variation in the incumbent party voteshare (the common dependent variable) to be a straightforward one. Yet, given the spread of forecasting work, and the coalescing around key explanatory variables, warrants giving Israel a try in forecasting its election results.

A first empirical step is to examine a correlation matrix of likely independent variables that could have an impact on explaining variation in the dependent variable, incumbent party voteshare. Below is such a matrix, for elections from the years 1973-2021 (N=15).

Table 1: Correlation Matrix of Potential Explanatory Variables

(Note on data: GDP Growth, Jobless Rate, and Inflation Rate are from the year prior to the election and come from the Israeli Central Bureau of Statistics; Incumbent Support is approximately one to two months prior to the election and comes from the Israeli National Election Studies.)

Out of these variables, the two most promising independent variables are the jobless rate from the previous year, and the support for the governing party. These results are similar to that of the only other Israeli election forecasting paper that I could locate ([9] who concludes that "[f]orecasting Israeli national elections has indeed proven to be a difficult task" (9)). The model to be estimated using multiple regression is as follows:

Incumbent Party Voteshare = f (Jobless Rate, Incumbent Party Support)

The parameter estimates for the model are as follows.

Table 2: Multiple Regression Model of Israeli Election Forecasts, 1973-2021

R-squared = .33

Standard Error of Estimate = 7.29

N = 15

Durbin-Watson (post-first differencing) = 2.70 (result indeterminate)

*** indicates statistical significance at the .15 level, one-tailed test, ** indicates statistical significance at the .10 level, one tailed test, and S.E.E. is the standard error of estimate.

The forecasting model performs in a modest manner. The R-squared statistic is low, at .33. The higher the R-squared statistic, the better, as it indicates how much variation in incumbent party voteshare is explained by the jobless rate and the extent to which the public supports the incumbent party. Forecasting models emphasize explaining variation in the dependent variable, as this will allow for a more accurate forecast. A rule of thumb for assessing the ability of forecasting equations to make creditable forecasts is an R-squared value in the .65 to .70 range, or 65% to 70% of variation in the dependent variable explained. The Israeli forecasting model, at .33, falls well short of that standard.

Additionally, while the slope coefficients are correctly signed, their significance levels are remarkably weak (although Lewis-Beck and Rice utilized the .15 level of significance in an early model of forecasting the French presidential and National Assembly elections; see [10]). From this perspective, it is hard to put a great deal of trust in the slope coefficients are different from zero.

Nevertheless, we will hazard a forecast for the incumbent party, the liberal Yesh Atid party, in the 1 November election. We plug in to the estimated multiple regression model current values of jobless rate of 5.0 and 20 percent support for Yesh Atid, solve the regression equation, and arrive at a forecast.

Voteshare for Yesh Atid = 18.87 - .78 (5.0) + .41 (20) = 22.97 percent.

This is relatively close to the actual voteshare for Yesh Atid, 17.78, so the forecast was at least relatively accurate, and within the standard error of estimate for this model of approximately seven percent.

Conclusion

Israeli election forecasts are few and far between. They require special caution, given the plethora of partisan options, the changing security situation (an interesting variable estimated by Leininger 2019 on terrorist attack fatalities may account for this), and the collapse of coalition governments. Even the COVID-19 era may play a role, though given its limited time frame it cannot be modeled the way that other independent variables can be in the forecasting enterprise. While the estimated model has some statistical weaknesses, such as stretching the bounds of statistical significance, it does have some accuracy in its forecasting potential, using variables that are available prior to the election itself. The model presented here is underspecified, based on the R-squared statistic, but it is a reasonable starting point for future work on forecasting Israeli elections.

References

1. Mongrain, Philippe, and Mary Stegmaier. (2024). “Introduction to Forecasting the 2024 U.S. Elections.” Special Issue on Forecasting the 2024 U.S. Elections. PS: Political Science 57(3):1-12.View

2. Burkhart, Ross E. (2023). “A Fractured Electorate? French Presidential Election Forecasting for 2022.” Journal of Political Science and Public Opinion 1(1):101-3. View

3. Belanger, Eric, and David Trotter. (2017). "Econometric Approaches to Forecasting." In The SAGE Handbook of Political Behaviour, eds. Kai Arzheimer, Jocelyn Evans, and Michael S. Lewis-Beck. Los Angeles: SAGE Reference. View

4. Rahat, Gideon, and Reuven Y. Hazan. (2008). "Israel: The Politics of an Extreme Electoral System." In The Politics of Electoral Systems, eds. Michael Gallagher and Paul Mitchell. New York: Oxford University Press. View

5. The Times of Israel." (2022). Annual poll: Public's faith in institutions, satisfaction in state of country dire." 6 January.

6. Sigelman, Lee. 1979. "Presidential Popularity and Presidential Elections." Public Opinion Quarterly 43:532-34 View

7. Lewis-Beck, Michael S., and Tom W. Rice. 1982. "Presidential Popularity and Presidential Vote." Public Opinion Quarterly 46:534-37.View

8. Burkhart, Ross E. 2018. "Was Anyone Right? Assessing U.S. Presidential Election Forecasts." The Blue Review, 5 February. View

9. Leininger, Arndt. (2019). "How unpredictable are Israeli parliamentary elections?" Typescript.

10. Lewis-Beck, Michael S., and Tom W. Rice. (1992). Forecasting Elections. Washington: CQ Press. View