the green line and the security fenceu.cs.biu.ac.il/~belenka/israeli_election.pdf · recent debate...
TRANSCRIPT
The Green Line and the Security Fence:
Finding a Psychological Factor in Israel’s National Elections
Ari Belenkiy∗∗∗∗
Mathematics Department, Bar-Ilan University, Ramat Gan 52900, Israel
with
Yosef Grisi
Economics Department, Bar-Ilan University, Ramat Gan 52900 Israel
Abstract. The results of the 1996 and 1999 elections, together with data from a 1995 domestic
census, give a clear snapshot of the political orientation of the Israeli population at the time. The
most important factors of Israeli political preferences: ethnic background, religion, nationality
were first recognized by S. Smooha (1978, 1993). Their relevance was confirmed in twelve
consecutive volumes of The Elections in Israel series, edited by A. Arian and M. Shamir. The
recent debate concentrated on the question of the importance of the social class factor.
Several authors since 1996 have pointed out the new opportunities for sociological
research presented by the 1996 Israeli electoral reform: the two-ballot system (separate votes for
the prime minister and political party) gave voters a chance to express their ‘global’ (security)
and ‘local’ (lifestyle) concerns separately. This paper focuses on the ‘global’ issues and related
methodological ones.
The major result is the mathematical description of a fifth factor, which we tentatively
call the psychological factor. We measured it by the voters’ proximity to the Green Line
(hereafter, GL), which separates the so-called “territories” (Judea, Samaria, and Gaza), populated
by Palestinian Arabs and Israeli settlers, from the rest of Israel. Applied to the aggregate data, the
results show: the closer to the GL a Jewish voter lived in 1996 and 1999, the more biased his vote
was toward the Right; roughly, the Right lost 3% of the vote for each 10 km away from the GL.
Though the “proximity” or “Distance to the GL” (further, DGL) parameter alone could not be the
ultimate measure of the psychological factor, it certainly touches on the essence of the problem as
being directly correlated with the time needed for a terrorist to penetrate Israel’s territory and
commit suicide. We also took care to separate the psychological factor from two purely
geographical factors: voters’ proximity to Israel’s external borders and the sea.
2
Under the supervision of the first author, a group of students in 2005 conducted a poll of
a group of 1,215 Israeli citizens to see if these issues could be clarified on the individual level,
avoiding the ecological inferences. The poll showed that the DGL variable appears to have the
most Wald-significant coefficient in the binary logistic analysis, several times greater than its
geographical analog. It is interesting that it has slightly decayed in magnitude with time, from
early (1996) to middle (1999) to more recent (2001) elections. We associate this decline with a
short period of peace from 1996 to 1999 and, after a new break of violence in 2000, an increased
feeling of security among the Israeli population since building the Security Fence came on the
agenda in 2001.
Introduction
In the introduction to his book, “Israel: Pluralism and Conflict,” Sammy Smooha, a
founding father of Israeli political science, wrote:
“The 4.6 million residents within ceasefire boundaries of Israel at the end of 1975 are
internally separated along five lines resulting in the following divisions: Palestinians –
Jews, Israeli Arabs – Jews, Druze – Christian – Moslem Israeli Arabs, religious –
nonreligious Jews, Oriental – Ashkenazi Jews…. Though there is a large and
expanding literature on Israel, with one or two minor exceptions none of the works has
focused on its multifaceted pluralistic structure. Furthermore, the five pluralistic
divisions are not given equal attention by Israeli sociologists. Palestinian-Jewish
relations, notwithstanding the headlines they capture, are rarely studied because of
their newness, sensitivity and perhaps inaccessibility.” (Smooha 1978, 2-3).
In this paper, we attack precisely the “inaccessible” issue: Israeli-Palestinian relations.
This issue left its distinct mark on the results of all Israeli elections. While it was known
to exist, until now it has eluded quantitative description.
True, the territories issue and performance of the Israeli leaders in the area of state
security, as seen by voters, were discussed constantly in the literature, and a steady
3
increase in the importance of these issues in elections from 1984 to 1999 was duly
observed (Shamir and Arian 1999: Table 2; The Elections in Israel-1999, Table 1.2; The
Elections in Israel-2001, 16: Fig. 2), but again, the quantitative measure of the voters’
reactions was missing. It is unclear, for example, how to deduce any meaningful
predictions from the fact that 90% of respondents in 1996 and 1999 said that the
territories issue “will influence my vote” (The Elections in Israel -1999: 20). Shalev and
Kis (2002) tried to distinguish a voting pattern in different types of localities, but looked
at the latter only from the class (income) point of view.
The tables and graphs that dominate the first eight volumes of the Elections in Israel
series (1984-1999) show only one parameter at time, although there are at least three
recognized factors of importance mentioned by Smooha: Oriental-Ashkenazi, religious-
nonreligious, and Jews-Israeli Arabs issues, which can be alternatively described as
ethnicity, religiosity, and nationality. Meantime, a fourth factor, so-called “class
cleavage,” gradually became of importance within Israeli society. To study the political
effects of any other factor, in particular the problem of Israeli-Palestinian relations, one
must find a way to control those four parameters. The search for a new technique to
examine the parameters other than tables and graphs was at a standstill when a lucky
event broke the impasse.
Data from the May 1996 Israeli elections and a November 1995 domestic census
provided a unique opportunity to analyze the political preferences of the different
segments of the Israeli population. Never in Israeli history had these two events been so
4
chronologically close. Elections that occurred earlier, in 1988 and 1992, and those
following in 1999, 2001, and 2003, though taking place at regular intervals, are three or
more years away from any census. As a result, only the elections in 1996 and, to some
extent, 1999, can be paired with reliable data for the overall Israeli population, though
they took place during a turbulent period of social and economic changes in Israel.
Though one more census, conducted in June 1983, could be similarly paired with the
1984 elections, a crucial factor was missing: the direct election of the prime minister,
which took place only in 1996, 1999, and 2001.
Shamir and Arian (1999) found that ethnicity and religiosity have decisive roles in the
outcome of those three elections; the same result for the 1999 and 2001 elections was
confirmed by Shalev and Kis (2002) and Andersen and Yaish (2003). Let us note, inter
alia, that although the latter two papers considered ethnicity and religiosity as control
parameters rather than the target of their research, they oversimplified their treatment of
the ethnicity factor by dividing the voters into Ashkenazim (European/American origin)
and Sephardim/Oriental Jews (African/Asian origin).1 First, this is imprecise
etymologically: Bulgarian and Romanian Jews are mainly Sephardim and vote
haphazardly. Second, it is not clear how Turkish Jews vote. Third, North American Jews,
though Ashkenazim, vote the opposite of European Jews: Right, not Left.
An important methodological insight made by the authors, who wrote after 1996, was
their pointing to unique opportunities provided by Israel’s 1996 electoral reform. While
the preferences of the Israeli voter are complex and inseparable (De Marchi and
5
Goemans 2001),2 the two-ballot system (separate votes for the prime minister and
political party3) gave the voters a chance to express their ‘global’ (security) and ‘local’
(lifestyle) concerns separately (see Israel At the Polls: 1996, 90, 127, 258). While Shalev
and Kis (2002) and Andersen and Yaish (2003) concentrated their analysis mostly around
‘local’ concerns (more precisely, analyzing the vote according to class subdivisions
within Israeli society), we shall focus on the ‘global’ one, the Israeli-Palestinian conflict,
which is basically a personal security or psychological issue.
In fact, the attitude toward this single issue gave birth to an inadequate Israeli political
nomenclature: Right and Left, where ‘Hawks’ and ‘Doves’ would be more accurate.4 One
can argue that there was a clear division between Right and Left, that is, between
‘Hawks’ and ‘Doves,’ even prior to 1996 since at least 1993, when the Oslo Accords
were signed. However, the appearance of such parties as Tsomet in 1992, or The Third
Way and Israel Ba’Aliya in 1996, with their mixed platforms on foreign politics, did
muddy the waters.5 Therefore, Anderson and Yaish (2003), in their study of ‘local
concerns,’ which they identified with ‘party vote,’ offered a rather complicated (probit6)
model wherein several pieces of the Israeli political spectrum were included and the
dependent variable was allowed to have a discrete set of values, not just two – for the two
major parties.7
Though different “third parties” offered a different ‘global’ security agenda, a separate
vote for the prime minister purified the voter’s attitude, separating his local concerns
from the security issue. One of the Oslo Accord signers, Shimon Peres, in 1996, and
6
Ehud Barak in 1999 and in 2001 were associated with the desire to make concessions to
Palestinians and therefore with the Israeli “Left,” while their rivals, Binyamin Netanyahu
in 1996 and 1999 and Ariel Sharon in 2001, opponents of the Oslo Accords, were
associated with the Israeli “Right.” Therefore, the dependent variable can be treated as
binary, which in turn, allows us to use a traditional, simpler regression model.8
Looking back, say to 1965 and 1969, when “split-ticket voting” was carried out when a
voter chose a party for Knesset and local council separately (Arian 1973, 183-6), one can
also discern the results we advocate here – but those results went unnoticed. We will
discuss this point later in the text.
The analysis of the Israeli political spectrum that we present in this paper is not unique.
After years of largely descriptive analysis, several authors carried out the multilevel
regression analysis. Shamir and Arian (1999) performed regression analysis of the 1999
elections, using an individual poll of nearly 800 respondents. Debating the strong and
weak aspects of preliminary polls, Shalev and Kis (2002) made regression analysis with
aggregate data from the Israeli Statistical Bureau, similar to data we used in this paper.
All of the above authors avoided treating the national (Jewish vs. Israeli Arab) factor
because of some “technical” problems: in one case there were no trustworthy polls
among Israeli Arabs; in another, they did not include all other relevant factors. The
problems with data and the way we posed the problem (as a “psychological factor” within
7
Jewish Israeli population) forced us to avoid discussing the national factor as well.9 We
attempted to address it in the individual poll, though unsuccessfully.
1. Key Question and Innovations
The key question was to find a mathematical way to describe the security factor such that
the variable behind it must be independent of the major trio, ethnicity, religiosity,
nationality, plus a social factor. Andersen and Yaish (2003) in their study introduced
several control parameters, among them the ‘urban locality,’ a binary variable. It must be
recognized immediately as inadequate for our purposes. It has a mixed content of
geography and culture and therefore fails to represent either of them. Its inadequacy to
represent “culture” is especially striking: small communities do not lack information
when compared with cities; besides, many kibbutzim can compete culturally with cities.10
In a more formal objection, one can say that an exact division between urban and rural
areas is non-existent in Israel. Introducing the urban locality seems a dubious attempt to
represent a standard ‘center-periphery’ division in a different way.11
It is not a surprise that urban locality did not feature significantly in Andersen-Yaish’s
regression model (2003).12 Moreover, the sign of the coefficient before this variable (see
ibid, 413, Table 3A) suggests that urban people prefer Rightist parties, which might be
true for Jerusalem and Southern Tel Aviv, but is in direct contradiction to what we know
about communities in Northern Tel Aviv and “red” (pro-Communist) Haifa. The failure
of urban locality forces us to suggest that another, more refined set of parameters might
be appropriate to comprehend Israeli reality.
8
In this paper we investigate what Andersen and Yaish neglected – the problem of a
security or psychological factor – and introduce a new parameter, the DGL. This is a
quantitative variable measured as the shortest distance from any locality to the GL. The
latter refers to the 1949 Armstice lines between Israel and its neighbors: Egypt, Jordan,
Lebanon, and Syria after the 1948 Arab-Israeli War, or Israeli War of Independence, and
encompasses Judea, Samaria, and Gaza. This variable was not provided in the census
data. We obtained it from the Web site of the Israeli Ministry of Interior.13
This variable grasps the essence of the “psychological” factor (hereafter, P-factor). Why
psychological? Because the distance is directly related to the time needed for a terrorist
group to infiltrate Israeli territory and reach this or that city. Before the Security Fence
was raised all along the perimeter of the Green Line in 2001-2003, terrorists could
penetrate Israel proper virtually through any place at the GL. After an informant's
warning, the Israeli police have some time to detect the terrorist – the time is inversely
proportional14 to the distance from the GL to the settlement, a possible target of the
terrorist. When, during 1996-2003, the country lived in expectation of new terrorist
attacks daily, the Fear on the streets was in the same (inverse) relation to the DGL. The
most distant places from the GL, like Eilat or Golan Heights settlements, did not feel this
Fear at all. True, the very first terrorist acts were committed in the capitals, Jerusalem and
Tel-Aviv, but later on, virtually every settlement became a target for terrorists, as follows
from the geography of suicide bombings (Haifa, Netania, Beer-Sheba, Hadera, etc.).
9
Since the 1993 Oslo Accords, the Left has been more inclined to make peaceful gestures
toward Palestinians. This includes not only transferring land and dismantling
unauthorized settlements, but also releasing prisoners and dismantling some military
checkpoints and roadblocks as acts of goodwill. These practices invariably led to easier
infiltration by terrorists into Israel proper and to new series of terrorist attacks and new
victims – which could not have added to the popularity of the Left, especially among
those living in proximity to the GL.15 To the contrary, the Right, enemy of the Oslo
Accords, has always been ready for an immediate and strong response, such as punishing
any acts of Palestinian violence by permitting new settlement activities, imposing new
checkpoints, and setting curfews. These actions could add to the feeling of security for
those living closer to the GL. In short, this is the same “Fear factor.” It is our goal to
show that the closer to the GL an Israeli Jewish voter lived during the late 1990s, the
more biased his/her vote was toward the Right.
One could ask “which came first, the chicken or the egg?” since the reverse chain of
reasoning is also viable – pro-Right Israelis could prefer settling in Jerusalem and beyond
the GL – in the Biblical cities like Hebron, Shechem, Bet El, Efrat, Jericho, and Bet
Lehem. This is correct, but the only visible reason behind such a preference is religiosity.
Regression analysis would take care of this problem automatically, using religiosity as a
control variable. The problem is that, unlike the situation with the individual poll, we do
NOT have an adequate representation for religiosity in the aggregate data. How can we
circumvent it? The argument is that for religious people, it makes sense to live IN a
historic Biblical place, not X km away from it! Therefore, it seems reasonable to exclude
10
the “Biblical” areas (and only them) from our analysis of aggregate data, which is
tantamount to excluding the GL settlements and Jerusalem (located at 3-5 km from the
GL). This solves the problem of the direction of causation by making the reverse
implausible: a move, say, from Haifa (40 km from GL) to Tel-Aviv (20 km to GL) hardly
would be seen as the desire to live closer to historic Biblical places!
Still, Jerusalem must be treated separately from the GL settlements, since it is already a
modern city, host to large, high-tech industries. A high-tech worker would have no a
priori preference to settle in Jerusalem, rather than in Tel-Aviv or Haifa. Therefore, we
decided to keep Jerusalem in the aggregate data analysis as well and report on differences
between the results that include it and exclude it.
There might be an additional reason to exclude settlements beyond the GL, even in the
analysis of the individual poll where we can control for religiosity. For those who live
beyond the GL, there are two reasons to vote Right. One is significantly lower taxes vs.
the rest of Israel and an opportunity to buy a house several times cheaper than inside the
GL.16 Israeli settlements also have enjoyed generous government financing for building
public facilities, special access roads, and roads that bypass Palestinian villages, and
industrial zones, in addition to the operation of schools and health clinics. This is a so-
called “Quality-of-Life” factor which might be a reason even for non-religious settlers to
vote Right, since the Extreme Left (“Peace Now”) since the 1990s has raised the question
of abolishing tax and security advantages for settlers.17 “Peace Now” is influential within
11
the Meretz Party, which was a part of the Leftist government coalition in 1992-1996 and
1999-2001. The Rightist government took the opposite stance.18
The question is whether the “Quality-of-Life” factor belongs to the P-factor. If one wants
to study the Fear (or rather “Fear-of-losing--life”) factor alone, since it is not obvious
how to separate it effectively from the former, the only solution seems to be to proceed
exactly as we did in the previous problem with direction of causation – to exclude the GL
settlements from the analysis. In case the “Quality-of-Life” factor is included in the P-
factor, the non-religious GL settlements might remain in the data. Actually, two sorts of
analysis can emerge – excluding or including GL settlements.
Arguably, the P-factor came into the limelight due to several terrorist suicide attacks in
Tel-Aviv and Jerusalem on the eve of the 1996 elections, but remarkably, it remained
underestimated until the 1999 elections, when the Left won by promising quick peace
after a comparatively peaceful lullaby during Netanyahu’s tenure in 1996-99. When this
promise failed, as a result of a new outbreak of violence in September 2000, the P-factor
came to the fore in the 2001 and 2003 elections, when the Right won a sweeping victory.
To mitigate the geographical flavor in the P-factor we introduced in parallel two
geographical factors: Location by Sea and Distance to the Borders of Israel (further: DB).
The first was measured as a dummy binary variable (1,0) and the second as the shortest
distance to the closest border with one of four countries: Egypt, Jordan, Syria, or
Lebanon. Both variables displayed a low significance for election results. This is
consistent with the logic of Israeli politics. Both Right and Left take serious heed to the
12
defense of the Israeli borders, and residents in the vicinity do not feel endangered by their
proximity. Location by Sea also assumes a “class” factor, since port cities in the past
were the bulwark of the working class. However, it is not a decisive factor, and, as we
shall see, two similar voters in Haifa and Ashdod have quite different priorities.
Because Israeli Arabs were expected to vote differently than the Jews, and because the
1995-1996 suicide explosions created psychological (personal security) problems within
the Jewish population alone, we decided not to introduce complicated, difficult to
interpret “mixed variables” but to restrict our analysis to the electoral areas with
predominantly (more than 90%) Jewish voters.19 This cost us about 200 points in the data
set (about 9%). Another, more technical, problem was to exclude the “outliers” (cities too
far away from the GL, like Eilat or the Golan Heights), and we restricted our analysis to
cities within 60 km of the GL. This cost us another 170 points of data (about 7%).
Both restrictions – 90% and 60km – seem arbitrary, but allowed us to focus on the major
problem more closely. The first parameter (90% Jewish voters) is reasonable in itself
since we cannot completely reject mixed communities with a mixed, but predominantly
Jewish, population, while the second (60km) includes the major Israeli cities, Haifa in
particular. Changing the latter parameter from the 60 km to 100 km range shows a slight
decline of the P-factor and its significance.
2. Difficulties
13
1. The first issue is that the census provides data about all residents of a particular
electoral area, but not all residents actually vote. Partial justification for our approach is
that the percentage of participation in the 1996 and 1999 elections was almost uniform
within statistical areas, about 80% (Israel At the Polls: 1996, 6; The Elections in Israel –
2001, 35).20 Therefore, we assumed that there was no significant bias in our analysis
under the plausible belief that representatives across all the groups abstained from voting
randomly.
2. Another, more general difficulty comes from the very nature of the data at hand.
Aggregate data (averaged over a statistical area) make any analysis liable to the so-called
“ecological fallacy” (see, e.g., Freedman 2001), in which one must restrict conclusions
about correlations of the results with the size of a particular group, and NOT with
individual preferences. A personal poll was necessary to confirm results at the individual
level and such a poll was conducted by the first author with his students in 2005.
Fortunately, this research with aggregate data is partially free from this problem as well.
The variable in which we are most interested, the DGL, is NOT an average. People vote
at the place where they live. Neither are the two geographical variables – DB and
Location by Sea – averages. This frees our major claim from the “ecological fallacy.”21
On the contrary, the other variables, representing controlled parameters (including ethnic
origin, income, religiosity, nationality, age and gender), all are represented by averages in
a statistical area. Therefore, we avoid any claims about the vote of an individual carrier of
14
one of these qualities.22 Still, it is a relief to know that our data both qualitatively (the
sign of regression coefficients) and quantitatively (their high significance) support
conclusions made by Shamir and Arian (1999), Shalev and Kis (2002), and Andersen and
Yaish (2003) about the major trio (ethnicity, religiosity, nationality) + social factor, while
pointing to several interesting clarifications at the same time. This, in turn, justifies our
correct choice of the control parameters in our major research directed to mathematically
represent the psychological (security) issue.23
3. Regarding Ethnic Background, we chose the smallest possible set of representatives:
either entire continents: Asia, Africa, North and South America, and Europe – or the
countries that represent their political preferences most prominently. True, several
countries, like Turkey, Romania, and Bulgaria, defy this simplified description, but they
do not influence much the results and after all we need this variable only as a control.
4. The question about religiosity does not exist in the census, and this is its major
deficiency regarding our analysis. We had to invent a plausible variable to represent it:
large families (7 or more members) could represent the religious group while single
people are more likely to be non-religious. This choice is justified a posteriori, since the
former group votes strongly Right while the latter group votes strongly anti-Right.
5. Though all variables, including dependent ones, are qualitative, except Income, after
aggregation they became “quantitative,” which makes the technical analysis much easier.
Therefore, to analyze census aggregate data, we chose linear regression instead of logistic
15
regression, since the sign and significance of linear regression coefficients allow for
direct interpretation.
6. There is a raging debate on exactly what “class” is. Andersen and Yaish (2003) argued
that “class” is defined solely by individual (family) income and occupation. Shalev and
Kis (2002) added to this pair the “population density.” We took the side of Andersen and
Yaish so as not to compute an extra parameter. As a proxy for “occupation” we used
“academic degree.” Though usually considered as “education,” an academic degree also
reflects “class” status within Israeli society.
7. In the study of P-factor, it would be too costly to discard the electoral areas with a
partial Jewish population, say, with 5-10% of Arabs. However, in such areas the figures
of the vote will not give adequate figures for Jewish voters we are seeking. Indeed, the
individual poll shows that the Arabs vote overwhelmingly Left. Such areas with 5-10%
Arab populations are pretty numerous in Jerusalem and Haifa. Therefore, we introduced a
correction coefficient for the dependent variable (vote for the Right): K = 100% / (100% -
% of Arab votes) which is adequate for the low percentage of Arab voters. Indeed, in the
area with 10% percent Arabs, the result of a 50% vote for the Right must be interpreted
as 55% of Jews voted Right – which is adequately represented by K=10/9 = 1.1.
8. To eliminate locations beyond the GL appeared to be an inevitable decision, since it is
unclear how to measure them properly. The “minus” sign was considered as an option,
but was not a successful one. For example, the city of Ariel is located in the center of the
16
Jewish block in Samaria, 17 km away from the GL, but far away from the Security
Fence, while several smaller settlements near the GL that appeared in the poll are much
closer to the Security fence and hence are more insecure. Therefore, the DGL for cities
beyond the GL first were set to zero, making them equally insecure.
3. Hypothesis
A zero hypothesis is that the major factors, trio (ethnicity, religiosity and nationality) +
social, are the only significant variables to explain the result of the Right-Left vote.
An alternative hypothesis is that there is at least one more significant variable, which we
designate as a psychological (security) factor.
4. Data Description
The technical part of our research was alleviated due to two fortunate circumstances. The
first was the availability of the powerful SPSS 10.0 Data Editor, which can perform
multivariable regression analysis. Another piece of luck came from being able to obtain
the special arrangement of data necessary for our research. The Israeli Central Bureau of
Statistics had prepared a special “integrated” electronic file for the 1995 census and 1996
elections data for 2,257 statistical areas (Electronic File with Results of Israeli 1996 and
1999 Elections and 1995 Census, 2003, hereinafter, Census data).24
The number of statistical areas came from the following arithmetic: There were 1,559
statistical areas comprising cities with population of more than 10,000, and 975 areas
17
made up of small communities. In the former, a city was subdivided into several
statistical areas to achieve an average number of residents, about 2,000-4,000 in one area.
It appears that both numbers had undergone an additional mixture: some 120 urban and
120 small statistical areas (each with small numbers of residents or voters) were grouped
together with another small one, within the same geographical territory. Therefore, this
nuisance factor does not influence our major parameters and conclusions.25
a. Description of variables and units of measurements
The dependent variables, Netanyahu_96 and Netanyahu_99, show the percentage of votes
cast for Netanyahu personally in 1996 and 1999 respectively.
1. Ethnic Background
This is represented by the percentage of those coming from the same country of origin
{Turkey, Iraq, Yemen, Iran, Morocco, Libya, Tunisia, Ethiopia, Romania, Bulgaria,
Germany, Poland, former USSR, Latin America, and North America} in a given
statistical area.26
2. Degree of Religiosity
This parameter is not reported in the census. We identified it by the number of people in
the family; variable Person_K shows families with K members. We use only Person_7+
(seven and more members) and Person_1 (one member), which we assume to represent
the two poles – very religious and secular.
18
3. Social Class
This parameter is represented jointly by Income and Occupation. Income here means
average income within the area, in NIS (New Israeli Shekels). We added to this two
variables: Rich and Poor – each represents a percentage of those that fall within the
wealthiest 10% and poorest 10%, respectively, of the Israeli population. Academic degree
played proxy for Occupation and was measured in aggregate data analysis as a
percentage of people with at least a first academic degree.27
4. Nationality
Variables – Jew, Muslim, Christian, and Druze – show the percentage of nationalities of
residents in the statistical area.
5. Inner Security (Psychology)
DGL is given as the shortest distance (measured in km) from the locality to the closest of
three GL segments: Judea, Samaria, or Gaza.
6. Geography
DB is given as the distance in km to the closest Israeli border.
Location by Sea is a dummy binary variable {1,0} that shows whether a statistical area is
located by the sea (Mediterranean or Red) or not.
7. Age
19
Israeli population was divided into 5 age groups: (18-24), (25-44), (45-64), (64-75),
(75+), each represented by a percentage.
8. Gender
Percentages of Males and Females. Census gave the number of those older than 15 years.
Let us stress here that we are not overly concerned with the exact influence of control
variables (which are aggregated), but only with their signs and their relevance (supported
by high T-statistics). As for psychological and geographical variables, we are concerned
with both the value of regression coefficients and their T-statistics.
Correlations
The strongest correlation, 0.6, is found between Income and Academic Degree. This
means that we have here an example of multicolinearity (Ramanathan, 2002: ch.5), and
these two variables seem to be interchangeable since both carry the same pro-Left bias.
The latter also has a strong correlation with different ethnic backgrounds: 0.4 for
Germans and North Americans, 0.23 for Polish, 0.2 for South Americans, nearly 0 for
Romanians and Bulgarians, and -0.2 for Asian and African Jews – all at the 1% level.
5. Psychological and Geographical Factors: simple analysis
It is not a great surprise that the DGL must be an important factor. In tiny Israel,
distances are ridiculously small. Jerusalem is located within 3 km of the border of Judea;
Beer-Sheba within 17 km of the border with Gaza; and Tel Aviv and Haifa are within 20
and 39 km, respectively, of the border of Samaria.28 It is rather a surprise that the P-factor
20
was never considered earlier! To illustrate this point by simple means, we prepared Table
1 that displays the pro-Right vote in 15 major Israeli cities, where only electoral areas
with a Jewish population higher than 90% were counted.29
The 15 cities were lined up according to their distance from the GL. All of the
settlements beyond the GL were assigned zero DGL.30 “# Stat areas” in Table 1 means
the number of electoral areas within a city, or, in effect, the relative size of the city.
City Green
line
Jeru
salem
Petah
Tikwa
Ashk
elon
Bnei
Brak
Neta
nia
Ramat
Gan
Beer
Sheba
Rishon
Lezion
Reho
vot
Tel
Aviv
Hol
on
Bat
Yam
Ash
dod
Hai
fa
#Stat_areas 82 143 44 20 32 46 41 41 41 26 129 46 39 36 59
DGL 0 3.2 9 10.8 14.3 15 15.6 17 17.3 17.3 20 20 23.5 29.6 39
Loc_by_Sea No No No Yes No Yes No No Yes No Yes No Yes Yes Yes
Dist_Border 32 62 49 65 67 65 49 69 56.7 72 69 74 69 31
1996 87.7 71 57.6 61 88 60.5 48.5 62 51 50 44 51 55.5 66.5 42
1999 79.4 65 50.5 60 87 54 39 55 42 43.4 36 44 48 58 33
Table 1. Pro-Netanyahu Vote vs. Psychological and Geographical Factors
in 15 Major Israeli Localities in 1996 Elections
At a glance, one can see extremely strong support for the Right near the GL and that it
gradually decreases going away from it. Of course, for those living in the settlements
beyond the GL, a vote for the Right was the obvious choice. Some irregularities versus
linear behavior also can be explained in light of what we learned earlier. The extremely
high percentage of pro-Right voters in Bnei Brak (15 km) could be explained by a
disproportional number of haredim (religious Jewish population). Another sharp rise in
percent of pro-Right voters in comparatively distant Ashdod (29 km) can be explained by
the ethnic component: Russian and Moroccan Jews comprise more than two-thirds of its
population. The results are shown as a graph (Figure 1):
21
Figure 1. Vote pro-Netanyahu vs. Distance to the Green Line for 15 Major Localities
Figure 1 is very telling. Drawing a straight line through Haifa (DGL = 40km, Right vote
in 1996 = 42%) and the center point of Greater Tel-Aviv31 (DGL = 20km, Right vote in
1996 = 48%) one can see the downward slope of about 3% per every 10km. A visible
jump upward of Jerusalem and the settlements near the GL are mainly due to the
religiosity factor, as we explained earlier, and we can even measure its intensity in
Jerusalem by comparing a projected 53% vs. 71% de facto.
Figure 1 also shows that in the 1999 elections, Netanyahu almost uniformly received 6%
to 8% fewer votes than in 1996, which confirms the existence of a firm slope that
depends on the DGL. This encourages performing the multivariable Linear Regression
analysis.
0 20 40 ……………..20km………………..…………... 40km
80% 60% 40%
Each city is represented by a pair of dots of the same color: dark for 1996 and bright for 1999. Color signifies location: Blue: sea Brown: mountain Green: valley Yellow: desert Black: Bnei Brak Dashed line: regression 1996 with slope –0.3 Dotted line: regression 1999 with slope –0.3
22
Historical Remark
Historically speaking, the same result could be discerned in the results of “split-ticket
voting,” when Israeli voters placed their personal concerns on the shoulders of local
municipalities rather than political parties, while the Knesset was likely associated with
the “global” security issue. The data (Arian, 1983: 186) for the 1965 and 1969 national
elections clearly point to greater concern for security near the GL (Jerusalem) than in a
comparatively secure locality (Ramat Gan). For example, in Jerusalem, the hawkish
Gahal party (future Likud party) received 28.3% to Knesset and 17.3% to municipality in
1965 and 28.3% and 16.9%, respectively, in 1969; while in Ramat Gan, the numbers
were reversed: 29.2% and 39.7% in 1965 and 29.7% and 39.6% in 1969. As expected, the
vote for the Alignment party (future Labor party) shows just the reverse relation between
Knesset and municipality votes. After discussing this fact, Arian (1983: 183) explained it
as the result of “party organization and electoral loyalties toward the mayors”32 and did
not proceed to analyze the geographical or psychological aspects of the facts.
6. Regression Analysis of Aggregate Data
A Five-Level Model
We used the linear regression program within the SPSS 10.0 for Windows Data Editor.
Regression analysis was performed in five steps. First were considered control variables:
1) demographic, 2) ethnic background, 3) religiosity, 4) social factors and only then were
added psychological and geographical factors. Were included 1,811 areas with Jews >
90%, and 1 km <DGL < 60 km. The R-square of the model reached 0.65-0.7 level but
behavior of F-statistic was not monotonic: it dropped at the fourth level and fell farther at
23
the fifth level. After adding social factors, many countries that represent Ethnic
Background became insignificant (North America, Germany, Bulgaria, etc.), so we chose
an alternative variant – to take the continent as a whole and pick the most noticeable
representative from every continent. Only at the fifth level did we add three variables
under investigation: DGL, Location by Sea, and DB. The form of the data in the Figure 1
resembles logarithmic function and suggests trying, not the linear distance DGL, but Log
DGL as a variable. All unmarked coefficients are at least *** significant. The factors that
did not feature significantly in the last two stages, like Gender, Academic Degree, DB
and Location by Sea, were entirely removed from the model.
1996 Elections
Category Variables I II III IV V
Constant 98.582 97.028 65.805 65.490 71.295 AGE25_44 -.816 -.946 -.372 -.314 -.320 AGE45_64 -1.362 -1.194 -.599 -.404 -.406
Demography
AGE65_74 -.176 .736 1.330 .814 .889 YEMEN 1.549 1.313 1.087 1.045
MAROCCO 1.246 .986 .637 .662 POLAND -2.733 -1.712 -1.062 -.973
Origin
LATIN_AMER -2.166 -1.569 -1.458 -1.528 PERSON1 -.312 -.475 -.513 Religiosity
PERSON7 1.553 1.272 1.201 POOR .108 .110** RICH -.163 -.160
CARS0 .078* .108* CARS1 .181 .211
Income
CARS2 -.295 -.290 Psychology LOG_DGL -3.048
R-square .103 .432 .576 .635 .644
F-statistics 69.5 196 272 223 216 Dependent Variable: Netanyahu_96
Table 2. Pro-Netanyahu Vote in 1996 Elections – Five-level Model
24
Demographic factors include Age and Gender. All the displayed Age groups’ votes were
*** (p < 0.001) significant; Age 74+ was a control group, while Age 18-25’ vote came
out insignificant. Gender appeared to be non-significant.
Ethnic background is represented by continents: Yemen for Asia, Morocco for Africa,
Poland for Europe, and Latin America for itself. All are **** (p < 0.0001) significant and
show the expected signs: the former two are pro Right, the latter two, pro-Left.
The Religiosity factor is represented by Person_1 and Person 7+ variables, which are
**** significant: the former has negative sign, the latter – positive. One can argue that
this not a surprise, since this factor correlates with the previous one, but the fact that the
ethnic background remains **** significant even after adding these two variables
strongly shows that another independent parameter exists. Their **** significance
ascertains that they grasp the essence of the Religiosity factor they represent.
Social factors are represented by Income and Academic Degree; the latter is insignificant.
The former splits into Poor and Rich and the number of cars in the family; people with 0
cars vote unclearly; the rest has expected signs of the coefficients: Poor and low middle
class (owners of 1 car) vote pro-Right, Rich and high middle class (owners of 2 cars) vote
pro-Left. F-statistics of the model drop after adding the social factor.
25
At the fifth level there were added the psychological and geographical factors: Log DGL
together with DB and Location by Sea. The first was *** significant while the other two
were insignificant. The coefficient -3 before the former shows that every 1 km farther
away from the GL, the Right loses 3/DGL % of votes; i.e., near Jerusalem, about 1%;
near Tel-Aviv, 3/20 = 0.15%; and near Haifa, 3/40 = 0.075% vote.
The R-square of the model reached 0.644 with final F-statistic at 216. Adding all the
countries for the Ethnic Background, not just several, improves R-square insignificantly
to 0.68, while the overall F-statistics drop from 216 to 148 (mainly because of the large
number of degrees of freedom = 26).
1999 Elections
Running the same set of variables against the variable Netanyahu_99, we obtained
practically the same results (Table 3). Again 1,811 localities were used, where Jews >
90% with 1 km < DGL < 60 km. The difference in Table 2 is that Polish background here
featured insignificantly and was dropped. We tried continents instead, but again, Europe
featured insignificantly and was dropped.
26
Category Variables I II III IV V
Constant 100.52 82.278 51.516 56.857 62.208 AGE25_44 -1.004 -1.079 -.460 -.417 -.421 AGE45_64 -1.554 -1.447 -.754 -.552 -.545
Demography AGE65_74 -.233 .220 1.008 .461 .544
ASIA .479 .364 .302 .287 AFRICA .477 .326 .227 .225
Origin
LATIN_AMER -1.308 -.880 -.996 -1.074 PERSON1 -.264 -.539 -.568 Religiosity
PERSON7 1.749 1.238 1.179 POOR .117 .118 RICH -.138 -.135
CARS0 .174 .198 CARS1 .116 .142**
Income
CARS2 -.369 -.366 Psychology LOG_DGL -2.676
R-square .145 .419 .594 .700 .707
F-statistics 101 217 329 322 309 Dependent Variable: Netanyahu_99
Table 3. Pro-Netanyahu Vote in 1999 Elections – Five-level Model
This time only the variable Cars_1 dropped in significance, while the overall picture is
practically the same: all the variables have the expected signs and are highly significant.
Again, F-statistics dropped at the fourth and fifth levels. Again, the DGL was ***
significant, while the DB and Location by Sea were insignificant. The coefficient -2.6
before the DGL shows that Fear was slightly reduced: every 1 km away from the GL, the
Right loses 2.6/DGL % of votes.
The R-square of the model reached 0.7 with final F-statistic at 309. Adding for Ethnic
Background the 16 countries of origin instead of continents does slightly improve R-
square to 0.75, but again causes final F-statistics to drop from 309 to 202.
27
Analysis of the Results
The fact that DB and Location by Sea are not significant might be interpreted in the
following way. The “center-periphery” problem might be considered non-existent in
Israel; this confirms the results of Andersen and Yaish (2003). There is no “periphery” in
Israel from a political point of view. Indeed, the Galilee (motherland of “periphery”
kibbutzim), “red” Haifa, and Northern Tel Aviv all voted for the Left, suggesting a
conclusive sample from the different regions of the country.
The fact that the DB is not significant shows that Israeli borders are not considered
dangerous by residents. This claim is supported by a separate regression analysis for
communities located within 20 km of the border, where all three variables feature as
insignificant. While this claim seems obvious with respect to borders with Egypt and
Jordan, the “calm” near the two other borders is remarkable. Though there was no peace
with Syria and Lebanon, in the eyes of the local residents, the Israeli occupation of the
Golan Heights and Southern Lebanon (until 2000) provided a reliable buffer against a
military threat. We strongly expect to see a rise in the significance of DB in the 2001
elections, after Israel retreated from Lebanon.
The fact that the same model works better for the 1999 than for the 1996 elections comes
as a surprise. Between 1996 and 1999, large numbers of new immigrants continued to
arrive in Israel, mainly from the former Soviet Union and Latin America, transforming
the ethnic background of some communities.33 In addition, this was the peak of a short-
lived economic boom, which changed income distribution in Israel in a volatile manner.34
28
8. Evidence from Individual-Level Data
The data from the poll conducted by the first author with his students in 2005 and
organized in electronic format as an Excel file by the second author was intended to
verify the results coming from the aggregate data. The individual poll has 1,215
respondents, consisting of persons from 127 localities, 36 of which are within the GL and
91 outside, in Israel proper. An advantage was to analyze the 2001 vote in addition to the
1996 and 1999 votes. We targeted the main socio-demographic characteristics that we
tested earlier: Religiosity, Gender, Age, and Academic Degree, together with the
psychological variable, DGL, and the geographical variable, the DB. We dropped the
Location at Sea since it strongly ** correlated to the latter two.
The percentages (rounded to the closest integer) are:
Number of Cars in 1996/1999/2001: 0 – 41/33/28%; 1 – 40/46/47%; 2 – 17/17/20%; 3 –
2/3/4%; 4 – 0.4/0.6/1.1%.
Number of Children in the Family in 1996/1999/2001: 0 – 37/34/29%; 1 – 10/11/12%; 2
– 16/15/16%; 3 – 15/16/17%; 4 – 11/12/13%; 5 – 6/6/6%; 6 – 2/2/2%; 7 – 2/2/2%; 8 or
more - 2/2/2%.
Number of Children in the Army in 1996/1999/2001: 0 – 77/73/75%; 1 – 17/19/19%; 2 –
5/7/5%; 3 or 4 – 1/1/1%.
Native-born in Israel - 56%; the rest had 32 countries of origin. 80% of our respondents
lived outside the GL (Israel proper); 25% lived by the sea. Average distances: to the GL –
11 km; to the Border – 54 km. Male participants amounted to 55% (52% among Jews),
29
while statistical data from the Israeli Interior Ministry Web site shows 49.5% males in
Israel (49% among Jews) in 1995-2005.35
Since it is known (see, e.g., Shalev and Kis 2002) that Israelis do not tell the truth when
asked about income in a poll, we substituted the number of cars in the family for the
Income variable. We thought that the number of children in the family and the number of
older children serving in the Israeli army might influence the individual vote, so we
included both numbers as independent variables.
The Religiosity variable acquired the meaning it always carries – strictness of
observance. We made it an ordinal with four values: 0 – secular (51%), 1 – traditional
(25%), 2 – orthodox (22%), 3 – ultra-orthodox (2%). The Central Bureau of Statistics
gives 45%, 38%, 10%, 7%, respectively, for the entire Israeli Jewish population in 2005.
The Academic degree variable also became ordinal with four values: 0 – high school
(56%), 1 – first degree (30%), 2 – second degree (11%), 3 – third degree (3%). The
Central Bureau of Statistics gives 58%, 22%, 20%, respectively, for the Israeli Jewish
population above age 15 in 2005; percentages for second and third degrees were
combined.
We have 95 Arab votes in the poll. All came from three cities, but all voted equivalently:
for the Left. Hence the variable Nationality (1-Arab, 0-Jew) was not meaningful, and as a
result, we excluded Arab voters from the regression analysis.
30
There remain 1,120 people between 22 and 86 years old who cast their vote at least once.
Rounded to the nearest integer for mean Age and median Age were: for all 1,120: 44/46y;
for those 1,082 older than 23y: 45/47y; for those 947 older than 26y: 48/49y.36 Hence,
after shifting back by 9, 6, and 4 years respectively, the true mean/median Ages were:
39/40y in 1996, 39/41y in 1999 and 40/42y in 2001. Israeli Interior Ministry statistics
show a median 28.5y, but over the total population, including the youngsters. Those
younger than 19 years old constituted about 37% of the total. Recalculating the averages
for eligible voters (18 years old and older), we found the median Age for the total falls in
the middle of the 35-44y old group, a good agreement with the data from our poll.
Again, as in the aggregate data analysis, the DGL variable was measured as the distance
to the closest point from Judea, Samaria, or Gaza. Since 90% of those participants living
within the GL were from the city of Ariel, and since the latter is not considered to be a
religious settlement, we decided to keep all of the entries in the analysis, setting the DGL
for them as zero. The negative feature of the poll data was that DGL variable correlated
with practically all other variables: Distance to the Border, Location at Sea, Religion,
etc., at the ** level. Therefore, we could not avoid the multicolinearity problem. Apart
from that, the results in all three elections show strongly the importance of the P-factor,
and, to lesser degree, the geographical variable.
Since the dependent variables Vote96, Vote99, and Vote01 are binary (1-Right, 0-Left),
we ran a binary logistic regression. There are many ways, however, to discuss its
31
goodness of fit – the best is to use the percentage of predictable outcomes. In all three
cases, elections in 1996, 1999, and 2001, the percentage of correct predictions was near
68% and R-square about 0.215. The results for individual variables are displayed in
Tables 4, 5 and 6.
Category Variables B S.E. Wald Sig. Exp(B)
Socio-Demography
Cars_1996 Child_Family Child_Army Acad_Degree
Age Gender
Religiosity Years_in_Israel
.328 -.043 -.082 -.041 .001 -.202 .349 -.002
.105
.053
.118
.095
.008
.150
.100
.005
9.773 .655 .481 .186 0.20
1.804 12.234 .128
.002
.418
.488
.666
.888
.179
.000
.721
1.388 .958 .921 .960 .999 .817
1.418 .998
Psychology DGL -.088 .009 88.070 .000 .916
Geography DB -.020 .006 10.289 .001 .980
Constant 2.200 .484 20.659 .000 9.021
Table 4. 1996 Elections; Individual Poll; Vote for the Right; 994 Respondents.
Socio-demographic variables contributed together about 10% to the initial 50%, while
Religion was the only *** significant variable. The cars variable had * significance. The
Age variable showed * significance only in 2001 elections (young people voted against
the Right), but not in 1996 and 1999, in agreement with the results of the 1996 and 1999
polls conducted by Shamir and Arian (1999: Table 2; The Elections in Israel-1999: 18-
19, Table 1.2). The variables Gender, Number of Children in the Family, Number of
Children in the Army, Academic Degree, and Years Spent in Israel do not become
significant even at the * level.
32
Category Variables B S.E. Wald Sig. Exp(B)
Socio-Demography
Cars_1999 Child_Family Child_Army Acad_Degree
Age Gender
Religiosity Years_in_Israel
.290 -.044 -.052 -.006 .007 -.233 .322 .007
.094
.050
.113
.091
.007
.140
.093
.005
9.626 .801 .214 .004
1.023 2.783
12.136 1.909
.002
.371
.644
.949
.312
.095
.000
.167
1.337 .957 .949 .994
1.007 .792
1.380 1.007
Psychology DGL -.083 .009 89.267 .000 .920
Geography DB -.020 .006 12.151 .001 .980
Constant 1.392 .422 10.876 .001 4.023
Table 5. 1999 Elections; Individual Poll; Vote for the Right; 1,081 Respondents
Adding the DGL variable in all three cases increased the result by about 7-10% – from
60% to 67-70%. Each time, this variable became significant at the *** level. The "odds
ratio" (the Exp(B) in the tables) equal to (for example) 0.92 case in 1999, suggests
diminution by 2% of the voters for Right for each km from the GL (since 0.92 = 0.48 /
0.52) which is remarkably consistent with the results of aggregate data analysis. Further,
adding the Distance to the Border variable in all three elections increased the result by
only a tiny 0.5-1%. The latter variable was significant at the ** or *** levels. Its
coefficient, equal to -0.02 in all three elections, was 4.4 times smaller in absolute value
than the one before the DGL in 1996, 4.1 times smaller in 1999, but only 3 times smaller
in the 2001 elections. This means that the P-factor gradually fell from 1996 to 2001, in
contrast to “security identified as a major issue before elections” (The Elections in Israel
– 2001, 16, fig. 2), which shows a significant drop of security concerns in 1999 that rose
anew in the 2001 elections.
33
Category Variables B S.E. Wald Sig. Exp(B)
Socio-Demography
Cars_2001 Child_Family Child_Army Acad_Degree
Age Gender
Religiosity Years_in_Israel
.208 -.038 -.134 -.050 .013 -.242 .380 .001
.083
.049
.107
.088
.006
.134
.092
.005
6.268 .614
1.568 .316
4.859 3.248
17.281 .070
.012
.433
.211
.574
.028
.072
.000
.791
1.231 .963 .875 .952
1.014 .785
1.463 1.001
Psychology DGL -.065 .008 70.228 .000 .937
Geography DB -.022 .005 16.077 .001 .978
Constant 1.313 .397 10.916 .001 3.718
Table 6. 2001 Elections; Individual Poll; Vote for the Right; 1,120 Respondents
Unfortunately, the data for the Ethnic variable (2-Sephardi, 1-Ashkenazi, 0-Arab) was
incomplete. Some 350 Jewish voters did not identify themselves properly: many may
have considered themselves as “mixed ethnic.” Hence, we dropped this variable from the
regression analysis. The straightforward bivariate correlation for 770 valid cases shows a
0.1** coefficient between the Ethnic variable and each Vote variable. This, of course,
means that Sephardim are more likely to vote Right.
The addition of the Ethnic Background variable did not improve the final score
significantly, only by about 1% each time. This seems strange at first glance; however,
the same effect for individual polls was observed by Shamir and Arian (1999: Table 2) in
their research on 1996 elections, where Ethnicity, which featured *** in the short (socio-
demographic) list of variables, lost one * after adding so-called “issues” and lost another
* after adding so-called “performances.” An even more striking effect was observed in
the 1999 elections (The Elections in Israel-2001, Table 1.2), where in the most general
34
model, the Ethnicity factor became only ** significant, but lost one * after adding issues,
and lost the last * after adding “performances.”
The correlation of individual votes in the 1996 and 1999 elections (those older than 26y,
including Arab votes) in our poll is 95.6%, which means that 4.4% of voters switched
sides, while in reality 6.6% voters changed their affinity to Netanyahu in the 1999
elections (from 50.5% in 1996 to 43.9% in 1999, see The Elections in Israel – 2001, 8).
The difference between 4.4% and 6.6% is not * significant.37
Further, our poll shows a 93.3% correlation of the individual vote in the 1999 and 2001
elections (those older than 23y, including Arab votes), which means that 6.7% of voters
changed sides. In the 2001 elections, Barak got 18.5% fewer votes, compared with the
1999 elections. However, in 2001 there was the lowest turnout for the elections ever
known: 2.7 million vs. 3.2 million in 1999 (The Elections in Israel-2001, 8). It is known
that many of Barak’s 1999 supporters, Arabs included, were reluctant to vote in 2001,
while our poll included only people who did vote in 1999 and in 2001. Despite this subtle
point, overall, we believe our poll represents fairly the entire Israeli population in its
diversity.
9. The Individual Poll vs. Aggregate Data
The poll highlighted the problems known as “ecological inference” regarding the Income
and Ethnic background variables.
35
1. The aggregate data analysis supports conclusions, derived by Shalev and Kis (2002,
Fig. 4.1) in a similar analysis, that the poor vote for the Right. The individual poll cannot
address this problem meaningfully since people do not report their income honestly. The
substitute for income – the Number of cars a family owns – in the poll shows a positive
coefficient in favor of the Right, but it is more likely that the Number of cars alone is
NOT an adequate measurement of Income.
2. According to the aggregate data analysis, the major role in the political choice of the
Israeli voter was played by his Ethnic background, or equivalently, by the way his/her
parents voted. This might imply that Israeli society is still a “gathering of exiles,” rather
than a cohesive new cultural entity. However, the individual-level poll did not support
this conclusion. The answer to this dichotomy (a new example of “ecological inference”)
comes from a well-known fact (Freedman 2001): people prefer to live near their social peers
rather than ethnic/cultural peers.
However, the aggregate data analysis highlighted a new fact: a simple division of the
Israeli population into Ashkenazim and Sephardim/Oriental is inadequate to discover
people’s political preferences. Turkish and South American Jews gravitate toward
European views, while North American Jews vote as do Oriental Jews. Religious fervor
is likely the most defining factor behind this division. When the poll was able to address
the latter factor comparatively adequately, the role of the Ethnic Background was
completely obliterated.
36
10. Major Conclusions: the Role of the P-factor
Summarizing, we addressed and partially solved several interesting methodological
problems, which for many years have been beyond the grasp of practitioners in the field.
The variable we introduced, the DGL, appears to capture the bias toward the Right of a
Jewish Israeli voter who resides near the GL, due to the obvious threat coming from
behind the GL. Among Palestinians, the level of (hypothetical) support for different types of
armed attacks remains high: 92% for attacks against soldiers, 92% for attacks against settlers, and
58% for attacks against civilians inside Israel.38
1. According to the aggregate data analysis for the 1996 and 1999 elections, the
psychological factor, as measured by the DGL, played a significant role in the outcome of
the 1996 and 1999 Israeli elections, though less significant compared with ethnic,
religious, and social factors. The quantitative claim is that for every 1 km away from the
GL inside Israeli territory, the Right loses 3/DGL% of the vote. A more refined
measurement should be found for cities inside the GL; for example, to the closest
dangerous place, or perhaps to the Security Fence.
According to the individual poll, the P-factor was *** significant in all three elections.
The "odds ratio" speaks of Right losing 2% of the vote every additional kilometer farther
from the GL. It was the most Wald-significant factor, several times more Wald-
significant than the geographical factor, Distance to the Border, and even more Wald-
significant than another well-known *** factor, Religiosity, though it is lower than the
latter in intensity. A scientist with the same income and color of yarmulke is more likely
37
to vote Right in Tel Aviv (20 km) than in Haifa (39 km). A non-religious dockworker is
more likely to vote Right in Ashkelon (11 km) than he would if he lived in Ashdod (29
km).
2. The analysis of the aggregate data involving two geographical variables (Distance to
the Border and Location by Sea), indicates that traditionally assumed center-periphery
bias is nonexistent in Israel, at least since 1996.
3. The direction of causation problem, which appeared along the way and is not
automatically resolved by doing regression analysis, was partially resolved by placing
certain restrictions on the DGL – excluding data for historic Biblical places within the GL
from the analysis.
11. Final Remarks and Open Problems
The P-factor came to the forefront in 1996 due to the suicide bombings in Jerusalem and
Tel Aviv in 1995-96, though likely it was implicit in all previous elections, as well. This
conclusion, expressing the importance of the P-factor, could be drawn from the results of
“split-ticket voting” of the past when Israeli voters placed their personal concerns on the
shoulders of the local municipalities, rather than political parties, and addressed their
psychological (security) concerns to the Knesset at large. The data presented in (Arian
1973) for 1965 and 1969 national elections clearly point to greater security concerns near
the GL (Jerusalem) than they do for a comparatively secure locality (Ramat Gan).
38
In September 2000, a new round of Palestinian terrorist suicide attacks interrupted the lull
of the late 1990s.39 The idea of constructing the Security Fence finally appeared on the
agenda in 2001. This is signaled by the reduction of the coefficient before the DGL, but
Ariel Sharon, then representing the Right, still got twice as many votes as Barak, on the
Left. Only the blind could miss the P-factor, but it would be important to check the
consistency of the coefficient before the DGL: (-3/DGL % per 1 km) with a reliable set of
control parameters.
Though by the next elections on January 28, 2003, the Security Fence had not yet been
completed, its construction was underway, supported by the entire Israeli political
spectrum.40 A feeling of increased security guaranteed another sweeping victory for
Sharon. Again, the importance of the P-factor would be of interest. We predict the
discovery of a drop of significance of the P-factor in the 2003 elections, and its possible
disappearance in the latest, March 2006 elections True, a more refined analysis might be
needed here, since the 2003 and 2006 elections lacked a major ingredient of our analysis,
the two-ballot system.41
The relevance of the “distance” parameter we have introduced here for the general
political or sociological discourse can be tested in the elections of countries where two
rival populations live in clearly defined areas, such as the French-speaking vs. the
English-speaking in Canada, or the Flemish and Walloons in Belgium.
Notes
42
∗ Corresponding author. E-mail [email protected]
39
1 This is the old approach in Israeli sociology; see Shamir and Arian 1983. According to Shamir
and Arian 1999: “Ethnicity effect is prominent as of 1977, but there is no clear trend.”
2 See also discussion about “cross cutting cleavages” in Israel At the Polls: 1996, 255.
3 Until 1996, the Israeli system was a one-ballot system with proportional party list representation
in the Knesset, a one-house parliament with 120 seats.
4 This is a well-known Israeli political misnomer; see, for example, Shamir and Arian 1999;
Andersen and Yaish 2003.
5 The Third Way was to give up the “territories,” although holding the Golan Heights, while
Israel Ba’Aliya did not clearly identify itself in the political spectrum (see Israel At the Polls:
1996, 136).
6 See, for example, its description in Ramanathan 2002 (ch. 12.2).
7 Andersen and Yaish (2003) included in the analysis the marginal parties, extreme Left and
extreme Right.
8 Like the one used by Shamir and Arian 1999, see also The Elections in Israel – 2001, 48-49.
9 Those who need to understand the vote in Arab sector may consult, for example, Israel At the
Polls: 1996, 103; The Elections in Israel – 2001, 55-103.
10 Until recently, every kibbutz had its own writer(s).
11 Even taking this as a hypothesis to verify the number of residents, a good quantitative variable
is a much better parameter.
12 In Model A, it is significant at 21% level; in Model B at 8.4% level.
13 This data seems unavailable by now; therefore, we had to measure it “by hand” for the poll by
measuring the distances with help of the Google Earth program. This caused some differences in
distances between the aggregate data and the poll data.
14 Either in arithmetic (-x) or multiplicative (1/x) sense. Here we discuss the former case only.
15 See some supporting evidence in the Report of the Special Committee to Investigate Israeli
Practices Affecting the Human Rights of the Palestinian People and Other Arabs of the Occupied
Territories (A/55/373/Add.1) (9 October 2000) at
http://domino.un.org/UNISPAL.NSF/5ba47a5c6cef541b802563e000493b8c/d6f6fd922cd766168
5256989005a3a80!OpenDocument 16 See, e.g., Internet website http://www.ynet.co.il/articles/0,7340,L-2227415,00.html of Israeli
newspaper Yediyot Achronot: settlers enjoyed 130 million NIS (c. $30 mln) in tax reductions in
2001; more particular: an average eligible tax payer settler in Judea and Samaria (total 34, 430
40
men) paid 6,456 NIS taxes less, in Gaza (total 1,890 men) – 8,934 NIS less than he would pay in
Israel proper.
17 “Peace Now” claims that 77% of settlers are those of “quality-of-life.” (see report at
http://www.peacenow.org/policy.asp?rid=&cid=3377). The figures in the report are not supported
by the table and clear-cut statistical analysis.
18 Foundation for the Middle East Peace reports, for example, that on December 13, 1996 (see
Web site at http://www.fmep.org/settlement_info/1996_settlement_timeline.html), the Netanyahu
cabinet approved the restoration of unspecified levels of benefits and subsidies to settlers and to
manufacturing, industrial, and commercial enterprises locating in settlements, which will now
enjoy “A”-level national priority area status. The Rabin government had earlier canceled or
reduced some of these incentives.
19 All of our predecessors excluded Arab voters from analysis for various reasons.
20 Participation in the 2001 Elections dropped to 62.3%.
21 We are grateful to Professor Gary King of Harvard for a discussion on this point.
22 True, if we knew the variances for every statistical area, it would be possible to infer some
statistical statements about individuals as well. Unfortunately, variances are not reported by
statistics, only averages. Still, though not reported, the variances might not be too large. People
prefer to live near their social peers. There are a few “mixed” communities, such as Ramat Gan,
but even in such a community within electoral areas the variance is likely to be small.
23 The first trio is now a standard reference point in every study; see, e.g., Friedlander et al 2002.
24 The Electronic File with the 1996 and 1999 election results was created by the group Mikum
and later sold to the Israeli Statistical Bureau. The number of electoral areas was higher than the
number of statistical areas, about 6,500 in number.
25 There were several more comparatively minor twists in the data, like special balloting urns for
soldiers and diplomats. Due to the small number of eligible voters among the Arabs in East
Jerusalem and the Druze villages on the Golan Heights, all of their electoral and statistical areas
were united in one. This does not affect our research since we are concerned with predominantly
Jewish areas.
26 The country of origin was assigned to the second generation according to how they identify
themselves, even though they were Israeli-born.
27 Israeli statisticians chose to record the number of years spent in study: 0-4, 5-8, 9-12, 13-15,
16+, which in fact prevents us from identifying the actual number of people holding BA/BS or
41
Masters degrees. We refer to the last group as holding BS degrees, though we certainly missed
some of the previous group, those listed with 15 years of study.
28 For Jerusalem, it was measured between the center of Jerusalem and the center of East
Jerusalem, whose status is uncertain.
29 Therefore, Jerusalem additionally lost 5 statistical areas out of 148; Tel-Aviv lost 27 (all of
Yaffo) out 156; while Haifa lost 26 out of 87.
30 This decision obliterates a more complex picture for communities beyond the Green line.
Jordan valley communities have always been seen, within the settlement movement, farther Left
than the communities closer to the Line, like Ariel and Ma'aleh Adumim.
31 This, in addition to Tel-Aviv proper, includes Ramat Gan, Holon, Bat Yam, Rishon Lezion and
probably Petach Tikwa and Rehovot.
32 Teddy Kolleck in Jerusalem and Abraham Krinitzi in Ramat Gan.
33 Governmental policy was to disperse new immigrants uniformly across the country.
34 In those years, the Israeli economy grew about 2% annually. Though the boom improved the
living standards of the entire population, redistributing the wealth through governmental
channels, the poor received their increase by artificial means, such as transfer payments to
religious institutions and the creation of government jobs, rather than making their fortunes by
starting new businesses. After the “bubble” collapsed, they became as poor as before.
35 See Web site for the Israel’s Ministry of Interior (Hebrew text): http://www.moin.gov.il
36 The standard deviation was 13-14 years.
37 The aggregate data show a ** 95.9%.correlation between votes in 1996 and 1999 elections.
38 Opinion poll # 3, conducted by the Palestinian Center for Policy & Survey Research, between
19-24 December 2001. The total sample size of this poll is 1357 Palestinians 18 years and older.
The margin of error is 3% and the non-response rate is 3%. See http://www.pcpsr.org/index.html.
39 The so-called “second Intifada.”
40 Despite strenuous and vocal resistance from the world community and the International Court
of Justice at The Hague, in particular. See, e.g., <http://www.conferenceofpresidents.org/fence.html>
41 The number of small splinter parties on the margin became so large in 1999 that in 2001, the
Knesset decided to return to the original one-ballot system.
42
42
References
1. Andersen R. and Yaish M. 2003. “Social cleavages, electoral reform and party choice: Israel’s
‘natural’ experiment.” Electoral Studies 22:399-423.
2. Arian A. 1973. The Choosing People. The Press of Case Western Reserve Univ. Cleveland
and London.
3. Electronic File with Results of Israeli 1996 and 1999 Elections and 1995 Census. 2003. State
of Israel, Central Bureau of Statistics, Jerusalem.
4. Freedman D.A. 2001. “Ecological inference and the ecological fallacy.” International
Encyclopedia of the Social and Behavioral Sciences. Oxford: Elsevier Science.
5. Friedlander D., Eisenbach Z., Ben-Moshe E., Lion-Elmakias L., Hleihel A., Lunievski S., and
Ben-Hur, D. 2002. Changes in Educational Attainments in Israel since the 1950s: The Effects
of Religion, Ethnicity and Family Characteristics. Hebrew Univ. of Jerusalem and Central
Bureau of Statistics.
6. Israel At the Polls: 1996. 1998. Edited by Elazar D.J. and Sandler S. London, Portland, OR:
Frank Cass Publishers.
7. De Marchi S., and Goemans H. 2001. “Bargaining and complex preferences: examining the
case of the Israeli electorate.” Durham, N.C.: Duke Univ. Paper presented at the Annual
Meeting of the American Political Science Association, San Francisco, 2001. Preprint.
8. Ramanathan R. 2002. Introductory Econometrics with Applications, 5th ed., Mason, Ohio:
South-Western, Thompson Learning.
9. Shamir M and Arian A. 1983. “The ethnic vote in Israel’s 1981 elections.” Electoral Studies 1:
315-331.
10. Shamir M. and Arian A. 1999. “Collective identity and electoral competition in Israel.”
American Political Science Review 93: 265-277.
11. Shalev M. with Sigal Kis. 2002. “Social Cleavages among non-Arab voters. A new analysis.”
In: The Elections in Israel – 1999. Arian A. and Shamir M., eds. Tel-Aviv: The Israel
Democracy Institute, pp. 67-96.
12. Smooha S. 1978. Israel: Pluralism and Conflict. London: Rutledge & Kegan Paul Ltd.;
Berkeley and Los Angeles: Univ. California Press.
13. The Elections in Israel – 2001. 2002. Arian A. and Shamir M., eds. Tel-Aviv: The Israel
Democracy Institute.