¹ Only one scientific survey of voters has been reported on previously; see Arian, Alan, “Electoral Choice in a Dominant Party System,” in The Elections in Israel–1969, ed. Arian, Alan (Jerusalem: Jerusalem Academic Press, 1973), pp. 187–201Google Scholar; Arian, Alan, The Choosing People (Cleveland: Case Western Reserve University Press, 1973)Google Scholar. The published analyses suffer from substantive and methodological flaws; for example, important variables, including sector of the economy, are ignored, analyses are based for the most part on bivariate tables with no calculation of measures of association, and the refusal rate concerning party choice was 30 percent.

² The selection of independent variables was influenced by the work of Lipset, Rokkan, and others, who show the widespread importance of such variables and provide a comparative framework into which this work may fit; see Lipset, S. M. and Rokkan, Stein, “Cleavage Structures, Party Systems, and Voter Alignments: An Introduction,” in Party Systems and Voter Alignments, ed. Lipset, Seymour M. and Rokkan, Stein (New York: Free Press, 1967), pp. 1–64Google Scholar; Rose, Richard and Urwin, Derek W., “Persistence and Change in Western Party Systems since 1945,” Political Studies, 18 (September 1970), 287–319CrossRefGoogle Scholar.

³ See, e.g., Goodman, Leo A., “A General Model for the Analysis of Surveys,” American Journal of Sociology, 77 (May 1972), 1035–86CrossRefGoogle Scholar; “Causal Analysis of Data from Panel Studies and Other Kinds of Surveys,” American Journal of Sociology, 78 (March 1973), 1135–91CrossRefGoogle Scholar; and “The Multivariate Analysis of Qualitative Data: Interactions Among Multiple Classifications,” Journal of the American Statistical Association, 65 (March 1970), 226–56CrossRefGoogle Scholar; Davis, James A., “Hierarchical Models for Significance Tests in Multivariate Contingency Tables: An Exegesis of Goodman's Recent Papers,” in Sociological Methodology 1973–1974, ed. Costner, Herbert L. (San Francisco: Jossey-Bass, 1974), pp. 189–231Google Scholar.

⁴ Knoke has used the methods to analyze party choice in the United States, but uses only dichotomous variables; see Knoke, David, “Religious Involvement and Political Behavior: A Log-Linear Analysis of White Americans, 1952–1968,” Sociological Quarterly, 15 (Winter 1974), 51–65CrossRefGoogle Scholar.

⁵ E.g., Alford, Robert R., Party and Society (Chicgo: Rand McNally, 1963)Google Scholar; Hamilton, Richard F., Class and Politics in the United States (New York: Wiley, 1972)Google Scholar.

⁶ This argument draws on: Lipset and Rokkan, “Cleavage Structures”; Murphy, Raymond J. and Morris, Richard T., “Occupational Situs, Subjective Class Identification, and Political Affiliation,” American Sociological Review, 26 (June 1961), 383–92CrossRefGoogle Scholar; Janowitz, Morris and Segal, David R., “Social Cleavage and Party Affiliation: Germany, Great Britain, and the United States,” American Journal of Sociology, 72 (May 1967), 601–18Google Scholar; and especially Lichtheim, George, “Class and Hierarchy: A Critique of Marx?” European Journal of Sociology, 5 (1964), 101–11CrossRefGoogle Scholar. Relationships between economic groups are non-hierarchical when there is no obvious way to order them relative to each other in terms of “less” and “more” or “higher” and “lower”; economic sectors as used in the census fit the definition.

⁷ Lichtheim; the possibility of economic conflict involving nonhierarchical dimensions is an implied criticism of Dahrendorf's model of conflict; see Dahrendorf, Ralf, Class and Class Conflict in Indust Society (Stanford: Stanford University Press, 1959)Google Scholar.

⁸ Burstein, Moshe, Self-Government of the Jews in Palestine since 1900 (Tel Aviv, 1934)Google Scholar; Eisenstadt, S. N., Israeli Society (New York: Basic Books, 1967)Google Scholar.

⁹ Although the impact of sector on political behavior has been discussed informally in work on Israel, neither Arian's work nor any other actually measures its impact.

¹⁰ See, e.g., Medding, Peter Y., Mapai in Israe (London: Cambridge University Press, 1972), Ch. 3Google Scholar.

¹¹ In the 1973 election, Gahal joined forces with two other right-of-centei parties to form Likud.

¹² E.g., Janowitz and Segal, “Social Cleavage and Party Affiliation”; Arian, Choosing People; Kies, Naomi, “Constituency Support and the Israeli Party System” (Ph.D. dissertation, M.I.T., 1969)Google Scholar.

¹³ Other methods, such as multiple discriminant function analysis, could be employed, but they too present problems. In multiple discriminant function analysis, for example, prediction of party choice is possible, but substantive interpretation of the coefficients is difficult because the predictor variables are not the original variables but rather linear combinations of them. See, e.g., Burstein, Paul, “Social Structure and Politics in Israel” (Ph.D. dissertation, Harvard University, 1973)Google Scholar.

¹⁴ See Goodman cited in footnote 3. Log-linear contingency table analysis is relatively new, and competing but very similar systems also exist; see Nerlove, Marc and Press, James, Univariate and Multivariate Log-linear and Logistic Models (Santa Monica: RAND, 1973)Google Scholar.

¹⁵ E.g., James A. Davis, “Hierarchical Models”; Davis, James A., “The Log Linear Analysis of Survey Replications,” in Social Indicator Models, ed. Land, Kenneth C. and Spilerman, Seymour (New York: Russell Sage Foundation, 1975), pp. 75–104Google Scholar.

¹⁶ Contrary to usage in some areas, variables are said to “interact” whenever they are related; thus, variables can interact in a two-way table.

¹⁷ Conceptually, the procedure is somewhat analogous to the elimination of possible causal paths in path analysis, where one aim is to determine which paths can be set to zero while still permitting reproduction of the original coefficients; see Leo A. Goodman, “Causal Analysis of Data”; Goodman, Leo A., “The Analysis of Multidimensional Contingency Tables When Some Variables are Posterior to Others: A Modified Path Analysis Approach,” Biometrika, 60 (1973), 179–92CrossRefGoogle Scholar.

¹⁸ Davis, “Hierarchical Models,” “Log Linear Analysis.”

¹⁹ It is theoretically possible to perform analyses comparable to path analyses, but doing so with polytomous variables is extremely laborious and complicated, and probably not worth the effort in the data set used below; see Goodman, “Causal Analysis”; “Analysis of Multidimensional Contingency Tables.”

²⁰ The Alignment is the modal choice of every group; it is merely relatively strong or weak in different groups in the sample. More detailed data including a finer breakdown of occupations and data for all parties do not contradict Table 1; both occupation and sector are related individually to party choice at better than the .001 level.

²¹ The 18 models consist of every logically possible combination of effects which would cause the cell frequencies to be unequal to each other, including the effects of the marginal distributions, of two-way relationships, and of higher order relationships, and all possible combinations thereof, plus the null model in which all cell counts are equal. (The number of models thus increases very rapidly with the number of variables; a four-variable system includes over a hundred possible models; obviously substantive considerations should be used a priori to select among the possibilities.) We eliminate ten possible models on the basis of our a priori knowledge that cell counts are not expected to be equal to each other, by our lack of interest in effects which are caused merely by skewed marginals on one or two of the variables, and by our lack of substantive concern with relationships not involving party choice.

²² Confirming the impressions derived from the significance testing of various models, no single relationship between party and occupation is strong, and the relationship between sector and party choice is not changed when the model including occupation is examined.

²³ That is, where the coefficient is significantly different from zero at the .05 level or better.

²⁴ It might be helpful at this point to make comparisons between the analyses presented and results found using more conventional techniques. Space limitations preclude detailed presentation of findings, but analyses parallel to all those presented here were carried out using dummy variable regression and multiple discriminant function analysis. Results of the regression analyses were not inconsistent with those described in this article; for example, occupation is not significantly related to party choice once sector is controlled, in four of the six equations (one equation for each combination of categories of the dependent variable).

However, the use of log-linear contingency table analysis gives the researcher three advantages over the user of standard regression techniques, two substantive and one practical. First, the Goodman system enables one to deal with all relationships as a system; one does not have a set of independently calculated regression equations where one has to guess about the relationships among them, but rather an interrelated system of equations. Second, it is easy to determine the nature of interactions among variables using the Goodman system, while it is extremely cumbersome to do so using ordinary regression analysis. Finally, from a practical standpoint, the Goodman system is extremely easy to use, once the model-testing features are understood; testing alternative models using regression analysis is extremely cumbersome by comparison.

Thus, it would have been difficult, though not impossible, to test for the specific interactions hypothesized above among sector, occupation, and party choice. Where many hypotheses involve interactions, as is the case throughout this paper, the Goodman system easily enables one to test hypotheses that would be extremely difficult to test using ordinary regression techniques.

²⁵ Matras, Judah, Social Change in Israel (Chicago: Aldine, 1965), Ch. 3Google Scholar; Smith, Herbert, “Analysis of Voting,” in Arian, , Elections in Israel, pp. 63–80Google Scholar; Lissak, Moshe, “Continuity and Change in the Voting Patterns of Oriental Jews,” in Arian, , Elections in Israel, pp. 264–77Google Scholar; Lissak, Moshe, Social Mobility in Israeli Society (Jerusalem: Israel Universities Press, 1969)Google Scholar; Arian, Choosing People, Ch. 3.

²⁶ It would be preferable to consider all the variables simultaneously, rather than in a series of three-variable systems, but the small number of cases precludes doing so.

²⁷ That is, chi-square for the model (PE)(PA)(EA) is 11.20 and that for the best-fitting simpler model, (PE)(EA), is 29.50. The difference is 18.30, which is significant at the .005 level with six degrees of freedom. The procedure for the other three-variable system is analogous.

²⁸ The data provide no evidence as to the possibility that differences between entire ethnic groups have been changing between elections, of course.

²⁹ The three-variable system ethnicity-party choice-sector was also analyzed. Although ethnicity is related to nativity, period of coming of age, occupation, and religiosity, it is not related to sector;hence, nothing of use could be learned from the system involving sector.

³⁰ See, e.g., Birnbaum, Ervin, The Politics of Compromise: State and Religion in Israel (Rutherford, N.J.: Farleigh Dickinson University Press, 1970)Google Scholar.

³¹ See, e.g., Liepelt, Klaus, “The Infra-Structure of Party Support in Germany and Austria,” in European Politics, ed. Dogan, Mattei and Rose, Richard (Boston: Little, Brown, 1971), pp. 183–202Google Scholar; Lijphart, Arend, The Politics of Accommodation (Berkeley: University of California Press, 1968), Ch. 2Google Scholar.

³² Arian, “Electoral Choice.”

³³ The statistical technique used by Arian—AID—and his choice of a dependent variable—voting for the Alignment as opposed to all other parties—makes a more precise statement impossible; see Arian, Choosing People, Chs. 3–5.

³⁴ Arian, Elections in Israel, Choosing People.

³⁵ See Arian, , Elections in Israel, p. 188Google Scholar; Laumann, Edward O., Bonds of Pluralism (New York: Wiley, 1973), p. 18Google Scholar.