A representative sample of more than three thousand members of a German political party were asked in which of eleven party related activities they had been involved. A one-dimensional Guttman scale was found with a maximal e = 17% or 536 violations in a declared zero cell. The construction and evaluation of this scale is described in some detail. We interpret the scale as ordering the activities with respect to the public commitment required on behalf of the party.
A two-dimensional MSA, e < 8%, for the same data found a solution with a fit of f = .979. As always, the first dimension counts the number of items endorsed, interpreted as the amount of involvement for the party. We interpret the poles of the qualitative dimension as 'demonstrating loyalty' versus 'support for the institution . The paths in the Hasse diagram could be given a temporal or biographical meaning (despite the fact that the data are not longitudinal) as different ways or careers from "easy" activities to those showing a very strong identification with this party.
Key words: Guttman scaling, one- and two-dimensional scalogram analysis, participation in political activities, identification with political parties
Introduction to the research problem
Modern representative democracy is unthinkable without political parties. They function as a democratic link between society and government. Rooted in society, political parties articulate the interests of social groups; organized in governmental institutions, they aggregate conflicting interests to political binding decisions. Political parties can fulfill their democratic function only if an active citizenry is willing to engage in them permanently, bringing this democratic link to life. During the last two decades the ability of parties to serve these functions has come under severe stress. The number of citizens interested in joining a political party has steadily declined; and the party members are less inclined to engage in party activities. At the same time, we saw an increase in citizen action committees and social movements, especially in the younger population as the new generation of party members. These diametrically opposed trends have been characterized as de-institutionalization of political participation, indicating a growing tendency of citizens to restrict their participation to a short term engagement and to single issues.
In this problem definition it was the aim of a large-scale party membership study of the Christian Democratic Union of Germany (CDU) to analyse the motivations for intra-party participation and the reasons for their decline (for details see Veen & Neu 1995). In the broader context of this study the specific aim of the following analysis is to elaborate the relationship between various modes of intra-party participation: Do different modes of party activity constitute separate dimensions, or are the "easier" forms connected to the more "difficult"? Could thus the preference for short-term and single-issue participation lead to an engagement on a permanent basis? Could, for example, members attending only party meetings also be expected to stand for elected office in the party organization or in a local or national parliament?
Methodologically this translates to the question whether the various forms of party activity are organized unidimensionally, being structured in a cumulative, hierarchical order (Milbrath 1965, Falke 1981). The alternative hypothesis states that the different modes of participation in the general public (Verba, Nie & Kim 1978; Verba, Schlozman & Brady 1995; Dalton 1996) could be separated empirically for intra-party participation, too.
For the test of these hypotheses, we could draw on the national representative sample of 3411 members of the CDU of Germany. The fieldwork of this standardized face-to-face interview was completed in 1993 in both the Old and New Bundeslander; the sampling institute was BASISRESEARCH. This study aimed to analyze the motives of members to join the party and to engage in various party activities, their modes of communication and information behavior, their evaluation of intra-party-democracy, and their support for party reforms. The variables on which we based our Feature Pattern Analysis were designed as dependent variables for the complex "party activity". In the interview, all respondents were asked to report whether they have participated in the following party activities (abbreviations in italics)3:
This distribution gives rise to the hypothesis that we should expect several groups of party activities differentiating the party membership. This is the starting point of our analysis.
Editing the data
While the number of participants of the total sample is 3411 we reduced this sample to those 3150 respondents without any missing data. FPA can be performed by just ignoring a vector with missing data leading to increased variability in the number of cases for the consistency tables (Feger, this volume). For a sample of the size as the one used here this would require a special program. But given our sample size we expect the same results.
Some general comments on the method of analysis
Feature Pattern Analysis (FPA, Feger 1994) is a general model to analyze response vectors generated by one or several sources. The variables recorded in the vector may be of any scale level, and the categories of items may be dichotomous or polytomous with or without any order defined over the categories. The main area of application is to attitude scaling, grid technique, semantic differential and similar procedures in the behavioral and social sciences where discrete observations of a low scale level are not infrequent.
The FPA model tries to decompose every response vector into contingencies. The analysis usually begins by using the contingency of the lowest possible order, i. e., bivariate. If the researcher decides that the result of the decomposition is not good enough contingencies of the next higher order may be tried.
A special case of the FPA model is the well known Guman scale. A Guttman scale is a one-dimensional structure ordering the items with increasing level of 'difficulty'. If a more difficult item is passed (agreed to etc.) then all items with lower difficulty should show the same response.
This structural definition implies a certain distribution of the responses over the cells of contingency tables. Every pair of dichotomous items defines a contingency table with four cells, the well known fourfold table. This table will be shown for the items A and B with the frequencies of our sample:
If the items A and B are a part of a Guttman scale, either A should be more difficult than B, or B more difficult than A. In our context, the difficulty of an item refers to the number of respondents indicating the behavior mentioned in the item. From the marginals we see that B is the more difficult item because only 1672 report bumper stickers while 1945 persons state the activity leaflets. Then, in a Guttman scale, there could be persons performing neither the activity of item A nor of B. The fourfold table reveals that there exist 945 members refraining from both activities. There also could or should be persons performing both activities, as 1412 respondents report. Some persons may perform the easier activity; there are 533 respondents with the answer pattern A = 1 and B = 0. However, there should be no person performing the more difficult activity but not the easier one. This implies that the cell A = 0 and B = 1 should show a frequency of zero, or, in technical terms, should be a zero cell. Contrary to our expectation, 260 responses are counted in this cell, or 8.25% of the sample analyzed here. The researcher could stop the analysis here deciding that this percentage is too high. As a matter of fact, the researcher should state in advance the value of the error threshold e indicating the maximum of error tolerable in this analysis.
It would not be fair to judge the fit of the solution by the small number of 1053 patterns explained because this number was not the criterion used to construct the solution. And such a criterion would not convince us because it would neglect the large amount of information contained in those patterns not accommodated. Our solution is the one giving the best representation of all bivariate contingencies in all patterns. An impression of the overall goodnessof-fit might be gained by the following calculation. With 3150 Ss and 55 contingency tables 173,250 observations exist. The sum of all errors (= sum of all cells in Tab. 2) is 12092 or 7% of all observations. But, after having emphasized this aspect, we nevertheless are motivated by the small number of patterns explained to try other solutions as well.
A tentative interpretation sees the items ordered as requiring increasing public commitment of the respondent in favor of the party. Performing the behavior described in the more difficult items means to expose oneself politically in front of others. The more difficult items refer to behavior allowing a person to be identified as being a convinced member of the CDU. At the time period in which the sample was taken the aspect of public commitment perhaps increased in importance because at that time the political parties in general had a very poor standing in Germany.
The next task is to evaluate this solution; the researcher has to decide whether the result is good enough. This can be done by using different criteria. As a first criterion we select the number of violations of the zero cell requirement. The solution chosen had to allow e = 17% or 536 violations (cases in the declared zero cell) as the maximum. As can be seen, the A x B table exhibits 260 violations. On the average, we observed 220 cases, corresponding to f = .93. Bassler (this volume) performed Monte Carlo studies with random data to evaluate f. For a one-dimensional Guttman scale with 7 items - the largest number of items studied thus far a mean f = .919 with a standard deviation of .028 was found. It seems to be well justified to expect an average f for 11 items to be somewhat lower, perhaps close to .9. Then, assuming approximately the same standard deviation, one could conclude that our Gunman scale is acceptable but not very good.
At least as important as testing the overall fit is the analysis of systematic deviations. For example, one might search for "poor" items which by their content or wording create irrelevant responding behavior. Here, we test the assumption that the number of errors, i. e., the cases in the declared zero cells, decreases with increasing distance between the items ordered according to their difficulty. Tab. 2 reports the error frequencies in the 55 declared zero cells.
The rows and columns of this matrix are ordered with increasing item difficulty. Two trends of the entries in this matrix may be noted: (1) The last row reports (in italics) the average number of the frequencies for those ten zero cells in which the column item is a part of the definition of these zero cells. E. g., for all cells with D, the average is 101.7 cases. As can be seen, the items with a position in the middle of the scale show a higher number of errors than the very easy and the very difficult items. When reacting to items in the middle of the scale, the respondents tended to react more inconsistently.
This reaction tendency has been observed with other persons, topics, and scaling methods as well. It is at least partially due to the fact that in the center of the scale more combinatorial possibilities exist (resulting from random permutation) than for categories at the ends of the scale. This does not rule out the possibility of an substantial interpretation. One possible interpretation is that the respondents categorize the party related behavior basically into three categories. This classification may be imposed on other categories that are more related to the content of the activities. A first category, covered by the items with low difficulty, pertains to behavior that easily could be performed. Abstaining from this behavior is highly visible in the membership of the party. The second category is that of behavior which is not extreme, and members are at liberty to show it or to avoid it without raising any specific reactions in general. Then this behavior may not show a very differentiated cognitive structure resulting from more intense discussion among members. Finally, the most difficult items draw the attention of other members; reporting an activity out of this category is a very conscious act.
(2) The second trend is revealed by comparing the values in the main diagonal with those in the adjacent diagonals. The average frequencies turn out to decrease from the main diagonal to the last value in cell DK = 37. The steepness of decent is impressive and implies that the second trend is by far the stronger one. The averages are:
From a method oriented point of view, we have obtained a measuring device leading to one or two personal scores which should be definitely more reliable and general than a single item. If only for this increased reliability alone, the relationship to other variables under study should be more clearly revealed. Guttman scaling provides information on the level of an ordinal scale. One may be tempted to use the results of the Thurstone-like scaling of the error matrix given in Tab. 2. This nonmetric procedure leads to interval scaled item values which could be used to derive individual scores. But we have no experience in this respect. On the other hand, ordinal scale values distinguishing between 12 levels of party engagement provide a sufficient degree of differentiation for most practical purposes.
3 The standardized response categories were: "have done and would definitely do again/ have done and would possibly do again/have done and would definitely not do again/have not done but would definitely do/ have not done and would possibly do/ have not done and would definitely not do". According to our interest to analyse the structure of actual behavior, these categories were dichotomized in "have done" (1) and "have not done" (0).
1. Bissler, C. (this volume).
2. Barnes, S., Kaase, M. et al. 1979. Political Action. Mass Participation in Five Countries. Beverly Hills: Sage.
3. Dalton, R. J. 1996. Citizen Politics. Public Opinion and Political Parties in Advanced Western Democracies. Chatham: Chatham House Publishers.
4. Falke, W. 1981. Die Mitglieder der CDU. Eine empirische Studie zum Verhaltnis von Mitglieder- and Organisationsstruktur. Berlin: Duncker & Humblot.
5. Feger, H. 1994. Structural Analysis of Co-occurrence Data. Aachen: Shaker.
6. Feger, H. (this volume). Some applications of FPA in the social and behavioral sciences.
7. Feger, H. & Klauer, K. C. (this volume). Essentials of Feature Pattern Analysis.
8. Lingoes, J. C. & Borg, I. 1979. Identifying spatial manifolds for interpretation. In: Lingoes, J. C., Roskam, E. E. & Borg, I. (eds.): Geometric Representations of Relational Data. Ann Arbor, Mich.: Mathesis Press, pp. 127 - 148.
9. Milbrath, L. W. 1965. Political Participation. How and Why do People Get Involved in Politics? Chicago: Rand Mc Nally.
10. Niedermayer, 0. 1989. Innerparteiliche Partizipation. Opladen: Westdeutscher Verlag.
11. Thurstone, L. L. & Chave, E. J. 1929. The Measurement of Attitude. Chicago: Univ. of Chicago Press.
12. Verba, S., Nie, N. H. & Kim, J. 1978. Participation and Political Equality. A Seven Nation Comparison. Cambridge: Cambridge University Press.
13. Verba, S., Schlozman, KL. & Brady, HI E. 1995. Voice and Equality. Civic Voluntarism in American Politics. Cambridge. Harvard University Press.
14. Veen, H.-J. & Neu, V. 1995. Politische Beteiligung in der Volkspartei. Interne Studien der Konrad-Adenauer-Stiftung, Nr. 113/1995, Sankt Augustin.
15. Wilkinson, L. 1990. SYSTAT: The System for Statistics. Evanston, Il.: SYSTAT, Inc.
WILHELM BURKLIN1, HUBERT FEGER2
1 Dr. Wilhelm Biin, Wirtschafts- und Sozialwissenschaftliche Fakultat, Universitat Potsdam, Am Neuen Palais 10, D-14468 Potsdam
2 Prof. Dr. Hubert Feger, Department of Psychology, Free University of Berlin*, Habelschwerdter Allee 45, D-14195 Berlin, Germany…