The question of whether urbanism and urban life have any political implications be asked in many different ways. The effects of urban environments to some extent be seen and analysed as a contextual effect. More than 50 years ago Herbert Tingsten (1937) was able to show how certain voting districts within Stockholm had a comparatively high percentage of Social Democratic votes, while other districts behaved differently according to the relation between the proportion of industrial workers and votes for the Social Democrats. This well-known pattern has been explained as a group effect caused by different norm systems among people belonging to certain strata.

This theory is basically in line with the notion that party attachment is primarily determined by social class and by influences from occupational and other groupings. According to this theory there is a more or less direct link between the group measured by social characteristics and the vote. From this basic theory of norms and group cohesion various subtheories have also been derived, for instance the theory of cross-pressure voting (Lazarsfeld et al. 1944). Other theories related to groupings take a point

Resumé

Using confirmatory factor analysis of nine survey perception items the author tests hypotheses derived from general theories in the field of electoral behavior - regarding fundamental latent structures of perceptions of party choice and party attachment in four types of Swedish environment defined by different degrees of urbanization. Significantly different structures have to be specified indicating greater group-dependent affective attachments in rural areas and attachments of greater dependence on the party-program contexts in urban districts. In the second and third sections of the paper these results are further explored through the construction of dichotomized typologies of voters. Loglinear analysis reveals that differences between environments, as with typologies, can be fully interpreted and explained by demographic educational differences in the composition of the electorate. Provided that the different composition has been brought about by general urbanization, the results seem to indicate the political effects of social change that can be further emphasized in the future.

of departure in Heider's theory of cognitive balance (Butler & Stokes 1974), or put special emphasis on basic cleavages within societies, cleavages that are knit to various group antagonisms and that are formed across different lines, social classes as well as religious groupings. (Lipset et al. 1967). According to this tradition, mainly a European one, these cleavages form the basis for the party system.

A quite different type of theory is based on the importance of political issues. Anthony Downs is often considered as a founder of models based on rational choice (Downs 1957). Politican man is thought of as a consumer of public policies, just as party strategists are considered as salesmen competing on a market. Although studies in the early 1960s gave very vague results according to predictions from this sort of theory (Converse 1964), many researchers today are willing to accept issue voting as a reality (Nie et al. 1976; Niemi & Weisberg 1984). Quite a few are also willing to accept the notion of ideology or value orientation as valid for at least certain societies (Inglehardt 1975).

Finally I would like to mention a third theory that might be regarded as a separate class: the Michigan school with its stressing of an affective attitudinal component directly knit to political parties. Important concepts derived form this theory are party identification, the normal vote, and attitudinal reactions towards candidates (Campbell et al. 1960).

As in many European countries, the Swedish party system is obviously closely related to social class and occupation, but more so in the past than is the case today. Taken from left to right the Communists some decades ago recruited working people in the big cities and northern parts of the country. Today it is less regionally based and is much less a working person's party.

The Social Democrats - 44.7 percent in the last election² - are basically a party of labor, although the party today incorporates substantial parts of the middle class and middle-range civil servants. In the middle of the system two parties compete. The Center Party, formerly the Farmers Party, has a basis in rural districts among farmers and the Peoples Party has a basis among white-collar workers in the private sector and among smaller entrepreneurs. Conservatives are the industrial people, private employers and employees.

This conventional picture is growing more conventional as time goes by. As with many European countries, electoral stueies show a progressively less clear-cut picture in which traditional political boundaries between occupational categories are fading out (Peterson 1977; Holmberg 1981). Class voting in the objectively defined categorical way declines, and party identification also declines slightly. As an alternative to traditional theories, models of issue voting and rational choice emerge (Borre et al. 1983; Holmberg 1984).

It is in this context that I would like to draw attention to a small data set
compiled in a non-academic environment. From data collected in spring
1985 I have chosen the following items:3

'Are the following of any consequences for your choice of party?'
(1) The Party leader?

(2) The Party's representative at the site?

(3) What the Party wants to achieve, the political goals?
(4) The Party's ideas on a single issue?

(5) That the Party can put force behind its promises?
(6) A sense of allegiance to the Party?
(7) How your family votes?
(8) Affiliation with union?

(9) The occupation you have, the social class to which you belong?

Each item could be answered in the following way:

Yes, absolutely
Yes

Ambivalent

No

No, absolutely not

In a first section of this paper I intend to use this data to explore whether the same factor structures lie behind these items in four different social environments in Sweden, namely within the population from rural areas, from small towns and agglomerations, from bigger cities, and from the three largest cities, Stockholm, Göteborg, and Malmö, here labeled metropolises. These four environments will somewhat inaccurately be referred to as the rural/urban dimension.

In a second section I intend to use two factor indices built from some of these items as instruments for the construction of a fourfold typology of voting patterns and use this typology to explore its relation to two other categorical variables, the party system and the rural/urban dimension. Finally, it is my intention to elaborate the relationship between the rural/ urban dimension and the typology of voting patterns.

Urbanism and Voting Perceptions - A Confirmatory Factor Analysis

Having presented the main topic of the paper and the data to which I have access, I now turn to the more specific problem of defining reasonable assumptions for this section. From a statistical point of view we obviously have to start by hypothesizing the presence of identical structures in the

four cultural settings. If the analysis does not support this assumption we can pursue the analysis in a tentative way by trying to find reasonable explanations. From a substantive point of view it would after all be more reasonable to hypothesize the opposite.

One reason for this is obvious. As has already been pointed out, the Swedish party system still has its bases in different types of social strata which in turn are located differently with respect to a rural/urban dimension. we add to such a fact the assumption that latent structures behind these items are in one way or another also related to the party system, it is reasonable to believe there should be different perceptions in different social environments. For the moment this short reflection will suffice and I will save some others for later in the paper.

Using a submodel from LISREL VI we specify the same starting model in the four subsamples, a model that can be described by the matrix below. In this matrix the three columns will indicate hypothesized latent structures and the figure's presence (1) or non-presence (0) of hypothesized significant relationships with these structures. The LISREL program will produce the estimates only for specified 'ones'.

The reasoning behind the choice of this model is as follows. The first two items deal with leadership. I simply assume that in general terms they might capture something of a personal element in the motives behind the party choice. Although no candidates or specific party leaders are mentioned it is reasonable to believe that such details are understood. Such a personal element is only present in these first two items.

The three last items can also quite easily be seen as a coherent group. These have a mutual denominator in the fact that they make references to various groupings, in the first place the family, secondly the unions, a primary and a secondary group. The third item increases the level of generality and opens a possibility for references to occupation as well as to social class.

That leaves us with four items. Three of these have a mutual property
in common through their references to party goals, single issues, and a

conditional constraint on party platforms very close to the meaning of credibility. The first item in this group of three has the greatest generality in the sense that it captures the totality of a political direction. Finally, there is an item about the 'feeling of allegiance to a party'. Obviously this wording has something to do with the concept of party identification even though our measurement deviates from the traditional way of operationalizingthe With respect to the outstanding position of this concept in the international literature it might be appropriate to elaborate on similarities and differences.

In the first place Hultgren's measurement obviously disregards the direction in the original concept. The similarities can only be found in the strength component. Second, our item only indirectly incorporates a measurement of strength; directly, it comprises an ordinal scale for awareness, scale symmetrically arranged in such a way that two positions on each side of a point of ambivalence single out varying degrees of presence and non-presence of allegiance respectively.

On the other hand it is quite clear that our item deals with an affective relation to a political party. Two positions obviously indicate presence of such a relation, two other positions indicate absence. Contrary to traditional measurements our item allows the respondents to discriminate between more than one position of independence, the equivalent of non-presence in the original theory.

The implications of this could be that our scale might in reality capture negative as well as positive aspects of an affective relation. In any case our way of measuring an affective relation in terms of allegiance, affinity, or attachment should be characterized by good validity due to the symmetrical nature of the scale. Yet the lack of direction, as well as other differences, makes the present author hesitate in a decision regarding the use of an identical terminology. I rather choose to label the item 'party allegiance'.

Our specification of the model has only one discrepancy from a straight simple structure. This discrepancy is motivated by a theoretical consideration the party-allegiance item. It is highly reasonable to hypothesize party allegiance being related to both a presumable structure for ideas and a structure for social class. This position is taken as a result of conflicting theories. According to the Michigan school the origins of party identification can be traced far back into childhood and early youth. It is a long-term orientation shaped by forces having to do with family, groups, and traditions. Consequently it should be related to a presumable latent structure for group forces.

According to findings in European countries there are reasons to believe that an identification or an affective orientation towards political parties also can be created through mechanisms having to do with value orientations and belief systems (Percheron et al. 1984). In the Swedish context Bo Sarlvik

had already shown in 1970 that political involvement was much more related to value orientations than to party choice (Sårlvik 1970). This makes it highly reasonable to assume that the party-allegiance item should also be related to a presumable latent structure for general political ideas and issues.

When running this model in all four subsamples we soon discover that it does not fit the data perfectly. Various modifications have to be made, not very large ones but nevertheless modifications that implicate deviations from the original assumptions. As a matter of fact the model does not work in any subsample. In all cases it is necessary to specify more relations.⁴

Let us now take a closer look at the fitted models in the different subsamples. I will present the results in three tables, one for each group of factor items. In this way comparisons over the urban/rural dimension will be easier to make.

Seemingly the items of the first factor are behaving in accordance with the predictions in all subsamples. No modifications in the original model have been necessary. Nevertheless some differences between subsamples can be observed. Even if the structure itself seems to be clear the values of parameters vary. This can be studied from the loadings as well as from the error terms. In rural areas and in small towns both items are equally reliable for the factor at the same time as the level of reliability decreases moving from rural districts to small agglomerations.

In bigger cities, on the other hand, the parameters start to diverge. The
reliability for the 'party leader' item continuously drops while the opposite
is true for the item concerning 'local leaders'. In big cities the latter item

has a comparatively small error term and a similar pattern holds for the
three biggest cities.

For the second factor (Table 2) the result is more ambiguous. In all subsamples the original model has been modified. Furthermore, these modifications were forced to be made in different ways. In rural areas 'single issues' had to be specified as related also to the third structure. The very same relation turns out to be significant also in small towns but in addition to such a change 'party goals' had to be related to the first factor. In big cities, the third group, two different things have occurred: the withdrawal of a relationship with the 'allegiance' item and the addition of a relationship with 'party goals'. In the metropolises the item 'party goals' also turns out to be related - negatively - to the first factor.

Apart from these modifications in model specification there seem to be substantial differences as regards the strength of parameters. Two items dominate the factor, 'party goals' and 'credibility'. 'Allegiance' decreases its importance in bigger cities and metropolises and so does the item 'single issues'. In the last group 'party goals' are of extreme importance for this factor.

A dominating tendency traceable through these results is related to the
interpretation of the general party-program orientation. The bigger the city

the more the factor must be seen as a structure for political ideas and programs. Since this tendency parallels a decrease in 'party allegiance' we have to add a more evaluative component to the interpretation of the structure the more we are removed from rural settings. The increasing tendency with regard to evaluations parallels a decreasing tendency with regard to affections.

Turning to the third factor (Table 3), further differences between the four environments can be found. The specified model fits the data only in one subsample, rural areas. In small towns 'family' has to be specified as related to all three factors; in big cities, 'social class' as related to two factors; and in metropolises, finally, 'unions' as related to two factors.

Family is a fairly reliable item in rural districts but loses its importance in the formation of the latent structure when moving to small agglomerations to big cities. In metropolises it regains its importance to a greater extent. Interestingly, this trait parallels an opposite trend for unions, an item highly reliable in small towns and comparatively reliable also in big cities.

Thus, again, we have to consider the possibility of perceiving a small shift in meaning and interpretation over the rural/urban dimension. Unions and social class dominate the structure in middle-sized agglomerations where we know traditional industrial companies with strong unions are most often located.

As a conclusion to this section, we find that different factor structures exist in varying social environments. People in rural areas do not perceive their attachment to a party in the same way as do people in agglomerations and bigger cities. Yet similarities are greater than differences between the four settings. Out of 40 predicted significant relationships, one was wrong. Out of 68 predicted zeros, nine were wrong. In spite of some deviations from a total identity, the dominating picture is one that depicts three structures and factors for generalized perceptions of forces for attachment and voting decisions. My guess is that most researchers would not argue too much against a conclusion in this direction. All items, after all, have their strongest relationship with the hypothesized factor. With this in mind we can now turn to the second section where the analysis can be pursued under other assumptions.

Before pursuing the analysis in its central aim, I would like to make some comments on the behavior of the 'party allegiance' item. I have earlier paid some attention to certain parallels between this item and the strength component in the concept of party identification. The performance of the item seems to indicate a certain compatibility with both the American and the European ways of theorizing on this matter. Our findings, however, are somewhat closer to the European variants.

Theoretically our three latent structures must be seen as generalized dimensions for perceptions of forces behind the attachment of an individual to a political party. Two such generalized dimensions are related to the party-allegiance item indicating an affective orientation towards a party. For one of these two structures this affective orientation parallels and is subordinated to a set of more evaluatively orientated items about programs, issues, and especially credibility.

For the second structure there are no indications of any evaluations. Most probably the second structure with its set of items concerning social groupings can be considered as an affective relation towards a party as well as towards these very groups. The structure might be interpreted in terms of the Michigan party-identification theory. In fact it is quite conceivable to regard the positive end of this structure as an indication of presence of increasing identification with a party through certain groups as intermediaries. so this structure should predict a minor increase in issue constraints but to such a moderate extent that the level of constraints is compatible with a causal link from party to ideas - i.e. the ideas should be seen as a consequence of party choice and not as a causal factor for the decision.

Following this line of reasoning the negative end of the structure can be interpreted as independence of groups as well as of parties. Since the structure is exclusively founded on group-related items, the negative end of the linear structure might perhaps also be seen not only as varying

degrees of a negation of group influences but also as negative affective
relations and negative identifications, a notion that seems to be relevant
for Sweden and Great Britain (Särlvik 1970, 1983).

Returning to the latent structure of a generalized perception of party programs, issues, and credibility as a foundation for party choice, the interpretation can be made in terms of identification with party ideas of a more general nature, a nature which might relate very closely to the concept of ideology. If so the structure should at a minimum be associated with a highly consistent pattern of increasing constraints on right/left issues. This is also the case. Furthermore, the higher the level of this structure the better party choice and especially bloc voting - i.e. choice of bourgeoise or socialist parties - can be predicted from idea structures (Brantgarde 1987). Since the allegiance item is rather strongly associated with this structure, an identification with party also seems to be traceable from sources having to do with what I presently would like to call images of party ideas. This finding seems to be in line with results from other European studies.

According to the party-identification theory, a party allegiance will grow out of learning processes within the family and other relatively intimate surrounding groups. These origins of party identification have been questioned by findings in France, where paired parent-child data seem to indicate a rather extensive absence of family influences on the political behavior of offsprings. The political learning processes seem to be much more independent of elder generations (Percheron et al.). Such an independence can also be concluded from our data. Party allegiance also seems to be formed by learning processes other than family and group influences. Once established, at least, in certain strata as indicated by our second structure there are no signs of any ties left with family and groups. Instead, party allegiance is heavily tied to the presence of perceptions of general ideas, indicated by party programs and credibility. This finding points in a rather different direction as far as the origins are concerned; a formation of opinions on the basis of involvement, knowledge, and evaluative towards politics.

A Typology for Party-Attachment Perceptions

Our findings so far seem to indicate some interesting differences between various social environments. With increasing urbanism certain significant changes occur in the way voters perceive their reasons for party choice or for their attachment to a party.

Let us now take up the same theme but from a slightly different point
of view. Let us start this section by assuming a common structure throughout

the whole population. A principal-component factor analysis always providesthe solution for such a situation in the way in which it neglects 'imperfect' data. Principal-component analysis extracts what is possible to extract. With a varimax rotation such an analysis can very easily bring us a solution that has several charming details. It can give us factors that can be treated as new variables as well as variables that are similar, constructed for the whole sample. Furthermore, these variables can be constructed orthogonally with zero intercorrelations, a property of convenience when constructing tables and making table comparisons.

Having done so I take the liberty of skipping a full description of the
results. The above confirmatory factor analysis gives most of these, with
the exception that our new factor solution is, of course, not that clean.

Let us now assume that these factors actually can be interpreted as different types of party-attachment perceptions and that each factor measures special type of attachment. Since the factors are constructed orthogonally can easily form a space in one plane. Let us take factor two and factor three. The first one is so much correlated to the others that the orthogonal solution is too much of a violation of reality.⁵

The crossing of these factors at their mean values - zero, that is - will bring us four quadrants and four different types of party attachment. Considering the properties of the original item scales it is now very natural to regard the upper half of each factor as the presence of a type of attachment and the lower half as non-presence. Since the second factor deals with party programs in general, issues, identification, and credibility, it might be regarded as a factor for idea attachment. The third factor deals with group attachment. High values on factor two and low values on factor three will bring about a combination that can be regarded as the presence of an attachment of idea reasons, high values on number three and low values on number two end up in group attachment. The combination of high values on both can be labeled 'group ideas'. Low values on both dimensions will here be labeled 'neither', although a title like 'independents' would have been an alternative.

This little game gives us a typology of four voting patterns: group idea
voters, idea voters, group voters and just 'neithers'. The terminology is
mine but basically the voters themselves have made the classification.6

Let us now turn to the political-party system. Is it reasonable to believe that these voting types are correlated with the Swedish party system in such a way that certain types are more predominant within certain political parties? I think there are very strong theoretical reasons for such an assumption. A label such as 'group voters' will, in the Swedish context as well as for most Western countries with a similar right/left party system, be associated with labor parties or parties to the right, while it is more difficult to tell how partisans in the middle label themselves.

Consider Table 4. Here I present the relation between the party system and the recently constructed typology. One note of care should be given. The dichotomy made at zero on the factor indices splits the sample into four groups of comparatively equal size. This is basically a function of the technique and should not be taken as a finding in the sense that there are so-and-so sized fractions of the electorate that can be labeled in this or that way. Only raw distributions can be treated in that way. For contingency analyses, however, it is possible to use these categories if one focuses on deviances from expected frequencies and relationships.

As assumed, there is a significant relation between our typology and the party system. The question is whether this relationship can be understood in terms of general knowledge. Within the Conservative Party, group ideavoters far below average and idea voters above. This is what could be expected. Group voters, on the other hand, are far below average and this is somewhat surprising. However, since we now are exploring people's psychological perceptions, results from processes in their minds, one should take some care when making comparisons with more objectively defined categories.

Take, for instance, the other finding that a similar pattern is also valid for the Communists. This leaves us with a picture dominated by two parties from positions that are conventionally regarded as extremist positions on a right/left party dimension. This U-shaped curve bears a strong resemblance a classical notion in electoral research (Campbell 1966; Särlvik 1970). The pursuing of this finding through the whole table gives a strong feeling that this resemblance is not coincidental. In fact all cell results point in the same direction: our factor of idea attachment behaves like a factor of political involvement would do. In fact the correlation between the two is fairly high.

The general pattern regarding the party system in relation to our rural/ urban dimension (Table 5) is that the two first parties, the Conservatives and the People's Party, are overrepresented in bigger cities including Stockholm, Göteborg, and Malmö. This is also true for the Communist Party. The Center Party can as expected be characterized as a rural party and Social Democrats have a stronghold in smaller towns and agglomerations the same time as they are fairly strong in all regional environments.

From the results to date, it is very easy to conclude that a presumable variation of voting patterns over different rural/urban categories as in Table 6 is a spurious effect which would be neutralized by the party system. The more urbanization, the more underrepresentation of group voters and the more överrepresentation of idea-voters is one of two dominating patterns from this table. Another one is the fact that the so-called 'neithers' are very evenly spread over all these regional categories. The lack of ideas, as well as group affiliations as motives for party choice, does not change as a result of associated factors with a rural/urban dimension or as a direct result of the same. A tentative conclusion from this would be that 'marginal

voters' or 'floating voters' have no regional concentration as to our four
categories of environments.

The remaining question for this section of the paper deals with the very clear possibility that the intercorrelation between party system and rural/ urbanism reduces the relationship from Table 6 to zero or to a value below a reasonable level of significance. To explore this I have undertaken a loglinear analysis.

As can be seen from Table 7 my fear for spuriousness was unmotivated. A very substantial reduction in the relationship occurs, chi-square is reduced by 50 percent, but the remaining effect is still highly significant. As can also be seen, the interaction effect is not significant but all main effects are. The model fits the data and LR chi-square returns a P-value of 0.064, Pearson's chi-square is equivalent to a P-value of 0.163.

As a summary of this section, we can conclude that people perceive their attachment to the Swedish parties according to a variety of patterns, some of which could be labeled issue- or ideology-related, class-related, or a combination of the two. A fourth pattern is constituted by the absence of both perceptions. The four patterns taken as a typology are significantly related to the existing party system as well as to a rural/urban dimension. Our initial finding from the confirmatory factor analysis cannot be solely explained as an effect from differing political cultures between the socioeconomically left/right organized parties of the early 1980s in Sweden. The perceptive variations must for some reason also occur within the parties and over cultural cleavages between social environments of different sizes.

Urbanism, Social Change, and Party Attachment

We have now reached a point where we have to reconsider the theoretical
aspects of our findings to date, as well as of our main dimensions. The

previous section was dedicated to the test of some models. Let us go through these once again. I started by establishing three significant relationships,one the rural/urban dimension and party attachment, another between the rural/urban dimension and the party system, and finally a relationship between party system and party attachment. The associations between these variables could be thought of as partly spurious or as of an intervening kind. In other words it is quite conceivable, for instance, to think of the first-mentioned relationship as reduced to zero when controlling for party system. Let me now make some reflections on the causal links and directions.

Let us say that the rural/urban dimension should actually be regarded as various categories of social environments. From a historical point of view social environment can easily be regarded as a causal factor for the party system through mechanisms having to do with the composition of the population as regards social structures. It is the basic environmental conditions which produce the social structure which in turn constitutes the foundation for the party system. On the other hand it is impossible to think of a causal link the other way around unless one incorporates long-term effects from policy implementations, an assumption which seems to be over-elaborate in this context. Social environment can also be thought of as a causal factor for the way people perceive their attachment to a party, while party attachment cannot influence social environment.

The link between party attachment and the party system is more ambiguous both causal directions are possible. It is here possible to think of a party attachment perception that over time has been developed through a variety of social influences, through family, friends, the educational system, and so on. If so it is reasonable to think that these perceptions influence the attitudinal relations to the party system and eventually the choice of party. It is also conceivable to think of these perceptions as formed by earlier experiences within the party system and as a party effect.

This reasoning leaves us with two alternative models.

and

where the last model includes a direct link between social environment and party attachment. The first model can be true only if the partial relationship between social environment and party attachment goes to zero. This did not happen in the loglinear analysis and therefore it can be ruled out.

Since this technique does not give us any details about reciprocity we
now have to add this possibility between party system and party attachment

which will give us a third model.

The model says that the social environment influences the party system, which in turn influences people's perceptions of motives for party attachment. also says that social environment directly influences people's perceptions and through these perceptions the social culture indirectly influences the party system.

The model could be pressed even further if we allow ourselves to regard the rural/urban dimension as a proxy for social change. It is a clear fact that social change in Sweden, like in most Western countries, is to a substantial degree reflected through the rural/urban dimension. Of course rural areas also have been affected by the modernization of society. The blessings of the modern welfare state as well as its problems are felt and known, and are also at least possible to grasp in the most peripheral parts of the country. Modern communications of various forms, commodities, and every modern convenience are fairly evenly distributed. Nevertheless, the old society is best mirrored by rural life and the modern society by urban and metropolitan life.

A view like this would in a tentative way tell us that social change produces changes in party system which in turn produces changes in partyattachment At the same time social change also produces changes in party attachment and voting patterns which in turn produces changes in party system.

This model operates, though, on a proxy level. Many questions concerning of social change out to be raised. What social changes have produced these effects? Is it possible to understand the differences or the changing patterns through an understanding of fairy superficial changes due to urbanization and modern welfare inventions?

The modernization of Swedish society has brought us large young cohorts with higher education. These cohorts have gone into new branches and new occupations which have primarily been scattered in urban environments, the least in the metropolitan areas. As a consequence rural parts of the country and minor sites have been emptied of younger people, females, and so on. These changes are known as general phenomena. The question I would like to raise is whether in a statistical sense this uneven distribution of sexes, educational levels, ages and occupations over the rural/urban dimension, an uneven distribution that can be traced as an effect of modernization processes, can help us to understand changes in party attachment and the party system. The way I will try to answer this and related questions is the same as earlier: through elaboration by loglinear modelling.

In order not to run out of cases in too many cells I will pursue two analyses, one for each pattern of attachment, which means that one analysis will be run on idea voters/non-idea voters, and another on group voters/ non-group voters. The typology of voting patterns will thus be decomposed into two dichotomies.

As predictors in both analyses I will use Sex (1,2), Age (1,2,3), Education (1,2,3), Occupation (1,2,3), and the Rural/urban dimension (1,2,3,4). Sex and age data are derived from public registers in the sampling procedure. Occupation, education, and the rural/urban categories are the results of self-classification.

Of all possible bivariate relationships or so-called main effects, three are insignificant: sex and education, sex and group voting, and rural/urban and group voting. The last relation generates a probability of 0.0755 and significance at the 10 percent level.

Our crucial urban/rural categories are related to all intervening predictors out of these intervening predictors only sex does not relate to group voting. It seems that uneven distributions with reference primarily to occupation, ages, and education in a sense therefore 'explains' or 'interprets' the originally established relation between rural/urban categories a perception of group voting.

I would also very briefly like to report that a similar analysis with the dichotomy idea/non-idea voting generates a somewhat different result. Here only education is directly related to the voting style while age and occupation generate insignificant results. Here, in addition, the relation between rural/urban and the voting style is being reduced to a significance level below 5 percent but above 10 percent.

As can be seen from Table 9 many identical relations can be found from
Table 8, which is as it should be when the same sample and sample size is
used.

Finally, I will translate those two models into figures under the assumption sex, age, education, and occupation can be seen as intervening variables between the rural/urban dimension and group voting and idea voting respectively. The term 'intervening' should be seen in the interpretative and as a result of an elaboration of the original relationship.

Since education and age simultaneously and directly are related to group voting, the direct bivariate effect between age and group voting has to be spurious to some extent. This can be interpreted as an indication of cohort effects since the reason for the spuriousness is probably associated with the higher educational levels in younger cohorts. Educational reforms of the 1950s and 1960s can ultimately be held responsible. These reforms have not affected people born in the 1910s, 19205, and early 19305.

A similar reasoning can be made regarding occupation seen as a test
variable in the triplet age-occupation-group voting. The number of whitecollarworkers
younger cohorts is greater than among older people,

once again indicating cohort effects through a similar sort of mechanism. The remaining direct effect from age, however, cannot be explained in this way. This link is open to various interpretations, one of which could be a true aging effect, a sense of increasing group/party allegiance/voting with increasing aging.

Social change has also brought about an increase in the perception of political ideas as guidance for the voting decision and for party attachment. This growth has primarily been generated by higher educational levels among younger cohorts.

Summary and Discussion

Contrary to traditional studies of the forces and determinants of party attachment, this paper exploits nine parallel items on various perceptions of motives for party choice. Similarly constructed, these items have been well suited for strict modelling by the usage of advanced techniques. From a theoretical point of view the items fit actual as well as general points of discussions in the field of electoral behavior.

In the first substantive section of the paper, I was able to show that these nine perceptions among the Swedish electorate actually form patterns that can be seen as generalized structures for three elements of influences, an element of ideas, a second element of group or social class, and a third element of party leadership. The general lines in the organization of items and structures were derived from theories in the field. The general fit between theories and structures seems to validate the conclusion that these structures mirror fundamental and generalized perceptions of party attachments. Given that, it seems reasonable to hypothesize that different attachment models hold for various segments of the Swedish electorate. These segments seem to be related to parties, environment, and basic social cleavages.

Provided that rural/urban differences can be seen as a proxy for social

change, it seems reasonable to hypothesize an ongoing shift in the electorate from a more class-based pattern of party attachment to a pattern based on programs and images of political ideas. Essential to this trend seems to be the entrance into the electorate of younger cohorts with high levels of education.

As a general conclusion, it can be stated that social changes, albeit in various forms and through various mechanisms, can ultimately be held responsible for the decline in the perceptions of the group as a basic force for the voting decision. Whether urbanism per se gives rise to a special way of life that can be thought of as a basic force in this process of decline is difficult to say. A relationship at the 10 percent level of significance still has to be explained but the extent to which this residual has something to do with any specificity inherent in urban environments is difficult to trace from this type of data.

Finally, the construction of similar factor structures in all social environments however, a violation of realities. That leaves us with some final comments on the results from the first section of this paper. Obviously metropolitan areas have a political culture other than that of smaller cultural settings. This phenomenon seems to be a consequence of processes through which social change has created different mixtures in occupations and educational levels. Still, it is a fact. These differences bring about slightly different cultures with a noticeable variation from the population as a whole.

Two basic characteristics for these differences can be traced. First, generally speaking urban political cultures seem to be more related to political parties through political ideas than smaller settings. Conversely, urban cultures are less related to parties through groups and class than their rural counterparts. Second, apart from that, urban political cultures seem to be less affectively orientated towards political parties while cultures in smaller settings are less evaluatively orientated.

---

NOTES

1. An earlier version of this paper was presented at the ECPR workshop on Comportement electorale urbain, Rimini, Italy, 5-10 April 1988.

---

2. The 1985 election result.

3. The data have been collected by a person who has dedicated his life to the measuring of opinions in an entrepreneurial way, Sten Hultgren, once founder of the leading Svvedish institute in this area, the SIFO. When retired he established a foundation, Stiftelsen for opinionsanalyser, and by financial support from major parties and organizations pursued six mail enquiries, over the period 1979 through 1985. Analyses of earlier enquiries in this series ofsurveys can be found in Boalt (1981, 1983, and 1984). Regarding consequences of sample losses in the various surveys, a rather thorough examination has been made in Brantgårde (1986). For the 1985 survey this examination does not reveal any severe deficiencies. With respect to reported voting in 1979 as well as in 1982, no significant deviations as measured by goodness of fit can be found from the marginals of the official election-result statistics. Also, in other ways the obtained sample of respondents seems to be very accurate.

4. The following measures are valid for the fitted models: Rural areas Small towns Big cities Metropolises Chi-sq. 28.99 22.80 32.73 20.55 Df 22 19 22 21 Probab. 0.145 0.246 0.066 0.487 GFI 0.972 0.991 0.969 0.984 N 222 568 226 267 Rsq in Xs 0.977 0.969 0.969 0.975

5. The correlations between factors in the confirmative factor analysis presented above are as foliows: Phi Rural areas Towns Cities Metropolises 1 x 2 0.559 0-373 0.527 0.568 1 x 3 0.383 0.366 0.248 0.524 2x3 0.189 0.063 0.044 0.068

6. Due to the international controversies regarding ideology, hard issues, easy issues, sophistication, and so on, I choose to make up own labels to avoid confusion (Niemi & Weisberg 1984).

ACKNOWLEDGEMENTS

Source for all analyses in this paper has been Svensk opinion 1985, collected in spring 1985 by mail questionnaire. Data have been made available through Swedish Social Science Data Service (Svensk Samhällsvetenskaplig Datatjänst, SSD) Göteborg University. Data have been produced by the Foundation for Opinion Research (Stiftelsen för opinionsundersökningar) under the direction of Sten Hultgren). Neither the SSD nor the principal investigator bear any responsibility for analyses and interpretations of data made in this paper. For general as well as more detailed advice on an earlier draft of this paper I am deeply grateful to Bo Särlvik, Professor of Political Science, Göteborg University.

REFERENCES

Boalt, G. & Bergryd, U. 1981. Political Value Patterns and Parties in Sweden. Uppsala:
Almquist & Wiksell.

Boalt, G. 1983. Competing Belief Systems. Uppsala: Almquist & Wiksell.

Brantgärde, L. 1986. 'Svenska Folket 1979-1985 - sex enkäter från Stiftelsen för opinionsstudier'.
SSD-rapport nr 2.

Brantgärde, L. 1987. 'Politiska värdestrukturer och perceptioner av orsaker til val av parti'.
Statsvetenskapliga institutionen, Göteborgs universitet (mimeograph).

Butler, D. & Stokes, D. 1974. Political Change in Britain, 2nd ed. London: Macmillan.

Campbell, A., Converse, P., Miller, W. & Stokes, D. 1960. The American Voter. New York:
Wiley.

Campbell, A. 1966. A la recherche d'un modéle en psychologie électorale comparative. Revue
Francaise de Sociologie 7.

Converse, P. 1964. 'The Nature of Etelief Systems in Mass Publics', in Apter, D. E., ed.,
Ideology and Discontent. New York: Free Press.

Holmberg, S. 1981. Svenska väljare. Stockholm: Liber.

Inglehart, R. 1977. The Silent Revolution. Princeton NJ: Princeton University Press.

Jöreskog, K. G. & Sörbom, D. 1984. 'Lisrel VI. Analysis of Linear Structural Relationships
by Maximum Likelihood, Instrumental Variables and Least Squares Method'. University
of Uppsala, Department of Statistics.

Lasarzfeld, P., Berelson, B. & Gaudet, H. 1944. The People's Choice. New York: Columbia
University Press.

Lipset, S. M. & Rokkan, S. 1967. Party Systems and Voter Alignments. New York: Free
Press.

Nie, N. H., Verba, S. & Petrocik, J. R. 1976. The Changing American Voter. Cambridge
MA: Harvard University Press.

Niemi, R. & Weisberg, H. (ed.). 1984. Controversies in Voting Behavior, 2nd ed. Washington
DC: Congressional Quarterly.

Percheron, A. & Jennings, M. K. 1984. 'Political Continuities in French Families', in Niemi,
R. & Weisberg, H. eds., Controversies in Voting Behavior, 2nd ed. Washington DC:
Congressional Quarterly, pp. 377-389.

Petterson, O. 1977. Väljarna och Valet. SCB Valundersökningar, Rapport 2. Stockholm:
Liber.

Särlvik, B. 1970. 'Electoral Behavior in the Swedish Multiparty System', Statsvetenskapliga
institutionen, Göteborgs universitet, Dissertation (mimeograph).

Särlvik, B. & Crewe, I. Decade of Dealignment. The Conservative Victory of 1979 and
Electoral Trends in the 19705. London: Cambridge University Press.

Tingsten, H. 1937. Political Behavior. Studies in Election Statistics. Stockholm: P. S. King &
Son.