Resumé

Debating explanations of electoral behavior, American scholars have focused on three main theories: The identification model relying on underlying loyalty of voters towards specific parties; the political agreement or proximity model assuming a rational calculation of parties' ideological positions or stand on salient political issues as the yardstick for choice of party; and the investment model relying on voters' ability to calculate which government alternative will bring most utility for the individual voter. Examining these theories with the use of Norwegian data from the elections of 1965, 1969 and 1977, we find that the identification model is far the most powerful in predicting individual voting behavior. This model also has an edge in explaining support for the individual parties and the total distribution of voters. However, at the major postwar government election in 1965, the investment model certainly is of importance, and at the election in 1977 the significance of the proximity model has increased.

Introduction

The last decade of electoral research has seen a major debate as to which
theoretical assumptions are the most fruitful in studies of political behavior
generally, and of voting behavior more specifically.

The traditional doctrine that voting mainly could be explained as a function of party identification has been strongly challenged during this debate. This debate has been developed primarily among American scholars, and has mainly been concerned with voting in two party systems (Miller et. al. 1976, Popkin et. al. 1976, Steeper & Teeter 1976, D. RePass 1976, Miller & Miller 1976). However, the debate has at least the same

---

* I would like to thank Warren Miller and Arthur Miller for encouraging discussions on theories of electoral behavior. For critical, but helpful comments I would like to thank H. Valen, Ph. Schievely, B. O. Aardal and O. Listhaug.

interest for European scholars working with voting in multi-party systems.

The point of departure for the criticism of an identification explanation of voting is the observed decline in the number of voters expressing identification or loyalty to either of the two major parties in USA, and also an increasing rate of defection even among those who express identification with a party.

This development has been explained by an increasing level of education and information about American politics within the electorate; this factor has been considered the foundation for higher ability to analyse politics in rational terms; to form theoretically or ideologically consistent opinions about politics (Downs 1957).

The important distinction between the two approaches is found in the identification model's assumption of lasting loyalties among the voters towards specific parties, while the rationality models assume voting to be entirely explained by voters' economic, social or cultural preferences. Voting is considered a rational act to the extent that individual preferences are compared with the political goals of parties according to some political yardstick, and the voter is able to decide which party has the closest position to his own.¹

Studies based on one or another economic-rational assumption disagree, however, on what yardstick is the most appropriate to explain voting behavior and what yardstick gives the best theoretical conception of voting, if voting is to be understood as an act of rational calculations (Popkin et. al. 1976, Miller et. al. 1976, Miller & Miller 1976, Kiewiet & Kinder 1979). Different rational models disagree not only on what aspects of the voter's rationality-calculus are of importance; the logical implications of the different models seem to be inconsistent, and give substantially different explanations as to why people vote for which party. The debate of recent years seems, however, to crystallize two major traditions for rational behavior studies; one tradition views party preferences as a function of political/ideological proximity or issues agreement between parties and voters (Miller et. al. 1976, Miller & Miller 1976). The second tradition is primarily concerned with party preferences as a function of voters' evaluation of what party will give the largest (economic) utility if in government position (Popkin et. al. 1976, RePass 1976, Steeper & Teeter 1976). The second tradition is concerned with policy performance as the important criterion of vote calculation, while the first tradition focuses on political profiles of parties.

Even if all three models have some common theoretical assumptions

about electoral behavior, there are still differences which make them important for research on political behavior. They differ in their assumption of electoral response to political events and to political parties, they differ on what political objects are of importance for voters' party evaluations, and they differ in the conceptual framework of analysis and thereby the substantive understanding they give of voters' behavior and party competition. (Table 1 gives the main differences between our three theories.)

The purpose of this paper is to study to what extent voters in a multiparty system conform to some of the assumptions and some of the predictions of what today are called the identification, the proximity and the investment theories of voting. We will also see whether changes in political conditions at different elections may help us understand the extent to which the electorate conforms to the assumptions of the different theories.

The Comparison of Theories

The three theories which will be analyzed are as mentioned mainly developed
and debated among American scholars of electoral research, and
are closely connected to the study of presidential elections.

This may be the reason for their focus on which model is the best, the most correct, or the most fruitful in analyzing electoral behaviour; that is, the reason for choosing one of the models. This may be reasonable for explaining voting in a two-party system. However, in party systems where one may find socialist, liberal, conservative, Christian, agrarian, regional, communist, and fascist parties, along with parties that cannot easily be labeled, the problem of choosing a theory or model might be a quite different matter. Obviously, there are reasons to expect explanations of voting behavior to be more complex under these conditions, and a fruitful approach may be to rely on more than one model.

Another matter that should be mentioned is that the three models examined, in spite of their dissimilarities, do not form inconsistent behavioral patterns. Obviously, we could find that large groups of voters conform to all three models. Most voters would be quite satisfied with voting for the party with which they identify, with expressing position agreement or ideological proximity with the same party, and with viewing their first political preference as the party most able to govern. Nevertheless, the question asked, and the explanation given for voting behaviour, would be different depending on the different theories.

Our focus will therefore concentrate on the relative importance of party

identification, of political agreement on issues, and on evaluation of government
performance in explaining party preferences of Norwegian voters.

We will assume that the different theories might vary in importance for helping us to understand the support of different parties, and we also assume that there might be differences in the relative importance of the theories as political circumstances change from one election to another.

The Norwegian party system since the 1960's should be appropriate as a test of the relative importance of the models, and of how their ability to explain voting behavior varies over time. The reasons for believing that all models are of some importance are of course found in major aspects of the party system:

- All parties are both by their foundation and by their voter support
closely connected to the sociological cleavage structure.

- All parties are connected to different ideological positions.

- Since the election in 1965, there have always been two distinct
governmental alternatives, a Social Democratic-Labor government
vs. a non-socialistic coalition government.

Under such circumstances, there are reasons to believe that all three models are of some importance in understanding the voting behaviour of the electorate, even if we expect the relative importance of the models to vary with the party studied and with changes in political circumstances.

Politics as Loyalty

A most common conclusion in all modern electoral research is that membership in different social categories correlates with voting for specific parties (Rokkan & Valen 1974). Membership in social, economic, ethnic, cultural and geographic categories seems not only to have been a major force in the mobilization process for political participation and the formation of distinct party systems; it also seems to have been the most important factor in explaining the stability of most multi-party systems (Rokkan 1967). The very slow changes inside these party systems can only be explained if long lasting identifications between members of different categories and corresponding parties are recognized.

An important presumption is, however, that societies are characterized by stability. Like all western societies, post-war Norway has been marked by a rate of social and geographical mobility unknown in earlier history. The main causes of this mobility are the expansion of the educational

system and the transformation of the Norwegian economy from a largely primary economy, to an industrial economy and finally to a modern service economy. With these objective changes there have also been changes in expectations for future social adherence. For the voter it is no longer only a matter of which group to belong to, but also of which way to head. Expectations may not even be concerned with one's own future position, but with that of one's children.

In societies with such trends of social and economic development, we might expect a breakdown of earlier social cleavages, and accordingly a break-down of social group identification and party loyalty. These expectations seem to be correct for the American electorate, but not so in Norway.

The proportion reporting an identification with a party among American voters has declined from more than 75% in 1956 to slightly above 60% in 1976 (Miller & Miller 1977). At the same time the rate of defection among those reporting to identify with a party has increased in the U.S.A. In Norway, on the other hand, identification rose from 65 % in 1957 to more than 70% in 1965, then dropped to nearly 60% in 1969, and then again steadily increased to approximately 70% in 1977 (Pettersen 1979). At the same time defection rates among Norwegian identifiers are much lower than among Americans.

Rokkan and Campbell have explained these differences by pointing to the social homogeneity of Norwegian parties as compared to the two major American parties (Rokkan & Campbell 1970), The 'wave pattern' in Norwegian identification also indicates that the conflict structure for different elections is of some importance for understanding identification among Norwegian voters. Especially the rise in identification in 1965 and 1977 seems to indicate some connection with the Labor party and the Conservatives as the main election opponents. It is increased identification for these two parties which explains the overall increase in identification. This gives us reason to believe that identification is a persistently better variable for explaining Norwegian voting than what we find in America, though we do expect differences among parties for identification as an explanatory factor of support.

Politics as Issue Agreement

The basic assumption behind the proximity theory is that individual voters as well as each party (in USA also each presidential candidate) occupy a specific ideological position, or take distinct positions on the salient political issues. The voters' interpretation of distances between their own

positions and the positions of the parties is the basis for calculating which
party is closest to their own position. The rational voter, of course, gives
his support to the party closest to, or least far from, his own position.

Since there is no reason to assume that a single voter will conceive one single party as the one closest on every issue, the voter must either decide what party is on the average the closest, or what party is the closest for those issues the voter feels have most importance (Miller et al. 1976, Miller & Miller 1976, Miller & Miller 1975).

Such decision-making by voters of course assumes a well-informed electorate, which is again the reason for introducing the proximity theory in societies with a comparatively high level of education for its population and a very high flow ofinformation through the media systems. Both in the USA and Norway media coverage of politics has increased, people have better access to old media, and the introduction of TV has made it possible to follow the political decision-making process in close view. The increase in information, and the higher average level of education, make us believe that voters are able to see sharper distinctions between parties and distinguish subtle differences between them, and thereby perceive the ideological cleavage structure between parties. And it is certainly the case that both among American and Norwegian voters there has been an increased ability among the electorate to express agreement with parties on issues and to differentiate party positions inside the party systems (Pettersen 1979, Nie et al. 1973).

Increased information and an increased ability to handle information could, however, have generated an increasing number of voters to become aware that parties other than their first preference have a closer stand on some issues. Increased information could in this way give the effect of an increased disruption of voters' issue sympathy. The question of interest under such circumstances is what issues seem to be vote-determining for the different parties. That is, to what extent does knowledge of a voter's stand on different issues make us able to predict which party he will choose?

Our analysis will not only give information on the saliency of different
cleavages; we will also illuminate what issues are of importance for the
support of different parties.

We will see, as well, to what extent the different parties are monolithic coalitions, to which voters give support through stands on one or a few issues. Or to what extent support is based on heterogeneous coalitions as regards stand-issues, where several issues give rise to the support of one party.

Politics as Performance Evaluations

An important pre-condition for the proximity model is the voter's ability to separate party positions either according to distinct ideological cleavages or stands on issues. This assumption is often criticized forgiving too much credit both to the consistency of voters' ideological understanding and to their knowledge of parties' stands on political issues. Therefore, the critics say, a model of proximity will not only give wrong information on electoral behavior, it will also hinder the development of a correct understanding of what voting behavior is all about.

According to this view, the model which should be subsituted for the proximity model is a model viewing voting, and other ways of political participation, as a calculation of utility and costs or 'as an investment in one or more collective goods made under conditions of uncertainty with costly and imperfect information' (Popkin et al. 1976, p. 780).

Voters' ability to choose between parties arises from their ability to calculate which party in government would give the most utility or the least hard times.² The benefits of participation must, however, be conceived as larger than the costs of participating; otherwise the rational voter would not find it worthwhile to participate.

It is easy to see that to some extent the investment theory of voting is an extension and revision of some of the assumptions behind proximity analysis of voting. In both cases the question of voters being able to decide which parties are closest to their own preferences is at stake, even if it might be argued that the two theories assume different preference scales, one ideological or based on political sympathies and the other based on a cost benefit analysis of economic utility.

Both models also assume rationality to be based on the information available to voters, and assume that the knowledge level of voters might be sophisticated or superficial. The main differences between the two theoretical approaches as regards explaining voting are first, the investor model's presentation of the costs involved in political participation. The main costs are connected with gathering relevant information of parties' political performance. The investment theory explicitly assumes information accumulation to be costly and assumes that most voters therefore use some 'shortcuts' to relevant political information like party labels, ideological labels or traditional expectations of different parties. These 'short-cuts' to information can be compared with the voters' political 'free information' from every day life. Knowledge about inflation from daily shopping, and knowledge about the economic and unemployment situation from reading newspapers or watching television are examples of such

free information.

Having introduced the concepts of informational short-cuts and free information, one may argue that the investment theory seems to be very close to the identification explanation of voting. However, neither the investment model nor the proximity model assumes, as identification voting does, a positive affective relationship between voters and parties. It is perfectly within the theoretical framework of these theories to have voters choosing the least disliked alternative.³ The cost benefit assumption behind investment voting allows for calculation which involves minimization of loss as well as maximization of utility. Proximity analyses allow for calculations of both relative and absolute distances between voters and parties.

A second difference between the proximity and investment models, one which is especially important for analyzing voting in multi-party systems, is that the political objects of attention for the two models would be slightly different. While proximity analysis of voting would be concerned with party ideologies (in US candidates also), the investment analysis of voting would be concerned with government performance (in US also candidate performance) and the conception of future government performance for parties competing for government position. This introduces a significant theoretical problem since only parties with a reasonable possibility of getting into government position could be of relevance for investment voters through the calculated utility of a hypothetical policy output. What we may call government-relevant parties should therefore be the only parties of concern for investor voters. This problem is of no importance in two-party systems, where a vote for a party automatically becomes a vote for a government alternative. Not so in multi-party systems. In most multi-party systems, as in Norway, there exist parties with no reasonable possibility of getting into government position.

During the 1950's and early 60' s, Norway had only one government alternative, as the non-socialistic parties were too much in disagreement with each other to form coalitions. In this period, it is difficult to understand how investment voting could explain voting for the non-socialist parties, even if they got between 40% and 50% of the vote. Not until the election of 1965 were the non-socialist parties able to form a pre-election alliance which gave promise to the voters of a new government in the event of a non-socialist majority in parliament. As a result of the elections of 1965 and 1969, 5 out of 7 parties were what we have called government-relevant parties. The proportion was reduced to 4 out of 10 by the early 70's. Given the assumptions of the investment theory of voting, there seems to

be no way of explaining support for some parties.

The government-relevant parties are, however, by far the largest ones, and a reasonable question therefore is, to what extent voting for the government-relevant parties must be explained in different terms than voting for the 'permanent opposition' parties.

Since 1965, Norway has had 3 types of government. Until 1971, a4 party coalition government existed, followed by a Labor Party government, which was the same party that governed from 1945 to 1965. Labor left the government position after the referendum on EEC membership in 1972, and a minor non-socialist government, consisting of Liberals, Christians and Agrarians, took over. The Labor Party came into office again after the election in 1973 and remained in office after the next election in 1977.

The Norwegian electorate, therefore, had the opportunity to experience the utility of several different government alternatives. The question of course is, to what extent these experiences provided the basis for an increase in the importance of evaluations of government performance as the premises for voting decisions.

Analyzing the importance of our three theories for explaining voting behavior among Norwegians, we shall depend on a set of variables expressing identification, political agreement and evaluation of political performance, and our analysis will tell:

1. Which variables have the largest impact on the individual voter's choice of party,
2. Which variables are important for the net aggregate election outcome, and
3. The importance of these variables for the support of the different parties.

Data and Statistics

The survey data which will be analysed have been collected by the Norwegian Electoral Research Program, Institute of Social Research, Oslo. These data were not gathered primarily for the investigation of theories like those focused on here. However, to a large extent, the data have proved very useful in examining different theoretical assumptions.

Analysis and debate of the three theories have, as mentioned, mainly occurred among American scholars. Differences in the political systems of the US and Norway, have, unfortunately, given rise to differences in survey questions which means that direct comparison between the two countries is impossible.

To examine the three theories for the Norwegian electorate, we need
three types ofinformation about the individual voter, in addition to information
on voting.

1) We need information on the self-declared party identification of the voters: to what extent voters have an affective attachment to a specific party, or whether they in fact feel independent of parties when facing the decision as to which party to vote for. In this analysis, the categories of strong identifier and weak identifier are merged into one category of identifiers as opposed to independents.

2) We need information on what parties the voters perceive as those closest to their own positions on the political issues involved. Here, the Norwegian data differ from the American data. The American surveys, utilizing a 7 point scale to measure political proximity or agreement, asked the voters to give their own position, plus their opinions on the two parties and on the candidates' positions on different ideological issues. The Norwegian voters were asked which party or parties they felt expressed their points of view with regard to the issues involved. Considering the number of parties in the Norwegian party system, a scaling technique would have been most uneconomic. Even if slightly different, both techniques give a measurement of political proximity or agreement.

A problem to face in both cases is the tendency for the importance of issues to change over time. This is, of course, one of the major properties of political conflicts, but gives the researcher few possibilities to compare proximity between voters and parties for the same issues over several elections. The survey always has to choose those issues which seem to be of most importance at the different elections. For the Norwegian electorate we have information on which party or parties voters feel express their opinions with regard to the following issues:

3) To investigate the investment theory of voting, we need information as to whether voters perceive a link between the policies of different governments and economic development, and to what extent voters believe a change of government would make any difference in the level of economic

development. In this case, for the elections of 1965, 1969 and 1977, we do have data on voters' perceptions as to whether government policies have caused unemployment and inflation and on their perceptions as to whether or not a change in government would make any difference in the unemployment and inflation rates.* (For 1965 we also have information on voters' perception of government policies as a cause of the housing shortage, and of whether a different government would make the housing market better or worse.) What we lack, primarily, are data which will indicate the extent to which voters feel that government policies are of importance for their private standard of living, and the extent to which changes in government would make any difference in their individual economic situation. For one election, 1969, we have such data and they will of course be analyzed. For the election in 1977, a question regarding the voters expectations of their future economic situation will also be analyzed. In this case we will analyze the voting pattern for those who expect an improvement or a worsening of their future economic situation.

A major problem in analyzing these data with multivariate statistical techniques, however, is that all the variables we depend on are nominal scale variables. To handle this problem, the original variables have been transformed to sets of dichotomous variables. For most variables this has been a straightforward categorization of voters for party xi as the 1 value while voters for other parties and abstainers get 0 value. Similarly identifiers with party xi or expressed agreement with Xi are assigned the value 1, non-identifiers or identifiers with other parties are assigned the value 0, and no agreement expressed or agreement with other parties than xi also get a value of 0.

On the questions of government performance respondents were allowed to choose which government they expected to do the best job, or to indicate that they felt there was no difference between the Labor-government alternative and the non-socialist coalition government. To solve this problem in constructing dichotomous variables we introduced one variable giving a value of 1 to Labor-government preferences, and a value of 0 to coalition government preferences and no difference responses. Then a new variable was constructed for which coalition government preferences were given a value of 1, and Labor-government preferences together with no-difference responses were given a value of 0. For government performance we will therefore have different variables in our equations as we analyze support for different parties. Obviously it would not make much sense to try to explain voting for the non-socialist parties by Labor government preferences or socialist voting by attitudes indicating that the

non-socialist parties would perform best in government.

Structuring our data in this way requires that we analyze the different parties separately. But the symmetry of the data gives a direct comparability between the importance of different variables for different parties and also across time.

The next problem is to choose which statistics to use. With our data
there are several available.

For the first part of our analysis we will depend on the Bi coefficients from regression analysis, since this would give us the most information in the most concentrated form. The Bi coefficients give the change in probability for the individual voter in choosing a specific party Yi if his score on variable xi is 1, and 0 for variables xj ... Xk. Then given the regression equation y=bo+ bi +bj .. „bk (since Xi, xj .. Xk all equal •1) the relative probability for choosing a specific party y given the bi value for a certain variable is y = bo + bi. The single B coefficients give the probability change for choosing party y given x as impact factor.⁵

The B's will, on the other hand, only give information on the probability of individuals choosing a specific party. To develop an aggregate statistic for the importance of the different variables we will have to take into consideration the effect of the proportion of respondents having a value of 1 for the different variables. The number of respondents identifying or expressing agreement with a party will, of course, vary from party to party and from one issue to another.

The individual probability of voting for a party, as a function of identification, political agreement or government evaluation, together with the size of the proportion identifying, expressing agreement or evaluating the government alternatives (the dichotomous means X), will give our aggregate net effect of different variables with our measurement of the product BiXi.

The BiXi will be much larger for the biggest parties while identification with a small party naturally will give a small net impact on the total outcome. To measure the relative importance of the different variables for the different parties we will calculate the BiXi-values as a percentage of the total support for different parties. This measurement (BiXi%) will give the relative importance of the variables for the support of different parties.

The Superiority of Loyalty and the Growing Importance of Issues

Since most Norwegian voters do identify with a specific party and very
few defect from their identification, it is hardly surprising to find identification

as the variable that best explains the probability for individual voting. Still the questions which must be answered are for which parties identification gives the best predictability, and for which parties issues and government evaluation have a certain marginal effect on voters' inclination to choose the party. Ourfocus willbe on which changes wefind from 1965 to 1977.

From Table 2 we may see the relative importance of the different variables at the election in 1965.⁶ Identification explains individual choices of party by far the most satisfactorily, compared with issue agreement or government evaluations, even if there are some differences among parties. Obviously the center parties are those where the relative importance of identification is greatest. With the Agrarian party as the leader there actually is a change in probability of voting for one of the center parties of from 75 % to more than 80%, knowing that a voter identifies with one of these parties.

The Conservatives and Left Socialists reach a 65 % level while Labor is below 60 %. (The Communists are down to a35% level, but in this case the small number of respondents introduces uncertainty.) Just a glance at the rest of the table tells the story of the extreme importance of identification.

Still, the question is to what extent parties which are relatively low on identification may compensate for this by gains from issue agreement or government performance evaluation. First we notice that the center parties seem fairly dependent on identification since no issue or government evaluation variable gives as much as a 10% increase in the probability of voting for these parties. As a matter of fact, in 1965 there seemed to be few other variables which could compensate for identifcation for voters for any party.

In 1965 we only find 3 variables giving more than 10% change in

probability voting for a party; that is agreement with the Left Socialist or the Conservatives on tax policies, and the belief among Labor voters that unemployment will increase if a non-socialist government should take over after the 1965 election. Except for these three variables there is little independent effect for other variables, even if the Agrarian and the Liberal voters also seemed to be somewhat affected by agreement on tax policies, as were Christians and Labor voters by agreement on social service policies. For the Conservatives the second most important issue seemed to be agreement on language policies, but the independent effect of this variable is only 7% probability change.

Looking at the B-coefficients for the analysis of the 1969 election (Table 3) gives the impression of major stability. After four years the only significant changes in the importance of identification have been for the Liberals and Christians. For the Liberals there has been a slight decrease in importance, while the Christians have become the most benefited party in 1969. For the rest of the parties the B-coefficients indicate stability in the effect identification has on individual voting from 1965 to 1969.

In 1969 we do find, however, a tendency towards larger effects of issue-_voting, while, as we have predicted, the importance of government evaluation has completely faded away. While only 3 issues had a probability effect of 10% or more in 1965, the number of issues which exceeded this level in 1969 had risen to 8. Tax policies again were the most important issue, giving a 10% increase for the Left Socialist, Labor Party and the Conservatives.

Remembering that 1969 was the year of renewal of Norwegian membership in NATO, it is no surprise to see that agreement on NATO membership also gives a substantial effect on voting for the Left Socialist and the Conservatives. At this election we also find a 10% probability increase for voters agreeing with the Left Socialists on development aids for voters agreeing with the Liberals on religion, and for voters agreeing with Labor on district development.

Close to the 10% level we also find agreement with the Christians and
Conservatives on religion, and district development for the Liberals.

The major pattern between the election in 1965 and 1969 is, however,
the continuing importance of identification.

Such changes as were found are marginal, mainly a slight increase in the number of issues which independently of identification have effect on voters' choice of party. More important, however, is the almost zeroeffect of government evaluation at this election. Obviously får too large a part of the electorate perceived no significant political difference between

the Labor government prior to the 1965 election and the post-election
non-socialist government, to make government performance an important
evaluation criteria for choice of party.

Looking at the pattern for the 1977 election we notice some important changes. While identification still is the most important variable overall, there has been a marked tendency towards less importance for this variable, even if the decrease is small for some parties, especially Labor. Compensating for this tendency is the sharp rise in the number of issues having an important effect on individual voting behaviour. By now the agreement between 14 issues and parties gives probability changes of more than 10%.

Again tax policies are of importance for the Left Socialists, Labor and the Conservatives, and also the Progressive Party is at this level. As in 1969 the question of NATO-membership seems to be of importance for the left socialist voters, and by 1977 we also find environmental protection and oil policies as important issues for the leftist voters. The political significance

of the abortion law is demonstrated through the large B's for the Liberal, Christian and New Liberal voters. Oil policies also are of political significance for the Agrarians and the New Liberals. Agreement on environmental protection is of importance for Conservatives and for voters of the Progressive party. The Agrarians, perhaps somewhat surprisingly, gain at the 10% level among voters agreeing on school policies.

This tendency of increased importance for issue agreement is also demonstrated by the large number of issues which comes very close to a 10% level of change in voting probability. Tax policies have importance for Liberal voters, school policies for New Liberals and Conservatives , environmental protection for Christians, oil policies for Liberals and the conflict over abortion law for the Left Socialists. Again we may notice that government performance evaluations are of hardly any significance.

At the individual level we may then conclude that even if identification is by far the best predictor for individual voting patterns, we have found a certain decline in the importance of this factor from the 1960's until the election in 1977. This development is paralleled by an increase in the effect of issue voting, even if voters' perception of what the salient issues are divides them inside parties as well as between parties.

When it comes to government evaluations, this theory seems without major significance for explaining voting patterns among individuals. Only in 1965 do we find one variable included in this theoretical framework giving a probability change of 10% (and one variable at the 6% level). Labor voters' fear of increased unemployment (and increased inflation) seems to be the only important instance of explanation based on government evaluation for those elections we have analyzed. This of course also fits well with our hypotheses; the 1965 election was to some extent a government election as the nonsocialist parties for the first time after the war had promised to form a government if they attained a parliamentary majority. Some Labor voters obviously feared the consequences of such a political take over, while it is slightly surprising that we do not find any effect of hope and optimism among nonsocialist voters. The best we get is a 4 % probability change among Conservative voters giving the existing Labor government the blame for inflation rates.

Explaining Party Support: Issues and Cleavage Structure

The B-coefficients only give information on the individual likelihood of
choosing a party. Obviously, although the B-coefficients for identification
with Labor were smaller than those for the Left Socialists in 1965 and

1969, they had a greater net impact on the election outcome. This, of course, is a consequence of the much larger proportion of the voters identifying with the Labor party compared to the Left Socialist or any other party.

The product of the proportion of voters with a positive association to a party for some specific variable (X), and the corresponding B value gives that specific factor's net aggregate impact on the election outcome. A BX value of, say, .010 would mean that a variable gave a net benefit of the total election outcome of 1 % for a specific party. But then the BX values for small parties naturally will all be very small since a relatively small proportion of voters tends to express agreement with these parties, and so the BX will not tell anything about the relative importance of variables for the strength or support of parties. This is a measure which we get from the BX-value as a percentage of the support of a party. If the Liberals have a product of proportion expressing agreement with the party on tax policies (10.3 of the voters) and a B coefficient for tax policies (B = .075), the BX product would be .008, meaning that agreement with the Liberals on tax policies at the election outcome level gave the Liberals a net benefit of 0.8%, which is hardly important for the total distribution of voters between parties. But if the Liberals at the election in 1965 got 10% of the votes, the BX product would tell us that 8% of support for this party could be accounted for by agreement on tax policies. That is, even small BX values may be significant for understanding the support of small parties, and of course the relative importance of the support effect of variables can be calculated the same way for larger parties. Negative BX coefficients tell us that even if voters tend to express agreement with a party, the net effect is (controlling for the effect of other variables) a benefit for other parties, which in this analysis we will not be able to specify. We shall mainly be concerned with the BX% values as an indication of which variables have the greatest relative importance for the support of different parties.

Table 5 gives the BX and BX% values for the election in 1965. All parties benefit most from identification, except for the Left Socialist where agreement on tax policies seemed to be slightly more important than identification in 1965. At this election we also see that the Agrarians and the Christians together with Labor are those who depend most on identification. (Identification with the Labor Party is of course the variable which gives the largest net aggregate impact with 22 % benefit, followed by identification with the Conservatives with a 6 % impact.) With reference to our theories we may notice that the Labor Party in 1965 gets additional

support from voters who either fear unemployment in the event of a non-socialist government or think that the Labor government is responsible for almost no unemployment in the country, together with issue agreement on social service policies.

The major governmental opponent at the election, the Conservatives, get their major advantage from agreement on tax policies, but again government performance evaluation is important since the party gets an 8% support from voters feeling that the Labor government is a major cause of inflation. Except for these two variables the support of Conservatives seems to be explained by agreement on NATO-membership. For the other non-socialist parties, government evaluation also seems to be of importance at the 1965 election. We may notice that the notion that the Labor government causes inflation also yields a substantial net support for the Liberals, while the Christians seem to benefit as much from voters hoping for a reduction in inflation in the case of a non-socialist government after the election. The issue agreement variables of importance for the center parties seem mainly to be connected to these parties' positions inside the Norwegian cleavage structure. The support for Agrarians is close to 10% for agreement on district development policies, and agreement on religious issues may explain 16% of the Christian support. The Liberal Party, the center party during the 60's, does not have one major support-explaining issue, but benefits from tax policies as well as from agreement on religious issues. 1965, then, seems to be an election in which all of our theories seem to be of some importance for explaining the support of different parties. Identification is obviously the most important, but a major part of the support for the different parties is explained by issues agreement, and except for the Left Socialists, government evaluation seems at least to be as important as issues in explaining support of party. Notice, however, that while Labor voters are inclined to evaluate

governments on employment performance, non-socialist voters seem to be more concerned with performance on inflation. It looks as if these two major groups of voters use a different yard-stick in evaluating government performance.

At the election in 1969, identification has lost some of its importance for every party except for the Agrarians, whereas the impact of issue-voting has increased substantially. Again remembering that this was the election at which renewal of NATO memberhsip was one of the salient issues, it is not surprising to see that both the Conservative and even to a larger extent the Left Socialist support may be explained by agreement on this issue.

Agreement on tax policies also explains a substantial part of the support
for the Conservatives (9.7%) and for the Left Socialists (15.3 %).

The party most dependent on issue agreement, however, is the Christians,
who in 1969 benefit at the 30% level from agreement on religious
issues.

For the other parties issue agreement is of less importance, even if the Agrarians and Liberals approach a 10 % benefit from agreement on district development, and the Liberals also get a similar benefit from agreement on religion.

Only the Left Socialists seem to have a major net gain among voters who agree on aid to developing countries, and this party is surprisingly the only party which seems to benefit from government evaluations in 1969; that is, among voters who think that the non-socialist government was the cause of the rise in unemployment from 1965 to 1969.

Otherwise, government evaluations are without any important impact either on the overall distribution of voters among parties (BX-coefficients) or on the relative importance of variables explaining support for the different parties.

The 70's were marked by much less stability compared to the 1960'₅. New issues were introduced, new parties were formed and old parties were split. At the same time we found an increasing mobility among Norwegian voters. This increased mobility should at least indicate a growing concern for issues in Norwegian politics, but it does not necessarily indicate less

identification among voters. This would mainly depend on what groups of voters caused the increase in mobility during the 1970'5. It could be a mobility increase among independent voters, where the identifiers stuck with their old party.

To what extent this is the case would be impossible to answer from our data, since we do not have the net shifts between parties for different groups of voters, but at least we can ascertain that the overall net impact of identification (BX) only fell for the Liberals, which of course is explained by the 1972 party split, and for the Agrarians who also were big losers at the 1977 election. Otherwise, the importance of identification is stable or slightly increasing in its aggregate net impact.

For the relative importance of different parties' support we mainly find the same impact as in the 1969 election, with the exception of Labor and the Liberals, where we find an increase in the importance of identification. The tendencies for these two parties are easily explained by the decrease in support from 1969 till 1977; this naturally emphasizes identification as long as we find some relative stability of the net impact. Now, the important question is of course whether we have had a rise in issue voting and in voting based on government alternatives. The most striking aspect of Table 7 is precisely the growing importance of issue-voting, whereas there is no trace of greater concern for government evaluations.

We may see the increased importance of tax policies for the support of at least four parties, the Left Socialists, Labor, the Conservatives and the Liberals. For one of the new parties, the Progressives, agreement on tax policies is by far the single most important variable for explaining support.

The NATO issue still holds its significance for the Left Socialists in 1977, and school policies have become important for both the Christians and the Conservatives, as has environmental protection for the Left Socialist Party.

The new issues which really provide a major explanation of party

support are, however, oil policies and the conflict over the abortion law. Three of the center parties, the Liberals, the New Liberals and Agrarians, along with the Left socialists, gain considerable support from voters who agree with these parties on oil policies. The same goes for the Left Socialists, Liberals and Christians on agreement regarding abortion legislation. The last party benefits only at the 3.0% level as a net aggregate impact, but this amounts to approximately 30% of the party's support.

Finally, we see that it is the smaller parties like the Left Socialists, Liberals and Progressive Party which to the largest extent depend on issue voting, indicating the inability of these parties to build up stable groups of voters giving these parties their main strength from one election to another. These parties depend heavily on their ability to communicate a political message important enough to be of electoral significance. Among the larger parties the Labor Party seems to be quite dependent on identification: the net aggregate effect for this variable has been at the 20% level for all three elections analysed, and the explained support at the 50% level. Labor does not seem to have one or a few issues which give marginal support in 1977; rather most issues are of some, but rather low, importance. For the Conservatives the importance of identification is much lower, but on the other hand agreement on tax policies seems to compensate for lack of identification effects. Even if the figures for the smaller center parties are uncertain due to the small number of cases, we may notice that the New Liberals gain much less than the 'old Liberals' from identification, and that agreement on abortion law is nearly as important as identification for them.

Looking at the overall trends in 1977, compared to the trends of the 1965 and 1969 election, we find some striking differences. While identification always has been the most important variable, it has become so to a lesser extent, and the relative importance of issue voting has been on the rise. More parties depend on issue voting and more issues seem to be of importance. An important point connected to issue voting is also that the question of which issues are of importance for the different parties seems to depend heavily on the parties' position inside the cleavage structure in Norwegian politics. Foreign policies and defense together with tax policies, both issues connected to the left/right conflict, are of importance for socialist and conservative parties. Parties concerned with the overexploitation of natural resources, mainly the Liberal and Agrarians, benefit from agreement on the exploitation on oil resources; while the traditional defenders of Christian values in Norwegian politics, the Christian People's party, benefit from the conflict over the abortion law.

Variables connected to theories of government evaluation seem to be of nearly no importance in 1977. Only at the election in 1965, an election which was the first post-war governmental election, and where expectations of major changes if a new government took over were dominant, did explanation based on the investment theory contribute to our understanding of the election outcome.

Conclusion

Evaluating our three theories on voting behavior, identification obviously is the best explanation in the sense that information on party identification gives the highest probability change for an individual in choosing a specific party, gives the highest aggregate net impact on election outcome, and is the relatively most important explanatory factor for the support of most parties. Still, from the 1960's to the election in 1977, identification has lost some of its edge in explanatory power to political proximity or issue agreement.

Especially at the election in 1977, issue agreement seems to explain a
substantial part of the support for different parties, even if the saliency
patterns of political issues vary for voters of the different parties.

The variation in vote-determining issues for the different parties seems to indicate that there is a major connection between parties' position inside the political cleavage structure of Norway and the importance of the effect of different issues on support for the parties.

We might have expected parties with the most distinct ideological profiles, where we also should expect to find the ideologically most homogenous voters, to be most dependent on identification. This is definitely not the case. Those parties most dominantly occupying centerpositions inside the political cleavage structure depend most on identification, while parties with a distinct socialistic or conservative profile are much more dependent on issue agreement. This tendency might be explained by the ability of ideologically distinct parties to express clear cut stands on most issues, and the voters' ability to recognize these stands, and then either express agreement or disagreement. Still, even for parties with distinct ideological profiles, there is a tendency to disagree on what the important issues are.

For the center parties, mainly the Christians and the Agrarians, their
dependency on identification may be explained by slightly different social
phenomena.

For the Christian voters, cultural and religious/church identification

obviously are of importance, and may give a direct effect on loyalty
towards a party perceived as devoted to defend religious values.

For the Agrarian voters there might be the feeling of adherence to a threatened social group. The post war economic development has reduced the economic importance, as well as the number, of farms and of the labor force inside the primary economy considerably. In both cases the feeling of membership in a threatened minority might bring solidarity and loyalty towards political institutions like parties.

As far as our three elections indicate, evaluation of government performance or investment voting seems to be of importance only at major government elections. Inourcasef/ie government election in 1965 had two alternatives, one based on a party which had continuously governed the country since 1945, the other a completely new alternative which of course was hard to evaluate on policy performance. The experience of nonsocialist government from 1965 to 1969 obviously must have given a not-so-much-of-a-difference perception among Norwegian voters, since all tendencies of government evaluations faded away in 1969. The government changes during the 1970's do not seem to have brought elements which might indicate major policy differences between the government alternatives.

This indicates that government performance evaluation only will have a significant theoretical force in Norwegian voting if conflicts among government alternatives rise to the forefront, at the expense of party politics. Otherwise it is likely that political differentials between governments only will have an indirect effect on voting through what impact they may have on identification or issue saliency and issue agreement.

---

NOTES

1 Authors who distinguish between voting according to some rational model and voting for party identification reasons often imply a real difference between the economic-rational voter and the irrational (or arational) voting of party identifiers. Identification voting is seen as less rational because party preferences are perceived not as deliberate choices, but more as preferences subtly transmitted from parents to children which accounts for the stability of party preferences across generations. Party identifiers, the argument goes, do not evaluate political preferences from election to election, but vote instead in order to express loyalty for a party in the absence of reflection upon questions of agreement, Utility or competence. This line of reasoning reflects a severe misunderstanding of the assumptions behind the different theories on voting. The different rationality theories assume a rationality based on specific political goals which, more often than not, connect to personal economic interests. Only given these goals is there any reason to assume rationality. The faet that some voters change party preference more often than others is in itself no indication that a rational calculating process has taken place. Similarly, consistent voting along party identification lines does not preclude calculations. Identifiers may have political goals of a quite different nature than those assumed to guide the economic-rational voter. Assuming goals of a broader scope, where there is concern among voters for social, cultural and economic development trends, the position of the different parties will be stable enough to provide voters with information on which party to choose in order to realize these goals. Given stable goals among voters the rational act may, of course, be to stick with the old party.

---

1 Authors who distinguish between voting according to some rational model and voting for party identification reasons often imply a real difference between the economic-rational voter and the irrational (or arational) voting of party identifiers. Identification voting is seen as less rational because party preferences are perceived not as deliberate choices, but more as preferences subtly transmitted from parents to children which accounts for the stability of party preferences across generations. Party identifiers, the argument goes, do not evaluate political preferences from election to election, but vote instead in order to express loyalty for a party in the absence of reflection upon questions of agreement, Utility or competence. This line of reasoning reflects a severe misunderstanding of the assumptions behind the different theories on voting. The different rationality theories assume a rationality based on specific political goals which, more often than not, connect to personal economic interests. Only given these goals is there any reason to assume rationality. The faet that some voters change party preference more often than others is in itself no indication that a rational calculating process has taken place. Similarly, consistent voting along party identification lines does not preclude calculations. Identifiers may have political goals of a quite different nature than those assumed to guide the economic-rational voter. Assuming goals of a broader scope, where there is concern among voters for social, cultural and economic development trends, the position of the different parties will be stable enough to provide voters with information on which party to choose in order to realize these goals. Given stable goals among voters the rational act may, of course, be to stick with the old party.

2 There has, however, been some debate on which yardsticks are used in calculating utility. Popkin et al. seem to rely on personal impact theories: where it is assumed that the more personally involved or affected by political action the voters are, the more likely will they respond politically. The 'pocket book' voter who feels his own economic situation affected by political action is one type, while we may think of the 'socioeconomic' voter on the other hand, being concerned by the well being of his nation, and less concerned by his private economic situation. The last position is represented by Kinder and Kiewiet; Sociotropic Politics, Yale University.

3 It should be noted by now that both proximity analysis of voting and investment explanations have difficulties in coping with voters' rationalization of political predispositions. That is, even if party identification may predispose voters to prefer a specific party, voters may still evaluate parties in terms of proximity or according to an utility function.

4 These are two of the issues which Popkin et. al. feel are most strongly connected to voters' perception of government performance.

5 Since our three models are neither logically inconsistent nor assume inconsistent behavioural patterns, we do expect quite large intercorrelations between our independent variables. This raises the problem of multicollinarity; which means that a large proportion of the explained variance (R²) cannot be indentified with specific independent variables. To measure the degree of multicollinarity we have used a technique usually known as 'R2R² delete' (J. Kmeta, 1971). We obtain the R2R² for all regression-equations with one independent variable omitted in turn. For the Labour party in 1965 this gives us 11 'R2R² delete'. Each 'R2R² delete' is substracted from our original R², and the 11 differences are summed. Since the divergence between our original R2R² and some of the 'R2R² delete' will be very small, due to the correlation between some of the independent variables, the sum of differences between the original R2R² and the 1 l'R² deletes' will be much smaller than our total explained variance. This sum of variance is the variance of the original regression equation that may be identified with specific independent variables. As an illustration we will here only show the differences between original and variable-specific variance for the Labor party for our three elections. Explained variance for Labor party Original R2R² Variable specific R2R² 1965 49.6% 23.1% 1969 49.9% 23.1% 1977 59.8% 20.9%

6 Space prevents us from commenting on our tables in any detail. Only the trends of major theoretical interest will be analysed.

REFERENCES

Downs, A. 1957. An Economic Theory of Democracy. New York: Harper & Row.

and Collective', Department of Political Science and Psychology, Yale University.
Kiewiet, D. and Kinder, D. 1979. 'Economic Discontent and Political Behaviour: The Role of

Miller, A. and Miller, W. 1975. 'Issues, Candidates and Partisan Divisions in the 1972
American Presidential Election', British Journal of Political Studies, Vol. 5.

Miller, A. Miller, W. Raine, A. and Brown, T. 1976. 'A Majority Party in Disarray: Policy
Polarization in the 1972 Election', American Political Science Review, Vol. 70, Nr. 3.

Miller, A. and Miller, W. 1976. 'Ideology in the 1972 Election: Myth or Reality - A Rejoinder',
American Political Science Review, Vol. 70, Nr. 3.

Miller, A. and Miller, W. 1977. 'Partisanship and Performance: 'Rational' Choice in the 1976
Presidential Election', Center for Political Studies, University of Michigan.

Miller, A. 1977. 'The Majority Party Reunited? A Summary Comparison of the 1972 and 1976
Election', in Fishel., ed., Political Parties and Elections in an Anti-Party Age, Indiana
University Press.

Nie, N. H. Verba, S. and Petrocik. 1976. The Changing American Voter. Cambridge:
Harvard University Press.

Pettersen, P. A. 1979. 'Voting: Identification, Proximity or Investment', Institute of Political
Science, University of Oslo/Center for Political Studies, University of Michigan.

Popkin, S. Gorman, J. Phillips, Ch. and Smith, J. 1976. 'What have you done for me lately?
Toward an Investment Theory of Voting', American Political Science Review, Vol. 70,
Nr. 3.

RePass, D. 1976. 'Political Methodologies in Disarray: Some Alternative Interpretations of
the 1972 Election', American Political Science Review, Vol. 70. Nr. 3.

Rokkan, S. 1967. 'Religion, Geography and Social Class', in Rokkan, S. and Lipset, S. ed.,
Party Systems and Voter Alignments, New York: Free Press.

Rokkan, S. and Campbell, A. 1970. 'Citizen Participation in Political Life: A Comparison of
Data for Norway and USA', in Rokkan S., Citizen, Election, Parties, Oslo: Universitetsforlaget.

Rokkan, S. and Valen, H. 1974. 'Conflict Structure and Mass Politics in a European
Periphery', in Rose, R., ed., Electoral Behavior, New York: Free Press.

Steeper, F. and Teeter, R. 1976. 'Comment on A Majority Party in Disarray', American
Political Science Review, Vol. 70, Nr. 3.