Resumé

For two decades researchers have been arguing whether profound changes were taking place in electoral politics or not. Surprisingly little attention has been paid to party identification in the European debate. In Norway, party identification has dropped dramatically since the mid-1980s. hypothesize that the strong psychological ties between citizens and parties, described as party identification, have become a rare occurrence for several reasons. The intense debate over EU membership weakened the formation of party identification in two periods linked to the referendums in 1972 and 1994, but we also believe the observed decline in party identification be a long-term change linked to the fading of the old cleavages and the decline of parties as mass organizations. The hypotheses have been tested on data from the Norwegian Electoral Surveys and the Referendum Surveys from 1972 and 1994. Although the hypotheses find support, alternative explanations cannot be ruled out at this stage.

Key Aspects of the Original Formulation: Family Heritage, Stability, and Cognitive 'Screen'

Party identification, meaning 'psychological identification with a party,' has been a key concept in electoral research since the publication of The American Voter in 1960 (Campbell et al. I960).¹ The popularity of the idea seems to spring both from its theoretical fertility and impressive empirical explanatory power, especially in explaining party vote. Less attention has been paid to other aspects of party identification as it was originally described by Campbell, Converse, Miller & Stokes (1960) in The American Voter, but in the last decade increased electoral volatility has triggered renewed interest in the phenomenon. One of the concept's key features has been its power to explain observed long-term stability in party preferences. This article discusses the roots of party identification, the interplay between vote and party identification, and the consequences of the alleged decline of party identification. Paradoxically, we believe that party identification

---

* Anders Todal Jenssen, Department of Sociology and Political Science, Norwegian University of Science and Technology, N-7055 Dragvoll, Norway. E-mail: andersti(a)bill.avh.unit.no

has become more interesting in multiparty systems as the number of voters with party identification has declined. The argument that party identificationin systems is inseparable from the vote has lost credibility. The answer to the question posed (rhetorically) by Budge et al. (1976, 11), i.e., 'how theoretically interesting is the statement that electors vote for the party to which they feel closest?' has become less obvious as the proportion of voters who feel very close to a party has decreased.

The authors of The American Voter were more concerned with the effects of party identification than with its roots. Party identification was described as being part of the family heritage, passed on from generation to generation. It is clear from the discussion that the authors envisioned a relatively stable society in which the homogeneous social environment surrounding the family bolstered its traditional party identification. However, reinforcement of party identification was mostly described as a product of the voter's own political behavior: 'identification intensifies as a function not of age per se, but rather as a function of the length of time that the individual has felt some generalized preference for a particular party and has repetitively voted for it' (Converse 1976, 13). The relationship between the partisan and the party was characterized as 'a lasting attachment' and maintained by profound 'loyalties' (Campbell et al. 1960, 67). Party identification was believed to be more stable than party preferences and vote (Butler & Stokes 1969, 40ff.). Moreover, the phenomenon was not confined to marginal groups of activists. Most voters were believed to harbor a party identification. The authors noted that 'independence of party is an ideal with some currency in our society,' but they went on to label many independents 'undercover partisans' (Campbell et al. 1960, 69). Once established, party identification voters make sense of the political sphere both as 'a supplier of cues' and as a 'perceptual screen' having 'marked effects on the internal [attitude] consistency' (ibid., 72, 76). In short, a party's agenda helped its followers identify the important political issues of the day, and the emotional strings made voters essentially immune to 'deviant' political propaganda. Not surprisingly, party identification was described as a stabilizing force in electoral politics, smoothening the upheavals of political rivalry.

Controversy

Despite its manifest success, or perhaps because of it, the party identificationmodel
became controversial. The model's relevance in
European multiparty systems has been questioned for different reasons.

Some have queried whether party identification was independent of voting and whether it was in faet more stable than the vote (Budge, Crew & Farlie 1976; Butler & Stokes 1969; Thomassen 1976; Kaase 1976; Johnston & Pattie 1996; Brynin & Sanders 1997). The directional component of party identification was seen as the root of the problem, but the critics also asked how the strength component was to be interpreted if no direction of the attitude could be specified. The meaning of 'independents' in Europe has also been questioned. Europeans do not register to vote and are therefore not asked to state whether they are party voters or independents (Butler & Stokes 1969). Nevertheless, numerous studies of European politics have included party identification (in combination with other variables).

Several studies have pointed out that party identification has been less stable than predicted in The American Voter. Short-term forces like new issues, presidential performance, candidate evaluations, and assessment of party platforms made citizens shift party identification (see Green & Palmquist 1990 and Schickler & Green 1997 for brief reviews). Gradually the Michigan model was 'opened' by including new variables and describing party identification as somewhat less rigid. To some critics this was insufficient. Popkin et al. (1976) presented an 'investment theory of voting' based on the work of Anthony Downs, declaring the strength of party identification 'a running tally of current party assessment.' They argued that the Michigan model could not be mended to fit the new political environment of the 1970s because the model had been wrong all along. Less harsh, but perhaps more influential, was Fiorina's version of the 'running tally' argument. According to Fiorina (1981, 76), party identification 'at any time is a function of party performance prior to that time.' A sense of devotion or loyalty can develop over time, and voters are unlikely to recalculate their party identification if they are not provoked. Fiorina emphasized the past over the present arguing that 'whether or not the model is rational, it is realistic' (ibid., 78). In Fiorina's reformulation, a stable party identification is the result of a stable political environment, not the cause of it. We consider Fiorina's reformulation too radical to be deemed another version of the theory (cf. Miller & Shanks 1996, 130). It makes a lot of difference when the status of a variable is changed from independent to dependent. Besides, there is little psychological commitment in a running tally.

The Michigan group has vigorously defended the essence of the original concept (Converse 1976; Converse & Pierce 1987; 1992; Miller 1991). Lately, new attitude stability models including estimates of measurement errors seem to indicate a much higher partisan stability than previous estimates (Green & Palmquist 1990; Schickler & Green 1997). Whether this has saved the classical model or not is a matter of debate.²

A Challenge to Party Identification: the EU issue

The EU issue stands out as the single most important political issue in Norway's postwar history. No other issue has challenged established party loyalties to the same extent. The question of EU membership has been discussed in four periods and has been settled twice by referendum, in 1972 and 1994. Three governments have left office (1971, 1972, and 1990) over the issue, and several parties have experienced intense internal opposition, even leading to party splits in some cases (Valen 1973; Listhaug et al. 1998). Many voters found themselves opposing 'their' party's stand on the issue, and many voters switched party in accordance with their position on EU membership at the elections following the 1972 referendum (Listhaug 1989, chap. 5) and before the 1994 referendum (Aardal & Valen 1995; Narud & Valenl996; Listhaug et al. 1998). The drop in party identification in 1972 and the subsequent election came as no surprise. More surprising was the rapid 'return to normalcy' (Valen 1976). By 1977, party identification had risen remarkably and by 1981, it had returned to the level of the 1960s (cf. Figure 1). The EU debate seemed to have had no lasting impression on party identification, not even in the younger cohorts entering the electorate at the time of the 1972 referendum (Listhaug 1989,152-53).

Based on these experiences, a lasting decline in party identification as a consequence of the 1994 referendum seemed unlikely (Aardal & Valen 1997). However, one could also argue that the ties between the voters and the parties could recover from one 'blow,' but that two blows would make a difference. Or, alternatively, that the ties between voters and parties had been weakened in the period between the two referendums due to other factors, so that the 1994 referendum might accelerate an already ongoing process of decline. A new question then arises: what were these 'other factors' that were enhanced by the 1994 referendum?

The Campbell-Valen Postulate

In their comparison of party identification in Norway and the United States, Campbell & Valen note that Norwegian parties seek to attract 'distinctive clienteles.' They find that party identification plays the same role in both polities, but with some interesting exceptions. One of them concerns class voting: 'In a party system having a close relationship between the parties and the social classes, it is difficult to isolate the independent influence which party identification has . . . one wonders if this [union membership and membership in the Labor Party] does not merely express in different form his basic identification with the working class' (Campbell & Valen 1961, 525). (The argument could have been extended to include all cleavages in the

Lipset-Rokkan model, but we will concentrate on the present argument for the sake of convenience). If party identification is rooted in class identification or, as Campbell & Valen argue - more or less embedded in it, we are led to regard the observed decline in party identification as a result of (or an aspect of) a decline in class identification.

Various empirical findings seem to support the argument. In his study of party loyalties in three European countries, Richardson (1991) found that those identifying with the 'old cleavage parties' were more stable voters and had more greatly developed partisan schemata than followers of other parties. In other words, cleavages seem to bolster party identification. Moreover, working class identification has declined in Norway since World War 11, especially during the 19705. The number of people with an immediate working class identification declined from 39 percent in 1969 to 24 percent in 1981. Immediate or 'conscious' middle class identification has remained relatively stable over the years with estimates fluctuating around 19 percent of the adult population (Valen 1981, chap. 7; Jenssen 1988). Although the decline in working class identification seems to have preceded the latest decline in party identification (cf. Figure 1), we will nevertheless investigate this relationship in some detail.

The Party Decline Thesis

The party decline thesis comes in various guises (Reiter 1989). Some argue that the electronic mass media have taken over as channels of political information. Others have argued that the political system, including the political parties, has been unable to deal with the rising expectations of affluence. According to yet others, the parties have become impotent compared with growing state bureaucracies. In many Western countries, party membership has declined. In the new politics literature, the postwar generations are described as well informed, competent, and engaged in elitedirectingpolitics contrast to the elite-directed politics of the 'old' politics). The new 'cognitively mobilized' and well-educated generations are not dependent on cues from the political parties; on the contrary, they take part in ad hoc politics to put pressure on the established political institutions, including the political parties. Their style of political action has been described as 'elite-challenging' and 'anti-establishment' (Dalton & Kuechler 1990). Most of the 'old' political parties are described as unresponsive, and, as a result, most Western democracies are believed to be in a state of dealignment, characterized by declining party identification (Inglehart 1977; 1981; 1990). Inglehart and his followers believe a realignment will follow, whereas others believe the dealignment will be essentially permanent (Dalton, Flanagan & Beck 1984; Jenssen 1993). Parties have simply become

less relevant to voters. Needless to say, the possible consequences of party
decline are portrayed as gloomy in the literature associated with the view.

A major study of the political parties in Norway concludes that '[activity is declining, and the ideological distinctiveness of the larger parties is blurred. Functionally their grip on the system is stronger than ever just when the public's trust in them is eroding' (Strøm & Svåsand 1997, 354-55). Party membership has declined rapidly from the mid-1980s (Svåsand et al. 1997, 108), and the level of activity among party members is 'extremely low' (Heidar 1994, 84). Perhaps the term 'withdrawal' fits the situation of the Norwegian political parties better than the concept 'decline.' The parties have made themselves independent of their members as far as economic resources and competent personnel are concerned. The rapid expansion of the electronic media in the 1980s has made them less reliant on their organizations for communication with the voters. This shift of communication channels may nevertheless be experienced by the voters as parties becoming more remote. Talking face to face with the local party member is very different from being talked to by the party leader from the screen.

Two studies of party identification in Norway have explicitly rejected the relevance of the party decline argument. Both Listhaug (1989, 167), studying the 1965-1985 period, and Berglund (1997, 342), adding data from the 1989 and 1993 elections, concluded that party identification in Norway moves in a 'cyclical pattern.' They argue that parties are able to adjust to changing circumstances, but that the adjustments are slow and gradual and, hence, will create periods of decline in partisanship before the parties catch up with the voters. We believe this argument is only partly valid. On the one hand, we should not underestimate the ability of parties to adjust to changing surroundings, but on the other, adjustment becomes increasingly difficult as issues come and go with increasing speed, in a situation where the great 'ideologies are dead,' and no new potent cleavages are emerging to replace the receding old cleavages. Circumstantial factors as well as sampling errors will always make estimates of the strength of party identification fluctuate. The question is which average it will fluctuate around. The use of the terminology 'cyclical pattern' (inaccurately) implies that the average is an everlasting constant.

Hypotheses

We believe there is such a thing as party identification as outlined in The American Voter, but that only a minority of Norwegian voters have developed the strong psychological attachment to a party with only secondary bonds to social groups (i.e., classes, religious denominations etc.),

as described in the original formulation. For many voters, group identity (defined by the social cleavages) was probably more important than party identification. As long as the political parties addressed voters by advocating group interests, group identity and party identity were almost inseparable. The party was believed to be the group's political instrument. The distinction became relevant as the cleavages started 'fading' and the political parties developed 'catch-all strategies.' Today's typical party identifier may, in fact, resemble the 'American Voter' more than most Norwegian voters expressing party identification 30 years ago. Today, loyalty lies, first and foremost, with the party. In other words, we may speak of two types of party identification, group-oriented party identification (rooted in social cleavages) and partyfocusedidentification outlined in The American Voter). Unfortunately, we do not have separate indicators for the two types of party identification; consequently, we can only discriminate the two types indirectly through their hypothesized relationship to crucial independent variables: We hypothesize that the gradual decline of cleavage politics has caused a decline in grouporientedpartisanship.

The Norwegian political parties have become smaller, richer, and more professional, but their ties to the citizens may have suffered in this process. Today, parties are more likely to estrange old voters due to lack of correction from the rank and file in the party and for the same reason, they are less likely to detect mood changes in the electorate or new emerging issues. In short, we hypothesize that the withdrawal of the parties since the mid-1980s has caused a decline in what we have labeled party-focused identification. the number of voters with strong party-focused identification is relatively small, we believe that the 'decline of parties' will lead to a drop in party identification in all generations. If, on the contrary, the original theory is still sound, we would expect the drop to be much more pronounced in the younger cohorts which have not had time to develop a strong party identification.

The withdrawal of parties may have contributed to a change in what people mean when they report thinking of themselves as partisan (as well as to the decline in the number reporting a firm affiliation). We are inclined to believe that reporting a strong party identification in the era of cleavage politics indicated a deeper and more consequential commitment than it does today. The loyal supporter was expected to adopt most of the party's ideological platform and to support the party through thick and thin. Today, even party leaders state their disagreement with their party's line from time to time, and lifelong commitments are rare, even among party members. Hence, we expect some of the characteristics assigned to the typical partisan in The American Voter to change. The political behavior of the average partisan may either converge with the behavior of the typical independent voter, or, perhaps more likely, both party identifiers and

independents will change their behavior in the same fashion. Of particular interest are differences in volatility and attitude constraint and stability. We expect volatility to be rising and attitude constraint and stability to be declining among partisans, even among the strong party identifiers.

The struggle over EU membership in the two years leading up to the EU referendum in 1994 has put many party loyalties to the test. We would expect a substantial drop in party identification as a consequence of the 1994 referendum, as experienced in 1972. However, most effects of the fierce debate seem to have been short-lived. In the local election in 1995, the main parties on the winning 'No' side suffered a severe setback (Jenssen 1995a, Table 9.1). Two years later, the parliamentary elections had a similar outcome. According to one survey, only 0.2 percent of the voters felt that EU membership was an important issue at the 1997 election (Jenssen 1997). In our data, two elections, 1973 and 1993 (as well as the two referendums), will be classified as dominated by the EU issue. In the 1973 election, two new parties emerged as a direct result of the referendum, and a third party was successfully launched because of the turmoil created by the EU issue (Bjørklund 1997, 77). In 1993, the two major 'No' parties campaigned passionately to win the number of representatives sufficient to block EU entry through parliamentary vote. 65 percent of the voters mentioned EU membership as an important issue according to the election survey (Aardal & Valen 1995, Table 3.1).

The elections of 1989 and 1997 were not dominated by the EU issue, nevertheless substantial volatility was recorded. In 1989, both the extreme left and right more than doubled their representation. Their success was related to green issues and a new issue, immigration, respectively. Neither of them tried to turn EU membership into a campaign issue. In 1997, the Christian People's Party scored on their appeal to traditional family values and became the largest non-socialist party in Stortinget. Both the two major 'Yes' parties and the two major 'No' parties had less success. These two elections may be classified as 'deviant,' but their deviance was hardly caused by the EU issue. We are inclined to see these two 'deviant' elections as indicators of a new normality marked by fading cleavages, rapid 'issue turnover,' high volatility, and modest party identification. It almost goes without saying that this hypothesis cannot be thoroughly tested. Two observations are insufficient to establish whether these changes have taken place or not.

The Ups and Downs of Party Identification in Norway 3

We will start our empirical investigation by discussing the changes in party identification briefly. Figure 1 displays the changes as recorded in the Norwegian Electoral Studies and other surveys.⁴ We can identify two periods of high party identification, namely the 1960s and the 1980s, and

two periods of low identification, the first associated with the referendum in 1972 and the following election, and a second period starting in 1989. The fierce debate over Norwegian EU membership 'explains' the low identification in 1993 and 1994, but cannot account for the decline from 1985 to 1989 and the even more dramatic drop (in terms of level) after 1994. Three different surveys recorded party identification (with the question posed in its traditional format) in the weeks before and after the 1997 election. The result reported in Figure 1 is the least extreme.⁵ According to one survey conducted two weeks before the election, only 14 percent reported strong party identification. In the second post-election survey, the proportion of strong identifiers was 17 percent. Although the disparities between the three surveys raise some methodological questions,⁶ we feel confident when we conclude that the 1997 election, unlike the one in 1977, was no 'return to normalcy.' On the contrary, our data seem to indicate that party identification in 1997 was well below the 1994 level. In 1977, many voters moved 'home' to the party they had supported in 1969. In 1997,

many voters seem to have moved on to yet another party (Statistics Norway 1998). It is important to note that the second period of decline started well before the debate leading up to the second EU referendum, but coincided with the decline in party membership (Svåsand et al. 1997, 108). Far from being a critical test, this finding nevertheless supports the party decline hypothesis.

It is also important to note that the results for 1965 and 1969 were obtained from pre-election surveys. For methodological reasons, these estimates may be too low compared with the results from the post-election surveys (cf. note 6). In other words, the impression that party identification was stronger in the early 1980s than in the 19605, conveyed by Figure 1, may be misleading.

Has the Substance of Party Identification Changed?

This section discusses three key aspects of party identification, the stability of the identification itself, its stabilizing effect on the vote, and the ideological of the party identifiers. If party identification is basically the same phenomenon today as it was in the 19605, we expect the behavior of the identifiers (especially the strong party identifiers) to change little over the years, whereas the political behavior of the independents may vary considerably.

In The American Voter, party identification is believed to be widespread, and the authors seriously discussed whether independent voters actually existed. Consequently, they included the category 'party leaners,' a category which, unfortunately, is not identified in Norwegian elections studies. Nevertheless, we believe the number of stable party identifiers ('weak' and 'strong') to be a valid indicator of partisan stability.

The general impression from Table 1 is that the level of stability in all
periods is well below the level one would expect if strong party-focused

identification had been a widespread phenomenon,⁷ and certainly below the level reported from elections in the US. According to Table 1, the changes in partisan stability follow the same pattern as the aggregate strength of party identification (Figure 1), but with less dramatic shifts. Stability goes down in periods of low party identification. The lowest stability is recorded for the two periods before the referendums, 1969-1973 and 1989-93. But the 1985-1989 period is also marked by instability. For reasons discussed above, this cannot be explained as a result of the EU issue. (Unfortunately the format of the party identification question was altered in the 1997 election survey, so we are unable to study the second 'deviant'period, 1993-1997.)

Even more important is the question of whether party identification stabilizes the vote or not. Due to the problems related to comparing stability of party identification and stability of the vote directly (for instance the status of independents and abstainers in such a comparison), we will study differences in volatility dependent on strength of party identification. The question we pose is simply: is the increased volatility in Norwegian elections a reflection of shifts in strength of party identification, or has the likelihood of party switching increased in all groups, irrespective of party identification? If we find that volatility has increased even among strong party identifiers, we would be inclined to believe that the meaning of party identification is undergoing change. Table 2 reports the net party shifts (shifts between parties) and gross party shifts (net party shifts + abstainers at one of the two elections):

The results reported in Table 2 leave very little doubt. The increased volatility is not limited to the group of independents. Even among the strong identifiers, volatility has become considerable. After 1985, one in five strong identifiers has not 'stuck to the party.' In the 19605, the number was one in ten. Nevertheless, there are systematic differences between the

three groups: strong identifiers are more loyal than weak identifiers, and weak identifiers are more loyal than independents (who have become more disloyal than loyal) at all points in time. On closer inspection, these differences are not constant. In periods of high volatility, the differences (ratios) between the three groups (strong-weak-independent) decrease because volatility increases relatively more among strong identifiers than among independents. This is, of course, partly due to the much higher volatility among independents in the first place.

Table 2 utilizes panel data instead of recall data to reduce systematic and unsystematic errors caused by imperfect memory. Comparing the estimates of party shifts from the panels (Table 2 above) with self-reported party shifts (not shown) yields one interesting observation. The divergences between the two tables seem unsystematic with one striking exception: Strong identifiers tend to underreport party shifts. Self-deception may also be a psychological mechanism typical of strong identifiers.

Listhaug (1989) has shown that there is a consistent association between the strength of party identification and the timing of the voting decision, the easiness of the decision and the ability to recognize differences among the parties: The strong identifiers make earlier voting decisions, they have fewer problems recognizing the differences between the parties and find it easier to make their voting decision than other voters. Listhaug concludes that 'party identification still has a strong and consistent function in the political perceptions and evaluations of the Norwegian voter' (Listhaug 1989, 168). The conclusion may well be correct, but we believe the correlations discussed Listhaug only support a more modest conclusion: partisans are more interested in politics than independent voters. A more critical test of the model would be to investigate whether partisans hold more ideologically consistent and stable political attitudes than independents. The available data limit our options with regard to period and attitudes. We have limited the test to two periods, 1977-81 and 1989-93, and attitudes linked to the ideological left-right dimension. The two periods are the earliest and latest periods available, the first characterized by high and increasing party identification, the second by low and declining identification. Attitudes linked to the left-right dimension were chosen for two reasons. There are more indicators available for this dimension than any other, and secondly, the left-right dimension has played a major role in all postwar elections, meaning that most voters are familiar with these issues.

The results are interesting. First of all, neither attitude constraint nor attitude stability among Norwegian voters is overwhelming. This result is in line with much prior international research. Second, the attitudes of those with strong party identification are more consistent and more stable than those of other voters. The exception is attitude constraint in 1993, perhaps a result of the ongoing EU debate. Prior research (Jenssen 1995b) indicates

that two ideological dimensions, the left-right dimension and the centerperipherydimension, fused during the EU campaign. The 'No' side mixed ideological symbols from the left and periphery positions, whereas the 'Yes' side linked symbols from the right and center (urban) positions, both more or less collapsing the two dimensions into one. This may have caused considerable confusion among ideologically aware citizens (cf. Zaller 1992).

Although the strong identifiers hold more stable attitudes and probably more consistent attitudes than other voters, they fall short of the stringent partisan described in The American Voter (Miller 1986). There are no profound between independents and those reporting moderate party identification. This cognitive aspect of what we have called party-focused identification seems to be confined to the group reporting strong party identification, at best. These results and the results reported in Table 2 support the argument that the meaning of party identification is changing. Not only is the number of party identifiers declining (see Figure 1), the political behavior of partisans is changing as well. Even the 'hard core' of strong identifiers is affected.

Working Class Identification and Identification with the Labor Party

The decline of the old cleavages in Norway is well documented (see, e.g.. Valen 1981; 1991; Valen & Aardal 1983; Listhaug 1989; 1997). Less attentionhas paid to the possible connection between cleavages and party identification. In the discussion above, we argued that the decline of cleavages may have led to a weakening of party identification. To test this

hypothesis carefully for all six traditional cleavages is impossible within
the scope of this article, so we will concentrate on the important labor
market cleavage.

Class voting has been stronger in Norway than in many Western countries. The formal and informal links between the working class, the labor unions, and the Labor Party have traditionally been strong. Gradually, however, the impact of class voting has declined both due to a shrinking of the traditional working class and decreasing support for the Labor Party within the remaining working class (Listhaug 1997). We can infer from our general hypothesis that the decline of working class consciousness (the psychological aspect of the cleavage) will be accompanied by and linked to a decline in identification with the Labor Party.

In the 1960s and early 19705, approximately 40 percent of the adult population actively identified themselves as working class compared to 24-25 percent in the 1980s and 19905. From 1965 to 1989, the proportion identifying with the Labor Party decreased from 40 percent to 25.8 percent (Berglund 1994, Table 4.1). These parallel developments are hardly a coincidence, but the process is more complex than these almost identical numbers suggest. The Labor Party has gained support from middle class voters and voters without a working class identification, and, at the same time, Labor has lost support within the (shrinking) group with a strong working class identification (ibid., 100-106). These developments add up to the pattern displayed in Table 4.

The correlations between working class identification and Labor Party
identification was at the 0.40 level in five elections, 1965, 1969, 1977, 1981,

and 1985. The two elections 'contaminated' by the EU issue deviate, but not to the same extent. The correlation in 1993 (0.26) is well below the 1973 level (0.37). The 1989 election stands out as well, as expected. The results from 1989 and 1993 indicate that about 50 percent of the conscious working class identify with the Labor Party. Labor Party identification is strongly related to Labor vote in all elections. At three elections (1977, 1981, and 1985), the magnitude of the correlations suggests that it is problematic to differentiate between a dependent and an independent variable. The other correlations between Labor Party identification and Labor vote are less problematic and 'save' the theory from accusations of tautology. The correlations between working class identification and Labor vote show the same pattern as the correlations in the first column in Table 4. If the three variables are envisioned as a simple causal chain, we can conclude that the variation in correlation between working class identification and Labor vote was primarily caused by the changes in association between class identificationand identification rather than the more stable correlations between party identification and vote. The result is illustrated in the last column in Table 4. In the 19605, about 30 percent of all voters identified with the working class and the Labor Party and voted for the party. In 1989, less than one in five had all three characteristics, and in 1993 only 16 percent fitted the description of the traditional Labor supporter. These results bolster our argument, at least with regard to class cleavage. The decline of identification with the Labor Party after 1981 seems to be intimatelylinked the decline of working class identification.

Long-Term Effects?

According to the classical theory, the strength of party identification increases with the number of successive elections the voter has supported the party, and, as the strength increases, the voter becomes less inclined to defect from the party - though if he does, he will be more likely to return to the party in subsequent elections (the 'homing' effect). This is not an effect of aging as such, it is more a question of political maturity. Nevertheless, we may call this an aging effect or life cycle effect. Under stable conditions, the age-strength correlations will - according to the theory - be constant and the average strength of party identification in the population will be fixed. This pattern may be disturbed by so-called period effects. A major event, like a scandal involving important institutions or politicians, may depress party identification for a period. The EU referendums in Norway seem to stand out as striking examples. Not only were the parties forced to share the political stage with political ad hoc movements, most of them faced trouble caused by opposition even among their most loyal voters.

When an issue is put to rest through a referendum, the parties are able to
regain their strength.

The life cycle effect is easily detected in Table 5. With some minor exceptions (1973, 1985, and 1997), the age-strength relationship seems to be almost perfectly linear. Some 'errors' (three out of 74 observations in this case) must be expected due to random sampling error. The period effect linked to the referendums is easily detectable across generations. Party identification is weaker in all generations in 1972 and 1993/94 than in 1969 and 1989. The difference in strength between the oldest and the youngest is largest in 1972 and 1994. The general depression of party identification caused by the EU issue seems to have hit the youngest generations somewhat harder. This observation is in line with much earlier research from the Michigan group. It is difficult for the young to form party identification in times of political turmoil.

Unlike period effects, cohort effects are long-term effects. Cohorts, or historical generations as they are sometimes called, are distinguished by their unique experiences in preadult years. These experiences are supposed to leave a lasting impact on a cohort. Mannheim (1972) went as far as to describe historical generations as social groups conscious of their common experience and common fate. If we are able to identify distinct cohorts, we should be able to describe the outcome of the generation replacement mechanism. If the cohorts that vanished were very different from the emerging cohorts with regard to the distribution of some attribute, the replacement will change the prevalence of the attribute in the population at large (everything else being equal). The 'withdrawal' of parties and decline

of cleavages may have had a lasting impact on the younger cohorts, or, more accurately, may have depressed the formation of party identification in these cohorts permanently. The strength of party identification may increase somewhat over the years (due to the life cycle effect), but it will never reach the same level of strength as the older cohorts.

The problem with the cohort argument has always been how to discern empirically between cohort effects, period effects, and life cycle effects. In some cases, the historical circumstances may come to the researcher's aid. If, for instance, historical changes favor the formation of stronger party identification (i.e., as political life calms down after a turbulent period), a cohort effect will (independent of all other mechanisms) produce an agestrength contrary to that produced by a life cycle effect. Unfortunately, we are in the opposite situation: the decline of party hypothesis a situation where members of the younger cohorts are believed to be less likely to form strong party identifications than the foregoing cohorts. Hence, we face an analytical problem, as cohort effects and life cycle effects may operate at the same time. We are unable to tell from Table 5 whether the observed pattern is the result of a cohort effect, a life cycle effect or both. The matter is further complicated by the presence of possible period effects. It may look like the younger cohorts (born after 1950) will never reach the level of party identification of those born before 1920, but we may also be witnessing a period effect from 1989 on.

Period effects are usually described as affecting all cohorts to the same extent, but in many instances this is unrealistic, especially when we are dealing with a socialization argument (like party identification theory). Older cohorts may interpret an incident as a minor deviance compared to prior experiences, whereas the youngest cohort may experience it as a major event. Thus, the youngest cohort may go through a 'formative experience' creating a cohort effect, where the older generations experience a short-term change that can be described as a period effect. The result for the youngest cohorts in 1972 and 1993/94 may be a case in point.

Table 5 does support the presence of all the described effects. This is unsatisfactory, however, especially with regard to the party decline argument. We would like to know whether we can identify an independent cohort effect in addition to the life cycle effect and the period effect. The obvious solution to this research problem is a multivariate model. The problem of cohort analysis has been described in statistical terms as an identification problem, as the independent variables age, cohort, and period are mathematically interdependent. Statistically, this problem has been solved by recoding the variables to reduce intercorrelations and applying statistical methods less sensitive to multicolinearity (Hagenaars 1990).^s This statistical convention highlights another question: As recoding of the variables becomes crucial, how can we reduce the number of categories

substantially and at the same time maintain theoretical validity? Regardless of recoding, the independent variables (especially age and cohort) will still be intercorrelated, so the chosen coding will influence the results from any multivariate analysis.

According to the authors of The American Voter, party identification is inherited through family and reinforced through the act of voting. Consequently, strength of party identification increases throughout a lifetime. Converse (1976, 44) argues that the age-strength relationship is not linear. According to him, gains in partisan strength will be greater early in life with gains per year decreasing over time. Inspired by this reasoning, we have coded age in three categories, 18 to 25, 26 to 40, and 41 to 79 years of age.

The cohort variable causes a special problem. Only respondents born between 1918 and 1945 are represented in every survey we want to use. Those born before 1918 gradually disappear and are completely missing in 1997. Those born after 1945 gradually enter the sample as they obtain the right to vote (cf. Table 5). Any result obtained for these two categories will not reflect the situation at all times to the same extent. The older cohort was fully represented in 1965, then it gradually fades. And, as it fades, the internal composition of cohort changes, becoming relatively younger in the sense that only the younger members of the cohort are present in the later surveys (Glenn 1977, 12). The youngest category also becomes relatively younger as it enters the electorate. Voters born after 1945 are fully represented only in the last survey (conducted 1997). To cope with this problem, we have coded those born before 1918 and after 1945 as separate cohorts to be discussed separately. That leaves us with those born 1919 to 1945. Ideally, the postwar generation would have been represented in all surveys analyzed, since a variety of theories describe the postwar generation as distinct compared to the prewar generations. The second best solution is to single out those born during the war because they had their most important formative years (15-25 years) after the war. (Unfortunately, from an analytic point of view, they were too young to experience the postwar affluence as something self-evident.) We would also like to identify the war generation and the generation that experienced the economic depression and class struggles of the 1920s and 19305. At the same time, we must keep an eye on the number of observations in each category at each point in time. The rather arbitrary decision was to make 1928 the cutting edge.

Normally, one of the extreme categories would be the reference category, but in this case we decided to use the 1929-1939 cohort as the reference category for reasons discussed above. The reference category should be present at all times, as other estimates are calculated as deviances from the reference category. This procedure decreases the likelihood of finding significant cohort effects.

The data used for this analysis is a combined file based on the Election Surveys 1965 to 1993, the two surveys from the 1972 and 1994 referendums on EU membership, as well as an opinion poll obtained after the 1997 election. The cases in the combined file are weighted so that the number of cases is identical for each year (the total number of cases is unchanged: N — 21071). The variable 'period' is based on the year of the survey. It is recoded into three values: 'normal elections' (1965, 1969, 1977, 1981, and 1985), referendums (1972 and 1994) and elections (1973 and 1993) dominated the EU issue (labeled 'EU issue'), and 'late elections' (1989 and 1997).

We proceed by blockwise multiple logistic regression. First, we enter the variable 'period' to account for the possible period effect caused by the two struggles over EU membership ('EU issue' and the hypothesized effect of party decline on the 'late elections'). In the second block, we enter the age variable representing the original theory of party identification, and in the last block we include cohort to investigate whether a cohort effect exists 'on top of the age and period effects. The results are given in Table 6.

The results from the first model including only the dummies representing the variable 'period' tell us two things. Both periods marked by the EU issue and the two late elections deviate from the 'normal' periods, and the late elections deviate more. (This is mainly due to the low estimates for strength of party identification in 1997. Cf. Figure 1 and Table 5.) We are led to believe that 'something' - unrelated to the EU issue - took place in 1989 and 1997. When age is included in the regression, the explanatory power is substantially improved according to the chi-square statistics. As expected from the original theory, party identification increases with age. However, the magnitude of the two coefficients indicates that the reinforcement of party identification is a more linear process than assumed by Converse (1976). When cohort is included in the model, the model is again significantly improved, but the improvement is not of the same magnitude as the prior improvement. Moreover, those born 1940-1945 and 1918-1928 are not significantly different from the reference category with regard to strength of party identification. So far the data support the classical theory. The oldest generation is significantly different (at the 0.01 level), but we cannot tell whether this is caused by the internal age change in the cohort or whether it is a true cohort effect. The same can be said for the youngest cohort (born after 1945). However, the result indicates that the youngest generation deviates more from the reference category than the oldest one. Due to the limitations in our cohort variable, the result is inconclusive with regard to the party decline hypothesis. However, the significant effect of the dummy variable 'late elections' favors the party decline argument.

To investigate the possibility of a cohort effect for the postwar generation further, we may drop the earlier elections from our data and hence increase the number of cohorts present at all points in time. By dropping the elections before 1981 and the 1972 referendum, we can add those born 1954 to 1963 to the cohorts present in all surveys. At the same time, we include both 'normal elections' (1981 and 1985), a period dominated by the EU issue (1993 and 1994), and the two 'late elections' (1989 and 1997). The coding of the cohort variable is adjusted so that those who experienced the hardship of the first postwar years are excluded from the postwar generation. The results of this analysis are shown in Table 7.

The results obtained with period as the single independent variable are not very different from the results displayed in Table 6. Both periods dominated by the EU issue and the late elections deviate significantly from the 'normal' elections, and the 'late elections' are the more deviant. The deviances as expressed by the regression coefficients are more articulated in this table than in Table 7. When age is entered into the model, the model is improved, but the improvement is smaller than in the previous analysis. The regression coefficients are statistically significant, but in this run, the differences between the reference category (18-25 years) and the 26-40 age

group are diminished compared to the analog coefficient in Table 6. The important change occurs when cohort is entered into the model. Age loses its independent explanatory power and all dummies representing cohort yield significant coefficients. Moreover, the model improvement is substantialcompared the results from Table 6.

How should this rather radical result be interpreted? The most obvious explanation seems to be that the applicability of the traditional party identification theory depends on the period under investigation. The theory fits the early elections better than the later. The age-strength relationship is sufficiently pronounced in the 1965-97 period to yield a consistent life cycle effect as hypothesized by the Michigan researchers (Table 6). When we confine our study to the second half of the period, the data fit the hypothesized cohort effect better than the hypothesized age effect. The results displayed in Table 7 are in line with the party decline argument, but these results are not beyond dispute. Although the age effect disappears when cohort is brought into the model (Table 7), this does not mean that the life cycle argument is invalid. Before cohort was entered into the model,

age had a significant effect, and strength increases with age at all points in time (cf. Table 5), even in 1997. Our data (Table 6) support the claim that many voters gradually developed psychological links to a party, at least before 1989. It seems unlikely that this form of political socialization has vanished completely in the 19905. One could also argue that the coding of the age variable is incorrect. A detailed examination of the age-strength relationship indicates that the relationship is almost perfectly linear (cf. also Table 5). However, this linear relationship is not in line with the original theory, which stresses the importance of early socialization.⁹

Conclusion

We still believe that the type of commitment to a political party as described in The American Voter exists, but this party-focused identification has probably never been as widespread in Norway as anticipated by Campbell and his colleagues for the US. Although we find that the strong identifiers hold more stable and consistent political attitudes and are less likely to switch parties than others, they do not fit the description of'homo politicus' envisioned in The American Voter. The results also show that the meaning of party identification is gradually changing. For example, the phenomenon of increased volatility does not affect independents only. On the contrary, even among the hard core of party identifiers, volatility has become notable. Data also support the arguments that party identification in Norway has been linked to the cleavage structure, and that the fading cleavages might contribute to a decline in party identification. Although our analysis has been confined to the class cleavage, we see no obvious reasons why the other cleavages should be less affected. The findings concerning the party decline argument are less straightforward. As anticipated, the EU issue strained the ties between many voters and the parties in 1993-94 as it did in 1972-73, but the 'deviant' results for the two 'late elections' in 1989 and 1997 were not caused by the EU issue. The cohort analysis, especially the results presented in Table 7, suggests that a long-term process may change the ties between parties and citizens more permanently. This finding is consistent with the party decline thesis, but can hardly be regarded as a definitive argument. First of all, two 'deviant' elections are hardly sufficient to conclude that the ties between citizens and parties have changed profoundly. Second, the 'party decline thesis' is not a theory or model in the strict sense of the word. If this argument is to be taken seriously, a consensus must be reached with regard to description of the processes believed to change the role of parties. Third, alternative explanations must be considered carefully, for instance, the possibility of a 'gray rebellion' causing increased volatility and consequently weakened party identification

among older voters. Such a phenomenon would account for the weak agestrengthrelationship 1989 and 1997 (cf. Table 5).¹⁰ Party identification will probably exist as long as there are political parties. Nevertheless, the data lead us to conclude that the phenomenon is changing both qualitatively and numerically. The group of firm 'believers' is shrinking and the commitmentamong partisans is not what it used to be. Recent findings (see Clarke & Stewart 1998, 363 for references) suggest that similar shifts are taking place in several Western democracies.

---

NOTES

1. The basic idea is probably as old as political parties themselves. Butler & Stokes (1969, 24) credited Graham Wallas' 'Human nature in Politics' (1910) as being the first description of the phenomenon. Campbell & Valen (1961) argue that Hume, Washington and Madison were well aware of the affective links between parties and their followers.

2. These models seem to indicate an almost perfect stability. Typically, they account for more than 90 percent of the variance in party identification on the latest point in time. In some cases (for instance Canada 1974-1979-1980), they even claim to account for 100 percent of the variance. Common sense tells us that these results may be methodological artifacts. As long as LISREL models include estimates of measurement errors but no estimates of correlations between these errors, the program seems to yield high estimates of attitude stability. In panel models, this is extremely important because there are obviously substantial correlations between the error terms since the same measurement instrument is (usually) used in all waves of the study. Hence, the realism of these models is questionable.

3. For a more comprehensive discussion of changes in party identification in the 1965-1993 period, see Listhaug(l9B9) and Berglund (1994).

4. The author would like to express his gratitude to Henry Valen and Bernt Aardal who have made such a tremendous effort to keep the Norwegian Electoral Surveys running for decades.

5. Both post-election surveys recorded 17 percent strong party identification. It is the number of weak party identifiers which makes the results different. The number in Figure 1 is 25 percent (data from Statistics Norway), whereas the second survey (by Opinion AS) reported only 14 percent weak identifiers. The second survey is used in the following analyses.

6. The data presented in Figure 1 originate from pre-election surveys, post-election surveys as well as referendum surveys. Holmberg (1994) and Holmberg & Weilbull (1997) have observed an 'electoral cycle' in Swedish politics: Party identification (as well as political awareness) tends to diminish between elections. In the 1988 election survey, 66 percent reported having a party identification (strong or weak). One year later, the figure was 53, and in 1990 the level reached an all time low with 48 percent. In the following election (1991), the 1988 level was almost restored. The pattern was repeated after the 1991 and 1994 elections. Hence, one must expect lower estimates of party identification between elections and prior to elections than in the weeks immediately following an election.

7. We disagree with Berglund (1994) who argues that a change from independent to party identification is in line with the party identification model. First of all, the move from independent to identification can be the second step in a move from identification with one party to identification with another. Second, the authors of The American Voter argue that the direction of party identification and, consequently, a strength component are established in the pre-adult years, and that the strength is reinforcedlater.

8. The autocorrelation problem is controlled as well by using logistic regression.

---

9. The linear age-strength relationship fits Fiorina's (1981) reformulation of the theory well. The authors of The American Voter gradually gave in to the critics of the family socialization hypothesis, but maintained that party identification 'crystallizes' during the age spån 25 to 35 (Miller & Shanks 1996, 131). Our coding schema covers this version of the argument as well as the original version.

10. The strength of party identification has declined more in the older cohorts between 1994 and 1997 than it did in the younger cohorts (cf. Table 5). Some years ago, a pensioners' party was formed in Norway. The party has already managed to win seats in town councils in spite of obvious organizational problems. Many older voters have reacted bitterly to the eldercare provided by the state, the problems in the health care system, and proposed and implemented changes in pension schemes. In the 1997 election, eldercare was one of the most prominent campaign issues. Whether this is a period effect or a new 'reversed' life cycle effect causing lowered party identification among older people remains to be seen.

REFERENCES

Aardal, B. & Valen, H. 1989. Velgere, partier og politisk avstand. Sosiale og økonomiske
studier 69. Oslo: Statistics Norway.

Aardal, B. & Valen, H. 1995. Konflikt og opinion. Oslo: NKS-forlaget.

Aardal, B. & Valen, H. 1997. 'The Norwegian Elections of 1989 and 1993: Norwegian Politics
in a Perspective,' in Støm, K. & Svåsand, L., eds., Challenges to political Parties. The Case
of Norway. Michigan: The University of Michigan Press.

Berglund, F. 1994. Partiidentifikasjon. En teoretisk og empirisk avklaring. Thesis. Department
of Political Science, University of Oslo.

Berglund, F. 1997. 'Partiidentifikasjon på vei ut? En kohortanalyse av norske velgeres
partitilhørighet,' Tidsskrift for samfunnsfor skning 38, 333-57.

Bjørklund, T. 1997. Om folkeafstemninger. Norge og Norden 1905-1994. Oslo: Universitetsforlaget.

Brynin, M. & Sanders D. 1997. 'Party identification, political preferences and material
conditions,' Party Politics 3, 53-77.

Budge, 1., Crew, I. & Farlie, D., eds. 1976. Party Identification and Beyond. New York: John
Wiley.

Butler, D. & Stokes, D. 1969. Political Change in Britain: Forces Shaping Electoral Choices.
London: Macmillan Press.

Campbell A., Converse, P., Miller, W. E. & Stokes, D. E. [1960]1964. The American Voter.
New York: John Wiley and Sons.

Campbell, A. & Valen H. 1961. 'Party Identification in Norway and in the United States,'
Public Opinion Quarterly 25, 505-25.

Clarke, H. D. & Stewart, M. C. 1998. 'The Decline of Parties in the Minds of Citizens,'
Annual Review of Political Science I, 357-78.

Converse, P. E. 1976. The Dynamics of Party Support. Cohort-Analyzing Party Identification.
Beverly Hills: Sage.

Converse, P. E. & Pierce, R. 1987. 'Measuring Partisanship,' Political Methodology 11,
143-66.

Converse, P. E. & Pierce, R. 1992. 'Partisanship and the Party System," Political Behavior
14, 239-59.

Dalton, R. J. & Kuechler, M., eds. 1990. Challenging the Political Order. New Social and
Political Movements in Western Democracies. Cambridge: Polity Press.

Dalton, R. J. , Beck, P. A. & Flanagan, S. C 1984. 'Electoral change in advanced industrial democracies,' in Dalton, R. J., Flanagan, S. C. & Beck, P. A., eds.. Electoral Change in Advanced Industrial Democracies: Realignment or Dealignment? Princeton: Princeton University Press.

Fiorina, M. 1981. Retrospective voting in American National Elections. New Haven: Yale
University Press.

Glenn, N. H. 1977. Cohort Analysis. Newbury Park: Sage.

Green, D. P. & Palmquist, B. 1990. 'Of Artifacts and Partisan Stability,' American Journal
of Political Science 34, 872-902.

Hagenaars, J. 1990. Categorical Longitudinal Data. Newbury Park: Sage.

Heidar, K. 1994. 'The polymorphic nature of party membership,' European Journal of
Political Research 25, 61-86.

Holmberg, S. 1994. 'Partierna tycker vi bast om i valtider,' in Holmberg, S. & Weibull, L.,
eds., Vägval. SOM-institute. Gøteborg University.

Holmberg, S. & Weibull L., eds. 1997. Ett missnøjt folk? SOM-undersøkningen 1996. SOMrapport
18, Gøteborg. SOM-institute. Gøteborg University.

Inglehart, R. 1977. The Silent Revolution. Changing Values and Political Styles Among the
Western Publics. Princeton: Princeton University Press.

Inglehart, R. 1981. 'Post-Materialism in an Environment of Insecurity,' American Political
Science Review 75, 880-900.

Inglehart, R. 1990. Culture Shift in Advanced Industrial Society. Princeton: Princeton
University Press.

Jenssen, A. T. 1988. 'Fra arbeiderparti til lønnstakerparti,' Vardøger 11, 11652.
Jenssen, A. T. 1993. Verdivalg. Nymassepolilikk i Norge. Oslo: Ad Notam Gyldendal.
Jenssen, A. T. 1995a. 'Etterspill,' in Jenssen, A. T. & Valen, H., eds., Brüssel midt imot. Oslo:
Ad Notam Gyldendal.

Jenssen, A. T. 1995b. 'Hva gjorde EU-striden med det ideologiske landskapet i Norge?'
Sosiologisk Tidsskrift 3, 213-29.

Jenssen, A. T. 1997. 'EU-saken i Norsk politikk 1972-1997 og i framtida.' Working paper
no. 18. The Referendum Project. Department of Sociology and Political Science.
Norwegian University of Science and Technology.

Johnston, R. J. & Pattie, C. J. 1996. 'The Strength of Party Identification Among the British
Electorate: An Exploration,' Electoral Studies 15, 295-309.

Kaase, M. 1976. 'Party Identification and Voting Behavior in the West German Electorate
of 1969,' in Budge, 1., Crew, I. & Farlie, D., eds., Party Identification and Beyond. New
York: John Wiley.

Listhaug, O. 1989: Citizens, Parties and Norwegian Electoral Politics 1957-1985. Trondheim:
Tapir.

Listhaug, O. 1997. 'The decline of class voting,' in Strøm, K. & Svåsand, L. G., eds.,
Challenges to Political Parlies: The Case of Norway. Michigan: The University of Michigan
Press.

Listhaug, 0., Homberg, S. & Sänkiaho, R. 1998. 'Partisanship and EU Choice,' in Jenssen,
A. T., Pesonen, P. & Gilljam, M., eds., To Join or Not to Join. Three Nordic Referendums
on Membership in the European Union. Oslo: Scandinavian University Press.

Mannheim, K. 1972. 'The problem of generations,' in Kecskemeti, P., ed., Essays on the
sociology of Knowledge. London: Routledge and Kegan Paul.

Miller, A. H. 1986. 'Partisan Cognitions in Transition,' in Lau, R. R. & Sears, D. 0., eds.,
Political Cognition. Hillsdale: Lawrence Erlbaum.

Miller, W. E. 1991. 'Party Identification, Realignment, and Party Voting: Back to Basics,'
American Political Science Review 85, 557-68.

Miller, W. E. & Shanks, J. M. 1996. The New American Voter. Cambridge: Harvard
University Press.

Narud, H. M. & Valen, H. 1996. 'Decline of electoral turnout. The case of Norway,'
European Journal of Political Research 29, 235-56.

Popkin, S., Philip, C. & Smith, J. A. 1976. 'What Have You Done for Me Lately? Toward
An Investment Theory of Voting,' The American Political Science Review 70, 779-805.

Reiter, H. L. 1989. 'Party Decline in the West. A Skeptic's View,' Journal of Theoretical
Politics 1,325-48.

Richardson, B. 1991. 'European Party Loyalties Revisited,' American Political Science Review
85, 757-75.

Schickler, E. & Green, D. P. 1997. 'The Stability of Party Identification in Western
Democracies. Results from Eight Panel Surveys,' Comparative Political Studies 30,
450-83.

Statistics Norway 1998. Ukens slat ist ikk No. 6. Oslo: Statistics Norway.

Støm, K. & Svåsand, L., eds. 1997. Challenges to Political Parties. The Case of Norway.
Michigan: The University of Michigan Press.

Svåsand, L., Strøm, K. & Rasch B. E. 1997. 'Change and Adaption in Party Organization,'
in Støm, K. & Svåsand, L., eds., Challenges to Political Parties. The Case of Norway.
Michigan: The University of Michigan Press.

Thomassen, J. 1976. 'Party identification as a cross-national concept: Its meaning in the
Netherlands,' in Budge, 1., Crew, I. & Farlie, D., eds., Party Identification and Beyond.
New York: John Wiley.

Valen, H. 1973. 'Norway: "No" to EEC,' Scandinavian Political Studies 8, 214-26.

Valen, H. 1981. Valg ogpolitikk. Oslo: NKS-forlaget

Valen, H. 1991. 'Norway,' in Franklin, M., Mackie, T. & Valen, H., eds., Electoral Change.
Responses to Evohing Social and Attitudinal Structures in Western Countries. Cambridge:
Cambridge University Press.

Valen, H. & Aardal, B. 1983. Et valg i perspektiv. En studie av stortingsvalget 1981. Sosiale
og økonomiske studier 54. Oslo: Statistics Norway.

Zaller, J. R. 1992. The nature and Origins of Mass Opinion. Cambridge: Cambridge University
Press.