Resumé

The aim is to analyze short-term fluctuations in Danish parliamentary party cohesion on the backdrop of an American electoral pattern in party cohesion. A Danish cycle is documented: party cohesion in relation to voting behavior is especially high just after an election, then it drops to rise again as election time approaches. A rational choice re-election model predicts the rise in party cohesion, but an obligation actualization model predicts the full cycle. Elections actualize Danish MPs' moral obligation to their party. Where American party cohesion drops in an election year, Danish party cohesion rises when an election approaches. This may be explained by different preferences in the American and Danish electorates: Danish voters value party cohesion per se, American voters do not.

Introduction

In his classic work on party cohesion, Ozbudun defined party cohesion as follows: '. . . the term cohesion suggests an objective condition of unity of action among party members, which may or may not be the function of disciplinary repressions' (1970, 305). Thus denned, the concept only refers to uniform action and degrees hereof. As it should be, it is absolutely silent about the factors that may have induced this condition.

Ozbudun noted that party cohesion can be analyzed both at the legislative party level and at the level of extra-parliamentary party organizations This article is only concerned with the legislative party level. Furthermore, the definition says nothing about what kinds of action must be uniform. Several aspects of action are possible candidates (cf. Worre 1970; Crowe 1983). One aspect of party cohesion which is easily measured is the extent to which politicians vote together with their party group in parliament, and this is the aspect which will be analyzed here.

There is a respectable amount of work in the field of party cohesion (for some references other than this article's see Maor 1997). Ozbudun's country comparative work has already been mentioned, but there are many others including some on fluctuations in party cohesion (Mughan 1990; Hurley &

---

* Asbjørn Skjæveland. Department of Political Science. University of Aarhus, Universitetsparken, Aarhus C. Denmark. E-mail: Skjaevelandl^ps.au.dk

Wilson 1989; Patterson & Caldeira 1988). The present analysis continues
this line of research by studying for the first time short-term fluctuations in
Danish party cohesion.

A party cohesion pattern related to the electoral term seems to exist in the American Congress: Party cohesion tends to be higher in the year after an election than in the year before the next. A party vote is a roll-call vote where a majority of Republicans vote against a majority of Democrats. A party unity score is the average proportion of party votes where members of a party follow a majority of their party. Harmel & Janda have looked at party unity scores for Democrats and Republicans in The House of Representatives and the Senate for the period 1954 through 1980. The four averages varied from 64 percent party unity to 69 percent. However, in election years, party unity was about four percentage points lower than in non-election years (Harmel & Janda 1982, chapter 6). A similar pattern has been found by Patterson & Caldeira (1988) for party votes in the period from 1949 through 1984. Harmel & Janda did not report levels of statistical significance, but Patterson & Caldeira did and the House pattern (but not the Senate pattern) reached an acceptable level of 0.06 (which is satisfactory the fact that it relates to population data (cf. Thomsen 1997)).

One explanation suggested by both Harmel & Janda (1982) and Patterson & Caldeira (1988) is that the re-election motive makes members of Congress worry more about what their voters want and less about what their party wants when an election is approaching.

The purpose of this article is to examine short-term fluctuations in Danish parliamentary party cohesion in relation to the electoral term: Can a pattern be identified? If so, how can it be modeled and how does it relate to the American pattern?

The rational choice approach assumes highly purposive actors, and from this assumption models of American and Danish party cohesion cycles based on an MP re-election motive can be deduced. The first question is whether the pattern to expect in Denmark is like the American one.

Models of a party cohesion cycle based on other motives can also be specified within the rational choice approach. It is a further purpose of the article to find out if one or more of these models fit the Danish data or if it would seem that in this particular case it is more fitting to apply a more sociologically oriented model. The sociological approach assumes inter alia that actors do what they feel or think they should do, i.e., they do what they deem appropriate (cf. Hall & Taylor 1996).

Rational choice re"-election models are proposed, but only as a starting point. After that, an electoral party cohesion cycle will be documented. Finally, various other models including an obligation actualization model will be discussed in relation to the Danish case.

Rational Choice Re-election Models for the US and Denmark

Not all rational choice theorists exclude altruistic concerns, but what has been termed 'mainstream rational choice' (Ward 1995, 79) operates on the assumption of egoistic motives exclusively, and this 'mainstream' or 'hard' rational choice features many important theorists (e.g., Downs 1957, 27-28; Dunleavy 1991, 165). Egoism makes MPs focus on their own re-election (among other things such as the ministerial limousine not directly included in the present model), and rational choice tells us, as mentioned, that they will do so in a strategic, means-efficient way (Ward 1995,79).

To get to a model, empirical assumptions about payoffs are needed, and when dealing with MPs hoping to get re-elected, this amounts to input about voter behavior. To predict a specific election cycle, two assumptions are needed. The first is shared by the US and Denmark, while the second is not.

The first is the well-established assumption that voters are myopic and
will tend to forget MP misbehavior at the beginning of an electoral term
(Nannestad & Paldam 1994).

The second states whether voters punish parties with low degrees of party cohesion and MPs causing this. Here it is necessary to take a detour around the general levels of party cohesion in the Danish parliament and the American Congress.

As the numbers already mentioned indicate, the American level of party cohesion is low. To cite newer data, in 1995 seven percent of the party votes which the Republicans could have cast (if they had all voted in all votes) were in disagreement with a majority of the party. The corresponding number the Democrats is 15 percent (Congressional Quarterly 1996, 245). In Denmark, the parties with the highest degree of party cohesion in 1994/ 95 had no votes deviating from their party majorities (the Unity Party, the Socialist People's Party and the Social Liberal Party; cf. Table 2). The party with the lowest degree of party cohesion (the Liberal Party) had 0.22 percent of all votes (which could have been cast had everybody always voted) against the party majority. In other words: party cohesion is very high in Denmark. In the period 1990/91 to 1995/96 there were breaks with the party line in at least one party in only five percent of all final divisions (cf. Table I).¹

The most important explanations of level of party cohesion identified by Ozbudun help to account for the difference between Denmark and the US. The most important, according to Ozbudun, is that the Danish political system is parliamentary, whereas the American is presidential (Ozbudun 1970, 355-63). Only in case of impeachment may the political life of the American president depend on the cohesion of his party. In Denmark, on

the other hand, the government party or parties and support parties must always be sufficiently cohesive to be able to vote down a vote of no confidence.The most important explanation, according to Ozbudun, is an aspect of political culture: the prevalent theory of representation in a society. American liberalism and individualism make the appropriate focus of representation the individual and the appropriate mode of representation a free mandate also in relation to the Congress member's party. A democraticradicalism materialized in the primary system may be added. All in all, the prevalent theories of representation do not support party cohesion. In Europe, and thereby Denmark, collectivism is much more influential than in the US, and collectivism sees the appropriate focus of representation as groups and classes, and the appropriate mode of representation is a mandate bound by the party (ibid., 363-79; see also Skjæveland 1997, 15-18, 31,102).

Of these two explanations, which contribute to the explanation of the difference of level of party cohesion between Denmark and the US, the second one is also pertinent to rational choice re-election models of the American and the Danish party cohesion cycles, because it tells whether to expect that voters punish low degrees of party cohesion. In the US they do not, and therefore the logical thing to do for a member of Congress who wants to be re-elected is to worry more about his voters' interests and policy positions and even less about his party's when an election approaches.

In Denmark voters do (presumably) punish MPs who cause a low degree of party cohesion. As a starting point, the assumption is that MPs are individually punished by voters for breaking with the party line, or at least that many MPs believe that voters will react in this way. Danish voters can, even though the electoral system is one of proportional representation, vote for an individual candidate within a party. The voters hereby have a say over which candidates get elected and which do not. The assumption is supported by the fact that only one independent MP has been elected since 1953 - Danish voters seem to strongly prefer team players (cf. Svensson 1982, 18-19). Furthermore, as documented later in the analysis, at least some MPs state a belief in the thought of nonconformist MPs being individually punished by the voters.²

From these assumptions, and assuming that various concerns from time to time tempt MPs to break the party line,³ the researcher can predict that in Denmark the degree of party cohesion will be relatively low immediately after an election, and that it will rise as election time approaches. In other words: the cycle predicted is the opposite of the one found in the American Congress.

This prediction is slightly complicated, but not fundamentally altered, by
the fact that elections in Denmark, as in many other countries (but not the
US), can be called before the end of the term (of four years). As time passes

after an election, not only does the risk that voters will remember misbehavior at the end of the term rise, so does the ceteris paribus risk of a more or less premature call for elections. So actually two logics work in favor of the predicted electoral cycle. One clarification is needed, though: since MPs in general do not know when elections will be held, they cannot take into account the period of time until the actual election, but only the date set by the constitution. Full cycles are only expected in full electoral periods; half electoral periods will only show the first half of the party cohesion cycle.

A Danish Party Cohesion Cycle Documented

The very high level of Danish party cohesion does not leave much room for variation. However, Tables 1 and 2 show considerable relative variation when party cohesion is measured in the negative terms of breach frequencies (when the breach frequency is low, party cohesion is high). An MP breaks the party line if his party group (defined by the majority) votes either for, neutral or against a proposal, and the MP casts one of the two other votes. Two kinds of breach frequencies have been calculated: the percentage of divisions with (one or more) deviating votes (Table 1) and the number of

deviating votes expressed as a percentage of all votes which could have been
cast if all MPs of all present parties were present and voted (Table 2).

Does the variation constitute a cycle and if so, is it the predicted one? The last full electoral term which has been completely reported is the one from 1990/91:2 (i.e., the 90/91 session after the election) to 1993/94. In that period, for final divisions on bills only, the breach frequency in the first year was 5.0 percent rising to 11.9 percent in the second year and then monotonously to 1.3 percent in the year when there had to be an election. Exactly the same pattern is found when not only final divisions on bills, but also final divisions on the two Danish non-bill decision types are included.

As Table 2 shows, the pattern also holds when deviant votes are counted instead of divisions with deviant votes. Of all the votes which could have been cast in discordance with a party, 0.05 percent broke with the party line in the first year after the election (90/91:2). Then the breach frequency rose to 0.13 percent in the second year, dropped to 0.10 percent in the third year, and 0.04 percent in the fourth and final year. All in all, there seems to be a party cohesion cycle. This cycle resembles the one predicted, but is not

identical to it because of the rise in breach frequency in the beginning of
theelectoral term (90/91:2-91/92).

In order to document that the cycle described by the aggregate data in Tables 1 and 2 is not simply a random phenomenon, further analysis is needed. First of all, it has been tested whether the variation in the aggregate population numbers in Table 2 is produced by a simple stochastic process (cf. Thomsen 1997). It is not: chi-square: P < 0.001. Second, since party cohesion is not least a party attribute, something resembling the aggregate cycle should be found for most of the individual parties. A perfect match cannot be expected though, since disaggregation to the party level for only one electoral term makes the numbers highly vulnerable to random influence.

According to Table 2, all parties conformed to the aggregate pattern in the beginning of the electoral term. It is the comparison of the three last years that blurs the picture. As an absolute minimum, one would expect a party to meet one of two demands: either the breach frequency was lower in the fourth year than the average of the three previous years, or the breach frequency dropped from the third to the fourth year. Eight parties were represented in parliament in the 1990-1994 term. Of these, six met the first demand and six met the other. Seven parties met at least one demand and thereby the minimum criterion. While all this is very nice, what really blurs the picture is the fact that the breach frequency kept rising from the second to the third year for six of the eight parties. Only the Center Democrats and the Social Democrats topped the second year.

Luckily, this aberration can be explained. The relatively high breach frequencies for the Socialist People's Party and the Social Liberal Party in 1992/93 were caused by the parliamentary divisions on The Maastricht- Edinburgh Treaty. Divisions in Denmark on proposals concerning ethics, local matters and EU integration have higher breach frequencies than divisions on other topics (Skjæveland 1997).⁴ Furthermore, introduction of proposals related to EU integration is only controlled by Danish actors to a limited degree. Therefore it seems appropriate to downplay the influence on the data from the Maastricht- Edinburgh Treaty.⁵

One more thing happened in 1992/93 which was of relevance to some parties' cohesion: The Conservative-Liberal government had to resign becauseof unlawful administration of a former Conservative minister of justice and related actions (the Tamil Scandal). In its place, a government consisting of the Social Democratic Party, the Social Liberal Party, the Christian People's Party and the Center Democrats was formed. Interview data as well as breach frequencies controlled (as accurately as possible) for the cohesion cycle suggest that participation in government ceteris paribus makes the degree of party cohesion go up in Denmark (ibid.) MPs act in a more responsible way because they do not want the defeat of their own

government, and because they belong to the lawmaking nucleus, which among other things means that they must send the right signals to potential lawmaking coalition partners. So one reason party cohesion dropped from 91/92 to 92/93 for the Liberals and the Conservatives is that they lost their government status and were opposition parties in the bulk of divisions in 92/93. But everything else is not always equal. Participation in government can have adverse effects, for instance when party members disagree on their party's participation in government. This seems to be the most likely explanation of the rise in the breach frequency for the Christian People's Party in the last two years of the election term. At least the party's difficulties choosing (the left) side are known in the extra-parliamentary organization and the party's electorate.⁶

Unfortunately, the Danish standard effect of participation in government may not only explain part of the aberration, but also contribute to the (supposedly) normal pattern of especially the Social Democratic Party. All in all, the explanations of the aberration may seem too ad hoc to some readers. So even though the party check has been fairly successful, a third check on the stability of the cycle would be comforting. Fortunately, there is a long tradition for calculating party cohesion in Denmark. As Table 3 shows, the numbers available from 1953 and onwards describe party cohesion final divisions (on bills only) as the proportion of divisions with one or more votes in discordance with a party line.

The party cohesion cycle is not a deterministic pattern. This is shown in the party level discussion and even in the aggregate numbers of Table 2 relating to the years 1994/95 to 95/96 (but not in the corresponding numbers in Table 1). Random factors can distort the normal pattern. By calculating average numbers for first, second, third and the relatively rare fourth years, the influence of these factors can be minimized. Table 3 documentsthat detected cycle is, in fact, a normal Danish pattern. On

average, 8.4 percent of first year divisions had deviating votes. The numbers are 12.6 percent for second years, 9.8 percent for third years and 6.3 percent for fourth years. Furthermore, Table 3 shows that the party cohesion cycle is also found in the periods 1953-1980 and 1980-1996 taken separately. The reason for splitting the period in two is that party cohesion rose steeply during the last years of the 1970s.⁸ Combined, Tables 1, 2 and 3 document a Danish party cohesion cycle.

Further Modeling of the Danish Party Cohesion Cycle

The fall in the degree of party cohesion from the first to the second year in the electoral term was not predicted by the re-election model. A model that (post-)predicts this fall therefore has the potential for greater predictive power than the re-election model. There are two possible and likely explanations do not exclude each other, but can both be fitted into a model.

The first one is a time lag effect: some first years come after a third or fourth year when election was expected with a certainty of 50 percent to 100 percent {ceteris paribus). In such years, party cohesion is generally high. MPs who are re-elected may adapt slowly to their post-election situation. This logic is supported by the time series data. To be supportive, breach frequency numbers must be lower after elections which could be expected than after elections which, in a formal sense, could not be expected.⁹ For the period before 1980, the breach frequency for first years after expected elections was 11.3 percent. For first years after unexpected elections, it was 13.4 percent. For the period after 1980, the corresponding numbers are 1.2 percent and 4.4 percent.

The time lag effect cannot be the complete explanation, though. Not even the average breach frequency numbers for first years after unexpected elections are as high as the breach frequency averages for the second year. A natural place to search for a supplementary explanation is in the factors which (at the microlevel) have been found to explain the very high general level of Danish party cohesion (Skjæveland 1997). These are a high degree of internal policy agreement in the party groups (cf. Wetterqvist 1996, 2); a heavy work load combined with specialization, which prevents MPs from forming independent views on all decisions (cf. Svensson 1982, 21); a moral obligation or commitment to the party and its electorate (cf. Crowe 1986; Ozbudun 1970, 356); and last, and possibly least, disciplinary sanctions broadly defined (cf. Crook & Hibbing 1985; Crowe 1986). On the contrary, individual re-election does not seem to play a significant role when an election is not on the horizon (cf. Ozbudun 1970, 356; Svensson 1982, 19),

nor does it look as if socialization related to seniority plays an important
role at any time (cf. Crowe 1986).10

The most promising of these factors in relation to the party cohesion cycle is the moral commitment to the party and the electorate. This is, among other things, an obligation to vote with the party. A not too bold conjecture would be that this obligation is felt most strongly when an election stands fresh in memory. After all, it was the party and its electorate that got the MP elected. The obligation may be 'forgotten' more easily in the second than in the first year after the election. To put the point in more general terms: an election may actualize the moral commitment to the party. This formulation makes the obligation factor a possible candidate also for the explanation of the rise in party cohesion at the end of the electoral term.

12 Danish MPs were interviewedon party cohesion, and nine were interviewed
on the cohesion cycle.11 At least two respondents suggest
resembling the obligation factor as the explanation of the
rise:

If one wants to have deviant viewpoints, it's best to have them at a time when it hasn't got any immediate consequence to the party, where there is time for the opinions to mature with the voters and the voters have time to think about them. Because, if one comes out the day before an election and disagrees strongly with the party then clearly it leaves the impression with the voters that the party doesn't know what it wants to do, and that is not convenient. I don't think one scrutinizes the calendar, but I firmly believe that it unconsciously influences one's need to be a rebel.

An election actualizes MPs' moral commitments to their parties before the election, when they foresee it, and the actualizing effect increases with the likelihood of elections being called. Similarly, an election actualizes MPs' moral obligations after the election, when they clearly remember it. This is a parsimonious model which beats the re-election model when measured in terms of predictive power. Furthermore, the time lag effect may be fitted in. If an election actualizes the commitment to the party, the actualizing effect ought to be bigger, if it is a predicted election which has been taken into consideration for some time. This could account for the time lag.

To get an idea of where the (post-)predictive success of the commitment actualization model leaves the re-election model, and more generally to get an idea of the possibility of making a rational choice model of the complete cohesion cycle, it is useful once again to look at the interview material. The following interview quotation exhausts (together with the previous quotation) the material on the rise in party cohesion at the end of the electoral term:

It isn't about party pressure. . . . First, what happens is that up to an election, much of the legislation and of the parties' announcements concentrate on the points where there is 100 percent agreement on the stuff that sells. That is one process. The other process is, of course, that you must consider when to detach yourself from the others. . . . Obviously, it is more

difficult to explain to the voters that you stand for a party if you have voted against that
party almost daily or every week during the last three months before the election. . . . You
see, in most parties it is only a relatively small part of the voters who vote locally and for a
local candidate, whereas the party line, the party leaders, the election manifestos, all those
things from the party in general convert people. And clearly, if a national campaign is run on
some viewpoints, and you flutter on a rostrum in Anyville and say the very opposite, then a
voter might say: Well, I happen to agree with the main line, that is why I vote for that party.
Now I will choose another candidate.

Individual re-election is actually suggested by two respondents as an explanation the rise. The trouble is that it seems extremely difficult to account for the high degree of party cohesion in the first year of the electoral term in strategic re-election terms. For instance, MPs might out of gratitude to their voters and their party do what they want them to do - vote with the party - but true acts of gratitude are not strategic; they relate to past events, not future goals. An option left to the rational choice researcher is then to operate with strategic acts resembling acts of gratitude. These acts, however, would have to be restricted in time to the first year in order to make the model fit the data and this would violate the assumption of myopic voters who, at election time, have forgotten what happened in the fist year after the last election.¹²

Another explanation, which would be true not only to 'hard' rational choice but also to the factors found to explain the generally high level of party cohesion in Denmark, could be based on disciplinary sanctions. Might it be the case that the party leaderships put stronger pressure on backbenchers when election time approaches? None of the nine respondents asked about the party cohesion cycle confirm this, and four MPs positively deny that pressure is increased before an election. Although this might be unpleasant to admit, the total lack in the interviews of any sign of increased pressure makes it unlikely that disciplinary sanctions cause the rise in the degree of party cohesion at the end of the electoral term. Nor is it in any way obvious why stronger than normal pressure should be exerted in the first year after an election.¹³

Actually, it is difficult to think of any strategic reasons for the high degree of party cohesion in the first year of the electoral term. This makes it very difficult for any rational choice model to gain the same level of predictive power as the commitment actualization model. The possibly most promising rational choice model is based on policy agreement¹⁴ and a party vote goal as suggested in the interview quotation as well as two other interviews. To focus on policy agreement is also in line with the explanation of the generally high level of party cohesion in Denmark. When election time approaches, the parties concentrate on proposals close to their politicalcore which there is less disagreement inside the party groups. As accordance is vital to party cohesion in final divisions, the degree of party cohesion naturally rises. This can be a voter-oriented logic in which high

party cohesion is only a side effect. Party core proposals are put to the fore because they are expected to win over voters. A high degree of party cohesion is an unintended consequence. But one respondent points to a slightly different logic where the party core proposals are a means to achieve the high degree of party cohesion needed in an election campaign. To make the model fit the beginning of the electoral term, it would have to be assumed that parties to a large extent keep their promises from the election campaign and therefore keep making proposals close to their political core. However, it will further have to be assumed that they will do so much more keenly in the first year after the election than in the second.

The individual MP's concern for policy per se makes the model fall outside the range of 'hard' rational choice, but the model faces a tougher problem also discussed in relation to the re-election model: as the policy agreement model is formulated now, in order to (post-)predict the first year cohesion level, it breaks the assumption of myopia in the electorate. There is no reason for a party to keep its promises especially keenly in the first year after the election when the myopia assumption says that voters will forget this promise keeping before the time of the next election. To be a worthy rational choice challenger of the obligation actualization model, this problem have to be solved without resorting to non-strategic assumptions such as true gratitude to the parties' voters.¹⁵

It remains to be seen if this problem can be solved within the rational choice approach or if another rational choice model can be specified which captures the complete cycle. However, true gratitude to the party and its voters rests comfortably in the obligation actualization model. This is because the obligation actualization model is a sociologically oriented model and because the sociological approach is not unbreakably tied to goaloriented

Conclusion

A Danish party cohesion cycle has been documented. The degree of party cohesion in final divisions in the Danish parliament is especially high in the first and fourth years of the electoral term. From the first to the second year, the degree of cohesion drops, after which it rises to the third and further on to the fourth year (if this year is reached).

All three explanatory factors - individual re-election, obligation actualizationand policy agreement - may be operative in at least a part of the electoral term as the interview material hints. It might also be the case that commitment to the party when an election approaches is, to some extent, instrumental to the individual MPs' hope for re-election (cf. Ozbudun 1970, 356).¹⁶ These thoughts, though, tend to downplay the criteria of

parsimony and predictive power to the possible benefit of empirical realism. So far, the parsimonious model with the greatest predictive power in relation to Denmark is not the individual re-election model, which (in its American version) fits the American data nicely, but the sociologically oriented obligationactualization An election actualizes the MPs' moral commitmentsto parties before and after an election, thereby raising party cohesion both in the beginning and in the end of the electoral term.

Comparing an American fixed two year electoral term¹⁷ and a Danish term of maximally four years is not without problems. Provided that it is correct to see the American cycle as a pre-election phenomenon where members Congress worry more or less about the next election, the American term should be compared with the last years of the Danish electoral term. Then the behavioral patterns of Danish and American national legislators are roughly opposite. They both worry more about what their voters want when an election approaches, but this makes American members of Congress concerned with their voters' policy positions and interests and Danish MPs more concerned about the party line. Therefore American party cohesion decreases as an election approaches whereas Danish party cohesion increases.

---

NOTES

1. The Danish and American numbers are not fully comparable because the American party unity score is only calculated for party votes and because of differences in pairing systems.

2. Empirical evidence on Danish voters punishing low degrees of party cohesion and nonconformist MPs is, admittedly, sparse. It may be the case that nonconformist MPs are only punished indirectly through the bad fate of their party in the following election. That parties are punished for a low degree of party cohesion is suggested in an empirical investigation (Thomsen 1984).

3. For a further discussion of this assumption, see note 12.

4. The aggregate pattern observed is also found when all divisions on proposals concerning local matters and EU integration are excluded.

5. On top of this, the Maastricht-Edinburgh Treaty implicated three highly linked divisions exactly the same persons breaking the party line in each division. This means that the breach frequency numbers for the Socialist People's Party and the Social Liberal Party in 1992/93 are somewhat exaggerated. Yet, other divisions in the period were linked, and the data have not been generally controlled for division linkage.

---

6. Some might claim that this argument makes government takeovers too good an explanation. would seem that both a rise and a fall in the degree of party cohesion could be explained by the faet that a party joined government. Although this is actually the case, one would normally expect the degree of party cohesion to rise when a Danish party joins government, and one would have to put up a special argument to support the opposite effect.

7. Depending on the source, short parliamentary sessions have either been neglected or included in the numbers for an adjacent parliamentary session. In at least two and at the most four instances, sessions have been placed in a suboptimal way. However, recalculations including worst case recalculations show that the general results are robust. The numbers for 1953/54 to 64/65 only cover bilis proposed by the government. period 1965/66 to 1970/71 is not included, as party cohesion has not been published for these years one by one. The numbers from 1971/72 to 1985/86 comprise all passed bilis, while the numbers for 1986/87 and onwards include all final divisions on bilis. The percentage bases only differ slightly. For example, 95 percent of the 239 final divisions on bilis in 1993/94 were on government proposals and 97 percent of all bilis were passed. The last years in the electoral term only comprise few instances, but every case comprises many divisions. A regression analysis controlled for the two time periods shows that the breach frequency in second years is significantly higher than in the other years (P < 0.001).

8. The best explanation so far for the rise in party cohesion from the first to the second period is that increased work load and specialization have forced MPs to take over the party group viewpoint more often than before. In many cases, they simply do not have the time to develop a stance of their own (Svensson 1982; Mikkelsen 1994; Skjæveland 1997).

9. Elections which are expected in a formal sense are defined as elections in third and fourth years. They also include the first election in the new one-chamber parliamentary system (in 1953), which was a necessary consequence of the amendment of the constitution Riges Grundlov 1953, §89). In the period analyzed, there were nine first years after expected elections and six after unexpected elections.

10. As an alternative to the re-election explanation, both Harmel & Janda (1982) and Patterson & Caldeira (1988) suggest that newly elected members of Congress may be less prone to deviate from the party majority in the first than in the second year. However, the Danish numbers seem to rule out both the possibility that the increase in party cohesion in the end of the term and the decrease in the beginning of the term are related to seniority.

11. The respondents were chosen from The Unity Party, The Social Democratic Party, The Center Democrats, The Conservative People's Party, The Progress Party and The Danish People's Party. Both front and backbenchers were included. The MPs from The Unity Party and the Danish People's Party were not interviewed specifically on the cohesion cycle, because their parties had not been represented in parliament for a full electoral term.

12 The Danish re-election model has another problem if it is to stay within 'hard' rational choice, namely how to specify the concerns that make MPs deviate from the party line. If the reason here is also re-election, then the model dynamics are lost (or, even worse, reversed as in the US). Both costs (the general punishment for breaking the party line) and benefits (expressing opinions that may be popular with some voters, possibly resulting in the MP's getting their vote) of breaking the party line would rise as an election approaches. The most obvious reason for breaking the party line, which is also supported in the interviews, is (non-instrumental) policy disagreement with the rest of the party. This fits nicely into a policy agreement model, and it is not at odds with the obligation actualization model, but the explanation does seem hard to understand in strictly egoistic terms.

13. Timmermans is close to predicting the full cycle when he states that '. . . internal party cohesion is likely to have the party leadership's special attention shortly before and after parliamentary elections. Before elections because party leaders are concerned with the image of their party toward the voters, and after elections because an openly divided party stands weak in coalition negotiations' (Timmermans 1998). This argument a strong disciplinary flavor, but it does not explain the relatively high level of party cohesion measured in the first year after all. Government formation has ended when parliament starts to vote on bilis and similar decisions, and it is for these decision types that party cohesion has been measured.

---

13. Timmermans is close to predicting the full cycle when he states that '. . . internal party cohesion is likely to have the party leadership's special attention shortly before and after parliamentary elections. Before elections because party leaders are concerned with the image of their party toward the voters, and after elections because an openly divided party stands weak in coalition negotiations' (Timmermans 1998). This argument a strong disciplinary flavor, but it does not explain the relatively high level of party cohesion measured in the first year after all. Government formation has ended when parliament starts to vote on bilis and similar decisions, and it is for these decision types that party cohesion has been measured.

14. Varying policy agreement is also suggested by Patterson & Caldeira as a possible explanation of the American pattern in party voting (Patterson & Caldeira 1988, 117).

15. With this model, excepting one factor, all explanations mentioned earlier in relation to the general Danish party cohesion level have been tried for models of the cycle. (The individual re-election factor has also been tested, even though it is not among the general explanations). The factor excepted is the combination of work load and specialization. The rather limited short-term fluctuations in work load do not correspond short-term fluctuations in party cohesion (Skjæveland 1997).

16. See note 2.

17. Individual members of the House of Representatives are elected for two years, whereas individual members of the Senate are elected for six years. However, the terms of approximately one third of the senators expire every two years.

ACKNOWLEDGMENTS

This article is based on my master thesis 'Ydre partisamstemmighed i Folketinget', especially chapter 5: 'Diakron komparation' (Skjæveland 1997). Many people at the Department of Political Science, University of Aarhus have made valuable comments on the article. I am grateful to the members of the section for Comparative Politics - particularly Erik Damgaard and Jørgen J. Poulsen - and to Jørgen Grønnegård Christensen, Asbjørn S. Nørgaard, Christoffer Green-Pedersen and Søren Risbjerg Thomsen, and to Annette B. Andersen for careful proofreading. I would also like to thank the editors and the referees for their valuable comments.

REFERENCES

Congressional Quarterly January 27, 1996.

Crook, S. B. & Hibbing, J. R. 1985. 'Congressional Reform and Party Discipline: The Effects
of Changes in the Seniority System on Party Loyalty in the US House of Representatives,'
Britisk Journal of Political Science 15, 207-26.

Crowe, E. 1983. 'Consensus and Structure in Legislative Norms: Party Discipline in the
House of Commons,' The Journal of Politics 45, 907-31.

Crowe, E. 1986. 'The Web of Authority: Party Loyalty and Social Control in the British
House of Commons,' Legislative Studies Quarterly 11, 161-85.

Damgaard, E. 1973. 'Party coalitions in Danish lawmaking 1953-1970.' European Journal of
Political Research 1, 35-66.

Damgaard, E. & Svensson, P. 1989. 'Who go ver
European Journal of'Political Research 17. 731-45.

Danmarks Riges Grundlov. 1953.

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

Hall, P. A. & Taylor, R. C. R. 1996 'Political Science and the Three New Institutionalisms.'
Political Studies 44, 936-57.

Harmel, R. & Janda, IC. 1982. Parties and their environments: limits to reform. New York:
Longman.

Hurley, P. A. & Wilson, R. K. 1989. 'Partisan Voting Patterns In the U.S. Senate, 1877-1986,'
Studies Quarterly 14, 225-50.

Maor, M. 1997. Political Parties & Party Systems. London: Routledge.

Mughan, A. 1990. 'Midterm Popularity And Governing Party Dissension In the House of
Commons, 1959-79,' Legislative Studies Quarterly 15, 341-56.

Nannestad, P. & Paldam, M. 1994. 'The VP-function: A survey of the literature on Vote
and Popularity Functions after 25 years/ Public Choice 79, 213^45.

Ozbudun, E. 1970. Party Cohesion in Western Democracies: A Causal Analysis. Sage
Professional Papers in Comparative Politics. Beverly Hills, California: Sage Publications.

Patterson, S. C. & Caldeira, G. A. 1988. 'Party Voting in the United States Congress,' British
Journal of Political Science 18, 111-31.

Pedersen, M. N. 1967. 'Consensus and Conflict in the Danish Folketing 1945-65,'
Scandinavian Political Studies 2, 143-66.

Skjæveland, A. 1997. 'Ydre partisamstemmighed i Folketinget,' Master thesis (unpublished),
Aarhus University: Department of Political Science.

Svensson, P. 1982. 'Party Cohesion in the Danish Parliament during the 19705,' Scandinavian
Political Studies 5, 17-42.

Thomsen, S. R. 1984. 'Udviklingen under forholdstalsvalgmåden (1920-84)' in Elklit, J. &
Tonsgaard, 0., eds., Valg og Vælgeradfærd. Århus: Forlaget Politica.

Thomsen, S. R. 1997, 'Om anvendelse af signifikanstests i ikke-stikprøve situationer,'
Teachingnote (unpublished), Aarhus University: Department of Political Science.

Timmermans, A. 1998. 'Policy Conflicts, Coalition Agreements, and Cabinet Governance,'
Paper, ECPR workshop, Warwick.

Ward, H. 1995. 'Rational Choice Theory' in Marsh, D. & Stoker, G., eds., Theory and
Methods in Political Science. Houndmills, Basingstoke, Hampshire and London: Macmillan
Press Ltd.

Wetterqvist, B. 1996. 'Partisammanhållingen i Riksdagen - en voteringsanalys av partisammanhållingen
riksmotet 1994/95 i jåmforelse med riksmotet 1969,' Seminar paper
(unpublished). Goteborg University: Statsvetenskapliga Institutionen.

Worre, T. 1970. 'Partigrupperne i Folketinget: Et magtcentrum i det danske politiske system,'
Økonomi og Politik 44, 143-88.