University of Aarhus

1. Introduction

In several fields in political science, recent years have witnessed what has been called a 'return to history' by a prominent participant observer.¹ Overt indicators of this development can be found in book titles and conference syllabuses;² more important, an increasing interest in the historical foundations of contemporary political institutions has found its expression in attempts at a more rigorous application of different kinds of diachronical approaches and at greater conceptual and methodological clarification.³ Moreover, important work has been done in several places to compile data archives including data on economic, social, and political factors back in history for different countries,⁴ which in turn has sparked off and facilitated inquiries into the historical dimension of politics.⁵

It has been pointed out that this development has ancient ancestors; Herodotus has been mentioned in this connection.⁶ However, the attitude of most political scientists toward history may presumably be in agreement with a kind of Aristotelian view, that '... the becoming is for the sake of the thing, not the thing for the sake of becoming.'⁷ A political scientist will usually not be interested in history merely for history's own sake. For example, the rationale for the intrusion of historical interest into the once programmatically nomothetic and ahistorical study of voting behavior has been sketched by Rokkan and Lipset⁸ in the following way, which, mutatis mutandis, may well be valid for the development in other fields of political science: '... as soon as we move into comparative analysis, we have to add a historical dimension. We simply cannot make sense of variations in current alignments without detailed data on differences in the sequences of party formation and in the character of the alternatives presented to the electorates before and after the extension of suffrage. We have to carry out our comparative analyses in several steps: we first have to consider the initial developments toward competitive politics and the institutionalization of mass elections, we next

must disentangle the constellation of cleavages and oppositions which produced the national system of mass organizations for electoral action, and then, and only then, we can make headway toward some understanding of the forces producing the current alignment of voters behind the historically given alternatives.'

Although a systematic and conceptually clarified approach to the history of political phenomena may be something of a novum in modern political science, the historical dimension of certain institutions in contemporary political systems has never been totally lost sight of. This is especially true of the study of parties and party systems. Here, more or less well-founded references to historical circumstances, have usually been abundant, since hypotheses and theories regarding parties and party systems - especially those forwarded by the institutional-morphological school-have always encompassed genetic explanations to a considerable degree ⁹ Neither does it seem difficult to intuitively grasp the reason for the relative preeminence of the historical aspect in this field of political science. In most Western countries, party systems of today are relics of the past in the sense that the party alternatives presented to the electorate and the relation of strength between them have remained constant over quite a long period. According to Rokkan and Lipset¹⁰ 'the party systems of the 1960's reflect, with jew but significant exceptions, the cleavage structures of the 1920'₅. This is a crucial characteristic of Western competitive politics in the age of "high mass consumption": the party alternatives, and in remarkably many cases the party organizations, are older than the majorities of the national electorates.' Thus, in studying parties and party systems, one is forced, time and again, to go back in history to search for the explanatory forces which once produced the constellation we are confronted with today.

In a study of the party system in the Danish Folketing 1945-68,^u Mogens N. Pedersen, Erik Damgaard, and the present author tried to sketch a rough map of the party space in the Folketing and found a remarkable degree of stability in the distances between the parties. As a natural consequence, we asked for a genetic explanation of the structure found. So far, virtually no research into the historical dimension of the Danish party system has been reported by Danish political scientists. Our knowledge of the history of Danish parties and the Danish party system is entirely based on the work of historians.¹² Therefore, it may be relevant to touch upon some problems connected with that kind of research in a Danish context.

The first problem confronting a political scientist who, for a better unders landing of some peculiar political phenomenon, decides to engage in historical research may well be how to escape the trap of infinite regress: If it is necessary to view a phenomenon in terms of its forerunners, where are we to stop in our search for the 'prime cause' and what should be included? The concept of relevance - always a manipulative one - rapidly becomes nebulous, as we move back in history. Thus, where are we supposed to set the beginning of the development which led to the Danish party system of today, and on which points of the development are we to concentrate our efforts?

According to Rokkan,13 'the terminus a quo is the conflict over the cultural

religious identity of the emerging nation state in the sixteenth century'. This refers explicitly to 'the study of the sequences in the establishment of the rules of the electoral games and in the formation of party alternatives'. Nevertheless, from the point of view of a Danish political scientist, it may well prove profitable to approach the study of the historical dimension of the Danish party system with a somewhat shorter time-period in mind.

Here, we may consider the year 1848 an appropriate terminus a quo by several standards. First, with the invocation of the Constitutional Convention, the Danish political system experienced a quite sudden and unprepared change from a kind of enlightened monarchical absolutism and bureaucratic paternalism to a representative system of government. The elections to the Constitutional Convention by near-universal manhood suffrage mark Denmark's first step into the age of mass politics. Next, and maybe of more direct importance for the study of the Danish party system, with the Constitutional Convention, the very first opportunity was given to political groupings to present themselves as alternatives to the electorate on a national level by means of representative action. If we want to understand the development of the Danish party system, we presumably cannot neglect the events in the Constitutional Convention, since we may hypothesize that it was the group formation in the Constitutional Convention which, to a considerable degree, provided the electorate with alternatives to choose from in the elections which followed and with a basis on which to judge the candidates' political positions. The importance of this aspect can be seen by the fact that not less than 114 of the 157 members of the Constitutional Convention later pursued a parliamentary career.

A second problem is the choice of suitable data. Here it must be remembered that the process of party development in Denmark did proceed in different arenas. At least three lines of development have to be distinguished. One can concentrate on the development in parliament, the organizational development, or the development in the electoral arena. Although closely related, the processes of party formation and development did not coincide in these three arenas. Thus, inferences from one arena to another are extremely dangerous, and the analysis should proceed separately for each of them by utilizing different sets of data.

As far as the development in the parliamentary arena is concerned, roll calls may seem to provide a valuable data set. By studying the divisions by suitable methods, one should be able to identify both the voting groups in a given period and the dimensions of conflict which structured the 'party space' at that time. Thus, both alternatives presented to voters and cleavage structures which produced them could be approached in that way. Although this prospect seems appealing at first glance, voting records have not been systematically exploited for the study of party development in Denmark or in other countries, with few notable exceptions.¹⁴

This may be because of the problems related to this kind of data, some of which will be discussed in a later section. Partly, it may simply reflect a certain reluctance on the side of the historians, who hold a near monopoly on the study in this field, towards the statistical devices necessary for that kind of analysis.

Finally, the danger of precipitating into plain historicism must be mentioned as a possible consequence of the new interest in the historical dimension of political phenomena. Not all researchers, even the most prominent, have always managed to avoid this trap.¹⁵

By focusing on legislative voting behavior in the Constitutional Convention 1848 -49 by means of a roll call analysis, this paper concentrates upon the parliamentary aspect of party development in the Danish political system at its very beginning. The main interest will be with the dimensions of conflict separating the voting groups in the roll calls. Beside the purely descriptive task, this paper seeks to explore the possibilities which modern legislative behavior research techniques may contain for future quantitative historical analysis of party development.

The Danish Constitutional Convention met on October 23, 1848. Its proceedings lasted until June 5, 1849. Of its 152 members, 114 were elected by universal manhood suffrage,¹⁶ and 38 were royal appointees. By-elections brought the number of people who had ever been members of the Constitutional Convention up to 157.

In the nine months of its existence, the Constitutional Convention passed only
three bills: the Constitution, a law on military service, and the electoral law. But
within the scope of these bills, a wide range of issues was raised and voted upon.

Historical descriptions of the Constitutional Convention and its work are abundant ¹⁷ Their qualities notwithstanding, they exhibit certain shortcomings from the viewpoint of a political scientist interested in the development of the Danish party system. First, all descriptions deal with the group-formation process in the Constitutional Convention in terms of purely unidimensional concepts. Normally, a rightleft dichotomy or a right-center-left trichotomy is applied.¹⁸ Although this may be a parsimonious and convenient way of description, it may conceal valuable information about the position of different groups with regard to the main dimensions of conflict in the Constitutional Convention and with regard to each other.¹⁹

Next, Danish historians are not agreed on what did produce the ordering of parties from left to right, i.e., the nature and character of the cleavages. Most stress the urban-rural conflict in the elections to the Constitutional Convention and implicitly in its proceedings; however, they do not agree as to the extent and the nature of this conflict. According to some historians, the social element, the class conflict, must be viewed as the central issue.₂₀ Others stress the political character of the conflict²¹ or tend to diminish its importance altogether.22

Moreover, the historians do not agree as to the number of groups existing in the Danish Constitutional Convention and the location of certain prominent members in one of those groups.²³ Somewhat different listings may be found in different works.

Finally, in virtually no case is one able to decide what kinds of data make up the
basis for the historians' different descriptions. Thus, the differences may well be
products of different data sets.

Despite these obvious shortcomings, enough is known about the proceedings of
the Constitutional Convention to allow us to check the external validity of the

findings reported in this paper. This fact makes the Constitutional Convention an
attractive object for a mainly exploratory quantitative historical research.

2. The Data: Problems of Validity

As mentioned before, this analysis is based primarily on the roll calls hi the Constitutional Convention. A total of 44 roll calls is recorded in the Beretninger om Forhandlingerne paa Rigsdagen, the Danish equivalent to Hansard, together with each member's stand on each roll call. Although it was possible to vote both 'aye', 'nay', and 'abstain', the last opportunity was so seldom used that the problem of how to handle abstentions is of minor importance. In the few cases this voting alternative actually was chosen, the abstainers are coded together with the absentees with a comparatively small loss of information.

In his study of voting patterns in the British House of Commons, Aydelotte24
adopts a very optimistic view on the quality and suitability of roll calls as data in
historical research in a passage which deserves quotation in full:

Yet there exists a source, a body of materials, on this subject which is not only readily available but is also fuller and more reliable than most sources which historians have at their disposal. This is the division lists in Hansard. The peculiar value of these lists is that they contain expressions of opinion upon important subjects from men who are no longer available for questioning. It does not strain terms too much to say that, imaginatively used, they can constitute a kind of questionnaire which one may submit to the dead and on which they will give us their replies. Further, the information about the votes of members of Parliament is perhaps more complete and more certain than anything else that is known about them. The social background, their economic interests and their relations with their constituents can be studied with profit, but on these matters the information is and always will be fragmentary and there are nuances which it is difficult now to recapture. By contrast, the stands which those men publicly adopted on the major issues of the day are documented by the division lists with a wealth of detail, repeated corroboration and, I have reason to believe, a relatively high degree of accuracy.

Presumably, most students of legislative behavior today would adopt a somewhat more cautious view on the qualities of roll calls as data. Normally, the problems in utilizing roll calls are considered abundant and the danger of arriving at artifacts produced by data or methods is ever present in this kind of analysis. Thus, if a student uses all roll calls at hand, he may be criticized for counting the most vital bills and the trivial matters as equal²⁵ and hereby distorting the 'true' picture. If he concentrates on selected 'key divisions'²⁶ - however defined -he

may be criticized for dramatizing relations in parliament.27 If he chooses roll calls
which fit a particular scale pattern, he may find himself exposed to criticism for
undue selectivity and ex post facto reasoning.

When it comes to making inferences from the result of roll call studies, the air is again filled with caveats: Voting on roll calls only constitutes one particular aspect of the parliamentary process, and not necessarily the most important;²⁸ the final vote does not tell us anything about compromising, coalition formation, and logrolling in previous stages of the passing of a bill;²⁹ the stand adopted on a particular roll call does not tell us anything about the intensity of the attitudes toward the issue involved;³⁰ the stand on a particular roll call may be produced by a vast array of forces influencing the individual member;³¹ voting together does not necessarily mean cooperation, hence voting groups isolated by statistical operations on roll call data need not reflect existing social groups³² -in short, interferences should not be made from studies of roll call voting to what went on in other areas of parliamentary behavior, to the forces which influenced the roll call behavior of representatives, or to the attitudes of the representatives toward the issues involved.

Although all these criticisms, or most of them, may be at least partly valid in many situations, so far they have not precluded legislative behavior research based on roll call data either in Denmark or in other countries. But in each particular case they call for a careful consideration of what we actually are studying when concentrating on the analysis of roll call data, and hence for a careful delineation of the scope of possible inferences.

Many of the possible criticisms listed above could easily be applied to this analysis. Thus, the roll calls on which it is based make up only a part, moreover a very minor part, of all divisions which took place in the Constitutional Convention. A great majority of divisions were taken by anonymous vote, without recording the names of the representatives and their stands on the issue. Further, we know that internal compromising and logrolling did take place in the Constitutional Convention and hence that the stand adopted in roll call voting does not necessarily reflect the personal conviction of the member concerned. Thus, a member of the Constitutional Convention noted in his diary after one roll call: This surely is the first constitutional, or better parliamentary comedy performed here. I'll bet that not ten of the assembly's members have supported this amendment out of conviction; most have followed their leaders ... and some have voted for it because they regarded it less evil than the bill.'³³ However, the choice of roll calls as data may be less arbitrary and more defensible than it may look like at first glance.

In the first place, the sole reliance on roll calls as data does not leave us with a hodge-podge of several inhomogeneous divisions, the only common trait of which is constituted by a common mode of division. When the debates in the Constitutional Convention are scrutinized - especially the debate on the proceedings for the assembly - we find many remarks and hints that roll call voting was to be reserved for decisions on matters of special importance, where the members had an obligation toward their constituents to make their position clear.³⁴ Thus, it is

among the roll call divisions we may expect to find the 'key divisions' of the Constitutional
Convention. A glance down the list of issues decided upon by roll call
voting tends to confirm that expectation, except for a few cases.

Moreover, if we are ready to skip the conception of 'what really did happen in the Constitutional Convention' as our key notion, we may argue that in studying the roll calls, we are focusing on a very central aspect of the proceedings of the Convention. Admittedly, the roll calls are too one-sided and incomplete a basis to allow inferences as to 'what really did happen'. But what they can tell us is at least what image each of the members of the Constitutional Convention was ready to show to his constituents and prospective voters.³⁵ Regardless of his innermost attitudes, logrolling, secret bargaining, and such things, which naturally were a part of the political process in the Constitutional Convention, by his voting in the particular roll calls the individual member did identify himself vis-a-vis the public with a certain position on the issues involved and in the perception of his voters grouped himself together with other members taking the same stand. In view of the fact that 114 of the Constitutional Convention's members pursued a parliamentary career later on, we may hypothesize that most members were aware of this fact. Furthermore, their awareness may well have been sharpened by the fact that the stand which some members of the consultative provincial assemblies (Raadgivende Stænderforsamlinger, 1835-1848) had adopted in the vote on the electoral procedure for the election to the Constitutional Convention had proved crucial for their fate at the elections.³⁶ In short, what we are concentrating upon in the study of the roll calls may be characterized as an important aspect of the defining and presenting of alternatives to the electorate - certainly not an unimportant aspect if the main interest is in the process of party development.

Besides data on voting behavior, background information on age, education, occupation, social status, and previous and following political career were collected for each member from biographical sources³⁷ and data on constituency background and election margins from the electoral statistics.

The difficulties involved in gathering and utilizing this kind of data hardly need any special elaboration.³⁸ Suffice it to say that they effectively inhibit all but very crude classifications. The possible loss of information incurred hereby must be judged against the fact that a detailed classification scheme more often than not tends to become totally inoperative in legislative behavior research, since crosstabulation will inevitably produce empty or near-empty cells as soon as there is control for only one variable.

3. Methods

To identify groups of legislators on the basis of voting behavior on a set of roll calls,
or alternatively groups of roll calls on the basis of similarity in voting patterns, a
great variety of clustering procedures has been developed and put to use in legislative

behavior research.39 Most of them differ somewhat with regard to levels of
measurement required and criteria applied in grouping the objects together. Different
methods, therefore, may easily lead to somewhat different results.

In the study reported here, Hierarchical Group Analysis as developed and described by Ward⁴⁰ has been used to identify voting groups in the Constitutional Convention. This method starts out from the total set of n objects, e.g. legislators, considering each of the objects as a group of size one and proceeds stepwise, reducing the number of groups from n to n-(n-l). At each stage in this iterative process two groups are combined to bring out the reduction in the number of groups. The decision as to which two groups are to be combined next is made on the basis of a previously defined 'objective function' or 'value reflecting function.'

For example, such a function could be the loss of information as expressed by
the error sum of squares of the form

where x* represents the score of the rth object on a particular variable. The grouping would have to be made in a way in which every step would minimize the 'loss' incurred by it. In the above example, that combination would be chosen which would result in the smallest: increase in the total error sum of squares. On the basis of the objective function, a 'loss function' can be defined as the deviation from the desirable as defined by the objective function. In the above example, if the 'objective' was an error sum of squares of value zero, the loss function would consist of the actual values of the error sum of squares through the iterative process. Thus, the method also yields a quantitative expression for the loss of information during the grouping process.

Since the procedure in each stage of the process utilizes the grouping which resulted from the last of the previous stages, it differs from the theoretically optimal way of building clusters. That way would be at the mth stage, where the n initial objects were to be clustered into n-m groups, to find that combination of objects which would minimize intragroup and maximize intergroup differences according to a previously defined measure. Unfortunately, that procedure would involve computational tasks which are prohibitive already with very minor problems. Ward's method constitutes a reasonable compromise which reduces the computational burden considerably, but at the same time it implies the risk of arriving at a final solution which is not optimal. This, however, may only be expected to happen when the natural clustering of the objects is relatively weak.

When used with the total within-group variation as value reflecting functions to be minimally increased hi each successive joining of groups, the Hierarchical Group Analysis will group legislators together according to the likeness of their stands on each roll call. If abstainers and absentees are given a numerical value in between the value assigned to 'aye' and 'nay' - say 0.5 if 'aye' is coded 1 and 'nay' 0 - they will be combined with members they do resemble most closely in the roll calls which they participated in. At the same tune, this grouping will

mean a loss of information which will be reflected by the value of the loss function.
Members who were absent in more than 50 percent of all roll calls should not
be included in the grouping, however.

Although it has not been used very much hi the study of voting behavior, Hierarchical Group Analysis seems to possess certain obvious advantages as a method for identifying voting groups from roll call data. In the first place, it is easily adapted to a computer (a bloc diagram is included in Ward⁴¹), can handle large sets of data, is moderately speedy,⁴² and hence quite economical to apply. Next, it provides some hints to the researcher as to when to stop the clustering procedure. The problem of defining a threshold value to terminate the joining of groups is a recurring one in cluster analysis⁴³ and must, of course, in all cases be solved on the basis of the researcher's personal judgment. Political scientists sometunes have been grossly mistaken with regard to the 'objectivity' of their methods in this respect.⁴⁴

In connection with Hierarchical Group Analysis, the loss function may be utilized to guide the decision as to where to stop. A very common picture of a loss function seems to be a steady increase at a moderate rate at the beginning and from a certain point, corresponding to a particular cycle of iteration, a dramatic 'rocketing up' of the function. If the loss values follow that pattern, it could seem rational to apply a kind of 'elbow-criterion' and stop clustering just before reaching the point where the loss of information begins climbing at greater pace. Finally, Hierarchical Group Analysis does not make specific assumptions with regard to the underlying distribution of the input data or their level of measurement. The disadvantages of this method first and foremost consist in the risk of arriving at non-optimal solutions. Moreover, it is not possible on the basis of the results to tell anything about the degree to which a member is associated with his particular group.

Hierarchical Group Analysis could of course also have been used to group together roll calls into issue-subsets according to the distribution of ayes and nays by simply transposing the data. Nevertheless, when it comes to identifying the groups of roll calls in terms of underlying dimensions it will be important to know to what degree a certain roll call is associated with a particular dimension. Therefore, a factor analysis was performed on the roll calls to obtain a picture of the conflict dimensions in the Constitutional Convention. Eigenvalues and Eigenvectors were extracted by principal component analysis and a rotation was performed according to the varimax-criterion. The Kaiser-criterion was used to determine the numbei of factors rotated. Factor scores were computed for each member using least squares estimation.

The application of factor analysis to roll call data may be open to criticism, since the data cannot meet the requirements of factor analysis with regard to level of measurement. Although other measures of distances or contingencies can be used,⁴⁵ factor analysis in its most common form starts out from correlation matrices, since these matrices possess the important property of being positive semidefinite, and hence require interval data, whereas roll call data are expressible

as ordinal numbers only. Therefore it may be argued that factor analysis should
not be used in legislative behavior research.

Nevertheless, factor analysis has been used and currently is widely used on purely ordinal data in a great number of cases, maybe even a majority.⁴⁶ What the risk of producing artifacts in that way actually is, nobody seems to know exactly. It seems to be quite a common assumption, however, that the impact of going from interval to alternative data does not matter much, if a clear structure does exist in the data.⁴⁷

4. Findings

From the clustering procedure and the application of the 'elbow-criterion' 11 clusters or (statistical) voting groups emerged, varying in size from 24 to six members. Seventeen members were excluded from the clustering because of their high frequency of absenteeism during roll calls.

As a partial answer to the question of external validity it may be taken that, although the number of clusters arrived at is at least twice the number of 'parties' usually found in the historians' descriptions of the Constitutional Convention, the two sets of classifications of the members are not altogether incompatible. If the results of the clustering are compared with the distribution of members on six parties made by Jensen,⁴⁸ Kendall's tau becomes 0.36. Thus the right people do appear in the right clusters to a considerable degree.

A preliminary attempt at characterizing the voting groups can be made by searching for 'pivot members,' i.e., prominent members of the Constitutional Convention with well-known political positions, and by using their location in the voting groups as indicators of the overall character of the group.

Thus, group 149I⁴⁹ contains a considerable number of members who are known for sure to have belonged to the group of 'peasant friends' (Bondevennerne) in the Constitutional Convention. The men normally considered the leaders of that group - I. A. Hansen, B. Christensen, and A. Gleerup - are conspicuously absent from group 1, however. They are to be found in group 9 instead, together with some other 'peasant friends'. This different location of the leadergroup and the majority of the 'peasant friends' seems noteworthy, since it may be taken as foreshadowing the conflict which split up the 'peasant friends' only a short time later in a struggle over the question whether the party was defined by its leaders or by its majority,⁵⁰ marking the first of a long series of splits on the left wing in Danish party politics. The two groups adopted different stands on roll call No. 14 on the exemption of clergymen and schoolteachers in rural districts from universal military service (group 1 11-9 for, group 9 0-12 against), on roll call No. 29 on the abolition of the nobility (group 1 17-6 for, group 2 2-10 against), on roll call No. 30 on court supervision of the administration (group 1 18-3 for, group 9 3-9 against), and on roll call No. 37 on the question on what should be included in the term "paid taxes' which would qualify for eligibility to the First Chamber under the new Constitution, group 1 voting 14-4 for a very

inclusive definition which would have been favorable especially to the peasantry, group 9 voting 4-6 against. Thus, group 1 and group 9 seem roughly to correspond to a radical and a moderate - or an uncompromising and a compromising - wing within the group of 'peasant friends'.

Three clusters - Nos. 5, 6, and 7 - contain well-known conservative members of the Constitutional Convention (v. Haven, David, Bishop Mynster, and Scavenius), the strongest adversaries of any constitutional change, in group 5, a leading figure of the ancien regime (Ørsted) together with J. E. Larsen in group 7, and Paludan-Miiller in group 6. These three groups took different stands on the following: roll call No. 12 on< a proposal to pay due regard to the domestic and civil status of draftees under extraordinary conscriptions, group 5 voting for (5-3), groups 6 and 7 against the proposal (2-5 and 8-9); on roll call No. 17 on a proposal to make it a royal prerogative to propose amendments to the Civil List, group 5 voting 6-2 for, group 6 1-8 against and group 7 2-15 against; on roll call No. 31 on the introduction of the jury system in grave kinds of criminal cases (group 5 0-7 against, group 6 10-0 for, group 7 8-6 for); and on roll call No. 35, the final vote on the constitution, group 5 splitting evenly 3-3, group 6 and group 7 voting for (10-0 and 10-1, respectively). Thus, group 5 may be identified with the most conservative opposition in the Constitutional Convention.

Group 11 lists the name of Grundtvig of whom it has been said later that '... he makes it his duty to form a minority and he does succeed admirably in that task',⁵¹ together with his close spiritual follower, Boisen, and some other members, known to contemporaries as the 'comets' or independents.

The remaining groups all list several well-known liberal names. The most prominent are, however, found in group 4, e.g. Krieger, Hall, and Andræ, and, somewhat surprising, Minister of War (until November 1848) Tscherning who by historians is listed alternatively as a 'peasant friend' or an independent, but by his voting behavior obviously placed himself in liberal company, thus making true in the Constitutional Convention what he declared to be his policy only a little later, just before the elections to the first Folketing under the new Constitution: 'If I get elected, and if a majority of peasant friends is returned, I will try to take as moderate a stand as possible and, so to speak, cling us to the ministry's back.'⁵²

Thus the picture of the voting groups in the Constitutional Convention seems to reveal a considerable degree of fractionalization and cross-cutting disagreement, together with the not uncommon pattern of relatively greater stability on what is normally considered the right and the left wing of the spectrum and a kind of morass in the middle.

The problem of the dimensions involved in the roll call voting in the Constitutional Convention may most expediently be approached from the concept of an urban-rural conflict dimension because of its preeminence not only in the historians' descriptions of that period, but also as an explanatory device in the study of the development of party systems.⁵³ We may hypothesize that, if an independent urban-rural dimension is present in the data, a factor structure will emerge from

principal component analysis and varimax rotation in which one factor will produce high loadings with roll calls, the contents of which have a clear bearing on the urban-rural issue, as e.g. roll call No. 27 on a proposal to ensure a special representation hi the First Chamber for Copenhagen and the cities vis-a-vis the rural districts, and roll calls Nos. 20, 22, 26, and 28, all of which deal with certain aspects of the extension of suffrage and the problem of representation, both questions loaded with antagonistic urban and rural interests. What we try to do hi the first place, then, is essentially to test the appropriateness of a particular model of the behavior in the Constitutional Convention on the roll call data.

From the factor analysis, a picture entirely consistent with that model did not emerge, however. Four Eigenvalues greater than 1.0 were extracted by the principal component analysis and preserved for the further analysis. The Eigenvalues are reported hi Table I, together with the cumulative percent of trace accounted for by the four roots.

Rotation according to the varimax-criterion resulted hi the pattern reported in
Table 11. Only loadings higher than 0.50 (sign ignored) are reported, and decimal
points have been omitted.

Some of the roll calls with high loadings on Factor I undoubtedly have a close relation to the urban-rural issue. This is true, e.g., of roll calls Nos. 22, 26, and 27. Roll call No. 16 can be accommodated within this pattern, too, although its content - the coronation ceremonies - on the surface does not seem to bear any resemblance to an urban-rural conflict issue. Nevertheless, since it involved the role of the church, in the perceptions of the members it may well have been linked to the question of religious liberty and established church which were closely bound up with the social aspects of the urban-rural issue, since different sectarian movements had their stronghold in the countryside and here especially in the lower social strata. No such interpretation seems possible, however, for the high loadings of roll calls No. 18 (on the proposal that the parliament should not be entitled to remain in session for more than three months a year without royal consent), No. 31 (on the introduction of the jury system in grave criminal cases), and No. 32 (a proposal to ban the use of funds assigned to churches, schools, or charitable institutions, for other purposes). Moreover, if the first factor was to be interpreted as an urban-rural conflict dimension, it would have to be explained why, for example, roll calls No. 20 (on suffrage and the economic qualifications) and No. 28 (on representation and suffrage) do not show high loadings on Factor I, although they undoubtedly are related by their content to an urban-rural conflict.

With regard to Factor 11, the picture seems a bit clearer. Roll calls No. 14 (on the exemption of clergymen and schoolteachers in rural districts from military service), No. 24 (on the abolition of the obligation to place oneself exclusively at the services of the incumbent of the parish), No. 37 (on the calculation of paid taxes as economic qualification for eligibility to the First Chamber), No. 39 (on the same issue), and Nos. 41 and 42 (on the daily allowance for the prospective members of parliament) all easily lend themselves to an interpretation in terms of divergent urban and rural interests, as does roll call No. 23 (on the King's obligatory membership of the established church) analogous to roll call No. 16. Roll call No. 15 (on a proposal to speed up the work in the committee which drafted the constitution) may also be interpreted in terms of divergent urban-rural interests, since it was the peasantry which stood to lose most from a delay of the reforms the constitution was intended to carry through. But again, if this factor is to be interpreted as an urban-rural dimension, it may be asked why several roll calls on issues apparently clearly related to that issue do not show high loadings on this factor, e.g. the above-mentioned roll calls Nos. 20 and 28.

Factors 111 and IV do not bear any resemblance to an urban-rural conflict dimension. Factor 111 singles out the only two roll calls with foreign policy content, whereas Factor IV separates the final vote on the constitution from all other roll calls. Thus, although it seems dubious to interpret either Factor I or Factor II or both as an urban-rural conflict dimension, the heterogeneity of the roll calls which show high loadings on these factors makes it extremely difficult to arrive at alternative interpretations.

One possible explanation of these difficulties could be the existence of a general
factor. The nature of the varimax rotation is such that it tends to destroy such

general factors during rotation. The inspection of the original loadings from principal
component analysis does not suggest this interpretation, however.

It may be interesting to note in this connection that Aydelotte did encounter the same problems when Guttman-scaling roll calls in the British House of Commons. The scales he arrived at were composed of very heterogeneous roll calls. This fact led him to suggest the following interpretation:⁵⁴ '... (T)he relationship between the issues can most easily be accounted for not in terms of their substantive content, but in terms of party maneuver and political opportunism: they are to be regarded not as ends in themselves but as means by which political leaders and politically active groups seek to promote their own interests: to build majorities, to retain power or to secure it and to mobilize support behind them for their purposes. Issues, then, need have no logical connection except for their convenience hi serving the purposes of party politics.'

Unlike Aydelotte, we may be able to perform some kind of test of this explanation. If it is true that the heterogeneity of the cluster of roll calls with high loadings on one factor expresses the maneuvers of parties and groups, it must be possible to arrive at an interpretation of the factors in terms of the groups instead of the issue-content of roll calls with high loadings. This implies the use of factor scores instead of factor loadings - together with our knowledge of the members and the groups - as means of identification.

In Figure 1 the 11 groups arrived at by Hierarchical Group Analysis are plotted
against Factors I and 11. The coordinates of each point, corresponding to a particular

group, are the coordinates of the centroids of the group-members' varimax
factor scores on Factors I and 11.

As can be seen from an inspection of the graph, compared with the preliminary characterization of the groups, a clearer picture emerges from the use of factor scores. Thus, the main conservative groups are found in the second quadrant with group 5 in the right-most position; all groups which contain well-known liberals are located relatively closely together in quadrant three; quadrant four, finally, is occupied by the two wings of the 'peasant friends' and by the 'comets'. This picture suggests a parsimonious interpretation of the first two factors: Factor I may be viewed as a liberal-conservative dimension, whereas Factor II is interpretable as an urban-rural dimension, dividing the members of groups in the Constitutional Convention.

From that finding, several substantive conclusions may be derived. In the first place it may be noted that, subject to the conditions of the factor analytical model, especially the orthogonality of the factors, the liberal-conservative dimension played a greater role in separating the members of the Constitutional Convention in roll calls than did the urban-rural conflict, since Factor I accounts for a greater percentage of the trace than does Factor 11. Thus, a wider split existed between members who wanted a constitutional change in liberal direction and members who were opposed to it than between members who wanted to promote the case of the peasantry and members who were affiliated with opposed interests.

Secondly, the findings seem to lend support to an adoption to the proceedings in the Constitutional Convention of Aydelotte's notion that party maneuver rather than the issue-content of the roll calls accounts for the voting patterns found. Hence, this may lead us to assess the overriding importance of relatively stable voting blocs in the roll calls in the Constitutional Convention vis-a-vis the notion nourished by some historians⁵⁵ of a kind of golden age where members carefully made up their minds on every single roll call and voted according to their conviction and the merits of the case. While this should not lead us to deny the importance of the content of the roll calls altogether, the voting patterns occurring in a significant number of them cannot be understood unless the importance of group maneuvers is taken into consideration.

Finally, the findings seem to back the view that the groups isolated by Hierarchical Group Analysis do not constitute artificial and casual structures. The pattern exhibited by their factor scores on the first two factors does not bear any resemblance to a casual structure. Thus, we may interpret these groups as social rather than statistical entities in the Constitutional Convention.

---

NOTES

1. Stein Rokkan, 'The Structuring of Mass Politics in the Smaller European Democracies: A Developmental Typology', Comparative Studies in Society and History, Vol. X (1967— 68), p. 178.

2. E.g. the Conference on the Use of Quantitative Methods in the Study of the History of Legislative Behavior, University of lowa, March 13—15, 1972.

---

3. Cf. Sylvia L. Thrupp, 'Diachronic Methods in Comparative Polities', in Robert T. Holdt and John E. Turner (eds.), The Methodology of Comparative Research, New York: The Free Press, 1970, pp. 345-358.

4. E.g. The Minnesota Political Data Archive (MPDA).

5. Cf. Richard A. Pride, Origins of Democracy: A Cross-National Study of Mobilization, Party Systems and Democratic Stability, Comparative Politics Series, Vol. 1, 1970.

6. Thrupp, op.cit., p. 345.

7. Aristotle, 'Part of Animals', 639b16-641b29, in Aristotle, Loeb Classical Library, Vol. XII, Cambridge, Mass.: Harvard University Press, 1967.

8. Seymour M. Lipset and Stein Rokkan, 'Cleavage Structures, Party Systems, and Voter Alignments: An Introduction', in Seymour M. Lipset and Stein Rokkan (eds.), Party Systems and Voter Alignments, New York: The Free Press, 1967, p. 2.

9. E.g. Maurice Duverger, Les Partis Politiques, Paris: Colin, 1951. 10. Lipset and Rokkan, op.cit., p. 50.

11. Mogens N. Pedersen, Erik Diungaard & P. Nannestad Olsen, 'Party Distances in the Danish Folketing 1945-68', Scandinavian Political Studies, Vol. 6, Oslo: Universitetsforlaget, 1971, pp. 87-106.

12. Some of them have, more or less explicitly, treated Danish politics in terms of changing relations amone oarties. e.e. Svend Torsen. Folkets Veie. Cooenhaeen: Fremad. 1953.

13. Rokkan, op.cit., p. 176.

14. E.g. Joel A. Silbey, The Shrine of Party, Pittsburgh: The University of Pittsburgh Press, 1967.

15. This is one of the points in Aaron Wildavsky's critique of Duverger; cf. Aaron Wildavsky, 'A Methodological Critique oi Duverger's Political Parties', The Journal of Politics 21 (1959).

16. All men above the age of thirty with their own households were entitled to vote.

17. Niels Neergaard, Under Junigrundloven, Vol. l, (Jopennagen, isyz, is usually regarded as the main work on this period. Of more recent treatments one could mention Roar Skovmand, Tolkestyrets Fødsel, 1830-1870', in John Danstrup and Hal Koch (eds.), Danmarks Historie, Vol. 11, Coperhagen: Politikens Forlag, 1964, pp. 271-288; Povl Engelstoft, 'Den grundlovgivende rigsforsamling', in Den Danske Rigsdag 1849—1949, Vol I, Copenhagen: Schultz, 1949, pp. 11-40; Povl Engelstoft, 'Det Danske Folk 1848-1864', in Schultz Danmarkshistorie, Vol. IV, Copenhagen: Schultz, 1942, pp. 634-688.

18. E.g. Engelstoft, op. cit., 1949.

19. For a discussion of unidimensional concepts of party distances, see Pedersen, Damgaard & Olsen, op.cit., passim.

20. This view is taken by, among others, Neergaard and subscribed to by Skovmand, op.cit., p. 274.

21. Engelstoft, op.cit., 1942, p. 663.

22. Bjørn Svensson, 'Politiske Hovedtræk', in Aa. Svendstorp and Chr. Torpe (eds), Billeder af Rigsdagens Historie, Vol. I, Odense: Codan, 1947, p. 26.

23. E.g. Tscherning (Minister of War until November 1848) and Grundtvig.

24. William O. Aydelotte, 'Voting Patterns in the British House of Commons in the 1840Y, Comparative Studies in Society and History, Vol. V (1962-63), pp. 134_—135.

25. Cf. William H. Riker, 'A Method for Determining the Significance of Roll Call Votes in Voting Bodies', in John C. Wahlke and Heinz Eulau (eds.), Legislative Behavior, Glencoe, 111.: The Free Press, 1959, pp. 377-384.

26. Robert A. Dahl (ed.), Political Oppositions in Western Democracies, New Haven: Yale University Press, 1966, p. 360.

27. Mogens N. Pedersen, 'Partiernes holdning ved vedtagelsen af regeringens lovforslag 1945-66', Historie 7, 3 (1967), pp. 421-425.

28. This fact is obvious in a Danish context where a majority of bills are passed by unanimous vote.

29. Niels Stjernquist and Bo Bjurulf, 'Partisammanhållning och partisamarbete', Statsvetenskaplig Tidsskrift 71, 4 (1968), p. 336.

30. Silbey, op.cit., p. 14.

31. Duncan Macßae, Jr., Issues and Parties in Legislative Voting, New York: Harper and Row, 1970, p. 5.

---

32. Ibid., pp. 208-209.

33. Quoted from Skovmand, op.cit., p. 286.

34. E.g. Beretninger om Forhandlingerne paa Rigsdagen, col. 186 (B. Christensen).

35. This is essentially the same argument as forwarded by Stjernquist and Bjurulf, op.cit., p. 336.

36. Members of the diets who had voted for the royal appointment of members to the Constitutional Convention were fiercely fought by the 'peasant friends' in the elections.

37. Standard biographical works are J. P. Jensen, Valgene til Rigsdagen i 80 Aar, Roskilde, 1929, and Victor Elberling, Rigsdagens Medlemmer gennem 100 År, Vols. I—111, Copenhagen: Schultz, 1949-1950.

38. Cf. Pride, op.cit., pp. 682-683.

39. Cf. Macßae, op.cit., pp. 39-41.

40. J. H. Ward, 'Hierarchical Grouping to Optimize an Objective Function', American Statistical Association Journal 58 (1963), pp. 236-244.

41. Ibid., p. 243.

42. Thus, the clustering of 157 subjects on 44 variables was performed in 24 CP seconds on a CDC 6400.

43. Macßae, op.cit., pp. 39-41.

44. E.g. Bruce M. Russet, 'Discovering Voting Groups in the United Nations', American Political Science Review 60 (1966), pp. 327-339.

45. P. Slater, 'The Factor Analysis of a Matrix of 2x2 Tables', Journal of the Royal Statistical Society 9 (1947), pp. 114-127.

46. K. Überla, Die Faktorenanalyse, Berlin: Springer Verlag, 1971, p. 302.

47. Ibid.

48. Jensen, op.cit.

49. The numbers are the numbers returned by the computer program.

50. Thorsen, op.cit., pp. 32-33.

51. Quoted from Skovmand, op.cit., p. 336.

52. Thorsen, op.cit., pp. 32-33.

53. Stein Rokkan, 'Norway: Numerical Democracy and Corporate Pluralism', in Dahl (ed.), op.cit., pp. 78-81.

54. Aydelotte, op.cit., p. 153.

55. E.g. Thorsen, op.cit., p. 32.