Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

As we noted in the introduction to this volume, constitutions are sets of rules that constrain the choices of policy makers. But they are also discourses that reveal the motivations that drove their drafters. Traditionally, constitutional analysis has adopted the first perspective where constitutions are viewed as contracts that define the rules of the game. Their analysis consists in describing these rules and their evolution (e.g., Congleton 2011), explaining the content of constitutions (Voigt 2011), and measuring the impact of rules on institutional and policy outcomes (e.g., Congleton and Swedenborg 2006; Persson and Tabellini 2003). The Veil of Ignorance Project (VOIP) adopts the second perspective. It looks at constitutions as discourses to infer the motivations of constitution drafters from the content of the very text they contributed to write (Imbeau 2009). In particular, the project aims at measuring the extent to which constitution drafters worked under uncertainty.

But this distinction between the contractual and the discursive approaches is not to be overstretched as both provide ample opportunities for cross-fertilization. For example, one of the connections between Congleton’s research and ours is the fine-grained nature of constitutional bargaining and the resultant power relationships that a minute reading reveals.

This chapter introduces to the theory and method of the VOIP project and presents some preliminary results. We proceed in three steps. First, we expose the theoretical foundations of the project based on Buchanan’s interaction approach. Second, we describe the content analysis method that we used to compare the discursive content of 16 constitutions. Third, we submit some of our empirical results to validity tests before concluding.

2 Social Interactions, Power Relations, and Uncertainty: A Theoretical Perspective

Summary: In this section, we argue that constitutions are the outcome of social interactions and that these interactions are best conceptualized as power relations.

Constitutional economics makes «a categorical distinction […] in the ultimate behavioral object of analytical attention» (Buchanan 2008: 281): choices made within constraints and the choice among constraints. Ordinary economics focuses on choices made within constraints; constitutional economics focuses on choices among constraints. This choice of constraints should be viewed, according to Buchanan, as an exchange. He wrote:

[I]ndividuals choose to impose constraints or limits on their own behavior primarily, even if not exclusively, as part of an exchange in which the restrictions of their own actions are sacrificed in return for the benefits that are anticipated from the reciprocally extended restrictions on the actions of others with whom they interact […]. So interpreted, the individual who joins in a collective decision to impose a generally applied constitutional rule is not, at base, acting differently from observed behavior in a setting that involves giving up one desired good, apples, for another desired good, oranges (Buchanan 2008: 282, emphasis in original).

Thus for Buchanan, constitutional economics as a research program is based on a catalectic perspective, a «science of exchange» among individuals. This perspective is to be contrasted with the mainstream perspective of choices made by utility-maximizing individuals where the main object of study is the allocation of scarce resources. «The elementary and basic approach that I suggest places ‘the theory of markets’ and not the ‘theory of resource allocation’ at the center stage» (Buchanan 1964: 217, quoted in Marciano 2009: 44). But for this research program to become a political economy program, it must also consider two other types of interactions—coercion and persuasion—that have formed the core of political science.

Indeed, the very definition of the state proposed by the German sociologist, Max Weber, as the organization with a monopoly on the legitimate use of violenceFootnote 1 shows how coercion is central to politics. Politics is «the authoritative allocation of values in society» (Easton 1960: 129, emphasis added). In this perspective, the choice of constraint in a society may be made through coercion. One may force others to «choose» such constraints. This is often the case, for example, when the military imposes a new constitution after a coup d’État.

Moreover, the choice of a constraining rule may be the result of persuasion, the other side of politics. Let’s turn to Buchanan again. He wrote:

The very term politics tends to conjure up a mental image of potential conflict among those persons who are members of the politically organized community. This conflict may be interpreted to be analogous to scientific disputes, in which separate participants or groups seek to convince one another of the truth of their advanced propositions. […] [P]olitics may seem, by its very nature, to involve conflict between and among individuals and groups within a polity (2008: 285).

Thus, influential people may persuade constitutional drafters to include a specific constraint on the behavior of future participants to collective decisions, such as a rule giving precedence to, or outlawing, a specific religious group.

2.1 Two Models of Decision-Making

Two psychological models of decision-making are implied in Buchanan’s argument about the opposition between an exchange approach and a choice approach or between a «theory of markets and a theory of resource allocation»: the choice model and the interaction model.

The choice model is the usual model we use in public choice analyses. It sees the outcome of the decision-making process (here, the constitutional document) as ensuing from the choices made by individuals. Typically, these choices are a function of the individual’s preferences, endowment, and constraints (see Fig. 1).

Fig. 1
figure 1

Casual relations in a choice model

Full size image

Buchanan provided a list of the constraints that limit the choices an individual can makeFootnote 2: Nature, history, past choices, custom and convention, other persons, rules and laws. For all practical purposes, the first four constraints cannot be easily manipulated in the context of the specific decision-making process over an issue like the drafting and the adoption of a constitution. One can certainly not remake history to erase past choices but one can perhaps alter their impact on an ongoing decision-making process through various rhetorical and heresthetic techniques. It may be difficult, for example, to reconsider a convention based on past choices concerning the decision rule once the decision-making process has started though this is certainly not impossible.Footnote 3 The manipulation of the last two types of constraint—other persons, and rules and laws—is easier and more often done. Constitutional choices are themselves made under constraints in addition to being about choosing constraints over future decisions.

Given this difficulty of manipulating constraints and assuming utility maximization as the unique preference of decision-makers, the main explanation of choice is endowment. This is the message Charles Beard conveyed. Here is how he put it:

Suppose […] that substantially all of the merchants, money lenders, security holders, manufacturers, shippers, capitalists, and financiers and their professional associates are to be found on one side in support of the Constitution and that substantially all or the major portion of the opposition came from the non-slaveholding farmers and the debtors—would it not be pretty conclusively demonstrated that our fundamental law was not the product of an abstraction known as ‘the whole people,’ but of a group of economic interests which must have expected beneficial results from its adoption? (Beard 2004 [1913]: 17)

More recently, the work by McGuire and Ohsfeldt proceeded from the same choice perspective. In their analysis of the ratification of the American constitution, they assumed that a delegate’s utility of voting on a constitutional proposal is a function of his personal interests and ideology as well as of his constituents’ interests and ideology. They showed that, at the ratifying stage, the support for the proposed constitution was significantly related to the interests and ideology of the delegates but, at the drafting stage, the constituents’ interests are a better predictor of a delegate’s choice than his own interests (McGuire and Ohsfeldt 1989; McGuire 1988). Stable preferences may be a good predictor of choice.

But the choice model reaches its limit when it fails to explain the outcome, i.e., the content of constitutions. Then, the interaction model proves to be useful. Indeed, as noted by Scheinkman, «[m]odels of social interaction seem particularly adapted to solving a pervasive problem in the social sciences, namely, the observation of large differences in outcomes in the absence of commensurate differences in fundamentals» (Scheinkman 2008: 1). William Riker provided a nice example of the effect of social interactions on the content of the American constitution. He told this fascinating story of how the constitutional provision for the Electoral College in the election of the president was created and adopted in the Federal Convention of 1787. He showed in particular how the interaction between Gouverneur Morris and his fellow delegates completely changed an early vote of 8-2 in favor of a provision for the national legislature to elect the president to a final vote of 9-2 in favor of an election by an electoral college. A choice model with stable preferences would be unable to account for this switch. But an interaction model could. Indeed, for Riker, Morris achieved this change through «heresthetics or the dynamic manipulation of the conditions of choice» (Riker 1984: 1).

In an interaction model, there are several individuals making the choice of entering or not into an interaction on the basis of the others’ choice. Adopting the interaction approach means endogenizing preferences. Contrary to the choice model where preferences are stable, an interaction model allows for changing preferences as each participant to the decision-making process can alter another’s incentive structure, i.e., his evaluation of the expected costs and benefits of a proposal. Thus, preferences may change depending on the structure of interactions. In this context, a decision-maker may use his or her resources to influence another’s choice so as to help produce the desired outcome. From this viewpoint, the content of a constitution is the result of the social interactions among the drafters (see Fig. 2).

Fig. 2
figure 2

Causal relations in an interaction model

Full size image

An individual’s endowment allows him or her to enter into interaction with another individual. In Buchanan’s exchange perspective, one individual exchanges his vote for another’s vote: «I will support a constraint over my future actions if you do the same». This is the usual vote trading or logrolling behavior. One uses one’s endowment (one’s vote) to make an exchange with another constitution-maker. But other resources can be put to use in this context. One may use coercion or persuasion to reach the same goal. Thus, facing uncertainty as to one’s future position in society, one could trade a constraint over one’s own future actions for a constraint over another’s future actions; or one could force another, or one could persuade another, to support the adoption of such a constraint depending on the resource that one is endowed with. Let us see how exchange, coercion, and persuasion can be conceived as three forms of power relations.

2.2 Power and Constitutional Choice: A Conceptual Framework

As the famous philosopher Bertrand Russell noted, “the fundamental concept in social science is Power, in the same sense in which Energy is the fundamental concept of physics….Footnote 4 Power, like energy, must be regarded as continually passing from any one of its forms into any other, and it should be the business of social science to seek the laws of such transformations” (Russell 1962: 10–12, quoted in Gordon 1999: 8). But there is a cruel lack of consensus on the conception of power in the social science literature.

This is the case among economists. Bardhan, for example, argues that «orthodox neoclassical economics fails to handle some of the key issues of power». He reviews several uses of the concept of power in economics and shows that «economics is not confined to the exercise of economic power [but] is often concerned with […] other forms of power, particularly political and ideological» (Bardhan 1991: 265). As Randall Bartlett, another economist, noted: «In economics, the study of power is clearly in a ‘prescientific’ state» (1989: 4).Footnote 5 Apart from references to monopoly power or to bargaining power, neoclassical economics, for example, generally ignores the concept of power only to state that it is absent from markets. Power and markets would be mutually exclusive… except for Herbert Simon who argued that submission to authority is what is being purchased in labor markets. Therefore power is something that can be traded! (Simon 1957, quoted in Bartlett 1989: 6) Institutional economists have a wider, more general, conception of power. Among them, Philip Klein is quite vocal about the attitude of mainstream economists vis-à-vis the concept of power. He wrote: «while [power] might be equally distributed, it is clearly unequally distributed among the participants in social interaction in the real world as viewed from the perspective of all the social sciences. That distribution […] is in reality the principal focus for analysis in all the social sciences except economics, where the studious avoidance of that reality has been honed, polished, and embroidered into a fine art. It remains enshrined there, rationalized as ‘a simple and logical starting point’ and as the ‘norm’ which the overwhelming thrust of standard theory almost never succeeds in moving» (Klein 1980: 873). Power conceived as coercion is assumed away. Another institutional economist, William Dugger, defines power as «the ability to tell other people what to do with some degree of certainty that they will do it. When power wielders must coerce others, power is tenuous and obvious. When coercion is unnecessary, power is secure and unnoticed» (Dugger 1980: 897). In this vision, power is not necessarily linked with coercion. It can appear in other apparels. Randall Bartlett has developed what seems to be the most elaborate theory of power in economics. He defines power as «The ability of one actor to alter the decisions made and/or welfare experienced by another actor relative to the choices that would have been made and/or welfare that would have been experienced had the first actor not existed or acted» (Bartlett 1989: 30). With this broad definition, he identifies four forms of power: decision power (simple economic power and decision control power), event power, agenda power, and value power (1989: 41 sq). Under the assumption of bounded rationality, he argues that economic activity is an exercise of power. In summary, the visions of power presented by mainstream economics depend on theoretical assumptions (power is «assumed away»), whereas institutional economists focus on the empirical object they are observing. We are confronted here with the classical trade-off between parsimony and realism. Some argue for the strict observance of neoclassical dogmas, often falsely thinking that assumptions are statements about truth; others question the orthodoxy when they feel that mainstream assumptions block the way toward a better understanding of reality.

Political science is also in a prescientific stage when it comes to the study of power. There is no dominant paradigm of power at the moment. But the literature on power in this discipline is much more developed than in economics.Footnote 6 As Klein noted:

«if economics customarily assumes away most of the implications of power, mainstream political science appears to begin by assuming that the distribution of power is the preeminent conditioner of how the political system operates. Power is the prime determinant of most of the decisions that lead to particular market prices. Power and wealth are mutually supportive. One is used to acquire more of the other. Together they seep through most of our attitude-forming institutions and, in the process, affect mightily what consumers want, whether and to what extent producers will produce efficiently, and for whom products will be produced. […] It is clearly misleading to say that ‘the market’ determines resource allocation, since power constellations have previously conditioned the market and determined the manner in which it will present choices» (Klein 1980: 883–4).

Despite the profusion of conceptions that one finds in the political science literature on power, one can readily identify a cumulative development in the contributions of Weber (1964) [1922], Russell (1962), Dahl (1957), Bachrach and Baratz (1962; 1963), Lukes (1974), and Dowding (1991).Footnote 7 In the rest of this section, we dwell on this literature to propose a conceptual framework which will help us see how uncertainty may impact on the content of constitutions. Doing so, we will intentionally depart from the prevailing assumptions of mainstream economics in hope that our conceptualization will help build a «political economy» of constitution-making.

Power and rational choice Bertrand Russell defined power as «the ability to produce intended effects» (Russell 1938). This definition implies intentionality. Reaching an effect by accident is not a manifestation of power but of luck (Dowding 1996: Chap. 3). More importantly, it also implies a disposition rather than an action. Indeed, this definition of power is ‘dispositional,’ writes Dowding (1996: 3), in the sense that it denotes a disposition of an agent to do something. To have power means to have the capacity or the ability to do something. Therefore, having power does not necessarily imply exercising power. One may well have the power to do something but choose not to do it. Power may remain potential, much like water behind a hydroelectric dam. Furthermore, the exercise of power does not imply action. One may exercise power by remaining silent, thus suggesting one’s approval; or one may prevent an issue from reaching the public agenda, thus discarding an option without any decision being taken about it. Bachrach and Baratz (1962; 1963) coined the term «nondecision» to refer to this possibility. Riker (1984) talked about heresthetics—i.e., «structuring the situation so that others accept it willingly»—to describe a mechanism through which nondecision is effective. Finally, Lukes (1974) defined a third-dimensional type of power where «A may exercise power over B […] by influencing, shaping or determining his very wants. Indeed, is it not the supreme exercise of power to get another or others to have the desires you want them to have—that is, to secure their compliance by controlling their thoughts and desires?» (Lukes 2005: 27) Power is a capacity that does not need to be actualized for it to exist.

Power as a capacity is based on resources with which its holder is endowed. To have the power to do something implies to control the resources necessary to exercise it. As mentioned above, power does not necessarily imply force, coercion, or authority. It may also be exercised through other means. Indeed, power as a capacity is recognizable in policy processes through three main resources which may be used to produce intended effects: force or authority, wealth or things of value, and knowledge or information with rhetoric. These define the sources of power.

Russell’s definition of power also implies a target of power, the «intended effects». The literature generally identifies two types of target: events or things, and persons. One may produce, or help produce, an effect by directly acting on things or events like casting one’s vote in an election. But one may also indirectly produce, or help produce, an effect by making another act on things or events. Thus in defining power, we need to distinguish two types of power relation: instrumental power or power to, and social power or power over (Göhler 2009). Instrumental power is the capacity of an agent to produce, or to help produce, an outcome by acting on events or things. Social power is the capacity of an agent deliberately to change the incentive structure of another agent so as to produce, or to help produce, an outcome (Dowding 1991: 48). In Fig. 2, social power corresponds to the arrow linking B’s choices to A’s preferences (or A’s choices to B’s preferences). B (A) has the ability to modify A’s (B’s) preferences, thus making A (B) choose what would make the outcome closer to what B (A) prefers. Instrumental power rather corresponds to the arrow linking A and B to the outcome. Thus through their choices—their vote, for example—A and B have the ability to help produce a desired outcome.

To modify A’s incentive structure, B must exercise instrumental power by using his resources. For example, B might use his authority or force to threaten A of a punishment, or B might offer A a reward in exchange for the desired behavior, or B might suggest arguments that could modify A’s beliefs about his costs and benefits. This is why we say that social power implies instrumental power. But the reverse is not true. A may exercise instrumental power without exercising social power. For example, in order to coerce a fast driver to slow down on the expressway (social power), a police officer needs to have the capacity to give the recalcitrant driver a fine and ultimately to put him in prison (instrumental power). But consider this other example. I may have the power to contribute to elect the prime minister through my vote (instrumental power) while having absolutely no capacity to make others do so (social power), if, for example, I am physically impaired to the point that contacting other people is almost impossible; I have (arguably very limited) instrumental power but no social power. Social power implies instrumental power.

Turning back to the interaction model of Fig. 2, we see that the source of B’s (A’s) choices is B’s (A’s) endowment. In terms of power relations, B (A) uses the power resources with which he/she is endowed either to help bring about a desired outcome or to make A (B) help bring it about. To be sure, to each resource corresponds a specific method and a main effect on the incentive structure of the target of social power (see Table 1). One may use the force or the authority one is endowed with to make another do what he would not do otherwise (or to prevent him from doing what he would otherwise) by increasing his costs through coercion or the threat of punishment. This is political power. Tribunals exercise political power when they threaten delinquents with fines or prison, thus increasing the cost of the unwanted behavior. Dictators do the same. But power may also be economic when a holder of wealth, or of something of value, makes another do what he would not do otherwise by offering a valued compensation in exchange for the wanted behavior. A minister of Finance, for example, makes wealthy people give up the use of part of their wealth by issuing government obligations, thus acting on the investor’s benefits. If the interest rate offered is high enough, the investor transfers part of her wealth to the government, something she would not do otherwise. Finally, power may be preceptorial when it is based on knowledge through persuasion. A clergyperson may use her knowledge to persuade a disciple that his benefit-to-cost ratio would be more favorable if he performed an action he would not otherwise—like detonating a bomb on the market place—or if he did not perform an action that he would otherwise—like having sex outside marriage. In these three cases, the process is the same. A power resource is used by an agent to change the incentive structure of the target agent through a specific method.

Table 1 The forms of power relations
Full size table

Social interactions therefore could be conceptualized to have nine forms, according to the resource used by the influencing agent (the one who exercises power) and the resources the influenced agent would be expected to use if he were to comply (cf. Table 2). Thus B, the minister of Finance, may use his authority, for example, to make A, the taxpayer, pay a tax. This is a type 2 interaction. Likewise, when the minister of Finance makes the investor buy government bonds by offering a sufficiently high interest rate, we have a type 5 interactionFootnote 8; or when an agent hides information from his principal to make the principal make a decision serving the interest of the agent (a principal-agent relation), we have a type 7 interaction; or when an entrepreneur bribes a politician to obtain a favorable decision, we have a type 4 interaction; etc.

Table 2 Types of social power relations
Full size table

Another source of choice in the interaction model of Fig. 2 is preferences. B’s choices are dictated by his preferences, i.e., his utility function. A power approach would lead us to assume that there are at least three arguments in an agent’s utility function: Agents want to maximize their authority, their wealth, and their knowledge relative to others. More accurately, they want to maximize the return they can get from the combination of their authority, wealth, and knowledge. Indeed, when they act over events or things—or when they exercise some form of instrumental power—agents essentially act over the distribution of power resources: F (the distribution of force or authority), W (the distribution of wealth or things of value), and K (the distribution of knowledge or of information and rhetoric). More precisely, the distribution of a power resource tells us who has more or less authority, wealth, or knowledge relative to others. For example, when a taxpayer pays his tax, he changes the distribution of wealth as he ends up having less and the government having more. Therefore, it is reasonable to consider that the outcomes that A and B ultimately look for are their respective positions on the distribution of authority, wealth, and knowledge. Considering the drafting of a constitution, for example, drafters and adopters want to include provisions that will optimize their future relative positions in those three distributions. Thus, we can conceptualize nine types of instrumental power by crossing the targeted distribution of power with the power resource. According to the power resource an agent uses and the distribution of these resources upon which she wants to impact, one may identify nine types of instrumental power (cf. Table 3).Footnote 9

Table 3 Types of instrumental power relations
Full size table

2.3 Power Relations Behind the «Veil of Ignorance»

Buchanan and Tullock argued that decision-makers choose differently when they are uncertain about their future position—when they stand behind a veil of ignorance—as compared to when they are relatively certain. The informational characteristics of the decision-making context makes them follow their own preferences under relative certainty but to move toward the preference of the majority—or the median preference—under uncertainty. As the theory has it, decision-making under uncertainty is typical of constitutional decision-making. Because constitution choices last longer, the uncertainty of decision-makers is more prevalent than in in-period choices. Hence, decision-makers tend to serve the preferences of the median voter as her future situation might later be closer to his than it now is.

However, in in-period choices, the veil of ignorance is lifted. The time horizon is shorter and the majority requirement for changing rules is less stringent, thus making it easier for one whose position will have deteriorated to change rules in the future. Therefore, when the decision-maker is relatively certain about his future position, she chooses according to her own preferences. Now, it is more difficult to gather a winning coalition under a more demanding constitutional decision rule when everybody follow their own preferences than when each tend to move toward the median position. Therefore, constitutions contain choices mainly corresponding to areas of uncertainty for constitution drafters as choices corresponding to areas of certainty fail to reach the required majority and thus are postponed to the in-period process.

In a power perspective, the preferences of agents are evaluated in terms of their power position in society. They use their power to maintain or improve their position, somewhat like the entrepreneur uses his wealth to produce more wealth (or to avoid losing too much). Therefore, uncertainty refers to the future power position of an agent: Will she be higher or lower in the future distribution of power? If she is uncertain about her future position, she will choose according to the median preference. If her co-deciders make the same evaluation concerning their own future position, the decision-making body will more easily arrive at a decision. Under relative certainty, the opposite will prevail. Constitution drafters will follow their own private interest and no constitutional decision will be made, the issue being postponed to the in-period process.

Now, it is quite obvious that constitution drafters occupy the higher part of the three main power distributions in society. They have more authority, more wealth, and more knowledge than the median individual in each of these distributions. Unless they are uncertain about their future position, they will work hard to protect or to improve their position. But if they think that the distribution of power is volatile and that they might drop toward the median position, then they will be careful to adopt rules that would protect them in the future.

This volatility of power positions is not equal from one distribution to the next. We can safely say that volatility is higher in the distribution of political power, especially in democratic regimes where majorities often shift with electoral results. When this occurs, a whole class of decision-makers changes position on the distribution of political power, some leaving, others entering government circles. Thus, constitutional drafters are quite uncertain about their future political position. But they are less uncertain about their economic power position. Without being absolutely certain, they expect to keep their economic position in the future and even to bequeath their wealth to their children. The distribution of wealth is much more stable than the distribution of authority but relatively less stable than the distribution of knowledge. Indeed, those who are considered as knowing—clergy persons in some societies, policy experts in others, etc.—occupy a preceptorial-power position that is quite stable. It takes a long time for a society to change the criteria it uses to determine what is true and what is good. Consequently, there is less uncertainty in preceptorial power than in economic power and less in economic power than in political power.

Our power theory of uncertainty in constitution-making is summarized in the following causal diagram:

The more opaque the veil of ignorance is in a constitution-making process, the more drafters will attend to the general interest, and the more constitutional documents resulting from this process will be concerned with (1) political power rather than economic or preceptorial power—because it is easier to reach a collective decision when most decision-makers attend to the general interest than when they attend to their own private interest, (2) social rather than instrumental power relations—because uncertain drafters will be more concerned with power over persons than over events or things—, and (3) negative (denying capacities) rather than positive (assigning capacities) power relations. It is expected that the motivations of drafters will crystallise into the constitutional text and that, by analyzing the actual content of a constitution, we can trace the effects of these motivations. Then, through an abductive reasoning,Footnote 10 we can infer, from the content of the text, the motivations of constitutional drafters and therefore the degree of their uncertainty. Thus, by measuring the extent to which a constitutional text insists relatively more on political (or social, or negative) power relations, we also measure the opacity of the veil of ignorance or the degree of uncertainty that prevailed when the constitutional text was being drafted.

Such is the objective we have set to ourselves in this project. We want to assess the validity of three definitional propositions related to the three dimensions of power relations we have identified:

Because uncertainty prevails in a constitution-making process, the following should be true:

  1. 1.

    There are more concerns for political power than for economic or preceptorial power in a constitution.

  2. 2.

    There are more concerns for social power than instrumental power in a constitution.

  3. 3.

    There are more concerns for negative than positive power relations in a constitution.

These propositions are «definitional» in the sense that, in the context where many constitutions are being compared on these dimensions, the variation in the intensity of concerns for each of these dimensions can be considered as an operational definition of the opacity of the veil of ignorance or the prevalence of uncertainty in a given constitution-making process.

3 Research Design

We follow the procedure designed by Imbeau (2009) and by Imbeau and Jacob (2011). We consider constitutional documents not as sets of rules but as discourses about power relations in society through which drafters reveal their preferences. If they attend to the general interest, they will refer more often to political power relations, i.e., to relations of authority, as compared to economic (exchange) and preceptorial (persuasion) power relations; they will also refer more often to social power than to instrumental power, and to negative than to positive power relations. Sixteen national constitutions were «content analyzed».

3.1 Content Analysis Procedure

Content analysis is a data generating technique involving two steps: unitizing and coding. The first step of the analysis consists in extracting meaningful contents from the constitutional document. The «walk-talk» perspective adopted here considers a constitutional document as a discourse rather than as a contract or a legal document. This means that we are not interested in the legal or contractual implications of the document but in its literal content. Following the conceptual framework provided above, we are looking for «power relations» defined as:

The description of the relationship between an agent and a capacity such that: (1) an agent has the capacity to do something (or is prevented from doing something), or (2) an agent has the capacity to make another agent do something or to prevent another agent from doing something (or is denied such capacity).

Therefore, in terms of the literal content of a constitutional document, we were looking for this particular discursive structure:

Each time we found this type of discursive structure, we had a «power relation» and we created a unit of analysis.

The second step in the content analysis procedure is the coding of the units of analysis on various dimensions (variables). Each unit of analysis was coded on three variables: source, type, and direction.

  • V1-Source of the power relation:

  • Which resource would the influencing «Agent» use to perform the action he has the capacity to perform or to make another perform an action: Authority, wealth, or knowledge?

  • V2-Type of the power relation (instrumental vs social power):

  • Could this action add (or remove) a possibility of choice to (from) the choice set of another explicitly identified agent? Yes (social power) or No (instrumental power).

  • V3-Direction of the power relation:

  • Positive (the «Agent» has the capacity to do something) or Negative (the «Agent» is denied the capacity to do something).

More details are provided in the codebook reproduced in an annex below.

Two senior undergraduate students in political science were trained to do the unitizing and the coding of excerpts from three constitutional texts, the constitutions of Canada, Belgium, and Cameroon. The training consisted in two steps. First, after they had read a first version of the codebook, the coders were assigned a common set of 30 paragraphs drawn from one constitution and were asked to unitize them independently of each other. Then, they would compare their unitizing decisions and discuss each of the discrepancies trying to formulate general rules that would solve them. When we felt that the level of agreement was satisfactory, we did a second attempt with a new set of paragraphs, discussing the discrepancies and amending the codebook. This process was repeated until the coders reached a Krippendorff’s alpha level larger than 0.80. We then proceeded to the second step and trained the coders to code each unit. We started with units commonly agreed upon in the first set of paragraphs and let the coders code them independently of each other. We then compared their coding, discussed the discrepancies, and amended the codebook. We repeated this process until we reached an K-alpha larger than 0.80.

3.2 Reliability Tests

Krippendorff (2004: 215) identifies three types of reliability in content analysis. Stability refers to the degree a unitizing or coding process is unchanging over time, or yields the same results on repeated trials. Instability is caused by intra-observer inconsistencies which «may be due to insecurity, carelessness, openness to distractions, difficulties in comprehending written instructions, or the tendency to relax performance standards when tired» (Ibid.). Measuring stability implies that coders reread and recode the same text after some time has elapsed (a test-retest process). Reproducibility «is the degree to which a process can be replicated by different analysts» (Ibid.). In other words, can different coders working in different locations with similar instruments come to the same results? Reproducibility is assessed through a test–test process. For Krippendorff, reproducibility is a stronger measure of reliability than stability. But accuracy is the strongest form of reliability. It measures «the degree to which a process conforms to its specification and yields what it is designed to yield» (Ibid.). Accuracy is assessed through a test-standard process in which the tested coding is compared to a coding that is taken to be correct.

We performed our reliability tests using Krippendorff’s alpha statistic (K-alpha) using Hayes’ SPSS macro (Hayes and Krippendorff 2007). In its simplest form, K-alpha is defined by

$$\alpha \, = 1 - \left( {D_{o} /D_{e} } \right)$$

where D o is the observed disagreement and D e the expected disagreement under pure chance. The statistic can be adapted for any number of coders and values and for any measurement level.Footnote 11 Krippendorff argues that the reliability tests should reach at least a value of 0.80 for a content analysis to be valid.

We tested stability several times at the beginning of the training process, and we consistently reached satisfactory results. Reproducibility was harder to reach. Actually, we used the training process to improve the k-alpha from its unsatisfactory value in the first attempt to the minimum level of 0.80. Once this satisfactory level was reached, we launched the actual content analysis.Footnote 12

The actual content analysis extended over six months.Footnote 13 At regular intervals during this period, we performed reliability tests, the results of which are reported in Table 4. The unitizing and coding are reliable as the mean k-alphas vary from 0.82 to 0.98.

Table 4 Inter-coder reliability tests (k-alpha)
Full size table

4 Results

4.1 Measuring the Opacity of the Veil

We performed a content analysis of the constitutions of 16 countries (see Table 5). The length in words of the constitutional texts varies from 2,643 words (Libya) to 26,917 (Greece).Footnote 14 On average, we extracted 564.6 power relations from each constitution (coefficient of variation = 41 %). This corresponds to a mean of 46.7 power relations per thousand words (density). The highest density was found in the American constitution (56.6), the lowest one in the Greek constitution (33.2). The oldest constitution is the American one, the youngest is the constitution of Niger adopted in 2010.

Table 5 Characteristics of constitutional documents
Full size table

We can deduce several measures of the opacity of the veil of ignorance (or, equivalently, the prevalence of uncertainty in constitution-making) from the coding performed on each power relation found in the constitutional documents. Four are reported in Table 6. They refer to variables V1, V2, and V3, defined above, and to a combination of V1 and V2.

Table 6 Measures of the opacity of the veil of ignorance (Ratio Freq./N)
Full size table

The first column reports the proportion of the total number of power relations that refer to authority as the main source of capacity for the agent holding power. In the Belgian constitution, for example, 84 % of the 572 power relations that we extracted refer to holders of authority (1 % refer to holders of wealth, 8 % to holders of knowledge; for 7 % of the power relations, it was not possible to make a precise decision on this issue. Note that these last results are not reported in the table). On average, 75 % of power relations refer to authority as the source of power. This result confirms Buchanan’s and Tullock’s intuition. Because they are more uncertain of their future political position, drafters tend to agree more often on contents that relate to political power rather than economic or preceptorial power. The cross-country variation is quite modest with a coefficient of variation of 7 %. The constitution that manifests the highest degree of uncertainty is the American constitution at 0.85. According to our assumptions, the drafters of the American constitution manifested more uncertainty than any other group of drafters in the sample. The lowest value is Switzerland’s at 0.62.

The second and third columns report the proportions of social and negative relations, respectively. As was argued above, social power relations (i.e., the capacity to act over another person) and negative power relation (i.e., the capacity to prevent someone from performing an action) denote a higher level of uncertainty than instrumental or positive power relations. The average proportion of cases of social relation is 0.47 suggesting that the discourse in constitutional texts is divided more or less equally between social and instrumental power relations. France has the highest score on this variable (0.55) and Switzerland the lowest (0.37). The average proportion of negative power relation is very low at 0.04. That means that almost all the power relations that were extracted denoted a positive direction. This means that constitutional texts tend to assign (positive power relation) rather than restrain (negative power relation) powers. Even though these proportion are low, there still is an important variation (CV = 58 %). The USA has the highest score which denotes more uncertainty, Cameroon and Canada the lowest.

Figure 3 shows the scatterplot of Source-Authority by Type-Social with four quadrants defined by the variable means. The diagram identifies one group of countries where, according to the two variables, the level of uncertainty was high (the north-east quadrant): USA, Estonia, Germany, Cameroon, and France. The southwest quadrant shows the countries where uncertainty was the lowest: Switzerland, Libya, Chad, Syria, and Niger. The other countries occupy a middle range position on one or the other variables.

Fig. 3
figure 3

Scatterplot of source-authority * type-social

Full size image

To translate the clusters of Fig. 3 into a continuous variable, we computed the ratio of the frequency of references to authority and to social power over the total number of power relations (fourth column of Table 6). This variable might be interpreted as a stronger version of our measure of the prevalence of uncertainty because it takes into consideration the two dimensions already mentioned. On average, 36 % of the power relations extracted from constitutional documents refer to authority and to social power. The variation is relatively modest with a coefficient of variation equal to 13 %. France, together with the USA, Germany, and Belgium exhibit the highest scores and Switzerland the lowest.

The correlations between these variables are reported in Table 7. They suggest that the combined variable Authority-Social is a summary of the variation in the two variables included in it. The weaker correlations between Direction-Negative and the other variables suggest that this variable maybe captures a different phenomenon. We will just ignore it.

Table 7 Correlation matrix
Full size table

4.2 Assessing the Validity of Our Measure of Uncertainty

Assessing the validity of a measure is a very difficult task, especially when we try to get at a phenomenon that has not been measured in the past, as it is the case here. We know of no other attempt at measuring the opacity of the veil of ignorance or the prevalence of uncertainty in constitution-making. Therefore, any assessment can only be tentative at this stage. Reliability is a necessary condition for validity. The inter-coder reliability tests reported above show that our results are reliable. But reliability is not sufficient to establish validity. Indeed, reliability means that most shots hit the same spot. But what if the shots consistently hit the same wrong spot? Assessing the validity of a measure implies that we assess the extent to which we hit the right spot, i.e., the outcome of the measurement process corresponds to the phenomenon that we are measuring.

A first type of validity test is face validity, a subjective evaluation of the quality of a measure: Does it reasonably measure what we are studying? Do we get what we would normally expect to get? Figure 3 and Table 6 give us a classification of countries in terms of the prevalence of uncertainty in constitution-making that seems reasonable. USA, Belgium, Estonia, Germany, and France cluster together at the higher end of the spectrum, whereas Chad, Syria, Niger, and Tunisia cluster together at the lower end. This is the kind of results that a first educated guess would suggest. Countries from the North should align with Buchanan and Tullock’s view more than countries from the South. But one case seems to be oddly located in this regard. Switzerland has the lowest score on V1V2 (0.26) at 2.37 standard deviations under the mean. According to the educated guess just mentioned, Switzerland should get a score much closer to that of Germany, for example. Our measure fails a face validity test.

This odd result for Switzerland reminds one of the position of Switzerland on Borrelli and Royed’s index of strength of government (1995: Table 4). In their paper, these authors tested the hypothesis linking government strength to change in fiscal deficit for 16 OECD countries over the 1959–1990 period. To perform this test, Borrelli and Royed had constructed an index of ‘strength of government.’ According to this index, Switzerland had the lowest score in government strength whereas it was one of the three countries which had never had a fiscal deficit in that period (the other two countries were, according to Borrelli and Royed’s figures, Norway and Finland). Borrelli and Royed noted «the extremely ‘weak’ value registered by Switzerland, which might strike some readers as undeserved […] Arguably, these numbers are an artifact of Switzerland’s traditionally ‘consociationalist’ politics […] rather than symptoms of inherent ‘weakness’» (Borrelli and Royed 1995: 245). On the face of it, the score for Switzerland was invalid. But Borrelli and Royed correctly chose not to adjust their index arbitrarily to make Switzerland appear ‘stronger.’ They ran regressions with and without Switzerland and found that their results were robust.

This strategy applied by Borrelli and Royed corresponds to a validity test called construct-validity. Here, we ask the question: How does the measure relate to another measure with which it should be theoretically related? In the case of uncertainty in constitution-making, we argued above that one source of uncertainty for constitution drafters is the possibility not to be in government anymore when comes the time to make decisions under the rules set by the constitution they are drafting. And we suggested that this uncertainty was higher in democratic settings. Therefore, uncertainty should be higher in democratic systems. Another source of uncertainty is the historical depth of a polity: The longer the history of living together, the lower the uncertainty. Using data from the Polity IV Project (Marshall et al. 2010), we explored these two hypotheses by regressing our measure of uncertainty (V1V2) on the variables ‘Polity’ and ‘Persist.’ The Polity variable ranges from +10 (very democratic regime) to −10 (very autocratic regime). The Persist variable is the number of years the regime has persisted without a recorded change in values on any of the six Polity component variables. The values of the two variables are reported in Table 9. The regression results are reported in Table 8.

Table 8 Regression results, dependent variable: uncertainty (V1V2) (std. error in parentheses)
Full size table

The first two regression models show that the level of democracy (‘Polity’) is positively related to uncertainty. More democratic countries show higher uncertainty in constitution-making. This result is not significant in the first model, but it is in the second one where the ‘Persist’ variable is introduced. In this model, the coefficients of the two variables are significant and have the appropriate sign. The R-square is relatively important at 0.35. This convincingly supports the contention that our measure is construct-valid.

The other four regression models are variants of the first two that show the robustness of the ‘Polity’ variable and that confirm the sign of the ‘Persist’ variable. In models 3 and 4, we arbitrarily adjusted the score of Switzerland on V1V2 to make it equal to that of Germany. Note that, per Table 9, Germany has a similar score on ‘Polity’ but a very different one on ‘Persist.’ The coefficient for ‘Polity’ is now significant in the bivariate model (model 3), and ‘Persist’ has a negative, though insignificant, coefficient in model 4. The R-squares are higher than in the first two models but of the same magnitude. In models 5 and 6, we simply dropped Switzerland from the sample and got similar figures. The differences that we note in models 3–6 do not justify a special treatment for Switzerland.

Table 9 Level of democracy (Polity) and regime persistence (Persist)
Full size table

The conclusion to draw from these tests is that our measure of uncertainty passes the test of construct-validity. There is no reason to be unhappy with Switzerland’s score. Quite the opposite! Like it was constructed, our measure of uncertainty not only confirms the hypothesis of higher uncertainty under democracy but it reveals the importance of historical depth. A longer history of living together prompts constitution drafters to be more confident in the future and to agree more easily on issues that involve economic or preceptorial powers.

5 Conclusion

This chapter presents the Veil of Ignorance Project. It views constitutional decisions as the outcome of social interactions among drafters as proposed by James Buchanan. In order to develop a political economic approach to constitutional design under uncertainty, it proposes a conceptual framework based on power relations. We assume that drafters want to maintain or to improve their political, economic, and preceptorial power position in society through specific constitutional contents. As they are more uncertain about their future political power position as opposed to their future economic or preceptorial power position, we infer that the level of uncertainty in the constitution-making process is higher when the content of a constitution refers more often to political power, to social as opposed to instrumental power, and to restraining rather than granting powers. Adopting a discursive approach to analyzing constitutions, we proposed a research design based on content analysis procedures. We show that there is a substantial variation in the level of uncertainty among 16 national constitutions and that this variation is explained by the level of democracy and the length of time since the last important constitutional change.

Decision-making under uncertainty is certainly an important issue when it comes to understanding why specific constitutional decisions are made. This project presents the first systematic empirical exploration of this issue. The results presented here suggest that a discursive approach to constitutional analysis may be fruitful as they highlight how a content analysis coupled with a theory of power relations can help understand the playing of uncertainty in constitutional design.