The problem of epistemic or knowledge-related discretion arises whenever knowledge of what is commanded or prohibited by the constitution is unreliable. The unreliability of knowledge and the discretion resulting from it are important aspects in the balancing of constitutional rights. A theory of epistemic discretion is capable of solving the problem of a proper allocation of burdens of proof. Greer has remarked that this aspect, although ‘not wholly inconsistent’ with Alexy’s model, was under-developed in his theory.1 This remark is correct in so far as the theory of epistemic discretion has only recently been integrated into the principles theory. As we will see, Robert Alexy has, in his Theory of Constitutional Rights, laid down an analysis of epistemic discretion that appears to raise a number of questions. These concern, for example, the relevance of empirical and normative knowledge in balancing, the function of reliability in the weight formula, and the correlation between epistemic discretion of the legislature and judicial review by a constitutional court. This chapter highlights some of the problematic issues and suggests several modifications of and completions to Alexy’s analysis.
This chapter focuses on an analysis of epistemic discretion in constitutional law. We will analyse epistemic discretion with regard to both normative and empirical knowledge and suggest several modifications and completions to Alexy’s analysis. We will develop a new classification balancing that complements Alexy’s weight formula. Classification balancing is necessary in order to decide how serious an infringement with constitutional rights is and how important conflicting principles are. This leads to a new type of discretion, the discretion of classification. We will identify the discretion of classification as the real epistemic discretion, whereas the use of this term so far has been deceptive.
We will first recapitulate on some basic elements of Alexy’s new principles theory of balancing (section II), followed by an analysis of empirical epistemic (section III) and normative epistemic (section IV) discretion, respectively. Finally, we will analyse the relation between discretion and judicial review by a constitutional court (section V). Here, we will argue that both aspects have to be strictly separated. The role of epistemic discretion has until the present day merely been examined in regard to a particular situation which is characterized by two elements: namely the relationship between the legislature and the constitutional court in matters of constitutional rights.2 A generally accepted theory applying to relations of control between other authorities, however, does not yet exist. We will argue that, contrary to Alexy’s position, the so-called formal principles are not relevant in order to establish the nature and the scope of epistemic discretion. Formal principles concern questions of competence only, whereas epistemic discretion arises just at the level of material principles. We will suggest an alternative model which allows for taking into consideration different relations of control between different public authorities or courts.
When balancing two competing principles, the second law of balancing does not only take into account the intensity of interference with the respective principle, as does the first law of balancing. It also considers—to the same extent—the certainty of the empirical knowledge underlying the interference.6 It can be combined with the first law of balancing in a way that in case of an increase of uncertainty, the ‘capacity of infringement or of defence’ of the principle concerned decreases.7 The weight formula can depict these correlations. Its basic form reads:
The variables Wi and Wj stand for the abstract weight of the two principles Pi and Pj; for reasons of simplification they will not be considered here. Ii and Ij stand for the respective intensity of interference by non-satisfaction of the principles; Ri and Rj, for the reliability of the respective empirical assumptions.
It is crucial to identify the variable R as a genuine epistemic operator, referring to the extent to which reliability or certainty of knowledge exists. R is concerned with reliability, not with probability. Only when taking this approach, phrases such as ‘x is to 100 per cent reliability probable to 20 per cent’ can be taken into account. Balancing of this kind is needed where legal argumentation presupposes the consideration of probabilities. This is the case, for instance, where risks have to be assessed or consequences are to be considered. Furthermore, only this solution allows for relating R to empirical as well as normative premises. A scale of probability, on the contrary, could sensibly only be related to empirical premises. Important aspects of the analysis of balancing could not be taken account of, namely situations of uncertainty regarding normative premises. When talking about ‘certainty’ or ‘reliability’, we will therefore refer not to probability, but to an epistemic mode.
1. The problem of scales
The problem of scales is concerned with the relation between reliability of empirical premises and judicial review. Alexy deduces the scale applied to empirical reliability, that is to the variable R, from the German Federal Constitutional Court’s intensity of review.8 He uses the following three epistemic classes: (r) reliable or certain, (p) plausible or defensible, and (e) not evidently false, and assigns the values r = 20, p = 2−1 and e = 2−2. Such a scale is a one of review, since the authority applying the classes, that is the reviewing authority, by means of these criteria is only able to examine the plausibility of an assessment by a different authority.
In consequence, the Federal Constitutional Court would merely review the assessment of the legislature as to the intensity of interference. The court would not, however, make an assessment on its own.9 Thus, the legislature would have an extremely wide discretion as regards assessment. Two arguments opposing this outcome can be put forward. First of all, this scale of review has specifically been developed to depict the relation of the legislature and the Federal Constitutional Court.10 It fails though, in the absence of a reviewing authority, or in cases in which the relation of review is of a different nature. The balancing of a court of first instance based on empirical uncertainties can, for example, be entirely substituted by a court of appeal. The Federal Constitutional Court will also apply a different scale depending on the matter in dispute, which could be a statute (relation of review to the legislature) or a judgment (relation of review to the ordinary courts). Subject to the characteristics of the relation in question, R would have to be scaled in different ways. This would diminish the universality of the second law of balancing.
Secondly and more importantly, Alexy’s scale of review cannot meet the requirements of the second law of balancing. The reliability of the premises underlying the interference has to be balanced with the intensity of interference and not merely with the plausibility of the respective assessments. The question on the degree of intensity of review needs to be looked at separately from the question of reliability.
We therefore suggest applying a scale according to the degree of reliability or certainty of the empirical assumptions. Corresponding to the triadic scale, the steps would be as follows: ‘reliability or certainty’ (c = 20), ‘average reliability or certainty’ (a = 2−1), and ‘unreliability or uncertainty’ (u = 2−2). This scale is to be called the reliability scale, as opposed to Alexy’s review scale.
The advantage of an approach referring to the degree of certainty lies in the universal validity of this scale as well as in its independence from specific relations of review. Every authority involved with balancing can, on its own account, determine the reliability of the premises underlying an interference with a constitutional right. Subject to Article 1(3) of the German Basic Law, it is even obliged to do so, for every authority has to come to a decision in accordance with constitutional rights, independently of subsequent review. It cannot delegate this responsibility to an authority that reviews the decision thereafter.11 The question of whether a legislative decision on the grounds of empirical uncertainty entirely falls under the ambit of its discretion or whether constitutional rights can to a certain extent set limits in this respect arises irrespective of possible subsequent review.
According to the reliability scale which is given preference here, relations of review—especially the relation of the legislature and the Constitutional Court—do not need to be taken into account when reconstructing epistemic discretion.12
2. The classification problem
The classification problem is concerned with the question of how to approach situations of empirical uncertainty when classifying the intensity of interference.
Shooting down an aircraft carrying passengers and causing their certain death is a serious interference with the right to life. In its decision on the Aviation Security Act, the Federal Constitutional Court emphasizes the seriousness of this interference by regarding the core of the right, protected by the guarantee of human dignity, to be affected.13 The importance of the competing principle, the state’s duty to protect the lives of the people not on board threatened by a terrorist attack, can certainly also be classified as high. In both cases empirical uncertainties do not exist.
It is crucial to note that in situations of § 14(3) of the Aviation Security Act, there are no empirical uncertainties regarding the classification of intensity of interference and the importance of the competing principle, respectively. Empirical uncertainties only exist as to whether the conditions required for such interference are met in the individual case, that is whether, for instance, a terrorist attack is given as a matter of fact or whether the radio contact is merely lost.
One can think of situations, however, where—contrary to the situation described above—empirical uncertainties are also given in regard to determining the intensity of interference per se. In general, two situations of empirical uncertainty have to be distinguished: the first situation of uncertainty regards the underlying premises of an interference, while the intensity of interference is certain, that is in terms of the weight formula, the value for Ii is fully reliable (situation of uncertainty no 1). In contrast, in the second situation the intensity of interference per se is uncertain (situation of uncertainty no 2).
Presuming the second situation of uncertainty to be far more common, it is also the one with greater practical importance. Empirical knowledge is often not sufficiently assured in order to precisely determine the extent and consequences of the detriment to a principle. Existing data, especially on potential long-term effects, can be very uncertain. In order to illustrate the second situation of uncertainty, we will consider the Federal Constitutional Court’s cannabis judgment already referred to by Alexy to explain the second law of balancing.14
When prohibiting the possession of cannabis, the classification of the intensity of interference with the freedom to act (Pi)as moderate (m)is empirically certain, whereas the classification of intensity of interference with the competing principle is uncertain. The legislature aims at protecting the health of the population against the dangers connected with the narcotic substance (Pj). The negative effects on health, caused by consuming cannabis, as well as the question of which protective effect can be attributed to criminal prohibition compared to other means of protection are—as matters of fact—uncertain. In consequence, the classification of intensity of ‘interference by non-satisfaction’15 of the duty to protect Pj is empirically unreliable.16 If cannabis endangered the health to only a minor extent, the interference by non-satisfaction with Pj would have to be classified as light (l). If, however, scientific evidence became available on serious harmful effects of consuming cannabis, taking into account the aspect of introducing people to drugs in general, the interference by non-satisfaction with the duty to protect would have to be classified as serious (s).
The second law of balancing does not offer any solution for these kinds of situation, because it requires a ‘certain’ classification of intensity of interference and merely considers uncertainties regarding the underlying premises of the interference. According to the second law of balancing, the degree of the existing uncertainty always needs to be related to a determined intensity of interference. The law of balancing does not, however, determine the intensity of interference per se. The relation between intensity of interference and the reliability of empirical premises can be illustrated by means of an indifference curve. The second law of balancing allows choosing any point on the indifference curve. Therefore, it does not clarify the issue of which of these points is to be chosen.
This reveals a fundamental lacuna in the present doctrine of balancing. We will illustrate this dilemma by means of the cannabis judgment. The Federal Constitutional Court held that the legislature’s original assessment of the health risks being serious is contentious today. The assumption of cannabis products not being dangerous, however, was also found to be unreliable.17 Assuming that both alternatives are likewise averagely unreliable (a = 2−1) leads to two differing balancing results, as can be seen in Formulas 15 and 16, respectively. If one relies on effects that are serious to health, the result is as follows:18
If, on the contrary, one relies on mere light effects to health, the result is as follows:
Formula 15 represents a stalemate. Thus, the prohibition of cannabis is neither commanded nor prohibited by constitutional rights. This allows for structural discretion of the legislature. In Formula 16, however, the freedom to act Pi takes precedence. A prohibition against the possession of cannabis can therefore not be constitutionally justified. Which intensity of interference with Pj, then, is to be given preference for balancing? The second law of balancing does not answer this question.
In order to solve the classification problem, seven possible approaches can be taken. We can either use the asserted, the most pessimistic, the most optimistic, the most uncertain, or the most certain intensity, as well as a point in the continua either between the most pessimistic and the most optimistic or between the most uncertain and the most certain intensity. This follows from the fact that balancing, in cases of empirical uncertainties, is based on two continua which are located at different levels. One continuum ranges from the most uncertain to the most certain value, the other ranges from the most pessimistic to the most optimistic value.
The most uncertain intensity can be excluded without difficulty. If a value is the most uncertain one, a value of higher epistemic certainty is always available. The latter value, then, has to be used in balancing.
Giving preference to the most optimistic intensity can also be ruled out. The most optimistic solution would be to assume the lightest interference possible. This, however, would not be compatible with the optimization character of principles. Principles demand that only assumptions are made which are most favourable to their optimization. In case of uncertainties regarding the degree of intensity of interference, the most favourable assumption for optimization is the most pessimistic classification. The most optimistic solution is at the other end of this continuum. It can therefore be ruled out.
That leaves five further options. Alexy does not address this problem, but his theory in its consequence has to give preference to the asserted intensity. This is true because Alexy deduces the scale from the Federal Constitutional Court’s intensity of review. Thus, he applies a scale of review; it is therefore always the intensity of interference asserted by the reviewed authority that is regarded.19
We have already given preference to the reliability scale which only differentiates according to the degree of reliability of empirical assumptions, irrespective of relations of review.20 Hence, when determining the intensity of interference in cases of empirical uncertainty, it can be concluded that we cannot rely simply on the interference asserted by a reviewed authority.
As mentioned above, the optimization character of principles suggests the most pessimistic classification. When the degree of intensity of interference is uncertain, the most favourable assumption for optimization of the constitutional right in question is the most pessimistic classification. Since the interference, therefore, tends to be assessed as serious, the reasons justifying the interference would have to be of comparatively heavy weight, according to the first law of balancing.
Nevertheless, arguments objecting to the preference of the most pessimistic classification can be put forward. In cases of decisions that require prognoses, the most pessimistic classification involves the risk of incapacity to act. Grave effects can often not be dismissed beyond doubt. Giving preference to the most pessimistic classification, even though less serious effects are more reliable, could in some cases lead to classifying the intensity of interference more pessimistic than needed. This would occur when more serious effects could not be dismissed beyond reasonable doubt.
This disadvantage would not arise, however, if one gave preference to the most certain classification. Yet, we can also put forward arguments against this approach. To differentiate degrees of reliability is often impossible, especially in cases of greatest uncertainty. Furthermore, the most certain solution cannot be given preference if the classifications are equally certain; this is often the case when coarse scales are used. This problem became relevant in the cannabis judgment, because the Federal Constitutional Court assessed both the danger of serious health risks as well as such a danger not to exist to be equally certain.21
Furthermore, a classification of an intensity of interference as merely unreliable (u) may not meet the requirements of the second law of balancing, since such classification takes a relatively high risk to be incorrect. In a case involving considerable threat to important constitutional rights, this risk can be assessed as being too high.
These considerations demonstrate that aspects of the most certain and the most pessimistic solution need to be combined when classifying the intensity of interference in cases of empirical uncertainty. The solution suggested here is balancing the most certain and the most pessimistic solution. Since this balancing leads to the classification of the intensity of interference, it will be called the classification balancing.
Classification balancing precedes the application of the weight formula and provides for its application, for it allows determining which values are to be assigned to Ii and Ri as well as to Ij and Rj.
The second law of balancing does not take account of classification balancing. It rather describes the correlation of the two elements to be balanced, namely is the most pessimistic classification (’the more serious an interference’) and reliability. In order to apply classification balancing, however, two or several possible intensities of interference, including their respective reliabilities, need to be considered. Thus, it is to be emphasized that classification balancing is not, as is the second law of balancing, concerned with the relation between the intensity of interference and its corresponding reliability, but with the relation between two or further intensities of interference, including their respective reliability.
An indication for classification balancing can be found in the cannabis judgment. One intuitively arrives at the conclusion that in case of equal reliability of two possible intensities of interference, it is always the pessimistic perspective that needs to be given preference. Only when taking this approach can the more serious effects that are imminent be averted. On the other hand, as soon as one arrives at a certain point of increased reliability of a mere light interference, it is convincing to rely on a light intensity of interference. One arrives at this point even faster if the (less reliable, but still imminent) pessimistic intensity of interference is merely valued as moderate and not as serious. The following rule can therefore be inferred:
The more reliable a more intensive classification of intensity of interference is, the more reliable must be a less intensive classification of intensity of interference.
This rule of balancing shall be called the law of classification. It has a heuristic relation to the weight formula. When the intensity of interference is uncertain, this rule allows one to decide with respect to a particular principle Pi on one of several intensities of interference Ii1, Ii2, …, Iin and their respective degrees of reliability Ril, Ri2, …, Rin. If the product of intensity of interference, reliability, and abstract weight is understood to be the ‘capacity of infringement or defence’ Ci of a principle, the law of classification directs the decision on one of several possible capacities of interference Ci1, Ci2,…, Cin.
In classification balancing, the abstract weight can be ignored, because the same principle is being examined on both sides—differing from the weight formula itself. The same value would have to be inserted in the nominator and the denominator for the abstract weight. They therefore cancel each other out. The following applies:
The law of classification directs the decision on selecting a concrete capacity of infringement relative to another possible capacity of infringement. The optimization character of principles requires giving preference to the pair of values which has the greater capacity of infringement. The law of classification means that the stronger a capacity of infringement Ci1 is, the more reliable must be a lower classification of intensity of interference, as part of a different capacity of infringement Ci2. The lower classification of intensity of interference, as part of Ci2, weakens Ci2 in relation to Ci1. Such deficiency could only be compensated or exceeded by a comparatively higher reliability.
If the value Cil, 2 is greater than 1, the values Iil and Ril are to be inserted in the weight formula. If the value Cil, 2 is below 1, the values Ii2 and Ri2 are to be inserted. In situations of Cil, 2 = 1, a stalemate occurs; discretion is given on whether the values for Ii and Ri summarized in Ci1 or in Ci2 are to be inserted into the weight formula.
The discretion that arises in cases of a stalemate between Cil and Ci2 has not been described previously. It shall be called the discretion of classification. The discretion of classification is the genuine empirical epistemic discretion. It arises when examining one single principle. It therefore precedes structural discretion, which can only arise in cases of a conflicting second principle. Thus, for the first time, the discretion of classification is identified as the true empirical epistemic discretion: only in cases of a classification stalemate can empirical epistemic discretion which relies on uncertainties regarding empirical premises be given.
Alexy’s definition of empirical epistemic discretion, in contrast, relies on the relation of review between the Federal Constitutional Court and the legislature.22 It is not inferred from the principles themselves, but emerges by inappropriately mixing the level of principles and the level of review. In his analysis of the prohibition of cannabis, Alexy relies without further explanation on the classification asserted by the legislature. In consequence, in cases of empirical uncertainty, the legislature could be free to decide which level of intensity of interference to apply in balancing. Its discretion of classification would be unlimited as far as the level of principles is concerned. This is incompatible with the legislature being bound to the constitution. Alexy considers empirical epistemic discretion to be too wide.
We have demonstrated for the first time that such an unlimited liberty of the legislature regarding classification does not exist—even in cases of empirical uncertainty. The limits can be established by applying the rules regarding classification balancing. Only in cases of a classification stalemate can the legislature take any decision on which capacity of interference to insert into the weight formula. These considerations refer to the level of principles. The solution may be different on the review level, however. The classification discretion may be modified, depending on different relations of review.23
In order to illustrate this novel balancing rule, we will consider different scenarios of the criminal prohibition of possessing cannabis. The question at issue is concerned with the capacity of interference of the principle to protect from health risks pj, as opposed to the interference with the freedom to act Ii. The latter is known and set to be moderate. The question here is therefore only as to the values of the denominator of the weight formula—that is, Ij and Rj.
In cases of average certainty (Rj1 = Rj2 = a = 2−1) of either light (Iij = l= 20) or serious health risks (Ij2 = s = 22), the value for Cj1, 2 is 4.
Thus, a serious interference has to be used in balancing. Contrary to the Federal Constitutional Court24 and to Alexy,25 who agreed with the court, the legislature did not have empirical epistemic discretion on the question of health protection.
If the threat of serious health risks (Ij1 = s= 4) is less reliable, but yet cannot be ruled out (R7 = u= 2−2), and if mere light health risks (Ij2 = l =20) are relatively certain (R2 = c = 20), a classification stalemate occurs:
It is then entirely for the authorities to decide on the intensity of interference. Because Cj1 and Cj2 have equal values, it is irrelevant which intensity of interference is relied on when balancing Pj with the freedom to act by means of the weight formula.
The position changes in favour of the optimistic approach, if (for instance, due to novel scientifically based knowledge) serious health effects can be entirely ruled out and—at most—only moderate negative effects to health exist (Ij1 = m= 21), and even this is uncertain (Rj1 = u= 2−2). This shall be based on the unmodified assumption of certain reliability (Rj2 = g=20) of only light intensity of interference with the right to health protection (Ij2 = l =20). Thus, the only difference between Formulas 19 and 20 is the value for Ij1 (s and m, respectively). In this case, the mere light interference by non-satisfaction of the duty to protect must be balanced with the freedom to act:
In sum: the classification problem is concerned with the question of which value to rely on for balancing when the intensity of interference is uncertain. This problem of choice needs to be solved prior to balancing by means of the weight formula. As described above, the problem can be solved by balancing imminent intensities of interference and their respective degrees of reliability. This balancing, which is directed by the law of classification, can be called classification balancing. Classification balancing is concerned with external justification of the values to be inserted in the weight formula.26 Only the correlation of different pairs of values is being examined, however. There is therefore a further external justification, in which again the pairs of values need to be justified independently from each other. If the result of classification balancing is a classification stalemate, genuine empirical epistemic discretion is given.
3. Dependency on perspectives
Rivers identified that the second law of balancing only considered one side of the weight formula.27 However, situations of uncertainties regarding both principles are possible. The weight formula takes this aspect into account as the variable R appears on both sides of the quotient. Thus, Rivers suggests rewriting the second law of balancing so that it would read analogous to the first law of balancing:
The greater the chance that one principle may be seriously infringed, the greater must be the chance that another principle is realised to a high degree.
His terminology of ‘chance’ does not perceive the epistemic character of the operator R. In addition, his attempt to phrase this relation of the second law of balancing has the effect that the relation between reliability and intensity of interference (the capacity of infringement or defence), which is expressed by the second law of balancing, is not illustrated anymore. Even if one elaborated Rivers’ suggestion in order to compensate this deficiency, it would only come to a description of the known weight formula28 Rivers puts both elements (reliability and intensity of interference) of one principle into the antecedent and the two corresponding elements of the colliding principle into the apodosis. If one stresses the relation of the two elements and includes the abstract weight to arrive at the complete picture, one gets the following formulation:
The greater the product of reliability, intensity of interference and abstract weight of one principle is, the greater must be the product of reliability, intensity of interference and abstract weight of the other principle.