Technical Legitimacy

The Empirical Appeal of Technical Legitimacy

A consistent finding in a wide range of empirical studies is that the tenets of technical legitimacy are powerful sources of empirical legitimacy for unelected bodies. Two examples of how this plays out in political discourse: The European Commission has defended the use of independent European Union (EU) agencies on the grounds that “their decisions are based on purely technical evaluations of very high quality and are not influenced by political or contingent considerations” (Commission of the European Communities 2002: 5). Alan Blinder has similarly bemoaned that government had become “too political.” He suggested that the delegation of more decisions to independent agencies would produce better policy decisions on “less political grounds” (Blinder, 1997: 126).

Similar sentiments are found in studies of citizen attitudes, ideological developments, institutional structures, and government organizations’ strategies of legitimization and reputation management. Citizens often have a preference for experts making political choices (Bertsou, 2021; Bertsou and Caramani, 2022; Hibbing and Theiss-Morse, 2002; Lavezzolo et al., 2021), at least under certain conditions and in certain policy stages. Representative party politics is increasingly challenged from a technocratic angle (Bickerton and Accetti, 2021; Caramani, 2017; Friedman, 2019). Unelected bodies are pervasive in modern society (Vibert, 2007). Epistemic logics occupy an increased space even in outright majoritarian venues like parliaments (Fischer, 2009). Depoliticization has been described as a large-scale trend in democracies (Fawcett et al., 2017). Institutions themselves might find it valuable to cultivate a perception of neutrality and technical expertise, as demonstrated in the literatures on knowledge use (Boswell, 2009; Littoz-Monnet, 2020; Sabatier, 1978; Weiss, 1979), risk-based regulation (Black, 2005; Hutter, 2005; Paul, 2021), and bureaucratic reputation (Busuioc and Rimkutė, 2020; Carpenter, 2010: 6; Maor, 2015: 19).

These authors study more than merely non-majoritarian institutions, suggesting that we are dealing with a more general logic. I do not claim that technical legitimacy is the only dimension of legitimacy applicable to non-majoritarian institutions. But in many instances, technical legitimacy is an empirical source of trust, reputation, and support for government institutions. Whether that trust and support is well founded, however, is a matter of normative analysis.

The Normative Debate over Technical Legitimacy

Some normative theorists are supportive of a variant of technical legitimacy. Fabienne Peter carves out “cognitive political authority” as a separate ground for legitimacy alongside the more established notion of “political will” (Peter, 2023). Others defend delegation to administrative experts on the grounds that it is required by good governance and the need for expertise (Heath, 2020; Sunstein and Vermeule, 2020 see a critical discussion in Lofthouse and Schaefer, 2025). Many more analysts, however, view the idea that certain institutions are legitimized by their technical, apolitical, and objective basis with skepticism. Claudio Radaelli reflects a general sentiment when he calls the insulation of a wide range of policy areas from majoritarian democratic procedures “fundamentally flawed from a normative point of view” (Radaelli, 1999a: 7). There are broadly two types of worries: one democratic and one epistemic.

The democratic worry is that delegation to non-majoritarian institutions inappropriately removes policy decisions from democratic debate and contestation, and that it fails to respect democratic equality since it gives more decision-making power to unelected experts than to ordinary citizens (Bellamy, 2010; Føllesdal and Hix, 2006: 11; Thatcher and Sweet, 2002: 19; Thoma, 2024). Moreover, non-majoritarian delegation may unduly narrow the sphere of democratic debate since it “closes down debate by technical fiat” (Moore, 2017: 38; see also Fawcett et al., 2017; Mair, 2013).

The epistemic worry centers on the issue that empirical knowledge is uncertain and that knowledge claims and epistemic authority therefore are poor sources of legitimacy. Real-world experts disagree with each other on the causes and solutions of social problems, and the social sciences have an especially poor predictive accuracy (Friedman, 2019; Radaelli, 1999a). If this is true, delegation to experts might ascribe an unwarranted cognitive authority to expertise and expert judgments (Holst and Molander, 2019: 545). Moreover, the separation of political from technical considerations, while conceptually clean, may prove untenable in practice (see, for example, Bellamy, 2010: 9; Busuioc, 2013: 256). If technical and political considerations are intertwined, the charge goes, it is difficult to legitimize institutions by reference to their purely technical basis.

It is important to realize the scope of such critiques. It is self-evidently appropriate, from any reasonable conception of democracy, to worry about the overstretching of technical legitimacy. It clearly cannot be the sole basis of political power in society (Beetham, 2013: 101; Radaelli, 1999a: 28). But a narrower version of technical legitimacy which can apply to a subset of clearly demarcated non-majoritarian institutions is also compatible with a wide range of theories of democracy.

A normative reconstruction of technical legitimacy must begin from the realization that societies choose to delegate authority to non-majoritarian bodies for a reason. These institutions’ independence from majoritarian democratic procedures is no coincidence; it is a core feature of their institutional design. The general logic is that of trusteeship (Knight and Johnson, 2011: 178). There might be instances where it is beneficial to establish institutions that do not share elected politicians’ preferences or incentive structures, and where that dealignment is part of the rationale for their existence (Majone, 2001: 104).

There are two main justifications for non-majoritarian delegation in the scholarly literature (see Majone, 2001). One is to enhance the credibility of policy commitments. The other is to leverage expertise and safeguard it from political interference. While analytically distinct, the justifications tend to overlap in practice.

The credible commitments argument is most clearly developed in neoclassical macroeconomics. The central idea is that certain government interventions in the economy, however well-intentioned, are bound to fail because of the rational expectations of market actors (cf. Stahl, 2021: 411; see Kydland and Prescott, 1977 for a classic example). Take monetary policy. The classic argument says that politicians are incentivized to use monetary policy for their short-term political gain, for instance, by enacting expansionary policy before an election, and market actors rationally expect this. The only way to credibly commit to price stability is to delegate monetary policy to an independent institution—an independent central bank. The logic of credible commitments extends to all instances of regulation where there is a time-inconsistency problem, that is, where an independent regulator would be seen as more credible due to its insulation from electoral politics (Jacobs, 2016; Maggetti, 2010: 3; Majone, 2001). In fact, independence from majoritarian politics is under this argument not only accepted; some claim it is required for such institutions to be legitimate. If this is true, delegation to non-majoritarian institutions is straightforwardly legitimized. It is the only way to achieve certain goals that we value as a society.

The second justification for non-majoritarian delegation is that independence might protect the epistemic integrity of institutions against undue political influence (Holst and Molander, 2019; Pettit, 2004; Steffek, 2015: 271). Here, too, delegation brings benefits that cannot be obtained if the institution were under majoritarian control. The key difference from the credible-commitments rationale is that the benefit to be brought about here is epistemic quality, not credibility. It could be possible to credibly commit to certain policy outcomes by delegating to institutions that lack an epistemic mechanism, like a lottery. But the epistemic justification places more substantive demands on the makeup of the institution and limits the scope of policy areas which can plausibly be delegated to non-majoritarian institutions.

The two justifications do not presuppose that politicians are any more stupid, selfish, or evil than the rest of us, nor that they are pure reelection-seeking machines without any consideration for the common good or the quality of their decisions. Rather, we only need to stipulate that politicians respond to incentives that might be overall reasonable, but which we might not want every part of government to respond to. Both justifications for non-majoritarian delegation only work where non-majoritarian institutions are set up to achieve ends which are difficult or impossible to realize under majoritarian democratic procedures. Where this is the case, we can conceive of delegation as a self-binding tool that politicians resort to precisely because they are concerned with the common good. It is in this case a measure that promotes genuine democratic self-rule, not one which undermines it.

Yet the two abstract rationales of credible commitments and epistemic integrity leave the important question on the table. Democracy entails a fundamental commitment to majoritarian representation and democratic equality which should only be deviated from if it realizes the goals of democracy in a more fundamental sense. It is necessary to spell out the conditions under which a deviation from the majoritarian principle would be permitted.

Existing attempts to specify those conditions fall short in important ways. One widespread specification builds on so-called output legitimacy, whereby institutions’ legitimacy is in part evaluated on the basis on the outcomes they produce (Scharpf, 1999). There is a trade-off between democratic input and output effectiveness, and output-oriented theorists hold that non-majoritarian institutions are legitimized more by their output than by their input credentials. By itself, however, the notion of output legitimacy does not get us much closer to specifying how those trade-offs should be made—or to justifying whether such trade-offs are at all permissible to make (Cordelli, 2020: 6–8; Zacka, 2022: 26). For this, a substantive normative criterion is needed. The normative criteria most often suggested in the literature—salience, (Pareto) efficiency, value freedom—rely on empirical premises that are either empirically contingent or outright implausible. Legitimacy requires a degree of robustness against counterfactuals (Pettit, 2012). It is not enough that a practice arbitrarily satisfies some normative criterion; it should be resilient against a reasonably wide range of counterfactual scenarios. ² Lacking such reliability, the criteria discussed in this section fail to provide the type of content-independent reasons for supporting an institution’s directives that legitimacy requires.

One influential version of output legitimacy relies on political salience. Scharpf’s (1999, 2003) functional definition of legitimacy holds that the “need for legitimation will vary with the salience of the interests that are at stake” (Scharpf 2003: 6). One may, for instance, argue that so long as there is broad political agreement and low public salience around an issue, output is a sufficient source of legitimacy (Moravcsik, 2002). The problem with this approach is that political agreement is a volatile thing and public salience may quickly change. A salience criterion does not secure that good outcomes are reliably secured (see Sandven, 2024: 552–53). Citizens are exposed to an undue risk of domination if institutions are empowered and granted independence from majoritarian procedures on the sole basis that they oversee a currently un-salient policy area.

A seemingly more principled approach to output-oriented legitimation is to draw a distinction between Pareto-improving and redistributive decisions. It has been common to argue that institutions take apolitical and value-free decisions insofar as they are Pareto-improving, and that this justifies delegation to such institutions (Majone, 1994; Tucker, 2018). Decisions approximating Pareto efficiency will “improve general welfare without violating significant interests” (Scharpf, 2003, 7) and therefore do not need input-oriented majoritarian legitimization. Pareto efficiency provides a tool to identify the class of decisions that can (or should) be delegated to non-majoritarian institutions. Even critics, although arguing that pure Pareto improvements are empirically rare and therefore unlikely to justify existing institutional structures (Føllesdal and Hix, 2006; Lord and Beetham, 2001), have accepted the premise that Pareto improvements have this kind of legitimizing force. ³

I believe the critique runs deeper and that we should reject the legitimizing appeal of Pareto efficiency altogether. The Pareto argument is in effect an argument about neutrality or value freedom, intended to separate technically neutral from politically value-laden decisions. For two reasons, it fails to do that. First, by focusing on the distributional shape of outcomes, it neglects all the ways values enter into the generation and evaluation of knowledge claims throughout an epistemic process, as I return to below. Second, by stipulating that Pareto improvements are value neutral, it rests on a too thin account of values. In a society that values equality, an outcome may be Pareto-improving and still run against society’s values whenever it increases inequality. Pareto improvements alone therefore do not offer any justification for delegation in my framework.

This leads to the final approach scholars and politicians often take when justifying non-majoritarian delegation: to rely on a more fundamental separation between technical and political decisions. So long as an institution only deals with issues of means or facts, not values and ends, delegation is justified (Berlin, 2002: 166; Christiano, 2012; Vibert, 2007). The idea goes under many names—for instance, the “transmission belt,” “agency instrumentalism,” or “division of labor” model (see, for example, Cordelli, 2020; Eriksen, 2021; Pamuk, 2021; Tucker, 2018; Zacka, 2022). Politicians deal with ends and values, while unelected bodies are restricted to purely technical issues of means. By contrast to above, this approach does not center on the distributional shape of an institution’s outcomes but on a deeper division between epistemic and value-based decisions. As I will return to at length, this fails to specify the conditions for technical legitimacy because it rests on an unrealistic notion of the value freedom of epistemic processes. If epistemic processes cannot be value free in the way the division-of-labor model requires, the model fails to offer a justification for non-majoritarian delegation.

Common Good

First, I submit that technical legitimacy requires that an institution is set up in service of a functionally restricted, at least minimally authorized, and theoretically plausible common good (see Scharpf, 2009: 188; Steffek, 2015: 274). Critics are skeptical toward the notion of a common good which specialists are better placed than citizens to bring about (Bellamy, 2010; Bickerton and Accetti, 2021: 3; Friedman, 2019). In fact, its reliance on a conception of the common good is often presented as a challenge against technical legitimacy, since it easily leads into guardianship, paternalism, or technocracy. If such a common good existed, majoritarian democracy would be entirely unnecessary since experts or guardians could establish “good policy” on our behalf (Downey, 2022: 8; Steffek, 2015: 273).

For technical legitimacy to hold up to the critiques, we need to make two admissions. First, we need to adopt a theory of democracy that includes an obligation toward the quality of policy decisions in addition to its obligation toward democratic equality and majoritarian representation. This is a move that purely proceduralist (e.g. Waldron, 1999) or realist (e.g. Shapiro, 2016: 75–76) theories would be hesitant to make. Second, if we adopt a two-dimensional definition of democracy, we need to allow that functionally demarcated common goods can, under certain circumstances, be identified.

Some scholars argue that they can. Steffek (2015) treats the common good (“public interest” in his terminology) as an ideal that it is possible and fruitful for institutions to pursue even though it may never be fully reached. It can in some instances be relatively uncontroversially assumed. Citizens probably have, for example, a “consistent and enduring aversion to plane crashes, food poisoning and buildings on fire.” It seems legitimate to have independent agencies regulate such areas (Steffek 2015: 275; this closely echoes Warren, 1996: 49). It is analytically fruitful to learn from this intuition and try to extrapolate what it is about such agencies that seems to bring up fewer legitimacy concerns.

Implicit in this statement is that any feasible appeal to the “common good” is functionally limited to a specific policy area. Independent institutions should not be set up in service of the common good in a broad sense—this seems impossible to identify and democratically problematic. Instead, we can draw a parallel with Warren’s (1996) account of functionally distinctive goods. For certain functions, we trust that authorities do their jobs well without our participation or control, so long as they only make decisions that are functionally specific to the good they are in service of (Warren 1996: 48–49). The functional specification does important work, as it places limits on the types of reasons that can permissibly guide an independent agency’s conduct. Authority becomes problematic when it oversteps its functional basis—for instance, when “questions of profitability compromise the safety of food or airline travel” (Warren 1996: 49). It is important to guard against such oversteps if technical legitimacy is to be normatively defensible.

Warren’s approach is appealing but is in one aspect too permissive. He relies on a dichotomy between “settled” and “political” issues, like the accounts relying on political salience discussed above. Holding salience as a criterion for delegation avoids the need for a more substantive conception of the common good. But, again, the fact that a topic is not currently contested is not robust against counterfactuals in the sense we would like from a normative theory of legitimacy. I argue, therefore, that technical legitimacy requires more than an issue being currently settled. We need an argument for why the good that an institution is set up to realize is in the interest of all citizens and not just one faction (Steffek, 2015: 272)—and why the good in question cannot be realized under ordinary majoritarian procedures. While the weaker criterion of low salience would allow delegation for purely managerial reasons whenever it makes governance simpler and people seem to be fine with it, the common good criterion offers a potentially democratic rationale for delegation (see Pettit, 2004; Steffek, 2015: 271).

In establishing an account of a common good, majoritarian consent and authorization remains important in the delegation stage—when citizens and politicians decide on what issues to delegate and what broad parameters to stake out in the institution’s mandate. But given that citizens’ and politicians’ direct control and influence over the daily activity of non-majoritarian institutions are ruled out by definition, we cannot rely on the ordinary majoritarian way of ensuring that institutions do not drift from the public’s idea of a common good: direct answerability to elected politicians. In this sense, my framework deviates slightly from scholars who have pointed out the need for ongoing democratic control of non-majoritarian institutions. In Warren’s framework, what grants legitimacy is the fact that citizens can at any time make the settled political and thereby reintroduce democracy (Warren, 1996: 57–58). Steffek similarly calls for “avenues of communication that can be activated, reliably and with ease, when the conception of the public interest on which the output side works, is called into question” (Steffek, 2015: 275). Downey (2021) argues for the “regular reassessment by the legislature of the terms and conditions of delegation” (p. 2). I agree with the spirit of such arguments. Legislatures should not cede more power than necessary to secure the functional good that delegation provides. But nor can they cede less. There is a need for emergency valves in the form of recall procedures, parliamentary overrides, and similar, but the entire point is that these should not be activated with too much ease. Levers of control that are too easily available might open the backdoor to the type of distorting incentives that delegation intends to block, undermining the rationale for delegation in the first place. This logic is similar to how constitutions work in many polities. They are more difficult to change than ordinary laws, but not impossible to change.

In place of a thin and procedural understanding of common goods, we therefore need a substantive common good criterion. This entails a move from actual authorization to hypothetical consent. Models of hypothetical consent may be simple or complex. The above reference to citizens’ shared aversion to plane crashes is a theoretical conception of a common good based on some fundamental assumptions about human psychology. So is a deontological theory of justice, a concept such as “human rights,” a Rawlsian veil of ignorance, or a utilitarian idea of aggregate utility (Bagg, 2021: 223). Proponents of central bank independence, for instance, do not claim that they have asked everyone what monetary policy they prefer. They claim that all citizens should have a rational self interest in the gains that central bank independence brings about (Steffek, 2015: 274), and that those gains come without any real costs (Grilli et al., 1991).

I do not wish to promote a substantive theory of the common good here. My argument is intended to be compatible with many different conceptions. This strategy is broadly in line with Samuel Bagg (2021), who points out that a “deliberately underspecified” approach would allow us to “be confident in identifying clear violations of the public interest, even if we remain hesitant to specify exactly what the public interest is in any detail” (Bagg 2021: 224; see also Steffek, 2015: 272–273). The idea of a common good can at a minimum serve as a “counterfactual contrast foil” (Steffek, 2015: 272). It offers a source of justifications and public reasons that can serve as a benchmark for assessing when, for instance, an agency oversteps its functional basis. Institutions themselves must also justify their decisions in terms of the common good they are set up to serve (this, too, resembles the public justification demand in Warren (1996)). There are some functions and actions that are clearly not contributions to any reasonable conception of a common good. Where this is the case, technical legitimacy will run into trouble.

Finally, even goods that in isolation are clearly in the best interest of all citizens come with trade-offs—for instance, between safety and effectiveness, costs, and environmental harms. The surest way to avoid plane crashes would be to ban aviation. Any workable definition of a common good must therefore be context-specific and with the relevant trade-offs as far as possible incorporated. Collectively figuring out the shape and parameters of a given society’s conception of common goods remains a crucial role of actual deliberation in the public sphere, while evaluating the appeal of such claims will be a job for (among very many others) political theorists.

Expertise

If there exists a common good in some field, non-majoritarian institutions need to be able to bring it about. And they need to be better placed to do so than laypeople or elected politicians. Technical legitimacy, in other words, requires expertise. The expertise criterion is an epistemic criterion. It would be risky to delegate power to an unelected body that did not possess any distinct expertise. For instance, we entrust a central bank with far-reaching powers in part because we trust that its experts, building on macroeconomic theory, are equipped to track the true state of the economy and to evaluate correctly the causal effects of their interventions.

Epistemic worries about expertise are common grounds for skepticism about technical legitimacy (Holst and Molander, 2019). I believe skeptics and optimists alike would agree on the following: Where expertise is more reliable by some measure (e.g. precision, expert consensus, historical track record), technical legitimacy will encounter fewer problems than where expertise is less reliable. So the point of contention is empirical, about how reliable we can reasonably expect expertise to be.

There are two dimensions of the expertise criterion. One deals with characteristics of the institutional context, the other with characteristics of the expertise itself. First, non-majoritarian institutions must be designed to minimize the potential for bias, groupthink, distorted incentives and epistemic blind spots. Holst and Molander (2019) suggest a set of mechanisms. Norms, incentives, and institutional measures must be in place which ensure that experts are selected and behave according to epistemic and truth-seeking goals and are not faced with distorting incentives. Experts should account for their reasons and decisions to their peers and to the broader public. Provisions for deliberation and cognitive diversity are also necessary. Police investigators, airline safety experts, and central bankers should be oriented toward truth and precision, not toward maximizing the number of convictions, the profitability of regulated companies, or future employment opportunities. Note how epistemic integrity relies on independence not only from politics but also from undue industry influence.

The institutional design criterion seeks to ensure that the expertise in any given institution is as reliable as it can be. But there might also be variation in the baseline reliability of expertise across different areas. This is an assumption that runs through many empirical studies of expertise use, under labels like “uncertainty,” “tractability,” or “hardness” (see, for example, Fjørtoft and Michailidou, 2021; Radaelli, 1999b; Rimkutė, 2015; Schrefler, 2010). I am agnostic as to how reliability should be operationalized. Generally speaking, expertise is more reliable when it is more likely to generate precise epistemic evaluations. This will depend on the intrinsic complexity of the object under study as well as the state of the field of the expertise that studies it. Many analysts draw up a hard–soft continuum with the social sciences at one end and the natural sciences at another. In the social sciences, “predictions are less reliable, well-established findings are fewer, and concept formation and measurement are more difficult” (Pamuk, 2021, 20; see also Friedman, 2019). If this is true, it would count against the reliability of social science expertise. In practice, the reliability of expertise should be evaluated by more fine-grained criteria than a broad division between scientific disciplines, but the general idea that some areas of expertise might be more reliable than others holds.

If there is variation in the reliability of expertise across fields, there exists a threshold of reliability below which technical legitimacy is inappropriate. The threshold for what counts as reliable-enough expertise may be procedurally determined in society or spelled out by a separate normative theory; the details are outside the scope of this article. I will only point out that allowing for this kind of variation means that critics must either argue empirically that no existing expertise meets the robustness threshold or normatively that the threshold should be so high that no existing expertise rises above it.

A persuasive version of the argument that existing expertise is below the threshold for robustness has been made by Friedman (2019). His critique of positivist technocracy rests on the argument that all predictions of social-scientific regularities are bound to fail due to “ideational heterogeneity,” and that this fundamentally undermines the technocratic ideology. He establishes the fallibility of social predictive claims at an ontological level. This move seemingly makes it impossible to establish the robustness of different fields of expertise. I would not go so far—against Friedman’s stipulation I would simply stake my ontological claim that some things are more reliably knowable than others, and especially so when we move further away from the social sciences which are Friedman’s primary concern. The principled feasibility of technical legitimacy must also be kept analytically distinct from whatever epistemic technique is currently in vogue, be it “positivism, neoclassical economics, or behavioral economics” (Friedman, 2020: 252).

These concessions soften the blow of Friedman’s critique, preserving, at least in principle, some space for technical legitimacy. I agree with Friedman that knowledge claims are always fallible, even in instances where they do not fail right now (Friedman, 2020: 248). But, in close parallel to what Zeynep Pamuk says about science, we do not need to believe that expertise is infallible to make good use of it (Pamuk, 2021: 14). Likewise, the fallibility of expertise does not block the possibility of technical legitimacy altogether. We should not ground institutions’ legitimacy in an unrealistic notion of experts’ objectivity or authority (for one critique see Lofthouse and Schaefer, 2025). Instead, we should strive to cultivate social processes and institutions that maximize institutions’ epistemic flourishing while also dealing with the risk of epistemic error. In practice, epistemic flourishing could be cultivated through a combination of expert-internal norms and an ethos of truth-seeking; mechanisms for reviewing experts’ judgments in external fora (e.g. institutional peer review); and institutional designs that foster diversity and deliberation (see, for example, Holst and Molander, 2019). The risk of epistemic error, on the other hand, calls for direct venues for value input into non-majoritarian institutions.

Appropriate Venues for Value Input

A core component of existing defenses of technical legitimacy is that non-majoritarian institutions are value free in their technical conduct. The value free ideal for non-majoritarian institutions is most clearly formulated in political discourse; academic commentators tend to be more skeptical toward the idea. Still, most existing defenses of non-majoritarian institutions’ legitimacy fall back on some notion of value freedom (Fjørtoft, 2024). The division of labor between value-free expertise and value-laden politics has been described as a “standard view” on the role of experts in policy development (Elster, 2024: 12) and the “dominant twentieth-century solution to the problem of expertise” (Pamuk, 2021: 7). The standard view is partly motivated by an epistemic concern to safeguard science against undue value influence, and partly by a concern to safeguard democracy against the encroachment of scientific and technical expertise (Pamuk, 2021: 7–8).

There is a close parallel between science and non-majoritarian institutions when it comes to value freedom. Both institutions are by design set up to respond to epistemic concerns. Non-majoritarian institutions sometimes carry out scientific research themselves, they often produce knowledge or make epistemic claims, and they often base their decisions on (and evaluate the credence of) existing research. This means that any arguments against the value-free ideal in science will transfer to non-majoritarian institutions as well (for the debate in philosophy of science, see Douglas, 2009; Hempel, 1965; Kitcher, 2001; Rudner, 1953).

A problem with the value-free ideal (and the corresponding standard view of expertise) is that it rests on an idealized view of knowledge. As Pamuk argues, scientific claims are uncertain, incomplete and fallible. Under uncertainty and incompleteness, “certain kinds of judgment that the division of labor model relegates to the expert domain are in fact value-laden ones” (Pamuk, 2021: 11). Choice under uncertainty always requires moral judgment. It is necessary to decide which errors one is more willing to accept or avoid, what amount of risk one is willing to accept, and so on. Scientists, when making such judgments, “move beyond what is justified by appeal to their superior knowledge.” And of course, an appeal to superior knowledge is exactly what technical legitimacy rests on.

How do values enter the internal stages of science? One mechanism is significance (Kitcher, 2001; Pamuk, 2021). Scientific discovery is not a random accumulation of truths; scientists aim to generate significant knowledge. And judgments of significance cannot be made based on data and evidence alone. They are determined with reference to a value scheme that is external to the epistemic process itself. Douglas (2009) makes a parallel but narrower argument building on the concept of inductive risk (see Fjørtoft, 2024 for an application of inductive risk to non-majoritarian institutions): Scientists unavoidably make value-informed decisions about which errors they are more willing to accept when, for instance, deciding on a threshold for hypothesis acceptance.

Note that the above arguments against the value-free ideal in science do not warrant radical skepticism toward the reliability of scientific or other epistemic claims nor an anything-goes attitude toward value involvement in non-majoritarian institutions. Just as in science, there are clearly objectionable roles that values could play in the workings of non-majoritarian institutions, and we should filter those out (Pamuk, 2021: 30). We cannot, to put it bluntly, accept an empirical claim only because we want it to be true (Douglas, 2021: 29).

The specification of mechanisms for value input will depend both on the function and expertise basis of the institution. There is a greater need for value inputs in institutions that deal with knowledge claims which are more uncertain or underdetermined. There is also a greater need for value inputs into institutions with more impact on the world—whether it is because they have extensive political powers, capacity for action, or operate in areas where errors have catastrophic consequences.

It is beyond the scope of this article to suggest in any detail concrete institutional mechanisms—but some general outlines are in order. The challenge for institutional design is to create mechanisms that make sure independent institutions are attuned to the appropriate societal values in their daily work without threatening the rationale for delegation in the first place. This again rules out the most obvious venue for value input: majoritarian democratic control. Procedures for direct layperson participation, advisory councils, and public hearings are possible alternatives. Many existing participatory arrangements are, however, demanding on people’s time and resources, are prone to capture by powerful interests, and might introduce power asymmetries between experts and lay citizens (Elliott, 2023; Moore, 2017; Pamuk, 2021). It might often be more fruitful, therefore, to rely on mechanisms that make sure the experts working in non-majoritarian institutions are themselves attuned to the values of the society they operate in. Schroeder (2021) suggests that democratic values should guide scientists’ value-judgments, which are arrived at through a process of which combines empirical inputs and moral reasoning. Pamuk (2021) suggests an adversarial science court which evaluates facts and values together, in a system of competing experts making their case in front of a citizen jury. Eriksen (2021) suggests a public reason model where experts ground their judgments in their institution’s mandate. Bagg (2021) suggests a combination of institutionally oppositional expertise, randomly selected citizen oversight juries, and direct leverage points for popular mobilization outside of the electoral cycle. What such suggestions have in common is that they are mechanisms to make the institutions and epistemic processes themselves attuned to societal values. They do not fall back on an untenable division-of-labor model nor direct majoritarian control.

In summary, technical legitimacy requires us—paradoxically—to abandon a strict value free ideal. This neither warrants an anything-goes approach to value input, direct democratic control over epistemic decisions, nor radical skepticism toward science and expertise. But it does require institutions to have mechanisms that make them attuned to the democratic values of the society they are part of, even in their inner technical workings.

No Common Good: The Case of Post-crisis Central Banking

Recall that the common good criterion requires that delegation serves an adequately narrow and specified common good for which there exists both baseline majoritarian authorization and a theoretical argument for why the good in question is in fact common and not just serving factional interests. Delegation to independent central banks is classically taken to fulfill this criterion. There has for the past several decades been a broad political consensus around delegation to independent central banks in countries across the world, backed by a macroeconomic argument for why independence serves a goal—credible commitments to an inflation target—which is unattainable if monetary policy were under direct political control. For the sake of this example, I stipulate that central banks are in fact governed by sound expertise and that their claim to value freedom is appropriate. I will instead take aim at the criterion that central banks should serve a clearly delineated and appropriately justified common good.

Jens van ’t Klooster (2020) argues that central bank independence is less obviously legitimate now than before the 2008 financial crisis. “Before 2008, central banks had clearly defined goals, which they could pursue by means of one instrument” (p. 596), but after the financial crisis, central banks “now have many more instruments to use in pursuit of a much less clearly defined set of goals” (van ’t Klooster 2020: 596). In addition to seeking price stability by controlling interest rates, central banks after the crisis have acted as lenders of last resort, as creditors to governments, and enacted policies of quantitative easing—the purchase of financial assets to counter low inflation (van ’t Klooster 2020: 594).

This shift toward a broader set of instruments and goals is a problem, according to the common good criterion, for two reasons. First, it makes it more difficult to identify when delegated authority oversteps the good it is set to provide. When central banks pursue a broad range of goals, they must inevitably make more trade-offs—which are today not adequately specified in most central banks’ existing mandates. Central bankers have defended their new activities by arguing that they fall under the price stability mandate, but this type of very broad interpretation of a mandate is problematic in my framework. Second, it is harder to make the theoretical argument that a new instrument like quantitative easing is a genuine common good than it is for price stability. Quantitative easing increases, for instance, the value of financial assets, which are disproportionately owned by the wealthiest 5 percent of households (van ’t Klooster, 2020: 594). Note that my (and van ‘t Klooster’s) argument is not that the underlying common good itself changed during the financial crisis. Central banks still appealed to price stability and couched their new conduct in old arguments. My point is rather that their claim to the same old common good of price stability became strained as they moved into new instruments and goals. An institution taking on new tasks, instruments and goals, especially when such a move challenges a substantial argument for why the institution serves a common good, should trigger a democratic reassessment of that institution’s terms of delegation. As discussed above, I do not believe that oversight mechanisms should be activated too easily, as this would undercut the rationale for delegation. But the development discussed here seems to rise above the threshold where a reassessment by elected politicians would be warranted.

Vulnerable Expertise: The Case of the FAA

Recall that there are two ways the expertise criterion may be violated. First, some areas of expertise may be inherently too uncertain to meet the reliability threshold necessary for technical legitimacy. Second, an agency may fail to provide the institutional structure necessary for epistemic incentives to guide the conduct of experts even where the underlying expertise is robust. I will discuss a case of the latter here. Aviation safety is a case of a relatively clear common good, building on a solid field of expertise, with plausibly only a limited need for value inputs. The expertise criterion can still be violated where the institutional structure introduces epistemic distortions.

The American aviation regulator, the Federal Aviation Administration (FAA), has been described as having effectively delegated the certification of new airplanes to the regulated companies themselves, through the practice of “self-regulation” (Schacter, 2021). This practice has come about for good reasons, as it is cost efficient and gives access to the comprehensive—perhaps even superior—expertise that industry employees hold. Self-regulation is widespread well beyond aviation (Schacter, 2021, 641). At the same time, it is easy to see how such a practice might lead the certification procedure to optimize for other things besides safety. Schacter (2021) argues that he FAA is forced even in its founding regulation to balance a concern for safety and a concern for the market (Schacter 2021: 649). In practice, this balancing exercise has often skewed toward the interests of the regulated industry. ⁴ For our purposes, it is important to note that the sources of capture are twofold. One is a close relationship between the regulator and the regulated industries, due to the well-known issues of “revolving doors” and so on. Another is Congress itself. Boeing enjoys significant lobbying power in Congress and has used this power to push for more self-regulation when the agency’s mandate has been renewed, even against FAA warnings that this could compromise safety (Kitroeff and Gelles, 2019; Robison, 2021: ch. 7). The expertise criterion requires protection against both kinds of influences.

Self-regulation has arguably contributed to disasters. Two Boeing 737 MAX 8 aircraft crashed within less than a year in 2018 and 2019, killing everyone on board. Investigators determined that a major cause of the crashes was a piece of software that had been developed to compensate for engineering deficiencies in the new aircraft—deficiencies which were introduced because Boeing wanted to keep the new aircraft as similar as possible to the old design, avoiding a lengthy and expensive certification process. According to the task force that investigated the crashes, the software that contributed to the accidents was analyzed almost exclusively by Boeing employees, with little oversight by FAA employees (Raso, 2019).

This case shows how a certain institutional structure effectively undercuts the expertise criterion of technical legitimacy. Experts employed by Boeing, even if they may very well be more competent than FAA employees, are exposed to incentives that might lead them to optimize for other things besides engineering quality and safety. ⁵ Reporters have documented the emergence of an organizational culture in Boeing which prioritizes the interests of shareholders over safety (Gates, 2024; see also Robison, 2021). The FAA was structurally unable to provide a corrective to such developments, making it a poor candidate for technical legitimacy. Again, this is the case even though the agency is set up in service of an uncontroversial common good (aviation safety), based on a field of expertise with a very strong empirical track record, and with a limited need for value inputs on a day-to-day basis.

No Value Inputs: The Case of Medicines Licensing

Medical licensing is a typical task of regulatory agencies around the world. Independent agencies evaluate whether a new drug is safe and effective enough to be granted market access through a technical-scientific evaluation of the tests and clinical trials that pharmaceutical companies must conduct to apply for authorization (Hauray and Urfalino, 2009). This seems like a good candidate for technical legitimacy. Safe drugs are plausibly a common good, also when balancing against issues of cost, as both consumers and producers benefit from a well-regulated market (Gehring and Krapohl, 2007, 211). Drug evaluations are based on “gold standard” randomized controlled trials in a highly developed scientific field. Accordingly, one might also believe that medical licensing is best seen as a value-free technical exercise where value inputs would be inappropriate.

This is in fact also a tendency. In the EU, the regime for medicines licensing emphasizes that decisions about marketing authorization should rest only on scientific evaluations of randomized controlled trials in expert committees that are independent from member states and the commission (Gehring and Krapohl, 2007; Hauray and Urfalino, 2009). A 1965 directive explicitly rejects the consideration of broader socioeconomic criteria like price and public health considerations in medicines licensing (Hauray and Urfalino, 2009: 443). Market authorization was carved out as a neutral and value-free domain in part because this was the only way to achieve European agreement on pharmaceuticals regulation—leaving the more contentious issues of public health and pricing to individual member states (Fjørtoft and Sandven, 2023; Pamuk, 2021: 76). As Pamuk (2021) points out, this retreat into value freedom is problematic because it “risks masking the ways advice is in fact nonneutral” (Pamuk, 2021: 76; see also Onoda, 2024). One thing is that experts are banned from taking broader societal considerations into account in their evaluations. This could plausibly be fixed by allowing politicians to have the final say over market authorization. A more fundamental challenge is that the scientific evaluations themselves unavoidably entail value choices due to the existence of inductive risk, as outlined earlier. When determining whether a drug is safe and effective, decisions over what type of error one is more willing to accept are unavoidable.

Medicine approval is not like aviation safety where more caution always saves more lives. If we are too slow bringing a new treatment to market (because of an overly slow and cautious approval process), lives could be lost. And, of course, the opposite is also true. Dangerous medicines that are allowed to market can do much more harm than good. Questions of the appropriate weighing of the chance of false positives against false negatives are therefore salient. The insistence on strict value freedom for medicines licensing makes it difficult to be explicit about decisions of inductive risk. Instead of being open about such trade-offs, medical experts are forced to couch their arguments in neutral scientific terms whenever choices under uncertainty occur throughout the epistemic process.