2. Preference-based freedom rankings and adaptive preferences

Consider:

Is Bob just as free in Job 2 as he is in Job 1?Footnote ⁵ Most political philosophers would answer in the negative. There is a consensus that adapting one’s preferences to constraints (as Bob does in Job 2) does not induce a state of freedom (Carter Reference Carter2019).Footnote ⁶ The freedom ranking of the available alternatives in Job 1 and 2 should not change merely as a result of Bob’s preference adaptation to the constraints he faces in Job 2. Yet, as I will show below, this well-grounded intuition poses a problem for preference-based freedom rankings. They either cannot account for the fact that adapting one’s preferences should not change the freedom ranking of Bob’s available options in Jobs 1 and 2 or they can do so at the cost of quite demanding assumptions. Before delineating CAP in more detail, let me clarify what I mean by ‘all-things-considered’ and ‘adaptive’ preferences respectively.

With a definition of all-things-considered and adaptive preferences in hand, I can illustrate CAP. As I will elucidate, preference-based freedom rankings suffer from CAP if they assume that an available alternative increases the freedom that a set offers if this alternative is ranked at least as good as the best alternative(s) in the set along one’s all-things-considered preferences.Footnote ⁸ To display how this shared assumption makes preference-based rankings suffer from CAP, I draw on Puppe and Xu’s (Reference Puppe and Xu2010) freedom ranking and define it over Bob’s all-things-considered preferences. Since the ranking can be specified to correspond to a variety of preference-based freedom rankings in the literature, employing it comes with no loss of generality.Footnote ⁹ The ranking is thus well-suited to show how CAP affects all preference-based freedom rankings that use a person’s all-things-considered preferences. Let me introduce the notation necessary to understand the ranking.

I denote by X the finite universal set of alternatives, here understood as alternatives that represent specific freedoms (to do something or be someone). The collection of all non-empty subsets of X is denoted by Z. Elements A, $B,\;$ etc. are sets (of mutually exclusive alternatives) that belong to Z. The notation $E\left( A \right) \subseteq A$ denotes a subset of A that includes all and only the so-called ‘essential’ elements. Essential elements are alternatives that strictly increase the freedom that the set A offers. The intuitive idea is that, if these alternatives were not in the set, the freedom of the set would decrease. The symbol $ \succcurlyeq $ denotes a freedom ranking (i.e. a binary relation that ranks elements in Z) and should be interpreted as ‘offers at least as much freedom as’. $ \succ $ is its asymmetric and $\sim $ its symmetric part. To explain CAP, I will define the essential elements in a set on the basis of a person’s all-things-considered preferences. Hence, $E\left( A \right),\;$ which denotes the essential elements in set A, will here designate the elements that are judged best in set A according to a person’s all-things-considered preferences on the elements in the set. I use the notation R to denote a binary relation on X that is complete, reflexive and transitive. xRy is interpreted as x being ‘at least as all-things-considered good as’ y. P is its asymmetric part and I is its symmetric one.

According to Puppe and Xu’s (Reference Puppe and Xu2010) freedom ranking, set A offers at least as much freedom as another set B if and only if A contains at least as many essential alternatives as the intersection between the union of A and B and B. I here define alternatives as ‘essential’ when they are a person’s favourite ones in the union of the two sets as judged by her all-things-considered preferences. So, how does Puppe and Xu’s freedom ranking, based on a person’s all-things-considered preferences, order Job 1 and Job 2?

Along Bob’s all-things-considered preferences in Job 1, ${C_1}$ offers him strictly more freedom than ${C_2}$ ( ${C_1} \succ {C_2}$ ). The reason is that, in Job 1, there is one ‘essential’ alternative available to Bob, as judged by his all-things-considered preferences (Doctor), while in Job 2 there are none. By simply adapting his preferences to constraints, Bob enjoys the same freedom in Job 1 and Job 2 ( ${C_1} \sim {\rm{\;}}{C_2}$ ). Along Bob’s adaptive preferences, both Job 1 and 2 contain one element that is best (Farmer). Yet, it seems to fly in the face of common sense that a person can be rendered freer only by adapting her preferences to the constraints she faces. A plausible freedom ranking should not change how Job 1 and Job 2 are ordered only as a result of a person’s preference adaptation. Hence, if, along Bob’s all-things-considered preferences in Job 1, ${C_1} \succ {C_2}$ , it should be the case that, even if he adapts his preferences to constraints, ${C_1} \succ {C_2}$ . But a freedom ranking that is based on Bob’s all-things-considered preferences cannot account for this. Bob’s case exemplifies a well-known problem: preference-based freedom rankings, defined over a person’s all-things-considered preferences, fall prey to CAP.

In response to CAP, scholars such as Jones and Sugden (Reference Jones and Sugden1982), Sugden (Reference Sugden1998, Reference Sugden2003), Pattanaik and Xu (Reference Pattanaik and Xu1998), Nehring and Puppe (Reference Nehring and Puppe1999) and Romero-Medina (Reference Romero-Medina2001) have suggested doing away with all-things-considered preferences as the informational basis of preference-based freedom rankings. They have proposed instead to base freedom rankings on a person’s reasonable preferences. Reasonable preferences are loosely understood as the preferences that it would be reasonable for a person to hold, or that a person could reasonably hold. Since adaptive preferences have been held unreasonable, freedom rankings based on reasonable preferences arguably escape CAP. Reasonable preferences have been defined either substantively (see e.g. Jones and Sugden Reference Jones and Sugden1982; Sugden Reference Sugden1998; Pattanaik and Xu Reference Pattanaik and Xu1998; Sugden Reference Sugden2003; Romero-Medina Reference Romero-Medina2001) or procedurally (see Binder Reference Binder2019) (for a complete overview, see Dowding and Van Hees (Reference Dowding, Van Hees, Anand, Pattanaik and Puppe2009) and Gravel (Reference Gravel, Peil and van Staveren2009)).

Under substantive understandings of reasonableness, a preference is reasonable if it is objectively or subjectively reasonable. It is objectively reasonable when it counts as such under an objective account of reasonableness. It is subjectively reasonable when it is judged reasonable in the light of the future preferences that an individual could hold, or the social group she belongs to or the preferences that a relevant percentage of people in society holds. Defining reasonable preferences substantively has been deemed problematic. However one specifies reasonableness, it is hard to come up with a definition that enjoys consensus (Van Hees and Wissenburg Reference Van Hees and Wissenburg1999; Bavetta and Guala Reference Bavetta and Guala2003; Bavetta and Peragine Reference Bavetta and Peragine2006; Binder Reference Binder2019). Binder (Reference Binder2019) has responded to the challenge by replacing substantive understandings of reasonableness with procedural ones. She proposed to determine what counts as a reasonable preference for an individual by looking at how her different values, or identities, rank the options. On her account, reasonable preferences are those that can be reasonably derived from a person’s identities or deeper values. She defines a person’s reasonable preferences in two steps. First, she delineates a ranking of options that orders an option x higher than another option y if, and only if, all the identities or deeper values that an individual has rank x as not lower than y. Then, she takes reasonable preferences to be the transitive completions of this ranking. Procedural definitions of reasonableness run into epistemic and empirical problems. They run into an epistemic challenge because they need to assume that a person not only has access to her deeper values or identities but also knows how each of these values or identities partially ranks the available options. And this is a demanding epistemic requirement. They are also vulnerable to an empirical challenge, since it is difficult, in practice, to elicit a person’s deeper values and identities.

To summarize, replacing all-things-considered preferences with reasonable ones as the informational basis for preference-based freedom rankings raises conceptual, epistemic and empirical challenges. On the one hand, substantive definitions of reasonableness are conceptually controversial. On the other hand, procedural definitions face epistemic and empirical challenges. Epistemic challenges, since a person may not have access to how each of her values and identities partially ranks the available options. Empirical challenges, since it is practically difficult for external observers to elicit such rankings.

As a result, preference-based approaches to freedom rankings face an impasse. If these rankings are based on a person’s all-things-considered preferences, they risk running into CAP. If they are based on a person’s reasonable preferences, they can escape CAP at the cost of conceptual, epistemic and empirical challenges.

Most scholars have attempted to resolve the impasse by concentrating on reasonable preferences. Here, instead, I will focus on all-things-considered preferences. I will argue that preference-based freedom rankings, defined on a person’s all-things-considered preferences, do not necessarily suffer from CAP. They do not if all-things-considered preferences are used as a source of information about the value that freedom has as a necessary condition for self-expression.

3. Freedom and its value for self-expression

I have claimed in Section 1 that accounting for the value that freedom has as a necessary condition for one’s self-expression allows freedom rankings to deal with CAP without needing reasonable preferences. In this section, I will explain what this value is and what increases it.

Freedom is considered valuable, among other things, since it is arguably a necessary condition to promote, recognize and exercise a person’s agency (Carter Reference Carter1995; Bavetta and Guala Reference Bavetta and Guala2003; Binder Reference Binder2019). Having freedom, it has been argued, enables one to say no to the options one wants to reject (Hurka Reference Hurka1987; Carter Reference Carter1995, Reference Carter1999). This ability, in turn, is often understood as a necessary prerequisite for a person’s agency (Hurka Reference Hurka1987; Carter Reference Carter1999). More specifically, as Scanlon (Reference Scanlon1998) has pointed out, the availability of alternatives that one can reject is a necessary condition to promote, recognize and exercise an instance of agency, self-expression.Footnote ¹¹ Self-expression amounts to the ability to express oneself (one’s cares, concerns and values) through the act of choosing. One can express one’s cares, concerns and values by choosing one’s job, or a present for one’s partner or the colour of one’s curtains. In all these instances, the act of choosing allows a person to express herself only if it is meaningful in some minimal sense. For instance, the choice of one’s job or a present for one’s partner is not self-expressive if it is made under coercion. As such, a necessary condition to choose in a self-expressive way is to have other choices available besides the chosen one, so that the person can reject alternative ways of expressing her cares and concerns. Let me call the value of freedom as a necessary condition for one’s self-expression the self-expressive value. Whenever I use the term self-expressive value of a set, I will refer to the value of a choice set for a person’s self-expression.

In this article, I will rely on two assumptions about what increases the self-expressive value of a set. First, I will assume that only alternatives that differ in some meaningful ways from those already available in a set increase its self-expressive value. Second, I will assume that meaningful difference between options cannot be captured without reference to an individual’s values. Let me justify the plausibility of these assumptions in turn.

Imagine that a person were to choose differently from how she does. These counterfactual choices would express a way of deliberating on, aggregating and resolving her cares and concerns that is different from her actual one. Unchosen options, therefore, can represent alternative ways of expressing one’s cares and concerns. But not all do. Indeed, some alternatives may be indistinguishable from what one chooses, others incomparable (I will expand on this in the next section). As a result, only alternatives that are different (in a way relevant to self-expression) from the chosen one in the set increase its self-expressive value. Given my definition of all-things-considered preferences, this implies that an alternative that is not different from the one that is judged all-things-considered best in the set does not increase self-expression.Footnote ¹² Consider Job 1 above. The availability of an unchosen career (Farmer) increases the self-expressive value of Bob’s set of jobs only if it provides him with an additional opportunity to express himself (his actual cares and concerns) by rejecting a career that conveys alternative cares and concerns. If Bob deemed the two careers (Farmer and Doctor) indistinguishable, it would be hard to see how the availability of a career in farming could increase the self-expressive value of Bob’s set of jobs. The reason is that choosing Doctor over Farmer would not allow Bob to better express the cares and concerns invested in being a Doctor by refusing those conveyed by being a Farmer.

Let me now justify my second assumption, which is that the difference between an alternative and one’s all-things-considered best in a set can only be defined with reference to a person’s preferences. As Sugden (Reference Sugden1998) has convincingly argued, whether two alternatives differ depends on what features of these alternatives are relevant to the context at hand. Here, I am interested in the difference between alternatives that is relevant to a person’s self-expression. Hence, whether two alternatives differ in a way that is relevant to a person’s self-expression crucially depends on the viewpoint of the person who is evaluating the options (Sugden Reference Sugden1998). If this is the case, Sugden (Reference Sugden1998) continues, whether two options differ can only be captured with reference to an individual’s values. To see this, consider the example of Bob again. Whether Farmer and Doctor differ in a way that is relevant to Bob’s self-expression depends on whether Bob deems them to be different. This hinges on what he considers relevant about the alternatives, which rests on his values. So, the possibility of rejecting Farmer in Job 1 increases the self-expressive value of the set only if Bob considers it different in a relevant way from Doctor, which ultimately depends on his values.

Importantly, making these two assumptions entails that self-expressive value cannot be measured in a value-neutral way. The reason is that these assumptions jointly entail that whether an alternative increases one’s self-expression depends on the values that a person holds. As a result, the value of a set of options for a person’s self-expression can only be measured with reference to a person’s values.Footnote ¹³

4. Capturing self-expressive value

How to judge whether the availability of an alternative increases the self-expressive value of a set? In this section, I show that little more than knowledge about a person’s all-things-considered preferences is needed. This lays the ground for the next section. There, I argue that it is unnecessary to delineate a reference set of reasonable preferences to overcome CAP. Freedom rankings based on all-things-considered preferences are immune to CAP as long as they account for self-expressive value.

5. Preference-based freedom rankings: a way out of the impasse

I claimed in Section 1 that it is unnecessary to rely on reasonable preferences to overcome CAP. I here argue for this claim. I show that freedom rankings based on a person’s all-things-considered preferences do not suffer from CAP if they account for self-expressive value.

Before turning to the remainder of the section, it is important to clarify what I mean by a preference-based freedom ranking that accounts for self-expressive value. I put forward a formal approach that is compatible with a class of families of value-based approaches to freedom (i.e. all preference-based approaches that measure the freedom that a person enjoys as partly dependent on self-expressive value and use one’s all-things-considered preferences to measure it).

In this section, I proceed in two stages. I first draw on the example at the beginning of the article and illustrate that the same freedom ranking (using the same source of information) can successfully deal with CAP if it accounts for self-expressive value. Hence, the exclusion of Bob’s adaptive preferences from the reference set of his preferences is unnecessary to overcome CAP. Then, I show that my argument holds water beyond this specific example.

Consider again:

CAP demands that the freedom ranking of the sets of alternatives in Job 1 and Job 2 does not change as a result of Bob’s preference adaptation. So, the freedom ranking should be ${C_1} \succ {C_2}$ both along Bob’s non-adaptive preferences and his adaptive ones.

To show why accounting for self-expressive value enables preference-based rankings to meet CAP, I will draw on Puppe and Xu’s (Reference Puppe and Xu2010) freedom ranking, with no loss of generality (see section 2). I define the set of essential alternatives as all the alternatives that contribute to Bob’s self-expression. I use Bob’s all-things-considered preferences to delineate them. As I will show, both when the set of essential alternatives is defined on information derived from Bob’s all-things-considered non-adaptive preferences (Job 1) and when it is defined on information from his adaptive preferences (Job 2): ${C_1} \succ {C_2}$ .

Let me first consider the case in which the set of essential elements is defined over Bob’s all-things-considered non-adaptive preferences and then that in which it is based on his adaptive ones.

It is self-evident that, if Puppe and Xu’s (Reference Puppe and Xu2010) freedom ranking accounts for self-expressive value on the basis of information derived from Bob’s non-adaptive preferences in Job 1, it ranks ${C_1}$ strictly higher than ${C_2}$ . ${C_1}$ contains two essential elements, while ${C_2}$ only one.Footnote ²¹

By accounting for self-expressive value, Puppe and Xu’s freedom ranking ranks ${C_1}$ strictly higher than ${C_2}$ also on the basis of information derived from Bob’s all-things-considered adaptive preferences. As I will illustrate, ${C_1}\;$ contains two essential elements, and ${C_2}$ only one, also along Bob’s adaptive preferences. This is because Farmer and Doctor can be compared, they are not indifferent to each other, and, I will show, Doctor is not dominatedly worse than Farmer according to Bob’s adaptive preferences.

Comparability and absence of indifference are trivially satisfied by Bob’s all-things-considered adaptive preferences. Doctor is all-things-considered worse than Farmer. The two alternatives are hence comparable (comparability) and not indifferent to each other (absence of indifference).

The satisfaction of absence of dominated worseness depends on whether, according to Bob’s adaptive preferences, Doctor is dominatedly worse than Farmer. Checking the satisfaction of this condition standardly requires knowing an individual’s cares and concerns (aggregated or unified to form the all-things-considered preference). In the specific case of adaptive preferences, however, this knowledge is unnecessary. In fact, as I will explain shortly, prominent understandings of adaptive preferences in the philosophical literature explicitly or implicitly grant the satisfaction of absence of dominated worseness. This means that an option that is downgraded as a result of one’s preference adaptation is not dominatedly worse than the adaptively all-things-considered best.

Let me elucidate this by considering the different understandings of adaptive preferences in three main literatures: those on (a) autonomy (Elster Reference Elster1983; Bovens Reference Bovens1992; Colburn Reference Colburn2011), (b) self-deception (Taylor Reference Taylor, Räikkä and Varelius2013) and (c) decision theory (Welsch Reference Welsch2005). Despite their differences (Khader Reference Khader2009; Dorsey Reference Dorsey2017), they all entail the satisfaction of absence of dominated worseness, implicitly or explicitly.

For Bovens (Reference Bovens1992: 73–74), adaptive preferences result from improper arbitration of all criterial judgments in favour of an alternative over another. So, the ranking of options according to each criterial judgement remains unchanged.Footnote ²² For Elster (Reference Elster1983), an individual who adapts her preferences would still choose her non-adaptively preferred option if this were available. This implies that the ranking of options according to her cares and concerns remains unchanged. Colburn (Reference Colburn2011) argues that people who adapt their preferences learn to change the aspects of the alternatives that they prioritize. So, again, the ranking of alternatives according to each aspect (equivalent to what I call ‘care and concern’) remains unchanged. Taylor (Reference Taylor, Räikkä and Varelius2013) defines adaptive preferences similarly to Elster (Reference Elster1983), but allows for the possibility that they result from a conscious process of self-deception (while for Elster adaptive preferences always form unconsciously). This implies that the ranking of alternatives along one’s cares and concerns has not changed. Welsch (Reference Welsch2005) models adaptive preferences as the result of a person changing the relative ‘emphasis’ that she attributes to the different goals she has. Just like Colburn (Reference Colburn2011), he assumes that the ranking of each alternative according to one’s cares and concerns does not change – only the weight attributed to each care and concern does.

Crucially, under all these understandings, Bob does not change the way in which the specific cares and concerns evaluate the options. According to Colburn and Welsch, he only changes how he deliberates about and trades off his cares and concerns. According to Bovens, Elster and Taylor, he does not change it all. For Bovens, he simply arbitrates improperly among his cares and concerns. For Taylor and Elster, he claims to prefer the adaptively preferred option to the non-adaptively preferred one, but in fact does not, since he would choose the latter if available. All understandings of adaptive preferences share the following feature: even along Bob’s adaptive preferences, some cares and concerns evaluate Doctor better than Farmer. Doctor is thus not ranked dominatedly worse than Farmer and absence of dominated worseness is satisfied.Footnote ²³

Bob’s all-things-considered adaptive preferences between Doctor and Farmer do not entail indifference and do not judge Doctor dominatedly worse than Farmer. If the essential alternatives in the union of ${C_1}$ and ${C_2}$ (intersected with ${C_1}$ and ${C_2}$ , respectively) are the alternatives that are comparable, not indifferent and not dominatedly worse than the best alternative in the set according to Bob’s all-things-considered adaptive preferences, then ${C_1} \succ {C_2}$ , according to Puppe and Xu’s (Reference Puppe and Xu2010) freedom ranking.Footnote ²⁴

As such, if the essential alternatives in Puppe and Xu’s (Reference Puppe and Xu2010) freedom ranking are identified as those that increase self-expressive value, the ranking stays ${C_1} \succ {C_2}$ , irrespective of whether the preferences used as an informational basis are adaptive.

My argument holds water beyond the specific example of Bob. Freedom rankings that are based on a person’s all-things-considered preferences and account for self-expressive value are immune to CAP. Even if these rankings are based on a person’s all-things-considered adaptive preferences, they rank the set that contains one’s non-adaptively favourite option(s) as offering more freedom than that without it (them). The reason is that, in my framework, an alternative increases the self-expressive value of a set if the relationship between that alternative and the best in the set satisfies the three conditions above. And adaptive preferences (between one’s non-adaptively preferred option and one’s adaptively preferred option) satisfy these three conditions. So, preference-based freedom rankings defined on all-things-considered preferences can deal with CAP without needing to individuate a reference set of reasonable preferences. There is thus a way to escape the impasse that either preference-based freedom rankings cannot escape CAP or they can at the cost of demanding conceptual, epistemic and empirical assumptions.

Let me close by discussing the reach of my proposal.

As I have clarified in the sections above, delineating the self-expressive value of a choice set standardly requires more knowledge than that of a person’s all-things-considered preferences (except for the case of adaptive preferences). While this illuminates that knowledge of one’s all-things-considered preferences is not sufficient to develop a preference-based freedom ranking that accounts for self-expressive value, it is important to stress two things. First, the knowledge required is much more limited than that needed when delineating reasonable preferences (see section 2). Second, the fact that additional limited knowledge is required does not undermine the claim that all-things-considered preferences can be used as an informational basis for preference-based freedom rankings in a much more uncontroversial way than usually assumed.

Another remark concerns phenomena of higher-order adaptation. These occur when a person deems an (unavailable) option dominatedly worse given how society has shaped the cares and concerns she holds. The availability of options that are dominatedly worse due to higher-order adaptation may contribute to a person’s freedom, and yet not increase a person’s self-expression.Footnote ²⁵ While accounting for self-expressive value provides preference-based freedom rankings with a way to escape CAP, it does not allow preference-based freedom rankings to elude another challenge, which is that of higher-order adaptation. Cases of higher-order adaptation are best understood as a distinct phenomenon from that of adaptive preferences. Theorists working on the topic (Khader Reference Khader2009; Christman Reference Christman, Veltman and Piper2014; Stoljar Reference Stoljar, Veltman and Piper2014) define them as a result of one’s adaptation to (or coping mechanism with) one’s oppressive environment. They usually understand this phenomenon as the non-autonomous adaptation of one’s desires to those oppressive conditions.Footnote ²⁶ They thus describe a phenomenon that is distinct and occurs on a different level. For this reason, I take it that phenomena of higher-order adaptation do not highlight a limitation to the framework I have developed, which addresses CAP. These phenomena however cast light on other challenges that preference-based freedom rankings can be vulnerable to.