There exists a tension between the competing principles of freedom of association (FoA) — the liberty to associate or dissociate with other people (Freedom of Association 2019) — and freedom from discrimination (FfD) — the liberty to not be (directly) discriminated against on the basis of race or gender (Discrimination 2020). This manifests itself in part in the economy, wherein businesses may wish to disassociate with people of a particular race or gender, a form of ‘direct’ discrimination (Becker 1971; Epstein 1992; Besson 1999). Rendering such practices illegal spurs a vast debate of perfectionism and state neutrality; however, insofar as there exists a solution that satisfies ‘principled compatibility’ in which both freedoms are fully attained, then the debate on the role of the state may be avoided in this domain. One key idea is that of Becker (1971): competitive markets with no association constraints will drive out discriminatory businesses, leading to freedom from discrimination. This paper will first outline ‘Beckerian discrimination’ and then argue that Becker’s ‘extended’ argument fails to prove principled compatibility: it favours freedom of association both de jure and de facto. Lastly, the paper addresses a forceful objection: it is rational for people to accept the model due to the ex-post de facto reduction in discrimination. I argue that such a reduction requires forceful compliance, enforcing the de jure prioritisation of freedom of association.

Becker did not aim to solve a principled debate: he studied the causes of the racial wage gap in his ‘taste-based’ model (1971). This essay will consider an ‘extended’ version of the argument, which argues that his model’s predictions can resolve the principled tension (henceforth, ‘Becker’s extended argument’) (Besson 1999). The core of the extended argument is this: insofar as employer discrimination is crowded out with FoA, the two freedoms are attained. Employer discrimination will be crowded out if such discriminatory businesses are outcompeted. This occurs if there is (1) free market entry, (2) perfect information and (3) a sufficient number of non-discriminatory employers (Becker 1971). The reason is that employers’ “taste for discrimination” leads to an arbitrage opportunity for non-discriminatory businesses, diverting profit from the former to the latter (Becker 1971: 14). This results as the discriminatory businesses only want to hire ‘unpreferred’ groups at a lower wage than other workers (for identical productivity), making it relatively cheaper for non-discriminatory businesses to hire such groups. In the same way as discriminatory hiring decisions limit the labour pool, creating inefficiency, limiting the pool of customers will lower profitability: for instance, in a perfectly competitive market, firms will exit in the long run, if they cannot cover their fixed costs.

Arrow argued that Becker’s model “predicts the absence of the phenomenon it was designed to explain” (Williams 2015: 7). Instead, Arrow posited that discrimination could exist in equilibrium with imperfect information, thus proposing a second model: ‘statistical discrimination’. None, however, have attempted to answer the following question: by fully accepting Becker’s model’s assumptions and predictions, will it attain principled compatibility? By assessing compatibility through a de jure and de facto framework, the extended argument fails on both accounts.

Firstly, the proposal makes FoA a principled priority because respecting a freedom de jure, either by its inclusion or exclusion, must necessarily appeal to a set of principles for its justification. The burden of proof is the following: does a de jure prioritisation necessarily imply a principled prioritisation?

Enshrining something into law implies a principled choice on the part of the authority that instated said law. This is because when making such a decision, the law-governing authority has deemed the law worthy of its implementation based on some metric. The decision metric is fundamentally principled in that whether the law seeks to (a) ‘practically’ change the state of the world (e.g. bettering the economy or pursuing that which is politically advantageous by, say, suppressing a minority) or (b) change the ‘principled’ character of agents within the nation (e.g. instil a sense of national identity through a distorted or prioritised history syllabus in schools), such law must necessarily invoke a set of values to motivate its existence. Thus, a de jure choice is a principled choice. Consequently, a decision not to enshrine something into law must be a principled choice, too, as it fails to satisfy one or several elements in the set of values. Note that even if there is a de facto equality of freedoms, the conclusion still holds if there is de jure inequality.

Secondly, Becker’s extended argument makes FoA a de facto priority as well. This is because it asks agents facing discrimination to tolerate it in the short-term (e.g. being restricted from visiting a restaurant) to gain the freedom from it in equilibrium. If individuals are not willing to do this, then its continued existence combined with a full appreciation of FoA implies a de facto prioritisation of FoA to FfD. Thus, the burden of this argument is to show (a) why at least one agent will not possess such tolerance and (b) why the remaining agents cannot remedy this bias.

Assume that the person facing discrimination strictly prefers FfD to FoA. While this need not be the case, it is understandable why it might be: discrimination based on your race and gender violates the dignity of your person, causing great emotional distress (Brooks 1999). As such, between the choice of the restaurant’s freedom to deny you service and the freedom from being denied, the latter is assumed to be strictly preferred. Note that I exclude all those deontologically motivated who would not tolerate discrimination, i.e. oppose discrimination today because it disrespects people qua people; this is to assume the best of Becker’s extended argument as previously outlined.

Given this preference, people will only tolerate short-term discrimination if they believe it will reduce their life-time discrimination. For some, this may not be the case: if people sufficiently discount the future, i.e. are sufficiently impatient, then the elusive future of a non-discriminatory market equilibrium carries no weight in their discrimination minimisation problem today. This occurs when the discounting factor tends to zero at, or before, the point at which the market attains equilibrium.

There is a wealth of literature that shows that people do indeed significantly discount the future, such as hyperbolic discounting (Laibson 1997). However, discounting future monetary valuations or utility is arguably different than that of discrimination; nevertheless, if we assume that discrimination eliminates the optimal consumption bundles (e.g. by being denied from a restaurant), then the absence of discrimination may be measured in added utility. As such, discrimination may, too, be indirectly subject to discounting. Note that this utility effect is in addition to the pain associated with the violation of one’s dignity.

In practice, discounting implies that the pain of being denied service at a restaurant because of your race or gender today is perceived as being greater than that you would face in a decade. As such, these individuals do not experience a reduction in discrimination when sufficiently discounting and will therefore not tolerate it today. Given their strict preference for FfD — a freedom they lack and do not experience a reduction in — this group experiences a de facto prioritisation of FoA to FfD, i.e. a ‘negative FfD bias’. This means that FoA, which is enjoyed by all, outweighs FfD, which is only sufficiently enjoyed by some of the discriminated group.

Even if one accepts the assumption that principled compatibility can still be attained if the negative FfD bias is cancelled out by a positive bias among other groups, this cannot remedy the extended argument either. Notice that individuals cannot experience a reverse prioritisation of FfD over FoA even with a discount rate below that which yields the negative FfD bias. With a discount rate of zero, FoA is already fully respected (de facto and de jure) throughout time, whereas, at its best, FfD may be sufficiently respected (de facto).

Considering the non-discriminated group, the model does not require them to over- or undervalue any of the freedoms: they face no trade-off of today versus equilibrium conditions akin to that of the discriminated group, sacrificing today’s pain for its absence in equilibrium. This can be shown by further breaking this group down into discriminating and non-discriminating. If people do not discriminate, there is no trade-off: they are free not to discriminate today and in equilibrium. Note that while this group will profit from the discriminatory businesses today and not in equilibrium, it is not a trade-off they face: it results from the choices of the discriminatory businesses. For the discriminatory businesses, they do not face any trade-off either as they have full FoA today and in equilibrium. In fact, it is by exercising this right that they exit the market in equilibrium. Although they face a trade-off of whether to discriminate, they do not face one of whether they can. As such, by aggregating over all groups, the extended argument favours FoA de facto.

In light of this argument, I anticipate the following objection: although people may not experience a reduction in discrimination ex-ante (an ‘irrational’ behaviour), they nevertheless will ex-post, assuming the counterfactual is the off-equilibrium path of discrimination. If this is true, then there exists no FfD deficiency ex-post, and this paper’s argument is constrained to the experience of discrimination, i.e. how an individual perceives the current and future levels of discrimination, thus triggering an immediate emotional response. The question is whether this is sufficient to disprove principled compatibility.

Even if one concedes the premise that the ex-post view is appropriate to assess the de facto criteria, the experience of current and future discrimination will exacerbate the de jure prioritisation by enforcing it. With a sufficiently high discount rate, an individual experiences future discrimination less potently today and is therefore not willing to tolerate short-term discrimination. If these experiences — rational or not — map onto a choice outside the set necessary for tolerance, individuals may wish to exercise their intolerance. If so, then the enforcement of the law is necessary to ensure compliance. For instance, suppose that the owner of a restaurant asks a person of a particular race or gender to leave their restaurant. While they may refuse to, they are unable to exercise it, i.e. state officials may remove them from the restaurant. Notice how this differs from the first substantive argument; the first argument posited that enshrining something in law implied a principled prioritisation, but it did not depend on enforcing said law.

Becker’s extended model, therefore, requires the enforcement of that which it aims to attain the absence of: discrimination. In all time periods prior to equilibrium, this will enforce the de jure prioritisation of FoA, forcing compliance among intolerant people. In equilibrium, there is no prioritisation in either direction: both FoA and FfD are fully observed. And so, even if one concedes the ex-post premise, they must accept an enforced de jure prioritisation, thus failing principled compatibility.

In conclusion, this paper has argued that Becker’s extended argument fails to prove principled compatibility: it favours freedom of association both de jure and de facto. The paper addressed the objection that individuals will, nevertheless, observe an ex-post de facto reduction in discrimination to which I argued that this fails as it implies a worsened, enforced de jure prioritisation. In hope of obtaining principled compatibility, further work may focus on similar analyses for alternative, advanced models, such as Arrow’s ‘statistical discrimination’. If this fails, the debate of whose right to uphold — the discriminatory restaurateur or the customer — seems inevitable.


Becker, G. 1971. The Economics of Discrimination. 2nd ed. Chicago: The University of Chicago Press.

Besson, S. 1999. Discrimination and Freedom of Contract. International Journal of Discrimination and the Law, [online] 3(4), pp.269-297. Available at: [Accessed 5 March 2021].

Brooks, R. 1999. Dignity and Discrimination: Toward A Pluralistic Understanding of Workplace Harassment. Georgetown University Law Center. Available at: [Accessed 8 May 2021].

Epstein, R. 1995. Forbidden Grounds: The Case against Employment Discrimination Laws. Cambridge, Mass.: Harvard University Press.

Laibson, D. 1997. Golden Eggs and Hyperbolic Discounting. The Quarterly Journal of Economics, [online] 112(2), pp.443-478. Available at: [Accessed 7 March 2021]. 2019. Freedom of Association. [online] Available at: [Accessed 5 March 2021]. 2020. Discrimination. [online] Available at: [Accessed 5 March 2021].

Williams, H. 2015. Discrimination: Theory. Massachusetts Institute of Technology.