The social science literature classifies discrimination as either taste-based (Becker, 1957) or belief-based, with either accurate beliefs (Phelps, 1972; Arrow, 1973) or inaccurate beliefs (Bohren et al., 2019a; Bohren et al., 2019b). This implies, that individuals either discriminate against minorities because of animus, or because of rational behavior based on (in)accurately perceived group differences. In this study, I want to dig deeper into the formation of these inaccurate beliefs and show that taste can play a role here, too.
Ultimately, I aim to introduce a fourth kind of discrimination - belief-based discrimination based on motivated reasoning (“motivated statistical discrimination”) - and show why it is important to differentiate between this type of discrimination and the well-established types above. In motivated statistical discrimination, individuals not only incorrectly perceive group differences, but they also have a motive to uphold or even strengthen these incorrect beliefs when confronted with challenging information. This does not mean that discrimination always remains unaffected by new information, but that the effect of information is ambiguous and that new information should be presented without the possibility for strategic interpretation.
This distinguishes motivated statistical discrimination from taste-based discrimination, in which any information treatment remains ineffective. It also distinguishes motivated statistical discrimination from the classical kinds of belief-based discrimination in which beliefs are either correct or in which information generally corrects beliefs as no motives prevent individuals from updating their beliefs. The finding that motives might drive rationalizing belief inaccuracies can have important effects for policy interventions, especially for information interventions against unfair treatments of minority groups.
I would like to show that (a) motivated statistical discrimination exists, (b) individuals strategically search for information in order to uphold (or strengthen) discrimination rationalizing beliefs, (c) correcting beliefs is not sufficient, as motives can still operate through biased information processing, and (d) limiting the scope for motivated interpretation of information can decrease discrimination based on motivated reasoning.
I set up an artificial hiring situation, in which ‘workers’ are presented to ‘employers’ who can decide whether or not they want to ‘hire’ a worker. Workers are subjects of a survey prior to the actual experiment. In this survey, they complete a productivity measure (assessment test) which is the basis on which the employers are paid if they hire a particular worker. In the actual experiment, the employers are allocated into one of six different treatments.
Does motivated discrimination exist and do individuals strategically acquire information?
Control : Employers are repeatedly presented with the profiles of two workers and are asked to hire one of these two. Additionally to some basic information about the two workers, they get the costless opportunity to acquire information signals about the productivity of each individual worker (e.g. ACT/SAT scores, high school grades, etc.) on a separate website.
Neutral : This is the same as the " Control " treatment, except that for employers in this treatment the part of the basic information that identifies the worker as a minority or majority worker is hidden. Instead workers’ group identities are framed neutrally (such as “circle worker” or “square worker”). In order to fix prior beliefs about the productivity of each worker, the employers are presented with previously elicited prior beliefs about the average productivity of the group to which the worker belongs (framed neutrally).
Does group level information alone not suffice to fight discrimination?
Control with corrected beliefs : This is the same as the " Control " treatment, except that employers in this treatment are given the true average productivity levels for each minority and majority group of workers. Employers still have the opportunity to acquire information signals about each individual worker’s productivity.
Neutral with corrected beliefs : This is the same as the " Control with corrected beliefs " treatment, except that in this treatment the part of the basic information that identifies the worker as a minority or majority worker is hidden.
Does group level information and limiting the scope for motivated information acquisition help?
Control with corrected beliefs and no information : This is the same as the " Control with corrected beliefs " treatment, except that employers in this treatment are not given the opportunity to gather information signals about each individual worker’s productivity on a separate website.
Neutral with corrected beliefs and no information : This is the same as the " Control with corrected beliefs and no information " treatment, except that in this treatment the part of the basic information that identifies the worker as a minority or majority worker is hidden.
I first set up a worker pool consisting of 200 workers to achieve sufficient group heterogeneity.
Additionally, as indicated by a power analysis (code available upon request) based on results of earlier studies (Bohren et al., 2019b; Chen & Heese, 2021), I expect to require approximately 130 participants for each of the six treatments. Together with the worker pool, this amounts to a total of about 980 participants.
The worker pool survey as well as the actual experiment are pre-tested and estimated to last around 55 minutes. I expect to pay a participation fee of 6.66£ (hourly rate of 7.50£). This amounts to a payment of 6.66£ * 980 = 6526.80£.
Additionally, I aim to pay incentives (for the workers to exhibit effort in the assessment test and for employers to hire workers according to utility maximizing principles) to 10% of all participants. These incentives amount to 10£ on average, leading to a total incentive payment of 98 * 10£ = 980£.
Including the Service Fee of 33% (2477.24£) and the VAT of 20% of the service fee (495.45£), I estimate the costs of this study to amount to 10479.49£.
I therefore apply for funding of up to 10.000£, but I would be happy about any share of the full amount.
Available on as_predicted: https://aspredicted.org/ry8hv.pdf
I aim to publish the final paper about this study in an open-access peer-reviewed general-interest journal. I additionally provide access to the experiment code, analysis code and data through my website and upon request.
Arrow, K. J. (1973): “The Theory of Discrimination,” in Discrimination in Labor Markets,
ed. by O. Ashenfelter and A. Rees, Princeton, NJ: Princeton University Press.
Becker, G. (1957): The Economics of Discrimination, Chicago: University of Chicago
Bohren, J. A., Imas, A., & Rosenberg, M. (2019a). The dynamics of discrimination: Theory and evidence. American economic review , 109 (10), 3395-3436.
Bohren, J. A., Haggag, K., Imas, A., & Pope, D. G. (2019b). Inaccurate statistical discrimination (No. w25935). National Bureau of Economic Research.
Chen, S., & Heese, C. (2020). Motivated Information Acquisition in Social Decisions (No. crctr224_2020_223v1). University of Bonn and University of Mannheim, Germany.
Phelps, E. S. (1972): “The Statistical Theory of Racism and Sexism,” American Economic
Review, 62, 659–661.