In the era of digital marketplaces and platform-based services, customer feedback systems have become a cornerstone for evaluating worker performance and shaping economic outcomes. However, new research from the University of Toronto’s Rotman School of Management reveals a concerning undercurrent within these ostensibly objective rating systems: subtle racial biases that significantly affect income inequality among service providers. The study, led by Professor Katherine DeCelles and published in Nature, delves into how multi-point rating scales can unintentionally perpetuate racial disparities, offering groundbreaking insights into the mechanisms behind these biases and potential solutions to mitigate them.
The research team analyzed data from an online home maintenance matching service that connects customers with contractors such as plumbers and electricians. Although the platform uses a five-point rating system to evaluate worker performance, the investigation uncovered that white workers consistently received higher average ratings than their non-white counterparts. Specifically, white workers secured top ratings 86.9% of the time, compared to 83.4% for workers identifying as non-white. While the numerical gap may seem marginal, its economic repercussions are profound given the platform’s revenue-sharing models that tie income directly to these ratings.
In platforms where compensation is linked to performance evaluations, even slight depressions in rating scores translate into significant income disparities. The researchers estimate that under the five-point rating system, non-white contractors earned only 91 cents for every dollar made by white workers. Such inequalities, embedded in the feedback structure, exacerbate existing societal inequities and spotlight the insidious role that subtle biases can play in ostensibly meritocratic systems. These biases are not overt acts of discrimination but often unconscious prejudices manifesting in ambiguous evaluative criteria.
Professor DeCelles emphasizes that the crux of the problem lies in the structural design of rating systems rather than individual intent. Unlike clear-cut forms of racism, where preferences are overt, these subtler biases operate beneath the surface, influencing judgments in ways the evaluator may be unaware of. For example, when users assess how good a worker was on a nuanced scale, their internal biases can skew perceptions, leading to systematic undervaluation of non-white workers’ performance despite equivalent job quality.
The study leverages a natural experiment enabled by the platform’s transition from a five-point rating scale to a simplified binary choice system: users were only asked whether they would hire the contractor again, effectively a thumbs-up or thumbs-down scenario. Remarkably, this change almost eliminated racial differences in top ratings. New workers who joined the platform after the switch experienced earnings parity regardless of race, highlighting how reducing the complexity of judgment parameters can neutralize bias.
Supplementing the real-world data, the team conducted controlled experiments with online participants who assessed artificial scenarios depicting contractor performance. These experiments corroborated the initial findings, demonstrating that two-point rating scales reduce racial disparities in evaluations even among individuals harboring subtle racial biases. Participants also self-reported that their evaluations felt less influenced by personal biases when pressed to give simple positive or negative feedback, versus discriminating on gradations of quality.
The cognitive mechanisms underlying these findings draw from psychological perspectives on decision-making. Ambiguous rating scales require evaluators to engage in subjective comparisons of “how good” a service was, creating fertile ground for biases to color interpretation. Conversely, dichotomous scales simplify decisions to binary judgments that are cognitively less demanding and more behaviorally anchored, thus limiting opportunities for implicit biases to distort evaluations.
These insights have profound implications for platform design and organizational policy. As gig economy platforms proliferate, the reliance on customer ratings to allocate work and income amplifies the need to ensure fairness and equity. Simplifying rating scales to binary measures not only curtails bias but also enhances clarity and transparency. Beyond scale design, regular audits of rating data and decoupling detailed customer feedback from compensation metrics can offer additional layers of protection against systematic discrimination.
Professor DeCelles and her colleagues advocate for an intentional reimagination of evaluation architectures on digital platforms. Since changing customer attitudes wholesale is a slow and fraught process, redesigning structural aspects of feedback systems presents a pragmatic pathway to mitigate inequality. Organizations can implement straightforward binary feedback mechanisms, conduct ongoing bias diagnostics, and create avenues for nuanced qualitative comments to complement basic ratings without affecting pay.
The research, co-authored by PhD candidate Demetrius Humes, Tristan Botelho from Yale University, and Sora Jun from Rice University, underscores the intersection of behavioral economics, social psychology, and organizational behavior. It pioneers an empirically driven approach to racial equity scholarship by linking micro-level cognitive biases in evaluations directly to macroeconomic outcomes like income disparity, thereby filling critical gaps in our understanding of discrimination dynamics in the digital age.
This study opens new vistas for policy makers, platform architects, and social scientists striving to build equitable digital labor markets. The findings also beckon further exploration into how other forms of bias—gender, age, or disability—may similarly be embedded within conventional rating and evaluation systems. As platform-mediated work becomes increasingly normative, ensuring that these environments foster fair opportunity is not merely an academic exercise but a vital societal imperative.
With racial equity at the forefront of global discourse, this research illuminates a tangible, actionable lever for reducing racial discrimination embedded within seemingly neutral systems. By recalibrating the granularity of performance ratings, digital platforms can disrupt cycles of bias-induced inequality and step closer towards truly inclusive digital marketplaces where workers are judged solely on their merits.
Subject of Research: People
Article Title: Scale dichotomization reduces customer racial discrimination and income inequality
News Publication Date: 19-Feb-2025
Web References:
References:
DeCelles, K., Humes, D., Botelho, T., & Jun, S. (2025). Scale dichotomization reduces customer racial discrimination and income inequality. Nature. DOI: 10.1038/s41586-025-08599-7
Image Credits: University of Toronto
Keywords: Racial discrimination, Business, Behavioral economics
