Perceptions of Race in the Labor Market

Discussion

Evan Rose

University of Chicago

2026-10-16

Overview

Broad acceptance of race as a constructed instead of essential social category

“a socially determined construction with complex biological associations” (Holland, 2008)

On the one hand, so what? Economists are typically less interested in philosophical fights taking place elsewhere in the humanities

Race is clearly real and can’t be just waved away as a “construct”

On the other hand, growing literature demonstrates that fluidity of these social categories matters for measurement and interpretation

Fact one: Racial identity is fluid and contested

Portrait of Barack Obama

President Obama: “I identify as African-American—that’s how I’m treated and that’s how I’m viewed. I’m proud of it”

2010 Pew Poll: 27% of Americans label Obama as Black (55% for those who identify as Black and 24% for whites)

Meaningful fraction of Americans switch racial identities over time (Liebler, 2017)

Fact two: Racial perceptions depend on non-racial factors

Faces in different clothing from Freeman et al. (2011)

Freeman et al. (2011): Identical faces categorized as Black more often in low-status clothes

Saperstein and Penner (2012): Racial coding sensitive to history of incarceration (cf. Charles and Guryan, 2011)

Gaddis (2017): Black names given by more educated mothers seen as less Black (e.g., Bria vs. Tamika)

Key point: Even if demographic labels are unchanged, observed characteristics affect how social identity is perceived (e.g., Monk Jr., 2021)

Fact three: Social identity is manipulable

Individuals constantly signal social identity with a variety of actions

  • e.g., by “whistling Vivaldi,” adopting new names and clothing, affecting patterns of speech, etc.

“Passing” is possible because racial identities are not strictly tied to innate physical characteristics

You are not Black because you are (in essence) Black; you are Black… because of how you act… because of how you juggle and combine many differently racialized and class(ed) actions (walking, talking, laughing, watching a movie, standing, emoting, partying) in an everyday matrix of performative possibilities (Jackson Jr., 2010)

The econ awakening

These ideas are increasingly penetrating the econosphere

Cornwell, Rivera, Schmutte (2017): wage changes in Brazil are associated with changes in racial categorization

Adukia, Hornbeck, Keniston, Lualdi (2026): wealth and literacy predict enumerator categorization conditional on skin tone in 1870 Census

Rose (2023): Race gaps in skin shade and distinctive speech in NLSY shrink with demographic, education, and AFQT controls, along with wage gaps

I view this paper as the richest empirical exploration of these ideas to date

What do we learn?

There is a lot in this paper, including a very beefy appendix!

What do I take away?

  1. Substantial mismatch between employer- and self-ascribed racial categories in Brazil, as well as across employers
  1. Nuanced relationship between skin tone and racial categorization, mediated by education level, cohort, and region
  1. Race gaps in cues (e.g., skin tone) shrink with “benchmarking” controls, complicating interpretation of residual gaps

Lots more in the paper

Some amazing facts buried in the appendix:

  1. In an AKM decomposition for employer-ascribed race categorization, firm FE explain 25% of variance in race categories! Approximately the same share as worker FE.

Would love to see more analysis of these firm effects, e.g.,

  1. Connection to wage premia
  2. Connection to employer size and industry
  3. Connection to employer racial composition
  1. Strong effects of prior wages on both employer- and self-declared race, even conditional on detailed controls

These labor market history variables seem like first-order inputs to the “race function,” especially when we’re considering wage gaps

Where is the room to learn more?

Two areas where I would like to learn more:

  1. Race as continuous vs. discrete categorization

  2. Endogenous inputs in the race function

Continuous categorization

Paper primarily treats racial categorization as a discretization of underlying latent perception:

\[ r^q = A_q(Z) + u^q \\ R^q = 1 \{ r^q \geq \tau_q \} \]

Implies that changing \(Z\) changes both categorization at the margin (\(Pr(R^q = 1)\)) and inframarginal perceptions (distribution of \(r^q \mid R^q = 1\))

Not obvious that people respond to discrete categories vs. how white/black someone seems

E.g., strong colorism gradient among people with same self- or employer-ascribed race

Continuous categorization

Perhaps we could learn more about effects of shifting cues among the inframarginal groups here

Allowing outcomes to depend on \(r^q\) vs. \(R^q\) also has obvious implications for the perception-normalized disparity (PND)

Analogy to IV is discretizing a continuous treatment; creates exclusion issues when instruments shift treatment along intensive margin

Authors are very careful to treat the PND as a measurement and not causal exercise, but I would like to see versions that attempt to normalize using differences in the distribution of \(r^q\) rather than \(Pr(r^q \geq \tau_q)\)

What is the race function?

Long history of race as an endogenously shaped identity (e.g., Coate and Loury, 1993; Steele, 2010; Austen-Smith and Fryer, 2005)

How much of the differences across self- and employer-ascribed categories, as well as changes in categorization across groups, reflects choices in how people present?

Obviously a hard problem, but interesting opportunities here:

  • Response of categories to categories of peers
  • Dynamics in categorization
  • Response of categorization to labor market conditions

Would bring a little more econ to the paper by thinking about the behavioral margins underlying the very interesting empirical facts

Bottom line

Impressive, interesting work

Incredibly detailed and credible analysis

Probably several more papers to write just using this merge