Race as a Bundle of Sticks:

Designs that Estimate Effects of Seemingly Immutable Characteristics

Omar Wasow

Asst Professor, Politics, UC Berkeley
(with Maya Sen, Harvard Kennedy School)

PS140O, 2/18/2025

“No causation without manipulation.”

– Holland (1986)

Can we randomly assign race?

Experiment with race as skin color?

What do we even mean by “race”?

The potential outcomes framework
- A large literature raising questions about causal inference with “immutable characteristics” (e.g., Rubin 1974, Holland 1986, Angrist 1996, Rubin 2005, Greiner 2010)
Race within the potential outcomes framework
- Problem 1: Race cannot be manipulated
- Problem 2: With race, everything is post-treatment
- Problem 3: Race is unstable

Can we estimate effects of race or sex?

Partial solution to problem of manipulability
- Identify an element of race that is both relevant to the research question and can be manipulated
- Traits that are constitutive and collinear with race while also mutable are well suited to causal inference

Name	% White	% Black	% Asian	% Hispanic
Yoder	98.11	0.14	0.18	0.68
Krueger	97.06	0.16	0.43	1.15
Washington	5.16	89.87	0.25	1.45
Jefferson	18.72	75.24	0.25	1.57
Zhang	0.61	0.09	98.16	0.16
Huang	1.03	0.09	96.83	0.29
Barajas	3.28	0.17	0.17	95.97
Zavala	4.10	0.17	0.16	95.10

“In Haiti, where most identify as Afro or Black, skin tone explains more economic variation than ethnoracial categories.”
“In contrast, in Guatemala, where Indigenous and Mestizo identities are more evenly distributed, ethnoracial categories play a larger role.”
“These patterns likely reflect each country’s racial classification history, labor systems, and institutionalized discrimination.”

Partial solution to problem of post-treatment bias
- Constitutive elements of race are “assigned” after conception or remain manipulable after conception
- Focusing on an element of race more tractable for research and policy

Hypothetical mutability of characteristics associated with race and ethnicity.

Thank you