Predicting race/ethnicity from name and geography (wru)

01

The method workspace

Follow the analysis.

Select a node to explore the method

Method mapInput → 3 steps → output

START WITHInput data

ARRIVE ATResults

Read step 01

Inside the methodReasoning, equations & implementation

01

Data prep

Load wru and the voter file

Step notesField note & discussion1

Data preparation — shapes the raw inputs into what the estimator expects.

What happens here

Load the package and the bundled example voter file containing surnames and geographic identifiers.

Reads fromNot specified Feeds into#2

Key code

# Install:  install.packages("wru")
library(wru)
data(voters)

Reference / docs ↗

Discussion on this step (0)

StatsOtterDiscussion starter · Editorial note

Before using this step, make its assumptions and its output explicit. A useful peer note explains what you would check, what could go wrong, or what would change your interpretation.

Questions to explore

Which assumption matters most here, and what evidence could challenge it?
What would you check before passing this output to the next step?

Peer discussion

Bring your experience, a question, or a different perspective.

02

Estimation

Predict race with BISG

Step notesField note & discussion1

The core estimate — where the causal quantity itself is computed.

What happens here

Call predict_race() to combine the surname likelihood with Census tract-level racial composition (Bayesian Improved Surname Geocoding).

Formula

P(R \mid S, G) = \frac{P(R \mid S)\, P(G \mid R)}{\sum_{r} P(r \mid S)\, P(G \mid r)}

Reads from#1 Feeds into#3

Key code

predict_race(voter.file = voters, census.geo = "tract",
             census.key = Sys.getenv("CENSUS_API_KEY"), party = "PID")

Reference / docs ↗

Discussion on this step (0)

StatsOtterDiscussion starter · Editorial note

Before using this step, make its assumptions and its output explicit. A useful peer note explains what you would check, what could go wrong, or what would change your interpretation.

Questions to explore

Which assumption matters most here, and what evidence could challenge it?
What would you check before passing this output to the next step?

Peer discussion

Bring your experience, a question, or a different perspective.

03

Reporting

Inspect posterior probabilities

Step notesField note & discussion1

Reporting — turn the numbers into a figure or table a reader can act on.

What happens here

The returned data.frame appends posterior race probabilities (pred.whi/bla/his/asi/oth) that sum to one per voter.

Reads from#2 Feeds intoNot specified

Key code

head(predict_race(voter.file = voters, surname.only = TRUE))

Reference / docs ↗

Discussion on this step (0)

StatsOtterDiscussion starter · Editorial note

Before using this step, make its assumptions and its output explicit. A useful peer note explains what you would check, what could go wrong, or what would change your interpretation.

Questions to explore

Which assumption matters most here, and what evidence could challenge it?
What would you check before passing this output to the next step?

Peer discussion

Bring your experience, a question, or a different perspective.

02

Data contract

Inputs & data

A data frame of individuals with surname (and optionally first/middle name) plus geographic identifiers (state, county, tract, or block FIPS codes).

Show data format & exampleHide example

surname	state	county	tract
Smith	NJ	021	000100
Garcia	CA	037	207103
Nguyen	TX	201	412900
Lee	NY	061	010300

03

Read the evidence

Results & interpretation

1 figure Workflow discussion 0

01FigurePredicted race/ethnicity probabilities (BISG) for the example voters.

Result figure rendered by StatsOtter from the package's documented example — unofficial community showcase; all credit to the original authors.

The estimate

\Pr(R\mid S,G)=\frac{\Pr(S\mid R)\,\Pr(R\mid G)}{\sum_{r}\Pr(S\mid r)\,\Pr(r\mid G)}

Context & interpretation

⚠️ Unofficial community showcase of wru (docs). Not affiliated with the authors — all credit to Kosuke Imai & coauthors; this summarizes public documentation.

What it does: wru (Who Are You) produces probabilistic predictions of an individual's racial/ethnic category when race is unobserved—common in voter files, administrative records, and audits of disparities. How it works: it applies Bayesian Improved Surname Geocoding (BISG), combining Census surname (and optionally first/middle-name) race distributions with the racial composition of the person's geographic unit (state, county, tract, or block) via Bayes' Rule. Newer versions add fully Bayesian name-and-geography models and embedding-based features. The core call predict_race() returns, per person, posterior probabilities of being White, Black, Hispanic, Asian, or Other. Assumptions: accuracy depends on the conditional independence of surname and geography given race, on correct and current Census reference tables, and on representative geocoding; predictions are population-level probabilities, not certainties, and can be biased for groups or regions where the reference data fit poorly.

What you get — Per-individual posterior probabilities for each racial/ethnic category (pred.whi, pred.bla, pred.his, pred.asi, pred.oth).

Example output

  surname state county  tract age sex pred.whi pred.bla pred.his pred.asi pred.oth
   Khanna    NJ    021 004000  29   0   0.0676   0.0043   0.0082   0.8668   0.0531
     Imai    NJ    021 004501  40   0   0.0812   0.0024   0.0689   0.7375   0.1100
  Velasco    NY    061 004800  33   0   0.0594   0.0026   0.8227   0.1051   0.0102
  Fifield    NJ    021 004501  27   0   0.9356   0.0022   0.0285   0.0078   0.0259
     Zhou    NJ    021 004501  28   1   0.0098   0.0018   0.0007   0.9820   0.0058
 Ratkovic    NJ    021 004000  35   0   0.9187   0.0108   0.0108   0.0108   0.0488

Links: package · paper

Follow the analysis.

Inputs & data

Results & interpretation

∑The estimate

Example output

Discuss the workflow (0)

The estimate