Arias et al 2008 — The Validity of Race and Hispanic Origin Reporting on Death Certificates in the United States

Summary

Using the National Longitudinal Mortality Study (NLMS) — 2.3 million Current Population Survey (CPS) respondents linked to National Vital Statistics System (NVSS) death certificates — Arias et al. evaluate the accuracy of race and Hispanic-origin reporting on U.S. death certificates relative to prior self-report. They find that White and Black certificates are unbiased (classification ratio [CR] $\approx 1.00$ ), Hispanic certificates are slightly inflated ( $CR = 1.05$ , validating the paradox), and Asian or Pacific Islander (API) certificates have improved to near-accuracy ( $CR = 1.07$ ). American Indian/Alaska Native (AIAN) certificates remain severely underreported ( $CR = 1.30$ , sensitivity $\approx 55\%$ ), reversing AIAN's apparent mortality advantage (age-adjusted death rate (AADR) ratio vs. White: $0.85 \to 1.11$ after correction) — showing AIAN have the worst mortality of any group.

Key Claims

Validation design: NLMS links CPS self-reported race/ethnicity (1973, 1978–1998) to NVSS death certificates; $252{,}627$ deaths; race sample $N = 161{,}302$ ; Hispanic sample $N = 114{,}869$ .
Three metrics: (1) Sensitivity = % of self-identified group correctly coded on death certificate (DC); (2) Predictive Value Positive (PVP) = % of DC-coded group who self-identified the same way; (3) Classification Ratio = CPS count / DC count — values $>1$ indicate undercounting on DC.
White/Black: $CR \approx 1.00$ ; sensitivity $\approx 99\%$ (White), $\approx 98\%$ (Black); no meaningful mortality effect.
AIAN: $CR = 1.30$ ; sensitivity $\approx 55\%$ (lowest of all groups; persistent across 1973–1998 sample periods); corrected AADR ratio vs. White changes from $0.85$ (apparent advantage) to $1.11$ (real disadvantage).
AIAN geography: CR ranges from $1.02$ (states with high coethnic concentration — AZ, NM, OK, MT, NE) to $1.63$ (low-concentration states); funeral directors use visible community cues to assign AIAN identity.
AIAN identity expansion: Unprecedented $>90\%$ increase in AIAN self-identification in CPS/census since 1960, driven by rising social acceptability of claiming mixed-ancestry AIAN identity; these individuals may not appear visibly AIAN to funeral directors, exacerbating the mismatch.
API: $CR = 1.07$ ; sensitivity $= 90\%$ and improving over the sample period; corrected AADR ratio vs. White changes from $0.60$ to $0.64$ (small effect).
Hispanic: $CR = 1.05$ ; overall AADR $= 79\%$ of non-Hispanic White (NHW) on raw data $\to 83\%$ after correction; Hispanic paradox is real, not a statistical artifact.
Hispanic nativity: Foreign-born $CR = 1.02$ vs. US-born $CR = 1.07$ ; funeral directors appear to use birthplace as a cue for Hispanic identity. By 1997, 23+ states had signed the National Health Interview Survey (NHIS) agreement that prompted funeral directors to ask the decedent's family about Hispanic origin — improving coverage.
Hispanic subgroups: Central/South American CR fell from $2.35$ to $1.05$ as coverage expanded; Puerto Rican $CR = 1.05$ ; Cuban $CR = 1.10$ . Note: the 2000 Census denominator is problematic for Hispanic subgroups because a wording change led $16\%$ of Hispanics (especially Central/South Americans, $+34.4\%$ ) to give a generic term ("Hispanic," "Latino") instead of a specific national-origin group.
Correction formula: $\text{AADR}^\text{corrected}_i = \text{AADR}^\text{raw}_i \times CR_i$ ; applies to numerator (DC race coding) and is equivalent to adjusting the denominator.
Reference to standards: Arias et al. 2008 cites Anderson and Rosenberg 1998 as the methodological basis for age-standardized rates used in the analysis.

Concepts Introduced or Extended

Entities Mentioned

Quotes

"Misclassification of race and Hispanic origin on death certificates creates a numerator-denominator mismatch that systematically biases race-specific mortality rates."

"After correcting for misclassification, the AIAN mortality advantage relative to the white population disappears and is replaced by a mortality disadvantage."

"The Hispanic mortality paradox is not an artifact of misclassification."

My Take

The AIAN finding is the most policy-consequential result. A $55\%$ sensitivity rate is catastrophic for any measurement purpose, and the mortality flip from apparent advantage to real disadvantage means that AIAN populations have been invisibilized in national health statistics for decades. The geographic CR variation ( $1.02$ vs. $1.63$ ) is methodologically illuminating: it shows that misclassification is driven by observability, not random error. The AIAN identity expansion paradox is underappreciated — rising ethnic self-identification increases the numerator of the CR precisely because funeral directors cannot track identity claims that are culturally rather than visually marked. The Hispanic subgroup denominator problem is a reminder that census instrument design has first-order effects on epidemiological conclusions.