Anderson and Arias 2003 — The Effect of Revised Populations on Mortality Statistics for the United States, 2000

Summary

U.S. mortality statistics for 2000 were initially published using 1990-based postcensal population estimates as the denominator. When the 2000 census revealed the true population was over 6 million larger, the National Center for Health Statistics (NCHS) recomputed all death rates and life expectancies using census-enumerated April 1, 2000 populations (bridged to single-race categories to match death certificate coding). The revised statistics are, in many cases, significantly different — particularly for minority populations — and supersede the originally published figures. 2000 data computed on 1990-based denominators are not comparable with 2001+ data.

Key Claims

Total U.S. crude death rate: 873.1 → 854.0 per 100,000 (−2.2%); age-adjusted: 872.0 → 869.0 (−0.3%).
Overall life expectancy at birth: 76.9 → 77.0 years (+0.1).
Direction of revision depends on which age groups dominate: 1990-based estimates were too small for most groups (death rates too high), but overestimated the very elderly (85+), causing death rates at oldest ages to rise under revised pop.
American Indian: crude rate −18.4%; 1990-based estimates dramatically underestimated this population for working ages, while overestimating the 85+ group (+51.2%).
Hispanic paradox partially attenuated: Hispanic age-adjusted rate +13.7% (males +13.8%, females +14.8%); the 1990-based estimates overestimated elderly Hispanics, inflating the denominator and deflating published rates — the Hispanic–non-Hispanic white mortality gap was substantially narrower than previously thought.
Black infants: death rate −5.3%; 1990-based estimates overestimated black infants.
Black males at older ages: large life expectancy losses under revised pop (up to −0.4 years at ages 65–94).
Cause-specific: causes concentrated in middle ages (homicide −3.3%, suicide −1.9%, unintentional injuries −1.7%) show larger reductions; causes at oldest ages (pneumonitis +1.7%, Alzheimer's +0.6%, stroke +0.2%) increase.
The 2000 multiracial census required "bridging" multiple-race counts back to single-race categories using an NCHS/Census Bureau algorithm; race-specific estimates carry uncertainty that is largest for small populations (American Indian and Alaska Native (AIAN)).

Concepts Introduced or Extended

Age Standardization — denominator revision problem: postcensal estimates can differ materially from census counts, creating a trend discontinuity at every decennial transition
Death Certificate Race-Ethnicity Misclassification — distinct numerator issue; this paper addresses the denominator side of the race-specific rate computation problem

Entities Mentioned

Quotes

"Revised death rates and life expectancies are, in many cases, significantly different from previously published mortality statistics calculated using 1990-based postcensal estimates for 2000."

"Previously published mortality statistics for 2000 using the 1990-based populations will not be comparable to the corresponding statistics that will be published for 2001."

My Take

This is an essential housekeeping paper that every researcher using U.S. mortality time series spanning the year 2000 needs to be aware of. The headline finding is modest at the aggregate level (−0.3% age-adjusted rate), but the subgroup-level corrections are large enough to matter for research conclusions — especially for American Indians and Hispanics. The Hispanic paradox case is the most analytically consequential: the "advantage" was partially an artifact of 1990-based postcensal overestimation of elderly Hispanics. The paper cannot easily disentangle race-bridging errors from true postcensal count errors, so the uncertainty in race-specific estimates (particularly AIAN) is noted but not quantified. The recurring nature of this problem (it happens at every decennial census) means researchers should always check which population base any published vital statistic uses.