Comparative Quality of Mortality Data Derived from Official Statistics, for both Historical and Cont

1
Comparative Quality of Mortality Data Derived
from Official Statistics, for both Historical and
Contemporary PopulationsJohn Wilmoth, Danzhen
You, and Bernardo QueirozDepartment of
DemographyUniversity of California, Berkeley

• Presentation for a Workshop on Adult Mortality
  in the Developing WorldMarin County,
  California, 9-11 July 2004

2
Objectives

• To perform an initial evaluation and comparison
  of the quality of mortality data for a selected
  group of countries already included in the HMD
• To compare the quality of data for a small group
  of developing countries, in part to determine
  whether some of them might be added in the future

3
Countries and Time Periods

• Sweden 1750-2000
• England Wales 1841-1991
• Japan 1898-2000
• France 1901-1999
• Taiwan 1905-2000
• Chile 1920-2002
• Mexico 1930-2000
• United States 1930-2000
• Costa Rica 1951-2000
• South Korea 1955-2000

4
Two Aspects of Data Quality

• Completeness of population coverage (in both
  census and death count data)
• Accuracy of reported age (digit preference, age
  exaggeration)

5
Two Sorts of Methods for Evaluating Data Quality

• Consistency among data sources
• Plausibility of observed patterns

6
Assessing Completeness

• Possibility of both under- and over-reporting
• Standard methods assume age invariance
• Measures of relative completeness

7
Standard Methods

• Growth Balance, Bennett-Horiuchi, etc.
• Typically used for less developed countries where
  data are known to be less complete (compared to
  countries in the HMD)
• Used to generate coverage estimates by country,
  sex, and intercensal time periods
• Mortality estimation based on adjusted data

8
Age Invariance in Completeness of Population
Coverage

• Completeness of Census 1
• Completeness of Census 2
• Completeness of death registration

9
Results Concerning Completeness of Population
Coverage

• Relative completeness of Census 1 compared to
  Census 2
• Relative completeness of death registration
  compared to Census 2

10
Estimates of Relative Completeness Census (black)
and Death Counts (color)Sweden, 1750-2000
11
Estimates of Relative Completeness Census (black)
and Death Counts (color)England Wales,
1841-1991
12
Estimates of Relative Completeness Census (black)
and Death Counts (color)Japan, 1898-2000
13
Estimates of Relative Completeness Census (black)
and Death Counts (color)France, 1901-1999
14
Estimates of Relative Completeness Census (black)
and Death Counts (color)Taiwan, 1905-2000
15
Estimates of Relative Completeness Census (black)
and Death Counts (color)Chile, 1920-2002
16
Estimates of Relative Completeness Census (black)
and Death Counts (color)Mexico, 1930-2000
17
Estimates of Relative Completeness Census (black)
and Death Counts (color)United States, 1930-2000
18
Estimates of Relative Completeness Census (black)
and Death Counts (color)Costa Rica, 1951-2000
19
Estimates of Relative Completeness Census (black)
and Death Counts (color)South Korea, 1955-2000
20
Types of Age Misreporting

• Digit preference (age heaping)
• Systematic age exaggeration
• Random errors

21
Indicators of Age Misreporting

• Index of age heaping
• Right-tail analysis
• Trends in extreme ages (previous studies)
• Matching studies (not considered here)

22
Digit Preference Census CountsEngland Wales
1911, , 1991
23
Digit Preference Death Counts England Wales
1911, , 1991
24
Age Heaping in Mexican Census Data
1940
1960
Graphs courtesy of Jean-Marie Robine, INSERM,
Montpellier, France
25
Ratio of Death Counts, Ages 70 to 60Four
countries, Males (blue) Females (red)
26
Ratio of Death Counts, Ages 70 to 60Four
countries, Males (blue) Females (red)
27
Ratio of Death Counts Ages 90 to 60
28
General Conclusions

• Need to develop better methods, especially with
  regard to migration
• Need to extend evaluation to include more
  countries

29
Specific Conclusions

• Sweden
• Consistency of population and death counts even
  for earliest time periods (from 1751)
• Some problems at older ages before 1860, but high
  quality afterwards (incl. extreme old age)
• England Wales
• Decrease in census completeness during 19th C.
  (but much better during 20th C.)
• High quality after 1911 (incl. extreme old age)

30
Specific Conclusions

• Japan
• Consistency of population and death counts for
  full 20th C. (except for gap due to WWII)
• High quality at older ages after 1950 (but for
  extreme old age, not until 1970)
• France
• Consistency of population and death counts for
  full 20th C. (with some uncertainties due to
  wars)
• High quality at older ages after 1946 (including
  extreme old age)

31
Specific Conclusions

• Taiwan
• Consistency of population and death counts for
  full 20th C. (except for gap at mid-century)
• No evidence of widespread age misreporting (but
  need to check further)
• Chile
• Major problems of inconsistency between
  population and death counts (except for 1990s)
• No evidence of widespread age misreporting (but
  need to check further)

32
Specific Conclusions

• Mexico
• Major problems of inconsistency between
  population and death counts (incl. recent
  decades)
• Pronounced age heaping in census data
• United States
• No obvious problems of inconsistency, but
  difficult to evaluate because of migration
• Nothing to add about age misreporting problems

33
Specific Conclusions

• Costa Rica
• Major problems of inconsistency between
  population and death counts (but with some
  improvement in recent decades)
• No evidence of widespread age misreporting (but
  need to check further)
• South Korea
• Consistency of census and deaths improves rapidly
  after 1950s
• Plausible data for 1970s and forward