HMD member-initiated meeting at the 2016 PAA conference March 30, 2016 Washington D.C.
The Human Mortality Database: a powerful resource of demography
Vladimir M. Shkolnikov, Dmitri Jdanov, Magali Barbieri, Domantas Jasilionis, Carl Boe HMD: General information
Collaboration Max Planck Institute for Department of Demography Demographic Research at the University of California, (MPIDR) Berkeley (UCB)
www.mortality.org
HMD Data Resource Profile in the International Journal of Epidemiology http://ije.oxfordjournals.org/content/44/5/1549 Support Max Planck Society (Germany), National Institute of Aging (USA), Institut national d'études démographiques (France), University of California at Berkeley (USA)
Outline of the presentation
• Reasons for and origins of the HMD
• What HMD does • Data problems • Enhancement of the methodology
• HMD-based studies
• Research teams • Reasons for and origins of the HMD
• What HMD does • Data problems • Enhancement of the methodology
• HMD-based studies
• Research teams Mortality convergence and expectation of convergence before the 1990s
1970s-80s: strong expectation of worldwide mortality convergence.
Gross analyses of international mortality trends by Keyfitz, Preston, Schoen, and Flieger suggested a mortality transition process: falling deaths at young ages, greater survival to old age, where people exposed to “degenerative” diseases, difficult to treat or prevent.
→ Expectation of rapid progress in high-mortality countries, via reduced young-age mortality and slower progress or stagnation in countries with already low mortality.
UN Population Division: 2.5 year gain in LEB every 5 years for countries with LEB<62, after which the 5-year gain decreases to 2 years. New phenomena: mortality divergence and steep progress at advanced ages Life expectancy divergence after 1970 Life expectancy divergence: - unexpected health crisis in communist and post-communist countries of the former USSR and CEE; - unexpected further progress in the established market economies (EME) Life expectancy at age 80 since 1880
Source: Timonin et al, 2015; Barbieri et al. 2015
Success in fight with CVD and other “degenerative” diseases led to spread of mortality reduction toward very old ages.
Source: Built on HMD data. Discovery of the linear life expectancy increase
Upper limits of life expectancy suggested by researchers in different years
The linear life expectancy increase inevitably suggests spread of mortality reduction toward very old and advanced ages. Source: Oeppen and Vaupel, 2002. New data requirements
Questions: What are the prospects of the longevity rise and population aging? What are the major components, determinants, and consequences of rising longevity and population aging? Demography addresses these questions through in-depth analyses and modeling of longevity and survival in human populations with a special emphasis on advanced (frontier) ages. Need for data that could reflect historical transformations of the mortality curve and the longevity revolution of the modern era by: - providing long-term continuous series without gaps or ruptures; - running up to the highest ages; - providing fine details according to age, time, and cohort dimensions; - ensuring sufficient quality and comparability across time and populations.
The international databases of the 1990s did not meet these criteria. HMD does. 1990s: V.Kannisto, R.Thatcher and J.Vaupel begin filling the gap
Roger Thatcher
Väinö Kannisto
James W. Vaupel
In 1994-96 Väinö Kannisto produced two books documenting advances in survival and longevity on the basis of data from 28 developed countries. The books contained numerous and detailed data tables. In 1988-2001 Thatcher, Vaupel and Kannisto published important works on old-age survival, assessment of data quality, and re- estimation of populations aged 80+. BMD and K-T DB: predecessors of HMD
The Berkeley Mortality Database launched in 1997 by John R. Wilmoth (Dept. of Demography at UCB). Four countries. Data up to age 110. Single-year divide by age, time, year of birth. Variety of age by time format: 1x1, 5x1, 5x5, …
The Kannisto-Thatcher database launched in 2001 MPIDR. 30 countries. Covers ages 80 to 110+. Follows the Kannisto’s approach for re- estimation of populations at ages 80+. • Reasons for and origins of the HMD
• What HMD does • Data problems • Enhancement of the methodology
• HMD-based studies
• Research teams HMD: basic facts
- Work began in autumn 2000 - Launched online in May 2002 with 17 country series - Now: 38 countries and 8 regions, 30,000+ users - Comparability across time and space - Continuous, long-term series without gaps or ruptures - Data by age, year, cohort, in age-by-time formats 1x1, 5x1, 1x5 etc. - Uniform data files compatible with stat. packages, research applications, and Excel - Detailed documentation on origins and quality of the data Processing of raw data into Lexis surface in the HMD
Raw Data Files Input Archive
Data checks Input Utilities Manual Work
Programs Programs Input for for Life Lexis Data calculation calculation Tables DB base of the Lexis of LTs DB
Data checks Data checks
WWW Processing of raw data into Lexis surface in the HMD England & Wales Lexis surfaces of period and cohort mortality What HMD does for its users. Work behind output data.
- Collects and provides official raw data for as many countries and years as possible at the highest possible level of detail. - Analyzes existing evaluations and literature and performs checks to ensure relevance, coverage, completeness, and consistency of the raw data. Males, FRG, 1990 1 data model - If needed, splits deaths at unknown age 0.9 original data from open age interval 0.8 and deaths in open-ended age intervals by 0.7 * Dx*i 0.6 S(x i)
0.5 D single-year ages. s(x) x*20 0.4 - If needed, splits deaths in 5-yr age groups 0.3 0.2 into single-year age intervals and further 0.1 0 70 75 80 85 90 95 100 105 splits single-year deaths by birth cohort. Age 4 x 10 Cumulative number of deaths 2.4 Proportion of deaths in lower triangle by IMR, Males age 0 x +1 2.2
0.90 x France data 1.8 Sweden fit Japan fit t t+1 t+2 France fit
0.85 1.6 x +1 1.4
0.80 x +1 1.2
0.75 x 1 Proportion in lower triangle lower in Proportion Cubic spline
0.70 0.8 x interpolation 0.6 0.65 t t+1 x -1 0 20 40 60 80 100 120 t t+1 0.005 0.010 0.020 0.050 0.100
Infant mortality rate (log scale) What HMD does for its user. Work behind the output data (cont.) Age x + 6 P(x+5,t+5) x + 5 D (x+4,t+3 - If official annual population estimates L x + 4 ) are not available or not fully reliable, x + 3 x + 2 constructs inter-, post- and pre- DU(x+1,t+1) x + 1 P(x,t) censal population estimates. x
t t+1 t+2 t+3 t+4 t+5 Time - Constructs more accurate population A - Official estimates / intercensal survival estimates at ages 80+ by the extinct B - Extinct cohorts C - Survivor ratio, SR90+ cohort method combined with the survivor ratio method.
- Computes period and cohort death rates and life tables. - Checks the output data for internal consistency and internal and external plausibility. What HMD does for its user. Work behind output data (cont. 2)
- Adjusts for territorial changes, changes of coverage, and population definition. - Provides data on important regions and sub-populations within countries (e.g. Germany East and Germany West, NZ Maori and NZ non-Maori etc.). - Provides additional estimates and adjustments for some countries: • Constructs mortality and population estimates over war periods for the total (civil + combat) populations. • Corrects problems at advanced ages by using additional higher- quality sources. • Makes country-specific adjustments to correct inconsistencies in time series. - Fully describes data origins, sources and highlights quality issues in the country-specific “Background and Documentation” files. - Provides special warnings pointing at problems which are not treated by the HMD methodology and remain in the output data. HMD: available data
Period life tables only Period and cohort life tables
1750 1800 1850 1900 1950 2000
Period and cohort mortality data series across time and populations Source: An updated version of the data map by Barbieri et al, 2015 • Reasons for and origins of the HMD
• What HMD does • Data problems • Enhancement of the methodology
• HMD-based studies
• Research teams Germany: implausible mortality trends at very old ages
West Germany East Germany 0.42 Males 0.42 Males Females Females 0.4 0.4 0.38 0.38
0.36 0.36 0.34 0.34
0.32 0.32
m90 0.3 0.3
0.28 0.28 0.26 0.26
0.24 0.24 0.22 0.22 0.2 0.2
1956 1960 1964 1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008
1956 1960 1964 1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008 Year Year
Trends in death rates at ages 90+, calculated from the official population estimates, for West and East Germany, males and females, 1956-2008. Germany: inflated population denominator at ages 90+ The problem was solved by using estimates of old-age populations by the Deutscher Rentenversicherung Bund (DRV) - the German Pension Scheme, and (later on) of the 2011 census. Ratio of DRV population by age to respective HMD estimates based on the official data, 2009.
West Germany East Germany 1.3 Males 1.3 Males Females Females 1.2 1.2
1.1 1.1
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
RatioDRVHMD population/ estimates 0.5 0.5 70 75 80 85 90 95 100 70 75 80 85 90 95 100 Age Age Portugal: correction of population series for the 1970s Current population estimates HMD inter-censal estimates
4 4 x 10 Population counts, year 1976 x 10 Population counts, year 1976
8 8
7 7 1976
6 6 1976
5 5 1975 1975
4
Population 4 Population
3 3
2 2
1 1
0 0 0 20 40 60 80 100 0 20 40 60 80 100 Age Age 23 of 32 Bulgaria: correction of population series over the 1990s and the 2000s
4700000 4700000
1984 1991 4500000 4500000 1992 1985 (census year) (census year) 4300000 4300000 Females 2000 MALES 4100000 FEMALES 4100000 Males
3900000 2001 3900000 census year
3700000 3700000
3500000 3500000 1980 1985 1990 1995 2000
1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 Trends in the total number of males and females. Bulgaria, 1961-2003. Official population estimates (left) and HMD data (right). An HMD candidate country Costa-Rica. Mortality understatement at old ages
Mortality rate ratios, males, 1970-2008 Costamx ratio,-Rica CRI/SWE, / Sweden males Costamx ratio,-Rica CRI/JPN, / Japan males
110 1102.0 2.0 105 105 100 100 1.8 1.8 95 95 90 90 85 851.6 1.6 80 80 75 75 70 701.4 1.4 65 65 60 60 1.2 1.2 55 55
Age
Age 50 50 45 451.0 1.0 40 40 35 35 0.8 0.8 Evidence of age 30 30 overstatement and 25 25 age heaping for the 20 200.6 0.6 15 15 whole series 10 10 0.4 5 50.4 0 0 1970 1980 1990 2000 2010 1970 1980 1990 2000 2010 Year Year Costa-Rica: overstated life expectancy at ages 65 and 80
Trends in male life expectancy at age 65 (left panel) and age 80 (right panel) in Costa Rica • Reasons for and origins of the HMD
• What HMD does • Data problems • Enhancement of the methodology
• HMD-based studies
• Research teams Beginning of the HMD Methods Protocol Revisions of the HMD MP
• Revisions 1-2 not published • Revision 3 (May 2002) is the first published version. The fist (published) HMD data were calculated according to the MP v.3 • Revision 4 (November 2005): ⁞ Changed method for splitting deaths into Lexis triangles; ⁞ Revised method for splitting open age interval; ⁞ Revised formula for population exposure; ⁞ Revised procedure for smoothing M(x). • Revision 5 (February 2007): ⁞ Various places through MP, changed "country"/"countries" to "country or area"/"population“; ⁞ Inaccuracies in some equations corrected; ⁞ Cubic spline method modified to split VV data. • Revision 6 (2016): ⁞ Changed method for calculating population exposures; ⁞ Changed method for calculating the mean age of infant death; ⁞ MP re-written in LaTEX • Revision 7 – work in progress
MP6: Population exposure accounting for variation in cohort’s birthdays MP6: New formula for a0 accounting for change in infant death distribution at low levels of mortality
Source: E.Andreev and Kingkade, 2015 Some ideas for MP7
- New inter-censal survival method accounting for uneven migration across time.
- Computation of death rates by Lexis triangles M(x, t, c). • Reasons for and origins of the HMD
• What HMD does • Data problems • Enhancement of the methodology
• HMD-based studies
• Research teams HMD citing
Total 2002-2015: All items - 2,244 Journal papers - 1,766 Studies on advances in survival with emphasis on old and very ages
The best-practice and country-specific Probabilities of death at ages 80 and 90 since life expectancies since 1846 1950
Source: Christensen, Doblhammer, Rau, Vaupel, 2009 Analyses and measures on mortality surfaces
Smoothed mortality surfaces for selected countries: 1950-2010
Source: Ouellette and Bourbeau, 2011. Computation of the cross-sectional average length of life (CAL) and of the average cohort life expectancy (ACLE)
Sources: Guillot, 2003; Schoen & Canudas-Romo, 2005. Measures sensitive to tails of the mortality distribution
Longevity measures expressing the A graphical depiction of the calculation of geometry of the survival curve the threshold age a†
IRA – Inner Rectangle Area, PMA-Premature Mortality Area, LEA – Longevity Extension Area, Source: Zhang and HA – Horizontalization Area, SPA – Shifting Vaupel, 2009 Potential Area, SR – Senescence Rectangle
Source: Ebeling and Rau, 2014 Use of HMD data and methods in methods protocols and software
Google Scholar shows 145 citations of the HMD Methods Protocol • Reasons for and origins of the HMD
• What HMD does • Data problems • Enhancement of the methodology
• HMD-based studies • Research teams See more on the Research Teams at http://www.mortality.org/Public/ResearchTeams.php John R. Wilmoth Vladimir M. Shkolnikov Magali Barbieri Dmitry Jdanov Founding Director, Director, MPIDR Associate Director, Head of the MPIDR UCB in 2000, now UN Head of the UCB Team, Team, MPIDR UCB&INED Max Planck Team Berkeley Team (members present and some former) (members present and some former) Gabriel Borges
Domantas Sebastian Dana Glei Jasilionis Kluesener Carl Boe Evgeny Andreev Tim Riffe Vladimir Canudas- Romo Pavel Grigoriev Eva Kibele Sigrid Gellers Celeste Winant Rembrandt Scholz Monica Lisa Yang Alexander