Content uploaded by Luis Rosero-Bixby
Author content
All content in this area was uploaded by Luis Rosero-Bixby
Content may be subject to copyright.
Demography, Volume 45-Number 3, August 2008: 673–691 673
T
THE EXCEPTIONALLY HIGH LIFE EXPECTANCY OF
COSTA RICAN NONAGENARIANS*
LUIS ROSERO-BIXBY
Robust data from a voter registry show that Costa Rican nonagenarians have an exceptionally
high live expectancy. Mortality at age 90 in Costa Rica is at least 14% lower than an average of 13
high-income countries. This advantage increases with age by 1% per year. Males have an additional
12% advantage. Age-90 life expectancy for males is 4.4 years, one-half year more than any other
country in the world. These estimates do not use problematic data on reported ages, but ages are
computed from birth dates in the Costa Rican birth-registration ledgers. Census data confi rm the
exceptionally high survival of elderly Costa Ricans, especially males. Comparisons with the United
States and Sweden show that the Costa Rican advantage comes mostly from reduced incidence of
cardiovascular diseases, coupled with a low prevalence of obesity, as the only available explana-
tory risk factor. Costa Rican nonagenarians are survivors of cohorts that underwent extremely harsh
health conditions when young, and their advantage might be just a heterogeneity in frailty effect that
might disappear in more recent cohorts. The availability of reliable estimates for the oldest-old in
low- income populations is extremely rare. These results may enlighten the debate over how harsh
early-life health conditions affect older-age mortality.
wo key fi ndings have emerged from recent studies of old-age mortality in humans
(Vaupel et al. 1998): (1) mortality rates are declining substantially, and (2) the increase of
death rates with age decelerates among the oldest-old. In the words of Vaupel et al. (1998),
these fi ndings are perplexing and hard to reconcile: according to evolutionary biology,
there is no possible selection against mutations occurring after reproduction and nurturing
have ceased. A possible explanation of the old-age deceleration is heterogeneity in frailty;
that is, as the frail die at early ages, the old tend to be a select subpopulation of the fi t-
test (Barbi, Caselli, and Vallin 2003; Horiuchi and Wilmoth 1998; Vaupel et al. 1998). In
turn, a possible explanation of the mortality decline at old ages is a cohort effect of past
improvements in health conditions at early ages; that is, recent improvements in health
status among the elderly would echo events that happened decades ago when cohorts were
young. These two explanations are somehow contradictory: does high, early-life mortality
make a cohort stronger by eliminating the frail, or does the cohort become weaker because
of accumulated injuries? An important scientifi c debate is taking place in this regard (Barbi
and Vaupel 2005; Finch and Crimmins 2004, 2005).
The heterogeneity in frailty argument has been mostly supported by mathematical and
simulation models (Vaupel, Manton, and Stallard 1979); by indirect evidence from genetic
homogeneous populations such as twins (Yashin and Iachine 1997); and by observations
in other species, such as the Mediterranean fruit fl y (Vaupel and Carey 1993). Indirect
methods have been developed to determine the existence of heterogeneity from cohort
mortality patterns (Manton, Stallard, and Vaupel 1981). Data showing low death rates at
old ages in low-income populations that saw harsh health conditions at young ages might
support the heterogeneity in frailty argument, given the prejudice that the poor cannot be
*Luis Rosero-Bixby, Centro Centroamericano de Población, Universidad de Costa Rica, Apartado 2060, San
Jose, Costa Rica; e-mail: Lrosero@ccp.ucr.ac.cr. The Wellcome Trust Foundation (Grant No. 072406/Z/03/Z) and
the Florida Ice and Farm Co. of Costa Rica provided support for this and other studies on aging in Costa Rica.
The Costa Rican Tribunal Supremo de Elecciones provided the databases. Daniel Antich from the Universidad de
Costa Rica provided assistance in processing the databases. Albert I. Hermalin from the University of Michigan
provided suggestions and encouragement to improve this manuscript. German Rodríguez from Princeton University
provided statistical advice to estimate the model.
674 Demography, Volume 45-Number 3, August 2008
healthy nor live longer. Coale and Kisker (1986) offered two possible explanations of the
observed mortality crossover at old ages among some socioeconomically disadvantaged
populations: selection in heterogeneity or bad data. They concluded that bad data was the
probable cause of these crossovers given the positive association between mortality in
childhood and at young adult ages, and mortality in old age that was observed in cohorts
with good data. A social selection argument, which has been used to explain good health
in low-income populations (especially immigrants), parallels the heterogeneity in frailty
argument. A well-known example of this is the “Hispanic paradox” (i.e., that Hispanics
have lower mortality than whites) in the United States, which some researchers explain by
several types of selection biases (Khlat and Darmon 2003; Palloni and Morenoff 2001).
Genetic makeup as well as nutrition, well-being, access to health care, lifestyles, and
environmental conditions (contemporary and past) are, of course, determinants of old-age
mortality above and beyond selection effects. The relative importance of these factors is,
however, unknown. In addition, determinants of mortality might act differently at old ages
than at young ages, challenging conventional wisdom that extrapolates to old ages what
has been observed for younger ages. For example, Okinawa displays exceptional longev-
ity even though it is one of the least developed regions in Japan (Cockerham and Yamori
2001). Another challenging example is that of Hispanics in the United States, who have
lower adult mortality than whites in spite of Hispanics’ lower socioeconomic status and
limited access to health care (Elo et al. 2004). And there is also the case of exceptional
longevity in Sardinia, Italy, where old-age life expectancy is higher than in the much
richer northern region of the country (Caselli and Lipsi 2006). Are elderly Okinawans,
U.S. Hispanics, and Sardinians really exceptions to the rule of a socioeconomic gradient
in mortality? Might it be that at old ages, the rules of survival are different than at young
ages? To what extent do poor health conditions early in life strengthen or weaken a cohort
at old ages? If dietary caloric restriction slows aging in other species (Roth et al. 2002),
could certain human populations that were undernourished when young have an advantage
for survival at old age?
An obstacle to answering these questions is the absence of suffi ciently accurate data
about old-age mortality in low-income populations. Costa Rica may be an exception. Since
1961, the United Nations has graded the Costa Rican vital statistics system—which was
established in 1883––as “complete” (United Nations 1961). The country also has a care-
fully kept population registry used for voting purposes. Costa Rica is one of 11 developing
countries whose vital registration statistics in 1995 are characterized by Hill et al. as both
complete (recording at least 90% of births and deaths) and accurate (producing mortality
estimates similar to those based on census and survey data) (Hill et al. 1999).
With its 4.5 million inhabitants, Costa Rica is the second most-densely populated coun-
try in the Continental Americas. (El Salvador ranks fi rst.) Located in the Central American
Isthmus, Costa Rica somehow escaped the wars and turbulences of the region in the 1980s
and has enjoyed political stability for many decades. In economic terms, Costa Rica does
not differ from the Latin American average. According to the World Bank (2006), its per
capita income is about $4,600 per year, compared with the $3,600 annual average for Latin
America. In terms of equity in income distribution, social security coverage, access to pub-
lic health services and sanitation, labor laws, and protection of the environment, Costa Rica
ranks among the highest in the Americas. Costa Rica has both a mixed economy with open
markets and government control of key areas, such as health, education, banking, energy,
communications, and insurance (Mesa-Lago 2000). The Human Development Index of the
United Nations ranks Costa Rica as 48th in the world and 4th in Latin America (after Chile,
Argentina, and Uruguay).
The country has essentially completed its demographic transition (World Bank 2006).
Its life expectancy is the second highest in the Americas (Canada is fi rst), which is higher
than in the United States. The total fertility rate of 2.00 in 2005 is lower than in the United
Exceptionally High Life Expectancy of Costa Rican Nonagenarians 675
States (2.04 births), and it is the second lowest in Latin America after Cuba. Costa Rica is
also one of the few Latin American countries with a substantial stock of international im-
migrants. Ten percent of the total population is foreign-born, a fi gure not that different from
the 12% foreign-born population of the United States (United Nations Population Division
2006). Because the demographic transition was so quick and recent, a population aging
process has not yet occurred: only 5.6% of Costa Ricans are aged 65 and older, although
this will change very quickly in the next few decades, surpassing 20% by the year 2050
(INEC and CCP 2002).
Offi cial life tables for 1995–2000 (Rosero-Bixby, Brenes-Camacho, and Collado-
Chaves 2004) suggest exceptionally high old-age longevity in Costa Rica. Comparing the
age-80 life expectancy in those tables with 13 high-income countries in exactly the same
period in a database kept at the Max Plank Institute (Kannisto et al. 1994) ,1 Costa Rican
males are the leaders with 8.2 years, followed by Japan with 7.6 years, and Iceland with 7.4
years. Costa Rican females, with 9.0 years of age-80 life expectancy, are in the middle of
this elite pack: for example, Japan has 10.0 years, and Iceland has 8.7 years.
Costa Rica is well known as a country with outstanding health indicators. For ex-
ample, it was included as one of the four study cases in the Rockefeller Foundation report
on “Good Health at Low Cost” (Halstead, Walsh, and Warren 1985). However, there is a
huge difference between having good health indicators and being the world-champion in
longevity, leaving clear grounds for skepticism. Past claims of exceptional longevity in
communities in the Andes and in the Caucasus have not resisted scientifi c scrutiny (Garson
1991). Demographers know well that age exaggeration among the elderly in censuses leads
to substantial infl ation of old-age populations and, consequently, underestimated mortality
rates (Coale and Kisker 1986; Preston, Elo, and Stewart 1999). With these antecedents,
academic circles may disregard this Costa Rican claim as just another case of “bad data.”
This article relies on new data of very high quality to validate the patterns observed in the
life tables and to obtain a more refi ned estimate of late-life longevity.
DATA AND METHODS
In an attempt to avoid data errors that have hampered studies of mortality of the oldest-old
in other populations (Garson 1991; Kannisto 1988), the estimates in this article do not use
conventional data sources. In particular, this article avoids using information on reported
age from censuses or vital statistics but instead uses the Costa Rican national population
voter registry, from which a database was created to study 24,400 Costa Rican nonage-
narians in 1983–2004. This database includes, essentially, extinct birth cohorts born in
1878–1903 and quasi-extinct cohorts born in 1904–1913.
The Supreme Electoral Tribunal (Tribunal Supremo de Elecciones) provided the voter
registry, which includes databases of births, naturalizations, and deaths as well as the voting
lists (the padrón) for the 1990, 1994, 1998, and 2002 elections. The computerized birth reg-
istry, which is supposed to include all ever-living Costa Ricans, includes individuals who
contacted the civil registration system since its computerization in 1970. Individuals con-
tacted the registry because of registration (or certifi cation requests) of vital events such as
births, deaths, or marriages as well as to obtain (or renew every 10 years) an identifi cation
card, or cédula. The databases of the registry are linked by the unique identifi cation (ID)
number that each Costa Rican is given as of birth registration or naturalization. Given that
this ID number also appears on the cédula, it is known in Costa Rica as the cédula number.
A “survival time” data set was created using STATA (Statacorp 2005) with information on
1. The 13 countries are Australia, England and Wales, Finland, France, West Germany, Iceland, Italy, Japan,
the Netherlands, Norway, Sweden, Switzerland, and the United States. Kannisto et al. (1994) judged that these
countries, with the exceptions of the United States and Australia, have “highly reliable data.” The data were taken
from the following Web site: http://www.demogr.mpg.de/databases/ktdb/.
676 Demography, Volume 45-Number 3, August 2008
sex as well as dates of birth, death, and likely place of residence in each election year for
all Costa Rican nonagenarians ever living in 1983–2004; the entry date is January 1, 1983
or the 90th birthday date, and the exit date is the date of death or October 30, 2004. Note
that ages (at death or at any time of observation) in this data set do not come from reports
but rather from computations based on dates documented in the registry.
Data Quality
The quality of the database of nonagenarians is crucial for this article. Three potential biases
may occur and need to be validated: (1) selection bias, if the registry does not include all
individuals and if those excluded have differential mortality; (2) underregistration of deaths,
which would result in an underestimation of death rates, as well as an overcount of individu-
als still alive, especially toward the end of the observation period; and (3) age-misreporting
biases, which, in other studies, underestimate mortality as result of age exaggerations.
How complete is the Costa Rican registry? It is almost impossible that a Costa Rican
adult lived in the country since 1970 without ever having his/her cédula and, thus, never
appearing in the registry. The cédula is required everywhere for all kinds of transactions,
public or private. Besides, no deceased can be buried (keep in mind that the great major-
ity of the studied nonagenarians have died) without a death certifi cate issued by either the
Civil Register offi ces or, in remote locations, by the Rural Guard. Thus, all the dead are in
the database, and a selection bias by exclusion of individuals from the registry is unlikely.
A cross section from the nonagenarian database revealed 5,900 people alive and aged 90
or older at the time of the 2000 census. The census count was 7,000, or about 20% more;
the percentage is similar by sex. This discrepancy does not come from defi ciency in the
registry but from overcounting in the census that is likely due to age exaggeration, as it was
reported in the evaluation of the 2000 census (INEC and CCP 2002). Moreover, a study of
the 1984 census estimated about a 50% overcount of population aged 80 and older, which
is also due to age exaggeration (MIDEPLAN, CELADE, and DGEC 1988).
The second potential bias, the infl ation in the count of people alive if the registry failed
to exclude some of the dead, is addressed by looking at cohorts that should be extinct.
Cohorts born in 1880–1895 were indeed extinct by 2004 in the registry. The maximum
age reached by any of the 24,400 nonagenarians was 109; three died at this age. If under-
registration of deaths occurred, one would see individuals still alive at age 120 or so, which
is not the case. Therefore, estimates for extinct cohorts are, by defi nition, free of error from
death underregistration. In addition, if the analysis showed that mortality in nonextinct
cohorts is not signifi cantly different from mortality in extinct cohorts, it would suggest that
there are not missing deaths in the two groups.
Regarding the third potential bias, as mentioned earlier, this article’s information about
age (at death or at any time during survival) does not come from reports but is instead com-
puted from the dates in birth and death certifi cates, avoiding the most problematic data error
in studies of mortality of the oldest old: age exaggeration that translates to underestimated
mortality. In addition, information within the ID number (cédula) that each Costa Rican
receives at birth allows for a second check of possible birth-date errors. This number is
given to each individual at birth registration (or naturalization) and includes the number of
the ledger and page where the person is registered. Because the ledgers are uniquely and
sequentially numbered since the beginning of the civil registration system in 1880, look-
ing at the ID number can establish the year when each individual was actually registered.
Those with timely registration—say, within a year of their stated birth date—cannot have
their age exaggerated. A person who appears timely registered could make consistent the
two years (birth and registration) only by moving ahead the birth year, never by moving it
back (which would cause the problematic age-exaggeration error). In contrast, those who
registered late (some of them as adults, including foreigners who are citizens by natural-
ization) may have reporting errors in their birth date that produce age exaggeration. For
Exceptionally High Life Expectancy of Costa Rican Nonagenarians 677
example, somebody born in 1920 and registered in 1960 can have a wrong birth year of,
say, 1900, which will result in 20-year age exaggeration. The analysis in this article tests
for signifi cant mortality differences by registration timing.
About 1,100 individuals (or 5%) in the database are centenarians: that is, they were
alive at their 100th birthday. Although this is a small fi gure, it is worth exploring its valid-
ity by checking the reliability criteria used by Kannisto (1988) in his article on centenar-
ians. Two indicators of data reliability in centenarians are applicable to the data in this
article (Kannisto 1988: table 1). First, deaths of those aged 105 and older as a percentage
of deaths at ages 100 or older are expected to be less than 5% and to be lower for men
than for women. In the Costa Rican database, this indicator is 5.2% for men and 5.9% for
women after excluding late registration births. These fi gures are borderline acceptable and
much better than in (for example) the United States (7% and 8% among whites, and 30%
among nonwhites), Spain (10% and 11%), and Portugal (16%) (Kannisto 1988: table 1).
Second, the probability of dying is expected to be higher at age 101 than at age 100, and
the ratio between them (q100 / q101) should be below unity. In the Costa Rican database, this
ratio was 0.94 for men and 0.79 for women, far lower than most populations in the Kan-
nisto article. For example, the ratio is 1.19 and 0.99 in Japan, 1.16 and 1.31 in Spain, 1.02
and 0.97 for whites in the United States, and 1.25 and 1.38 for nonwhites in the United
States (Kannisto 1988: table 1).
Characteristics of the Nonagenarians in the Database
The database of nonagenarians rendered about 101,000 person-years for 24,400 individuals
born from 1878 to 1913 (Table 1). More than two-thirds of the observation segments cor-
respond to the 1994–2004 period. Almost all individuals born before 1904 are deceased;
these individuals were dubbed “extinct cohorts.” Mean age at death is 93.8 years, ranging
from 96.1 in the oldest cohorts to 92.9 in the youngest one. These fi gures and trends are,
however, severely biased by censoring effects: left-censoring for oldest cohorts because
observation starts in 1983, and right-censoring for the youngest because observation stops
in 2004. They are not good indicators of life expectancy. From those born in 1904–1913,
Table 1. Selected Data on Costa Rican Nonagenarians, 1983–2004
Birth Cohort
___________________________________________
Total 1878–1893 1894–1903 1904–1913
Number of Individuals 24,438 2,150 7,692 14,596
Number of Observed Years
Total 101,439 8,778 38,981 53,680
In 1983–1993 33,409 8,611 24,798 0
In 1994–2004 68,030 167 14,183 53,680
Mean Number of Observed Years 4.15 4.08 5.07 3.68
Deceased (%) 73 100 97 57
Mean Age at Death 93.8 96.1 94.4 92.9
Mean Observed Age 92.7 94.9 93.2 92.0
Female Ratio 1.28 1.20 1.24 1.33
Late Birth Registry (%) 17 36 17 13
Central Region (%) 71 76 71 69
Source: National Registry of the Tribunal Supremo de Elecciones.
678 Demography, Volume 45-Number 3, August 2008
43% were still alive at closing date in 2004. Each individual was observed a little more
than four years on average. The mean observed age is about 93 years. The female to male
ratio is 1.28, with an increasing trend in more recent cohorts, which indicates that the sex
gap in mortality is widening. The proportion of late-registered births is 17% overall and
substantially higher (36%) in the cohorts born before 1893. Almost three-fourths of the
observations correspond to the Central region.
Estimates of Mortality and Life Expectancy
This article estimates the mortality rates of Costa Rican nonagenarians using the “extinct
cohort” method, which was developed for European countries by Vincent (1951) and
was used in the United States by (among others) Rosenwaike (1981). The denominators
for the rates in this article based on microdata are exact counts of person-years lived in
each age.2 Survival-time routines in the STATA software facilitated these computations
( Statacorp 2005).
The observed age-sex rates are summarized and smoothed out using a three-parameter
relational model of mortality adapted from the Coale’s model for marital fertility (Coale
1977). The death rate m at age x and sex d (dummy variable equal to 1 for males) is mod-
eled as a function of an old-age standard mortality schedule V of high-income countries;
I refer to this as the Kannisto-Thatcher standard.3 The modeled death rate is a product of
the standard V and the parameters M, denoting the relative level of mortality of females at
age 90; A, representing the effect of aging above and beyond the standard schedule; and S,
representing the effect of sex above and beyond the standard schedule. Values of 1 for the
parameters indicate a behavior identical to the standard schedule. In symbols,
mxd = Vxd MA(x – 90)Sd.
This model is preferable to a parametric hazard regression model for two reasons:
(1) parameters are meaningful in substantive terms and not just in mathematical terms;
and (2) death rates are not forced to follow a mathematical function, such as Gompertz
or Weibull, but are allowed to adjust to patterns observed in other populations; this is an
analytic strategy with a long tradition in demographic modeling that includes the Louis
Henry model of natural fertility, the Ansley Coale models for fertility and nuptiality, and
the William Brass models for survival (Brass 1971; Coale 1977; Henry 1972).
This article estimates model parameters using Poisson regression, following to
Rodríguez and Cleland (1988), who estimated the analogous Coale/Page model for fertility
using this log-linear regression model. Given that the mortality rate m is the ratio of the
count of deaths Y and the number of person-years of exposure N, the Poisson regression
models expected number of deaths E[Y] as the dependent variable:
E[Yxd ] = [NxdVxd] exp[b0 + b1(x – 90) + b2d],
2. For example, a man born on February 1, 1900 and deceased on May 1, 1995 will enter into observation
in 1990. He will contribute 11 person-months to the denominator of the rate in that year and age 90; one person-
month to age 90, year 1991; 11 months to age 91, year 1991; and so on until his fi nal segment of three months
at age 95 and year 1995, which ends in a death and contributes 1 to the rate’s numerator. If this person were still
alive at the end of the observation period on October 30, 2004, his last segment would be 9 months at age 104,
ending in censoring.
3. I averaged the 1992–1998 data for the 13 countries listed in footnote 1 to defi ne an old-age standard
mortality schedule for high-income countries. Appendix Table A1 shows rates in the standard schedule along with
observed rates in the Costa Rican database of 20,000 nonagenarians, which were estimated using an exact account
of person-years in the denominator.
Exceptionally High Life Expectancy of Costa Rican Nonagenarians 679
where the product NV is an offset term (McCullagh and Nelder 1989), and bi are the esti-
mated regression coeffi cients, which when exponentiated, render the M, A, and S parameters,
respectively. The model’s parameters bi were estimated using STATA (Statacorp 2005).4
To investigate mortality covariates, the three parameters of the model are also esti-
mated for subgroups defi ned by variables of interest. Those variables are included as addi-
tional terms in the Poisson regression model as well as their interactions with x and d. The
only additional variables available in the database of nonagenarians are the calendar year
of observation (1983–2004); whether the individual was registered in a timely manner at
birth (an indication that age is error-free); the place of residence, which is a time-varying
variable (for each age-segment, the most recent voting place listed on the padrón), and the
month of birth (a proxy to assess the effect of early life-health conditions). Preliminary
analyses showed that most geographic variations are captured by the dummy variable
“residence in the Central region,” which includes the capital city.
Causes of Death
Six broad groups of causes of death, plus a residual category, are defi ned as follows, with
the codes from the 9th International Classifi cation of Diseases (ICD-9) listed in parentheses
after each cause: (1) communicable diseases (1–139, 460–490); (2) cancer (140–239); (3)
cardiovascular diseases (390–459); (4) Chronic respiratory diseases (491–519); (5) diabe-
tes (250); and (6) accidents and violence (800–999). The information on causes of death
comes from the vital statistics system because the voter registry does not have these data.
Age-specifi c mortality rates were computed for the six groups of causes of death for the
period 1990–1999. For comparative purposes, standardized death rates were also computed
for the United States (white population only) and Sweden for the period 1994–1996. Data
disaggregated by age and causes of death were not readily available for ages 85 or older
in these or other countries. The comparison thus refers to the group aged 85 or older and
uses the “indirect” procedure of standardization (Shryock and Siegel 1976), with the Costa
Rican rates as the standard.5
RESULTS
The age-specifi c death rates from the Costa Rican database are substantially lower than
the average of 13 high-income countries (listed in footnote 1)—the Kannisto-Thatcher
4. The data set for estimating the model is of the survival-time type, with censoring at the end of 2004 and
entry to observation at the 90th birthday (or January 1983 for individuals older than 90 and alive at that time). Each
observation was split into single age units to properly model the effect of age as time-varying covariate. The use
of Poisson regression for grouped data generated with the STATA command “strate” is a logical choice because
the dependent variable is a count of deaths in each age and the exposure is the number of person-years observed.
A problem with these grouped data is that the sample size is infl ated: each individual is counted several times, one
in each age until death. Standard errors were estimated with STATA regression models using individual-level data
and “robust” estimates, which take into account that information has been replicated for each person. Statisticians
have been using Poisson regression to fi t survival models for decades, and even Cox’s partial likelihood approach
has been shown to be a form of Poisson regression (Clayton and Cuzick 1985; Whitehead 1980). In the present
case, it is not necessary to prove the count is Poisson (the 0–1 death outcome at the individual level might not be)
but just that the likelihoods of the survival and Poisson models are equivalent, which has been demonstrated by
Holford (1980) and by Laird and Olivier (1981). Some statisticians refer to this approach to fi t survival models as
the “Poisson trick.” Regarding the naive issue of equality of mean and variance as requirement to use Poisson, it is
true that the variance equals the mean in a Poisson distribution, but estimates obtained by maximizing the Poisson
likelihood are optimal under the weaker condition that the variance is proportional to the mean, which is another
standard result in generalized linear models (Wedderburn 1974). The standard errors are typically underestimated
with overdispersed data, but one can estimate the proportionality factor via Pearson’s chi-square or by using robust
standard errors, as done here.
5. The specifi c data sources on causes of death were as follows: for Costa Rica, the death data base provided
by the National Statistical and Census Institute (INEC) and available online at http://censos.ccp.ucr.ac.cr/; for the
United States, the WONDER system of the Centers for Disease Control at http://wonder.cdc.gov/; and for Sweden,
information provided by Professor Charli Eriksson from data in the Swedish National Institute of Public Health.
680 Demography, Volume 45-Number 3, August 2008
standard (see Figure 1). The Costa Rican advantage is larger for males, which means a
narrower sex gap in Costa Rica, analogous to one observed in Sardinia, Italy (Robine
et al. 2006). The rates increase with age, with a slope resembling that of the standard.
In populations with bad data at these ages, one usually observes fl at curves. There is
some deceleration in the increases at advanced ages—a phenomenon observed in other
populations and species as well, which is the subject of intense scrutiny (Horiuchi and
Wilmoth 1998).
The three-parameter model provides a reasonable adjustment of the Costa Rican rates
in Figure 1. Smoothing the rates with the model seems necessary to eliminate large, random
fl uctuations. The 95% confi dence intervals (CI) illustrate that the observed rates become
highly unreliable by age 98 and beyond because of random errors originated in small num-
bers of observations. The relational model in this article corrects these probably random
fl uctuations and purposely imposes a monotonic pattern of increasing rates with age, as
observed in 13 developed populations.
Table 2 shows the three parameters of the mortality model estimated for Costa Rican
nonagenarians and used to smooth the rates in Figure 1. The M parameter is estimated at
0.829; that is, Costa Rica has 17% lower mortality at age 90 than the Kannisto-Thatcher
standard for high-income countries. The A parameter is estimated at 0.989; aging occurs
more slowly in Costa Rica than in the standard, at a rate 1.1% slower for each extra year of
age. The S parameter came out as 0.878; Costa Rican males have an additional advantage
of 12% lower-than-expected death rates.
Figure 1. Observed and Adjusted Age-Specifi c Death Rates: Costa Rica (1983–2004) and Kannisto-
atcher Average (1990–1999)
200
400
600
800
1,000
Death Rate (log scale)
90 95 100 105
Age
Males
90 95 100 105
Age
Females
Adjusted
Kannisto
Observed
95% Confidence interval
Exceptionally High Life Expectancy of Costa Rican Nonagenarians 681
Table 2. ree Models Describing Mortality of Costa Rican Nonagenarians Estimated With Poisson
Regression, Robust Estimates: Models Diff er by the Explanatory Variables Included
Model 1 Model 2 Model 3
_______________________ ______________________ _______________________
95% 95% 95%
Confi dence Confi dence Confi dence
Parameter Interval Parameter Interval Parameter Interval
M Level 0.829 0.808–0.851 0.847 0.814–0.882 0.859 0.833–0.886
S Male = 1 0.878 0.853–0.905 0.888 0.862–0.915 0.874 0.848–0.904
A Age (90 = 0) 0.989 0.984–0.994 0.991 0.985–0.997 0.990 0.984–0.997
Eff ects on M
Late registry 0.985 0.947–1.026 0.937 0.887–0.990
Year (1995 = 0) 0.991 0.987–0.996 0.996 0.992–1.000
Non-Central region 0.892 0.863–0.922 0.851 0.809–0.894
Non-extinct cohort 1.005 0.955–1.058
Born in March or April 1.036 0.996–1.078 1.070 1.009–1.133
Eff ects on S
Late registry 1.088 1.006–1.177
Eff ects on A
Year (1995 = 0) 0.998 0.997–0.999
Non-Central region 1.013 1.001–1.024
Born in March or April 0.990 0.977–1.004
Eff ects on Year
Non-Central region 1.007 1.001–1.013
Notes: e eff ects on sex, age and year were estimated by including the respective interaction variable in the model. See the
text for an explanation of the parameters.
Table 2 shows estimates for two additional models. The second model allows for varia-
tion in mortality levels (the M parameter) with fi ve additional variables. It is useful just to
show that there is no signifi cant difference in mortality of extinct and non-extinct cohorts,
which is an assurance that there are no death underregistration errors. By including statisti-
cal interactions,6 the third model also allows variation in the sex and age effects. Being a
late-birth registry (a proxy for potential age errors) reduces mortality by 6%, but this ef-
fect occurs only among women, as shown by the interaction effect with sex. There is also
a signifi cant trend of mortality reduction over time of 0.4% per year, but this trend occurs
only in the Central region (as shown by the region-year interaction) and dissipates at older
ages (age-year interaction). The non-Central regions have 15% lower mortality by 1995.
Given that those regions are the most remote and least developed, one wonders whether
this apparent advantage may come from data errors. The advantage of non-Central regions,
however, disappears at older ages and more recent times, as shown by the corresponding
interactions. Finally, those born in March or April have a 7% higher mortality at age 90,
a disadvantage that diminishes with age. Analogous effects of month of birth observed in
other populations have been taken as indication of the direct correlation between early-life
conditions (in utero and neonatal) and old-age health, linked to the shortage of food during
6. Only signifi cant effects are included in the third regression. Signifi cance was tested by looking at the varia-
tion in the log-likelihood ratio when the variable and its relevant interactions are included in the model.
682 Demography, Volume 45-Number 3, August 2008
winter.7 Costa Rica does not see that kind of food shortage, but seasonality does occur for
other factors, particularly those linked to the dry season that goes from January to April. In
particular, conceptions and diarrhea used to peak in January and February, months in which
people also used to be very busy harvesting coffee and celebrating the extra income from
this and other harvests and the dry season. It may be that infections in the fi nal months of
pregnancy that hampered in utero development affect the health of these newborn babies
even when they reach very old ages.
Restricting the analysis only to the Central region and timely registered births assures
high-quality estimates, although these may be conservative. With these two restrictions, the
parameters for aging (A) and sex (S) are about the same as in the simple model presented
earlier. The parameter for mortality level (M) is a bit higher, and the advantage for Costa
Rica declines from 17% to 14%. The exceptional longevity of Costa Ricans does not seem
sensitive to this refi nement. Figure 2 shows life expectancy by age originally estimated
with the observed death rates, as well as that estimated with the rates from the model and
restricted to timely registered births and the Central region. No important differences are
seen between the two series up to age 102. The fi gure also shows that while Costa Rican
females differ little from those in Japan and the United States, Costa Rican nonagenarian
7. Adult mortality is higher for those born in spring: April to June in the northern hemisphere, and October
to December in the southern hemisphere (Doblhammer 2004).
Note: “Costa Rica, corrected” refers to Costa Rican estimates based on death rates smoothed with the regression model and
restricted to timely registered births and residence in Central region.
Figure 2. Life Expectancy, by Age and Sex: Costa Rica (1983–2004), the United States (whites only,
1990–1995), and Japan (1990–1995)
1
2
3
4
5
90 95 100 105 90 95 100 105
Males Females
Costa Rica, observed Costa Rica, corrected United States Japan
Life Expectancy (years)
Age Age
Exceptionally High Life Expectancy of Costa Rican Nonagenarians 683
males have a one-half year advantage in life expectancy at all ages. By age 100, Costa
Rican males have 2.7 years of life expectancy, and females have 2.8 years; in the United
States (whites only), males have 2.2 years, and females have 2.4 years. However, after
about age 100, comparisons must be taken cautiously because of uncertainties originated
in the small number of observations in Costa Rica and the sensitivity of the estimates to
small variations in computation methods or in data errors.
Age-90 life expectancy—a summary of the mortality of nonagenarians—in this new
data set resulted in 4.7 and 4.4 years for women and men, respectively—fi gures almost
identical to those in the offi cial life tables for 1995–2000. Figure 3 compares my estimate
for Costa Rica for the period 1983–2004 (central year 1994) with high-income countries
in the aforementioned Kannisto-Thatcher database for the period 1992–1998 (with 1995
as the central year). The estimates in the fi gure are thus contemporary to the same period.
Costa Rican males have the highest life expectancy, which is one-half year more than
the United States, Japan, Australia, and Iceland. Costa Rican females are essentially tied
in fi rst place with Japan and the United States. The sex gap in life expectancy is notori-
ously smaller in Costa Rica: 0.3 years at age 90 compared with 1.1 year for France or the
United States (see Figure 3).
Data on causes of death may help to understand the Costa Rican advantage. Cardiovas-
cular diseases (CVDs) are, by far, the leading cause of death, accounting for nearly 50%
of all deaths of nonagenarians. Chronic respiratory diseases (mostly “other chronic airway
Figure 3. Age-90 Life Expectancy, by Sex, for Selected Countries Ordered by Female Life Expectancy:
Costa Rica (1983–2004) and Other Countries (1990–1995)
012345
Age-90 Life Expectancy (in years)
Finland
West Germany
Norway
Netherlands
Sweden
Switzerland
Iceland
England and Wales
Italy
France
Australia
United States
Costa Rica
Japan
Males Females
684 Demography, Volume 45-Number 3, August 2008
obstructive diseases”), communicable diseases (mostly bronchopneumonia and pneumo-
nia), and cancer have similar importance, each accounting about 12% of old-age deaths. A
comparison with the United States and Sweden in Figure 4 points out that the Costa Rican
advantage is mostly due to its lower CVD mortality. The age-adjusted rate of mortality
from CVD, at ages 85 and older, is 20% lower in Costa Rica than in the United States
and 30% lower than in Sweden. In turn, mortality by communicable diseases is similar to
that in the United States and lower than in Sweden. By contrast, Costa Rican elders have
substantially higher mortality from chronic respiratory diseases and accidents (huge rate
ratios on the order of 200%–400%). Cancer is another pathology from which Costa Ricans
have slightly higher mortality rates than do U.S. and Swedish citizens (about 15% higher),
mostly attributable to stomach cancer.
DISCUSSION
Fresh data from a population registry kept in Costa Rica for voting purposes confi rms
early estimates of exceptional longevity of its elders. Life expectancy for nonagenarian
males is one-half year more in Costa Rica than in any other country, with reliable statistics.
Although this life expectancy is still less than that for females, the difference is only 0.3
years, which is the smallest recorded by national populations at these mortality levels.
Figure 4. Mortality Rate Ratio by Cause of Death in Costa Rica Relative to the United States
and Sweden for ose Aged 85 and Older in the 1990s: Indirect Standardization by Age
and Sex
.5 1 2 4
Costa Rican Mortality Rate Ratio
Residual
Cardiovascular diseases
Communicable diseases
All causes
Diabetes
Cancer
Accidents and violence
Respiratory diseases
Relative to the United States Relative to Sweden
Exceptionally High Life Expectancy of Costa Rican Nonagenarians 685
Are these fi gures valid or just a product of “bad data”? How could they be valid,
considering that well-being and health services in Costa Rica are far behind those in high-
income countries?
This article is mostly concerned with the validity of the estimates. By not basing the
estimates on self-reported age data, it avoids age exaggeration, which is the most problem-
atic and pervasive data error at old ages. By double-checking birth dates and excluding
individuals whose birth dates are fully documented but whose registration did not occur
close to birth, almost no possibility of age errors exists. The only possibility of error would
come from massive identity suplantation of older and deceased individuals by younger
ones. Such a massive fraud does not seem plausible.
Estimates in this article for extinct cohorts are free from death underregistration errors:
cohorts were not extinct if there was a failure to register some deaths. Independent census
data, which are also free of missing-death errors, confi rm that survival of Costa Rican
non agenarians may be exceptional. Table 3 shows that although the percentage of non-
agenarians in the total population is not impressive (0.2%) because of very rapid population
growth in the past, the nonagenarian rate is exceptionally high for Costa Rican males: more
than twofold those of France, Italy, Sweden, or the United States, and four times higher than
in Russia. Following common practice in demographic analyses of centenarians (see, e.g.,
Robine and Paccaud 2005), the nonagenarian rate was defi ned as the ratio of the population
aged 90 and older in 2000 to the population 60 and older in 1970: that is, a 30-year cohort
survival proportion among the elderly and assuming null migration. A problem with these
census data is age exaggeration that might infl ate the nonagenarian rate. An independent
evaluation of the 2000 census found that age exaggeration did indeed exist, especially
among the oldest-old.8 Correction of the age-reporting errors reduces the Costa Rican nona-
genarian rate from 6.9% to 5.6% for males (Table 3, second row), but this fi gure continues
to be more than twice as large as in France, Italy, and Sweden. The nonagenarian sex ratio
in Table 3 is also exceptional in Costa Rica: there are only 30% or 40% more women than
8. I compared the census-reported age with the age in the national identifi cation card—the cédula—in a
sample of 7,400 seniors. Among those aged 90 and older, about 30% of the individuals exaggerated their age by
more than six years on average, compared with 10% by those in their 60s (Rosero-Bixby et al. 2004).
Table 3. Nonagenarians in the Population, Cohort Nonagenarian Rate,a and Female to
Male Ratio in Nonagenarians: Costa Rica and Selected Countries, Circa 2000
In Population (%) Cohort Rate (%)
Female
____________________ ___________________
Country Female Male Female Male Ratio
Costa Rica, Observed 0.22 0.17 9.2 6.9 1.4
Costa Rica, Corrected 0.19 0.14 7.4 5.6 1.3
France 1.09 0.33 6.1 2.5 3.5
Italy 0.90 0.31 5.4 2.3 3.0
Japan 0.77 0.26 8.3 3.3 3.0
Russia 0.39 0.10 2.8 1.6 4.2
Sweden 1.05 0.37 5.6 2.2 2.9
United States 0.78 0.26 6.9 2.9 3.1
Sources: Data for Costa Rica are from the country’s 2000 census, observed and corrected fi gures, correction in
(INEC and CCP 2002). Data for other countries are from the Human Mortality Database (http://www.mortality
.org).
aNonagenarian rate = the population aged 90 and older in 2000 / Population aged 60 and older in 1970.
686 Demography, Volume 45-Number 3, August 2008
men, compared with the 200% or 300% excess of women in the other countries in the table.
This result corroborates this article’s fi nding that the mortality sex gap among Costa Rican
nonagenarians is substantially smaller than in other countries. It is reassuring to reach the
same result with two independent data sources. This Costa Rican peculiarity has also been
observed in the island of Sardinia of Italy (Robine et al. 2006).
According to the World Bank (2006), by 2004, Costa Rica had a per capita gross na-
tional income of about US$4,700 and a health expenditure of $310. These fi gures are about
one-tenth those in high-income countries. In the United States, these amounts were $41,400
and $5,700, respectively. Indicators of health services, such as per capita physicians and
hospital beds, are also substantially lower in Costa Rica: they equate to only one-third the
number of U.S. physicians and one-tenth the number of Japanese beds. It is perplexing that
a country with these modest levels of well-being, health investments, and infrastructure
may be the one with the highest life expectancy among the elderly.
Broad explanations of Costa Rica’s health achievements in the literature include
the orientation of the government toward equity and social development, with large so-
cial investments being possible, in part, because of the absence of military expenditures
(Rosero-Bixby 1991). The 1949 constitution abolished the armed forces. Investments in
education and the very high coverage of health insurance are often mentioned as key fac-
tors (Caldwell 1986). Health insurance covers 82% of the population, including the 9%
population deemed destitute, whose insurance is paid by the government (Rosero-Bixby
2004). Provision of primary health care, particularly to remote or poor populations, has
a quantifi able impact on death rates, especially among children (Rosero-Bixby 1986). A
17-year follow-up of a group of Costa Rican elderly has shown no meaningful differences
in survival by socio economic condition (education or wealth) nor by being covered by the
national health insurance9 (Rosero-Bixby, Dow, and Lacle 2005); this suggests that the
Costa Rican advantage at old ages may be present across the entire society, with no clear-
cut health interventions or classic socioeconomic gradients as explanation.
Data on causes of death suggested that the Costa Rican advantage comes mostly
from CVDs. However, the comparison with Sweden and the United States must be taken
cautiously because differences may be an artifact from variations in how causes of death
are registered in each country, as well as from age misreporting errors or possibly under-
registration of deaths. The data for this comparison are regular data from vital statistics,
which are good albeit not perfect in Costa Rica. However, it is worth noting that an early
study (Rosero-Bixby 1996) among young adults found similar patterns. For example, it
found that mortality by heart disease among males aged 25–74 is 42% lower in Costa Rica
than in the United States. The CVD advantage of Costa Rican males does not seem to oc-
cur only among the oldest old, and this advantage is so large that its being the result of bad
data is hard to believe.
In Sardinia, another place with exceptional old-age longevity among males and a small
sex gap, Caselli and Lipsi (2006) also found that low CVD mortality explains the survival
advantage of elderly Sardinians compared with other Italians.
Another suggestive result regarding causes of death is that old-age mortality by com-
municable diseases is similar to that in the United States and is lower than in Sweden. This
result somehow confi rms the high level of development of the current Costa Rican primary
health care system, which other studies have identifi ed as an important factor for the low
mortality at earlier ages in Costa Rica (Rosero-Bixby 1986, 2004).
One can safely assume that the lower CVD mortality of elderly Costa Rican males
does not come from access to superb health care. Costa Rica has a good health care sys-
tem, especially at the primary level, with almost universal coverage. However, the Costa
9. There seems to be a selection bias in this lack of insurance effect because the frail may tend to seek out
insurance coverage more frequently.
Exceptionally High Life Expectancy of Costa Rican Nonagenarians 687
Rican health care system is not comparable with the health infrastructure of Sweden and
the United States, especially considering the access to health care that Medicare provides to
the elderly in the United States. So what are the preventive, genetic, or behavioral factors
that protect Costa Ricans from high CVD rates? Table 4 provides some hints by compar-
ing selected markers from National Health and Nutrition Examination Survey (NHANES)
2001–2002 in the United States and results from an ongoing study, Costa Rica: Estudio
de Longevidad y Envejecimiento Saludable (Costa Rican Study of Longevity and Healthy
Aging; CRELES). Smoking, past and present, is not a factor among males, nor is high
blood pressure or elevated cholesterol or triglycerides levels. It does not seem that Costa
Ricans have the genes or a diet that reduce these risk factors. The only lowered risk fac-
tor for which Costa Rican males have a clear advantage is a lesser prevalence of obesity.
Prevalence of obesity in Costa Rican males is two-thirds that found in the United States.
This probably results in the signifi cantly lower prevalence of uncontrolled diabetes in males
as measured by the glycohemoglobin level, the only other factor in Table 4 that shows a
Costa Rican advantage. Other factors that may be worth investigating are levels of stress,
support networks, and the like.
This article does not have an answer to the question of why elderly Costa Ricans do so
well. It could be a genetic factor, lifestyle, social factors, or the environment. It could also
be just a heterogeneity in frailty effect. Costa Rican nonagenarians are true survivors of
cohorts that underwent extremely harsh health conditions when young. For example, they
survived infant mortality rates in the range of 250 per thousand prevalent in Costa Rica in
the early twentieth century. Malaria, tuberculosis, and diarrheic diseases decimated these
Table 4. Proportion of ose Aged 60–90 Who Suff er From Selected Risk Factors, by Sex: Costa
Rica (2005) and United States (2001–2002)
Males Females
_____________________ _____________________
Costa United Costa United
Risk Factor Rica States Rica States
Obese: BMI ≥ 30 .16 .22 * .31 .27
Waist ≥ 94/80cm, Male/Female .48 .80 * .86 .87
Ever Smoked .68 .66 * .22 .42 *
Currently Smokes .14 .07 * .04 .06 *
High Blood Pressure, Diastolic > 90 .37 .04 * .41 .03 *
High Blood Pressure, Systolic > 140 .67 .38 * .69 .51 *
HDL Cholesterol ≤ 40/50 mg/dl, Male/Female .46 .31 * .59 .28 *
Total Cholesterol > 250 .14 .07 * .26 .16 *
Triglycerides ≥ 150 mg/dl .43 .44 .48 .50
Glycohemoglobin ≥ 6.5% .12 .14 * .20 .11 *
Average N 1,176 557 1,410 607
Mean Age 75 75 75 75
Notes: Figures are age-adjusted proportions with logistic regression to age 75. Figures in gray boxes indicate risk factors for
which Costa Ricans have a signifi cant advantage.
Sources: For the United States, data come from the Centers for Disease Control (CDC), National Health and Nutrition
Examination Survey, 2001–2002 (http://www.cdc.gov/nchs/nhanes.htm). For Costa Rica, the data come from the study “Costa
Rican Study of Longevity and Healthy Aging” (CRELES), Centro Americano de Población, Universidad de Costa Rica (http://
ccp.ucr.ac.cr/creles).
*Diff erence is signifi cant at p < .05.
688 Demography, Volume 45-Number 3, August 2008
cohorts when they were young. It looks like the selection-of-the-fi ttest effect prevailed over
the weakening effect. In addition, modern health evils, such as obesity and a sedentary
lifestyle, are less common among them. Finally, a reasonably good health care system is
currently protecting them from dying of communicable diseases.
These explanations, however, say nothing regarding why the Costa Rican advantage
occurs mostly among males, or why the sex gap in mortality is so small. The only thing
known so far is that this population exhibits low cardiovascular mortality and that Costa
Rican males of these ages are thin. Comparatively, Costa Rican women tend to be obese,
which perhaps is due to their high fertility in the past; each extra pregnancy usually increases
mother’s weight, as shown, for example, by Arroyo et al. (1995) for Mexican women.
If the high longevity of elderly Costa Ricans is mostly a result of a selection process
of the less frail, this may be an ephemeral advantage that may disappear as more frail indi-
viduals reach old ages, thanks to the rapid progress that took place in the past.10 Analyses
in other low-income countries with adequately robust data are needed to see whether the
early harsh conditions generally faced 80–100 years ago in these countries lead to similar
patterns of low mortality at old ages.
10. Life expectancy at birth in Costa Rica rose from 46 to 63 years from 1940–1960, which means a gain of 19
hours of life every single day in a 20-year period. In the 1970s, there were again gains at the same staggering speed,
which raised life expectancy to 73 years in 1980. In 2000, life expectancy was 78 years (Rosero-Bixby 2004).
Appendix Table A1. Age and Sex Mortality Rates per 1,000 Population
Kannisto- atcher Costa Rica (1983–2004)
_____________________ _____________________________________________
Age Males Females Males Na Females Na
90 231 178 166 9,391 149 11,780
91 253 198 181 7,704 160 9,790
92 277 220 195 5,925 175 7,652
93 302 243 211 4,529 203 5,891
94 328 268 220 3,430 211 4,419
95 357 295 271 2,505 227 3,286
96 387 323 264 1,781 259 2,359
97 419 352 285 1,236 284 1,656
98 453 382 351 844 303 1,123
99 489 412 326 561 303 756
100 526 444 335 362 274 515
101 556 482 355 236 346 344
102 582 495 405 146 358 218
103 635 518 365 77 397 128
104 721 553 246 49 330 79
105 853 604 274 29 369 49
106 1,054 672 344 17 441 25
aN = person-years observed.
Exceptionally High Life Expectancy of Costa Rican Nonagenarians 689
REFERENCES
Arroyo, P., H. Avila-Rosas, V. Fernandez, E. Casanueva, and D. Galvan. 1995. “Parity and the
Prevalence of Overweight.” International Journal of Gynaecology and Obstetrics 48:269–72.
Barbi, E., G. Caselli, and J. Vallin. 2003. “Trajectories of Extreme Survival in Heterogeneous
Populations.” Population (English ed.) 58:43–65.
Barbi, E. and J.W. Vaupel. 2005. “Comment on ‘Infl ammatory Exposure and Historical Changes in
Human Life-Spans.’” Science 308:1743.
Brass, W. 1971. “On the Scale of Mortality.” Pp. 69–110 in Biological Aspects of Mortality. London:
Taylor and Francis Ltd.
Caldwell, J.C. 1986. “Routes to Low Mortality in Poor Countries.” Population and Development
Review 12:171–220.
Caselli, G. and R.M. Lipsi. 2006. “Survival Differences Among the Oldest Old in Sardinia: Who,
What, Where, and Why?” Demographic Research, Vol. 14, article 13:267–94. Available online at
http://www.demographic-research.org/volumes/vol14/13.
Clayton, D.G. and J. Cuzick. 1985. “The EM Algorithm for Cox’s Regression Model Using GLIM.”
Applied Statistics 34:148–56.
Coale, A.J. 1977. “The Development of New Models of Nuptiality and Fertility.” Population numéro
spécial:131–54.
Coale, A.J. and E.E. Kisker. 1986. “Mortality Crossovers: Reality or Bad Data.” Population Studies
40:389–401.
Cockerham, W.C. and Y. Yamori. 2001. “Okinawa: An Exception to the Social Gradient of Life
Expectancy in Japan.” Asia and Pacifi c Journal of Clinical Nutrition 10:154–58.
Doblhammer, G. 2004. The Late Life Legacy of Very Early Life. Berlin: Springer-Verlag.
Elo, I.T., C.M. Turra, B. Kestenbaum, and B.R. Ferguson. 2004. “Mortality Among Elderly Hispanics
in the United States.” Demography 41:109–20.
Finch, C.E. and E.M. Crimmins. 2004. “Infl ammatory Exposure and Historical Changes in Human
Life.” Science 305:1736–39.
———. 2005. “Response to Comment on ‘Infl ammatory Exposure and Historical Changes in Human
Life-Spans.’” Science 308:1743.
Garson, L.K. 1991. “The Centenarian Question: Old-Age Mortality in the Soviet Union, 1897 to
1970.” Population Studies 45:265–78.
Halstead, S.B., J.A. Walsh, and K.S. Warren. 1985. “Good Health at Low Cost.” Proceedings
of a Conference Held at the Bellagio Conference Center. New York: The Rockefeller Founda-
tion.
Henry, L. 1972. On the Measurement of Human Fertility: Selected Writings. Amsterdam: Elsevier.
Hill, K., R. Pande, M. Mahy, and G. Jones. 1999. Trends in Child Mortality in the Developing World:
1960 to 1996. New York: UNICEF.
Holford, T.R. 1980. “The Analysis of Rates and of Survivorship Using Log-Linear Models.” Biomet-
rics 36:299–305.
Horiuchi, S. and J.R. Wilmoth. 1998. “Deceleration in the Age Pattern of Mortality at Older Ages.”
Demography 35:391–412.
Instituto Nacional de Estadística y Censos (INEC) and Centro Centroamericano de Población (CCP).
2002. Costa Rica: Estimaciones y proyecciones de población 1970–2050 actualizadas al año 2000
y evaluación del censo del 2000 y otras fuentes de información [Costa Rica: Population estimates
and projections 1970–2050, updated by the year 2000, and evaluation of the 2000 census and other
data sources]. San José, Costa Rica: INEC.
Kannisto, V. 1988. “On the Survival of Centenarians and the Span of Life.” Population Studies
42:389–406.
Kannisto, V., J. Lauritsen, A.R. Thatcher, and J.W. Vaupel. 1994. “Reductions in Mortality at Ad-
vanced Ages: Several Decades of Evidence From 27 Countries.” Population and Development
Review 20:793–810.
690 Demography, Volume 45-Number 3, August 2008
Khlat, M. and N. Darmon. 2003. “Is There a Mediterranean Migrants Mortality Paradox in Europe?”
International Journal of Epidemiology 32:1115–18.
Laird, N. and D. Olivier. 1981. “Covariance Analysis of Censored Survival Data Using Log-linear
Analysis Techniques.” Journal of the American Statistical Association 76:231–40.
Manton, K.G., E. Stallard, and J.W. Vaupel. 1981. “Methods for Comparing the Mortality Experience
of Heterogeneous Populations.” Demography 18:389–410.
McCullagh, P. and J.A. Nelder. 1989. Generalized Linear Models. London: Chapman and Hall.
Mesa-Lago, C. 2000. Market, Socialist, and Mixed Economies: Comparative Policy and Performance—
Chile, Cuba, and Costa Rica. Baltimore, MD: Johns Hopkins University Press.
Ministerio de Planifi cacion Nacional y Politica Economica (MIDEPLAN), U.N. Centro Latinoameri-
cano de Demografi a (CELADE), and Direccion General de Estadistica y Censos (DGEC). 1988.
Costa Rica. Estimaciones y Proyecciones de Población 1950–2025 [Costa Rica. Population Esti-
mates and Projections 1950–2025]. San José, Costa Rica: Imprenta Nacional.
Palloni, A. and J. Morenoff. 2001. “Interpreting the Paradoxical in the Hispanic Paradox: Demograph-
ic and Epidemiological Approaches.” Annals of the New York Academy of Sciences 954:140–74.
Preston, S.H., I.T. Elo, and Q. Stewart. 1999. “Effects of Age Misreporting on Mortality Estimates at
Older Ages.” Population Studies 53:165–77.
Robine, J.M., G. Caselli, D. Rasulo, and A. Cournil. 2006. “Differentials in the Femininity Ratio
Among Centenarians: Variations Between Northern and Southern Italy From 1870.” Population
Studies 60:99–113.
Robine, J.M. and F. Paccaud. 2005. “Nonagenarians and Centenarians in Switzerland, 1860–2001: A
Demographic Analysis.” Journal of Epidemiology and Community Health 59:31–37.
Rodríguez, G. and J. Cleland. 1988. “Modelling Marital Fertility by Age and Duration: An Empirical
Appraisal of the Page Model.” Population Studies 42:241–57.
Rosenwaike, I. 1981. “A Note on New Estimates of the Mortality of the Extreme Aged.” Demography
18:257–66.
Rosero-Bixby, L. 1986. “Infant Mortality in Costa Rica: Explaining the Recent Decline.” Studies in
Family Planning 17:57–65.
———. 1991. “Socioeconomic Development, Health Interventions, and Mortality Decline in Costa
Rica.” Scandinavian Journal of Social Medicine (Suppl.) 46:33–42.
———. 1996. “The Decline in Adult Mortality in Costa Rica.” Pp. 166–95 in Adult Mortality in Latin
America, edited by J. Chackiel, L. Ruzicka, and I. Timœus. Oxford, United Kingdom: Oxford
University Press.
———. 2004. “Evaluación del impacto de la reforma del sector salud en Costa Rica” [Impact
evaluation of the health sector reform in Costa Rica]. Revista Panamericana de salud Pública
15:94–103.
Rosero-Bixby, L., G. Brenes-Camacho, and A. Collado-Chaves. 2004. “Tablas de vida para cálculo
actuarial de rentas vitalicias y retiro programado. Costa Rica circa 2000” [Life tables for actuarial
computations of life-long rents and programmed retirement. Costa Rica circa 2000]. Población
y Salud En Mesoamérica (Revista Electrónica) Vol. 1, article 4. Available online at http://ccp.ucr
.ac.cr/revista/index.htm.
Rosero-Bixby, L., W.H. Dow, and A. Lacle. 2005. “Insurance and Other Determinants of Elderly
Longevity in a Costa Rican Panel.” Journal of Biosocial Sciences 37:705–20.
Roth, G.S., M.A. Lane, D.K. Ingram, J.A. Mattison, D. Elahi, J.D. Tobin, D. Muller, and E.J. Metter.
2002. “Biomarkers of Caloric Restriction May Predict Longevity in Humans.” Science 297:811.
Shryock, H.S. and J.S. Siegel. 1976. The Methods and Materials of Demography. New York: Aca-
demic Press.
Statacorp. 2005. Stata Statistical Software: Release 9.0. College Station, TX: Stata Corporation.
United Nations. 1961. Demographic Yearbook. New York: United Nations.
United Nations Population Division. 2006. Trends in Total Migrant Stock: The 2005 Revision. New
York: United Nations.
Exceptionally High Life Expectancy of Costa Rican Nonagenarians 691
Vaupel, J.W. and J.R. Carey. 1993. “Compositional Interpretations of Medfl y Mortality.” Science
260:1666–67.
Vaupel, J.W., J.R. Carey, K. Christensen, T.E. Johnson, A.I. Yashin, N.V. Holm, I.A. Iachine, V. Kan-
nisto, A.A. Khazaeli, P. Liedo, V.D. Longo, Y. Zeng, K.G. Manton, and J.W. Curtsinger. 1998.
“Biodemographic Trajectories of Longevity.” Science 280:855–60.
Vaupel, J.W., K.G. Manton, and E. Stallard. 1979. “The Impact of Heterogeneity in Individual Frailty
on the Dynamics of Mortality.” Demography 16:439–54.
Vincent, P. 1951. “La mortalité des Vieillards” [The mortality of old people]. Population 6:
181–204.
Wedderburn, R.W.M. 1974. “Quasi-Likelihood Functions, Generalized Linear Models, and the Gauss-
Newton Method.” Biometrika 61:439–47.
Whitehead, J. 1980. “Fitting Cox’s Regression Model to Survival Data Using GLIM.” Applied Sta-
tistics 29:268–75.
World Bank. 2006. “World Development Report 2006: Equity and Development.” New York: Oxford
University Press.
Yashin, A.I. and I.A. Iachine. 1997. “How Frailty Models Can Be Used for Evaluating Longevity
Limits: Taking Advantage of an Interdisciplinary Approach.” Demography 34:31–48.
