- Email: [email protected]
Life Table
Life Management, Deelopmental Psychology of Lerner R M, Busch-Rossnagel A (eds.) 1981 Indiiduals as Producers of Their Deelopment: A Life-span Perspectie. Academic Press, New York Mischel W 1968 Personality and Assessment. Wiley, New York Neugarten B L 1968 The awareness of middle age: a reader in social psychology. In: Neugarten B L (ed.) Middle Age and Aging. University of Chicago Press, Chicago, pp. 22–8 Staudinger U M, Marsiske M, Baltes P B 1995 Resilience and reserve capacity in later adulthood: potentials and limits of development across the life span. In: Cicchetti D, Cohen D (eds.) Deelopmental Psychopathology. Vol. 2: Risk, Disorder, and Adaptation. Wiley, New York, pp. 801–47 Watson J B 1925 Behaiorism. Norton, New York
A. M. Freund
Life Table The life table is perhaps the oldest tool developed for the analysis of survival patterns in human and other populations. Its roots go back at least to the famous book on the English Bills of Mortality that John Graunt published in 1662, a book that is sometimes cited as the beginning of systematic statistical science. Life table techniques are described in detail in most introductory textbooks on the methods of actuarial statistics, biostatistics, demography, and epidemiology (see, e.g., Chiang (1984), Elandt-Johnson and Johnson (1980), Manton and Stallard (1984), Preston et al. (2001). In this article the basic notions of life-table construction are explained with a focus on issues that are sometimes underemphasized in other accounts. More of the underlying mathematical theory can be found in Hoem (1998) and in the textbooks. For the history of the topic, consult Seal (1977), Smith and Keyfitz (1977), and Dupa# quier (1996).
1. A Cohort Life Table Classical presentations of life-table techniques typically start with data in a format like that of the first four columns in Table 1. These particular data come from the Eritrean Demographic and Health Survey of 1995 and show the survival up through their fourth birthday of 7,098 girls whose births and deaths were recorded in the survey. This data set is different from normal textbook examples in two ways. First, the data are given for intervals of single-month segments instead of the usual single-year intervals. Second, the present data are subject to strong heaping in the reported age at death (for the girls that died). In part this is due to a typical inherent inaccuracy of age reporting in societies like the data source at the time of data 8832
collection, and in part to the fact that (probably for that very reason) only integer years of age attained was asked except for very young ages at death. By contrast, an individual child’s age when observation ended (age at censoring) could be computed accurately to the month, for in these data a child’s record was censored only because the child’s parent was interviewed; the age of a live child could then be calculated from its date of birth. This unconventional data set is used in order to illuminate how a number of practical issues that confront a demographer can be handled in a first analysis of mortality. A more complete investigation would subsequently use intensity-regression methods from event-history analysis (see Eent History Analysis: Applications), as was done by Woldemicael (1999). If there had not been any heaping in the reported ages at death, one would calculate Rx l Sxk" (DxjWx) as an approximate number of person# of exposure at age x. Here S is the number of months x children who survive to exact age x months, Dx is the number of deaths recorded between exact ages x and xj1 months, and Wx is the corresponding number of records that were censored during this interval. The approximation consists in assigning deaths and interviews to in the middle of the month on average. It has become known as the actuarial method. The next step would be to compute occurrence\exposure rates Dx\Rx (see Demographic Techniques: LEXIS Diagram). The age heaping is abundantly evident in the plot of the ‘raw’ death rates (converted to ‘deaths per 1,000 person-years’) in Table 1. Such age heaping must have been caused by a shift of the real age at death to a reported near-by ‘round’ anchor age like three months, six months, or multiples of 12 months. The deaths in some interval [x ka, x jb] around each anchor age x have largely !been ! ! to each age in the assigned to the anchor age instead of interval, but the total number of deaths bt= −aDx +t in ! such an interval should be largely correct. Similarly, the total exposures bt=−aRx +t in the interval should be about right, even though! the individual term in the latter sum may be wide off the mark. Under these circumstances it is more sensible to compute death rates for groups of reported ages and list them in the row for the corresponding anchor age. By this reasoning, % % m l Dx\ Rx $ x=# x=# is entered for age three months, ( ( m l Dx\ Rx ' x=& x=&
Life Table is entered for age six months, and so on, as indicated in column 7 of Table 1. For the intervening ages x, rates mx have been computed by linear interpolation. The corresponding curve of rates is given with the label ‘grouped’ in Table 1. The rest of the life-table computations are based on the following simple mathematical theory. Each individual child i has a lifetime Ti with some distribution function F(t) l PoTi tq (with F(0) l 0), a probability density f (t) l dF(t)\dt, and a force of mortality or death intensity µ(t) l f(t)\o1kF(t)q. The various lifetimes Ti are independent of each other and they have the same distribution F(t). Simple integration gives F(x) l 1kexp ok x µ(t) dtq. The death prob! ability is qx l PoTi xj1 Q Ti xq l oF(xj1)kF(x)q\F(x) from which is derived
&
" qx l 1kexp ok µ(xjt) dtq !
(1)
If one approximates the function µ(xjt) by the constant mx for 0 t 1 (the assumption of piecewise constancy), then (1) gives qx l 1ke−mx, which has been used to compute the next-to-last column in Table 1. Finally, a life-table surial function is defined as %(x) l %(0)o1kF(x)q, with %(0) l 100 000. This is the expected number of survivors to exact age x out of an original cohort of %(0) initial individuals. Simple manipulation with the definition of qx gives %(xj1) l %(x)(1kqx),
for x l 0, 1, …
(2)
This is a practical recursive formula by which we have computed the final column in Table 1. From the final entry in the %(x) column in Table 1 we see that more than 15 percent of the original cohort of girl babies would die before age 4 (years) according to our data, because 1k0.84859 l 0.15141. The mean number of months lived between birth and age n is eo : l E min(Ti, n) l !n
&!%(t) dt\%(0) n
(3)
If one partitions the integral here into a sum of n terms x %(t) dt for x l 1, 2, …, n, use the trapezoidal rule x− of "numerical integration to see that this integral is approximately equal to " o%(xk1)j%(x)q, and collect # numerical approximation terms suitably, we get the 1
eo : n # ! n
5
4
3
2
n 1 1 %(x)k k %(n) 67 \%(0) 2 2 8 x=!
(4)
The girls in the Eritrean data set lost five months of life on average before they reached their fourth birthday, for according to Eqn. (4) with n l 48 months they experienced a mean lifetime of 43 months.
2. Other Life-table Procedures Life-table construction consists in the estimation of parameters and tabulation of functions like those above from empirical data. The data can be for age at death to individuals, as in the illustrative example, but they can also be observations of duration until recovery from an illness, of intervals between births, of time until breakdown of some piece of machinery, or of any other positive duration variable. In general, a Ti is the duration until occurrence of some event that ends individual survival in a given status. The method of estimation depends on the character of the data. If accurate (ungrouped) individual-level data are available, then the Kaplan–Meier estimator (see Eent History Analysis: Applications) can be used to estimate %x for all relevant x, and estimates of the other lifetable functions can then be computed subsequently. Alternatively a segment of the Nelson–Aalen estimator can be used to estimate each integral " µ(xjt) dt, Eqn. (1) can then be used to estimate q ! each integer x, and the rest of the computationsx for follow suit. Sometimes the force of mortality is represented by some function h(x; θ), where θ is a vector of parameters. For instance, actuaries often use the classical Gompertz–Makeham function h(x; a, b, c) l ajbcxfor the force of mortality in their life tables (see Demographic Models; Demographic Analysis: Probabilistic Approach). From any given schedule of death probabilities q , q , q , …, the %(x) table is easily computed using the ! " # Eqn. (2). Much of the effort in life-table recursive construction is, therefore, concentrated on providing a schedule oqxq. So far, this account has assumed that the data come from a group of independent individuals who have all been observed in parallel, essentially a cohort that is followed from a significant starting point (namely from birth in our empirical example) and which is diminished over time due to decrements (attrition) caused by the risk in question and also subject to reduction due to censoring. The empirical example displayed data only for the first four years of life of such a cohort. One would have had to follow the cohort until the end of the cohort’s total life to produce a complete cohort life table. It is more common to compute a qx schedule from data collected for the members of a population during a limited time period and to use the mechanics of life-table construction to produce a period life table from the qx values. If real mortality patterns are tied to cohorts, individuals who live at widely differing ages in the observational period do not normally have the same risk structure, and the period life table is taken to reflect the patterns of a 8833
Life Table Table 1 Life table computations for Eritrean girls aged 0 through 4 years Age x in months 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48c
Observed survivors
Deaths Dx
Censored Wx
Exposures Rx
Raw rate Dx\Rx per 1000
7098 6875 6788 6713 6634 6591 6548 6456 6412 6355 6310 6271 6224 6050 6010 5972 5954 5914 5880 5821 5792 5762 5746 5717 5697 5524 5490 5458 5427 5400 5365 5332 5293 5265 5229 5205 5177 5046 5008 4988 4956 4925 4895 4865 4841 4815 4797 4777 4761
208 48 36 43 14 21 50 21 26 19 16 10 144 6 7 3 5 3 25 5 2 0 0 2 148 0 0 0 0 0 0 0 0 0 0 0 109 0 0 0 0 0 0 0 0 0 0 0 46
15 39 39 36 29 22 42 23 31 26 23 37 30 34 31 15 35 31 34 24 28 16 29 18 25 34 32 31 27 35 33 39 28 36 24 28 22 38 20 32 31 30 30 24 26 18 20 16 20
6986.5 6831.5 6750.5 6673.5 6612.5 6569.5 6502 6434 6383.5 6332.5 6290.5 6247.5 6137 6030 5991 5963 5934 5897 5850.5 5806.5 5777 5754 5731.5 5707 5610.5 5507 5474 5442.5 5413.5 5382.5 5348.5 5312.5 5279 5247 5217 5191 5111.5 5027 4998 4972 4940.5 4910 4880 4853 4828 4806 4787 4769 4728
357.26 84.32 64.00 77.32 25.41 38.36 92.28 39.17 48.88 36.00 30.52 19.21 281.57 11.94 14.02 6.04 10.11 6.10 16.40 10.33 4.15 0.00 0.00 4.21 316.55 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 255.89 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 116.75
a: Computed as sum of deaths\sum of exposures for age groups indicated (Used for rows where age groups are indicated) b: Computed by linear interpolation c: Additional rows for ages 49–60 months have been deleted
8834
Age grouping (months) 0 1 2–4 5–7
8–17
18–30
31–42
43–54
Grouped rates mx
Death probab. 1000 qx
Life table survivors %x
357.26a 84.32a 70.01b 55.70a 56.00b 56.30b 56.60a 54.98b 53.35b 51.73b 50.11b 48.48b 46.86a 45.45b 44.05b 42.64b 41.24b 39.83b 38.43b 37.02b 35.62b 34.21b 32.81b 31.40b 30.00a 29.28b 28.57b 27.85b 27.14b 26.42b 25.71b 24.99b 24.27b 23.56b 22.84b 22.13b 21.41a 20.44b 19.48b 18.51b 17.54b 16.57b 15.60b 14.63b 13.66b 12.70b 11.73b 10.76b 9.79a
29.33 7.00 5.82 4.63 4.66 4.68 4.71 4.57 4.44 4.30 4.17 4.03 3.90 3.78 3.66 3.55 3.43 3.31 3.20 3.08 2.96 2.85 2.73 2.61 2.50 2.44 2.38 2.32 2.26 2.20 2.14 2.08 2.02 1.96 1.90 1.84 1.78 1.70 1.62 1.54 1.46 1.38 1.30 1.22 1.14 1.06 0.98 0.90 0.82
100 000 97 067 96 387 95 826 95 383 94 939 94 494 94 050 93 620 93 204 92 803 92 417 92 044 91 685 91 339 91 004 90 681 90 370 90 071 89 783 89 506 89 241 88 987 88 744 88 512 88 291 88 076 87 866 87 663 87 465 87 272 87 086 86 904 86 729 86 559 86 394 86 235 86 081 85 935 85 795 85 663 85 538 85 420 85 309 85 205 85 108 85 018 84 935 84 859
Life Table synthetic (fictitious) cohort exposed to the risks of the period at their various ages. Whichever way the life-table survivor column has been produced, Eqn. (4) can always be used to calculate mean numbers of time-units lived under the risk for which the life-table was constructed. If the time unit is a year and n is chosen so large that no one lives more than n years (symbolized by letting n l _ and %(_) l 0), eo : _ is called the life expectancy of the ! life table and is normally denoted eo . The subscript 0 ! for a newborn indicates that the calculation is made individual. Similar computations can be made at any age x and one gets expected remaining lifetimes of the form eox: l n %(xjt) dt\%(x), with approximation ! to Eqn. (4). formulas nsimilar
3. Multiple-decrement tables When suitable data are available, the risk of death (or force of mortality) may be partitioned according to mutually exclusive causes of death. Let µk(t) be death intensity or force of mortality for cause k at age t, and let the total force of mortality be µ(t) l k µk(t). Then the probability that an individual who is alive at age x will die from cause k before age xj1 is qx(k) l
&!" p µ (xjt) dt
(5)
t x k
where t px is the probability of surviving to age xjt for an individual who is alive at age x. For given risk intensities oµk(:)q, q(k) can be computed by numerical x integration in Eqn. (5). A corresponding column of q(k) x values may then be added to the life table for each k, and other cause-specific life table functions may also be computed. Note that q(k) is influenced by risk x intensities other than µk(:), for 1
p l exp t x
&
5
t
4
3
2
k µj(xjs) ds j ! 6 7 8
where the exponent depends on all µj(:), including those for j k. Several further life-table functions can be defined by formal reduction or elimination of one or more of the risk intensities in formulas like Eqn. (5). In particular, a single-decrement life table can be computed for each cause k in order to show what the life table would look like in the hypothetical case where this cause were the only one operating in the study population and where it did so with the risk function estimated from the data. The purpose is to see the ‘pure’ effect of the risk in question on a fictitious cohort without interference from other causes. (The decrement probabilities and other features of the single-decrement life table do not depend on any other intensities than the one for which
the table is computed.) This does not mean that one should believe that the total risk intensity can actually be reduced to the risk in focus or that this risk operates independently of other causes. Most single-decrement life-table functions have straightforward interpretations. If, for instance, %k(x) denotes the corresponding survival function, then %k(x)\%k(0) is the probability of surviving to age x in the fictitious table. Suppose, for example, that the study population is a group of unpartnered individuals subject to the competing risks of marital and nonmarital union formation as well as to the risk of death. One can then compute a single-decrement life table based on the risk of transition into a nonmarital union alone and find the probability that an individual would end up in such a union by age 50, say, if marriage were no alternative and death did not operate, even though in reality marriages continue to be formed by group members and unpartnered individuals continue to die. In such a table one may get %k(_) 0 because the event never occurs to some individuals, even if it operates alone. (We will all die but some individuals never enter a nonmarital partnership even if they live to age 100.) Life expectancies eo : computed in the three"& n be interpreted as the mean decrement life table would number of years (say) lived in the unpartnered state between ages 15 and 15jn. In the single-decrement life table for nonmarital-union formation, one possibility is to compute the mean age of entry into a nonmarital union only for those who do enter such a union by age n, that is, the conditional mean
&!% (x) dx\o% (0)k% (n)q n
k
k
k
(6)
Conversely one can compute a cause-deleted life table by eliminating one or more of the cause intensities in the formulas. To ‘remove’ cause k, introduce µ−k(t) l µ(t)kµk(t) and construct a ‘normal’ life table by replacing µ(t) by µ−k(t) everywhere. This would show what the life table would look like if it were possible to eliminate cause k without changing the risk of any other cause. It could be used for example to see how partnering probabilities would appear if the partnering process were undisturbed by mortality. One would then remove the force of mortality from the life-table formulas and operate only with the sum µ−k(t) of the intensities of marital and nonmarital union formation in the computations. Note that this is again a purely hypothetical construct, in fact it is counterfactual for mortality cannot of course be completely removed. Life expectancies can of course be calculated for a cause-deleted life table as for any other life table. If %−k(x) is the table’s survival function and %−k(_) l 0, then a formula for the cause-eliminated life expectancy is e(−k) l _%−k(x) dx\%−k(0). The classical applications ! to cause-of-death ! were removal, and the difference 8835
Life Table e(−k)keo was interpreted as the (fictitious) gain in life ! ! expectancy produced by removal of cause k. For example, Westergaard (1907) found that according to the English life tables of 1881–90 life expectancy for females would increase by 2.7 years if tuberculosis deaths could be eliminated, by a year if cancer deaths could be removed, and by 1.1 years if deaths due to diarrhea and dysentery could be deleted separately. The effects were not additive, and if all three cause-ofdeath groups could be removed, the gain would be 5.6 years and not 4.8. The life expectancy for females in the original table was eo l 47.2 years. ! See also: Age Structure; Life Course: Sociological Aspects
Bibliography Chiang C L 1984 The Life Table and its Applications. Krieger, Malabar, FL Dupa# quier J 1996 L’inention de la table de mortaliteT . Presses universitaires de France, Paris Elandt-Johnson R C, Johnson N L 1980 Surial Models and Data Analysis. Wiley, New York Graunt J 1662 Natural and Political Obserations Made upon the Bills of Mortality. Roycroft, London Hoem J M 1998 The life table. In: Armitage P, Colton T (eds.) Encyclopedia of Biostatistics. Wiley, Chichester, pp. 2235–9 Manton K G, Stallard E 1984 Recent Trends in Mortality Analysis. Academic Press, Orlando Preston S H, Heuveline P, Guillot M 2001 Demography: Measuring and Modeling Populations. Blackwell, Oxford, UK Seal H 1977 Studies in history of probability and statistics, 35: Multiple decrements or competing risks. Biometrika 64(3): 429–39 Smith D, Keyfitz N 1977 Mathematical Demography: Selected Papers. Springer-Verlag, Heidelberg, Germany Westergaard H 1907 The horoscope of the population in the twentieth century. Compte-Rendu de la XI Session de l’Institut international de Statistique aT Copenhague du 26 au 31 aout 1907 Woldemicael G 1999 Infant and child mortality in Eritrea: Levels, trends, and determinants. Ph.D. thesis, Stockholm University
J. M. Hoem Copyright # 2001 Elsevier Science Ltd. All rights reserved.
Lifelong Learning and its Support with New Media: Cultural Concerns
has become an integral and irremovable part of work activities. Learning is a new form of labor, and working often is (and needs to be) a collaborative effort among colleagues and peers. In the emerging information society, an educated person will be someone who is willing and able to consider learning as a lifelong process. More and more knowledge, especially advanced knowledge, is acquired well past the age of formal schooling, and in many situations through educational processes that do not center on traditional schools (Illich 1971). Lifelong learning is more than adult education, which often is restricted to providing people with opportunities to engage in (school-like) learning activities during their adult life. The challenge for lifelong learning is to fundamentally rethink learning, teaching, and education for the information age in an attempt to change mindsets. It involves and engages learners of all ages in acquiring and applying knowledge and skills in the context of authentic, selfdirected problems, and it exploits the possibilities offered by new media. Lifelong learning has emerged as one of the major challenges for the worldwide knowledge society of the future. A variety of recent events supports this claim: (a) 1996 was the ‘European Year of Lifelong Learning,’ (b) UNESCO (United Nations Educational, Scientific and Cultural Organization) has included ‘Lifetime Education’ as one of the key issues in its planning, and (c) the G7\G8 group of countries has named ‘Lifelong Learning’ as a main strategy in the fight against unemployment. Despite this great interest, there are few encompassing efforts to tackle the problem in a coherent way. Lifelong learning is comprehensive; it cannot be investigated in isolation by looking just at one small part of it, such as K-12 education, university education, or worker re-education.
2. Problems in the Information Age 2.1 Lack of Creatiity and Innoation Societies and countries of the future will be successful not ‘because their people work harder, but because they work smarter.’ Creativity and innovation are considered essential capabilities for working smarter in knowledge societies (Drucker 1994); thus, an important challenge is how these capabilities can be learned and practiced. An implicit assumption is made that self-directed and lifelong learning can influence the creativity and innovation potential of individuals, groups, organizations, and countries (Dohmen 1999).
1. Introduction Learning needs to be examined across the lifespan because traditional notions of a divided lifetime— education followed by work—are no longer tenable (Gardner 1991). Professional activity has become so knowledge-intensive and fluid in content that learning
2.2 Coping with Change Most people see schooling as a period of their lives that prepares them for work in a profession or for a change of career. This view has not enabled people to
8836
International Encyclopedia of the Social & Behavioral Sciences
ISBN: 0-08-043076-7
A. M. Freund
Life Table The life table is perhaps the oldest tool developed for the analysis of survival patterns in human and other populations. Its roots go back at least to the famous book on the English Bills of Mortality that John Graunt published in 1662, a book that is sometimes cited as the beginning of systematic statistical science. Life table techniques are described in detail in most introductory textbooks on the methods of actuarial statistics, biostatistics, demography, and epidemiology (see, e.g., Chiang (1984), Elandt-Johnson and Johnson (1980), Manton and Stallard (1984), Preston et al. (2001). In this article the basic notions of life-table construction are explained with a focus on issues that are sometimes underemphasized in other accounts. More of the underlying mathematical theory can be found in Hoem (1998) and in the textbooks. For the history of the topic, consult Seal (1977), Smith and Keyfitz (1977), and Dupa# quier (1996).
1. A Cohort Life Table Classical presentations of life-table techniques typically start with data in a format like that of the first four columns in Table 1. These particular data come from the Eritrean Demographic and Health Survey of 1995 and show the survival up through their fourth birthday of 7,098 girls whose births and deaths were recorded in the survey. This data set is different from normal textbook examples in two ways. First, the data are given for intervals of single-month segments instead of the usual single-year intervals. Second, the present data are subject to strong heaping in the reported age at death (for the girls that died). In part this is due to a typical inherent inaccuracy of age reporting in societies like the data source at the time of data 8832
collection, and in part to the fact that (probably for that very reason) only integer years of age attained was asked except for very young ages at death. By contrast, an individual child’s age when observation ended (age at censoring) could be computed accurately to the month, for in these data a child’s record was censored only because the child’s parent was interviewed; the age of a live child could then be calculated from its date of birth. This unconventional data set is used in order to illuminate how a number of practical issues that confront a demographer can be handled in a first analysis of mortality. A more complete investigation would subsequently use intensity-regression methods from event-history analysis (see Eent History Analysis: Applications), as was done by Woldemicael (1999). If there had not been any heaping in the reported ages at death, one would calculate Rx l Sxk" (DxjWx) as an approximate number of person# of exposure at age x. Here S is the number of months x children who survive to exact age x months, Dx is the number of deaths recorded between exact ages x and xj1 months, and Wx is the corresponding number of records that were censored during this interval. The approximation consists in assigning deaths and interviews to in the middle of the month on average. It has become known as the actuarial method. The next step would be to compute occurrence\exposure rates Dx\Rx (see Demographic Techniques: LEXIS Diagram). The age heaping is abundantly evident in the plot of the ‘raw’ death rates (converted to ‘deaths per 1,000 person-years’) in Table 1. Such age heaping must have been caused by a shift of the real age at death to a reported near-by ‘round’ anchor age like three months, six months, or multiples of 12 months. The deaths in some interval [x ka, x jb] around each anchor age x have largely !been ! ! to each age in the assigned to the anchor age instead of interval, but the total number of deaths bt= −aDx +t in ! such an interval should be largely correct. Similarly, the total exposures bt=−aRx +t in the interval should be about right, even though! the individual term in the latter sum may be wide off the mark. Under these circumstances it is more sensible to compute death rates for groups of reported ages and list them in the row for the corresponding anchor age. By this reasoning, % % m l Dx\ Rx $ x=# x=# is entered for age three months, ( ( m l Dx\ Rx ' x=& x=&
Life Table is entered for age six months, and so on, as indicated in column 7 of Table 1. For the intervening ages x, rates mx have been computed by linear interpolation. The corresponding curve of rates is given with the label ‘grouped’ in Table 1. The rest of the life-table computations are based on the following simple mathematical theory. Each individual child i has a lifetime Ti with some distribution function F(t) l PoTi tq (with F(0) l 0), a probability density f (t) l dF(t)\dt, and a force of mortality or death intensity µ(t) l f(t)\o1kF(t)q. The various lifetimes Ti are independent of each other and they have the same distribution F(t). Simple integration gives F(x) l 1kexp ok x µ(t) dtq. The death prob! ability is qx l PoTi xj1 Q Ti xq l oF(xj1)kF(x)q\F(x) from which is derived
&
" qx l 1kexp ok µ(xjt) dtq !
(1)
If one approximates the function µ(xjt) by the constant mx for 0 t 1 (the assumption of piecewise constancy), then (1) gives qx l 1ke−mx, which has been used to compute the next-to-last column in Table 1. Finally, a life-table surial function is defined as %(x) l %(0)o1kF(x)q, with %(0) l 100 000. This is the expected number of survivors to exact age x out of an original cohort of %(0) initial individuals. Simple manipulation with the definition of qx gives %(xj1) l %(x)(1kqx),
for x l 0, 1, …
(2)
This is a practical recursive formula by which we have computed the final column in Table 1. From the final entry in the %(x) column in Table 1 we see that more than 15 percent of the original cohort of girl babies would die before age 4 (years) according to our data, because 1k0.84859 l 0.15141. The mean number of months lived between birth and age n is eo : l E min(Ti, n) l !n
&!%(t) dt\%(0) n
(3)
If one partitions the integral here into a sum of n terms x %(t) dt for x l 1, 2, …, n, use the trapezoidal rule x− of "numerical integration to see that this integral is approximately equal to " o%(xk1)j%(x)q, and collect # numerical approximation terms suitably, we get the 1
eo : n # ! n
5
4
3
2
n 1 1 %(x)k k %(n) 67 \%(0) 2 2 8 x=!
(4)
The girls in the Eritrean data set lost five months of life on average before they reached their fourth birthday, for according to Eqn. (4) with n l 48 months they experienced a mean lifetime of 43 months.
2. Other Life-table Procedures Life-table construction consists in the estimation of parameters and tabulation of functions like those above from empirical data. The data can be for age at death to individuals, as in the illustrative example, but they can also be observations of duration until recovery from an illness, of intervals between births, of time until breakdown of some piece of machinery, or of any other positive duration variable. In general, a Ti is the duration until occurrence of some event that ends individual survival in a given status. The method of estimation depends on the character of the data. If accurate (ungrouped) individual-level data are available, then the Kaplan–Meier estimator (see Eent History Analysis: Applications) can be used to estimate %x for all relevant x, and estimates of the other lifetable functions can then be computed subsequently. Alternatively a segment of the Nelson–Aalen estimator can be used to estimate each integral " µ(xjt) dt, Eqn. (1) can then be used to estimate q ! each integer x, and the rest of the computationsx for follow suit. Sometimes the force of mortality is represented by some function h(x; θ), where θ is a vector of parameters. For instance, actuaries often use the classical Gompertz–Makeham function h(x; a, b, c) l ajbcxfor the force of mortality in their life tables (see Demographic Models; Demographic Analysis: Probabilistic Approach). From any given schedule of death probabilities q , q , q , …, the %(x) table is easily computed using the ! " # Eqn. (2). Much of the effort in life-table recursive construction is, therefore, concentrated on providing a schedule oqxq. So far, this account has assumed that the data come from a group of independent individuals who have all been observed in parallel, essentially a cohort that is followed from a significant starting point (namely from birth in our empirical example) and which is diminished over time due to decrements (attrition) caused by the risk in question and also subject to reduction due to censoring. The empirical example displayed data only for the first four years of life of such a cohort. One would have had to follow the cohort until the end of the cohort’s total life to produce a complete cohort life table. It is more common to compute a qx schedule from data collected for the members of a population during a limited time period and to use the mechanics of life-table construction to produce a period life table from the qx values. If real mortality patterns are tied to cohorts, individuals who live at widely differing ages in the observational period do not normally have the same risk structure, and the period life table is taken to reflect the patterns of a 8833
Life Table Table 1 Life table computations for Eritrean girls aged 0 through 4 years Age x in months 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48c
Observed survivors
Deaths Dx
Censored Wx
Exposures Rx
Raw rate Dx\Rx per 1000
7098 6875 6788 6713 6634 6591 6548 6456 6412 6355 6310 6271 6224 6050 6010 5972 5954 5914 5880 5821 5792 5762 5746 5717 5697 5524 5490 5458 5427 5400 5365 5332 5293 5265 5229 5205 5177 5046 5008 4988 4956 4925 4895 4865 4841 4815 4797 4777 4761
208 48 36 43 14 21 50 21 26 19 16 10 144 6 7 3 5 3 25 5 2 0 0 2 148 0 0 0 0 0 0 0 0 0 0 0 109 0 0 0 0 0 0 0 0 0 0 0 46
15 39 39 36 29 22 42 23 31 26 23 37 30 34 31 15 35 31 34 24 28 16 29 18 25 34 32 31 27 35 33 39 28 36 24 28 22 38 20 32 31 30 30 24 26 18 20 16 20
6986.5 6831.5 6750.5 6673.5 6612.5 6569.5 6502 6434 6383.5 6332.5 6290.5 6247.5 6137 6030 5991 5963 5934 5897 5850.5 5806.5 5777 5754 5731.5 5707 5610.5 5507 5474 5442.5 5413.5 5382.5 5348.5 5312.5 5279 5247 5217 5191 5111.5 5027 4998 4972 4940.5 4910 4880 4853 4828 4806 4787 4769 4728
357.26 84.32 64.00 77.32 25.41 38.36 92.28 39.17 48.88 36.00 30.52 19.21 281.57 11.94 14.02 6.04 10.11 6.10 16.40 10.33 4.15 0.00 0.00 4.21 316.55 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 255.89 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 116.75
a: Computed as sum of deaths\sum of exposures for age groups indicated (Used for rows where age groups are indicated) b: Computed by linear interpolation c: Additional rows for ages 49–60 months have been deleted
8834
Age grouping (months) 0 1 2–4 5–7
8–17
18–30
31–42
43–54
Grouped rates mx
Death probab. 1000 qx
Life table survivors %x
357.26a 84.32a 70.01b 55.70a 56.00b 56.30b 56.60a 54.98b 53.35b 51.73b 50.11b 48.48b 46.86a 45.45b 44.05b 42.64b 41.24b 39.83b 38.43b 37.02b 35.62b 34.21b 32.81b 31.40b 30.00a 29.28b 28.57b 27.85b 27.14b 26.42b 25.71b 24.99b 24.27b 23.56b 22.84b 22.13b 21.41a 20.44b 19.48b 18.51b 17.54b 16.57b 15.60b 14.63b 13.66b 12.70b 11.73b 10.76b 9.79a
29.33 7.00 5.82 4.63 4.66 4.68 4.71 4.57 4.44 4.30 4.17 4.03 3.90 3.78 3.66 3.55 3.43 3.31 3.20 3.08 2.96 2.85 2.73 2.61 2.50 2.44 2.38 2.32 2.26 2.20 2.14 2.08 2.02 1.96 1.90 1.84 1.78 1.70 1.62 1.54 1.46 1.38 1.30 1.22 1.14 1.06 0.98 0.90 0.82
100 000 97 067 96 387 95 826 95 383 94 939 94 494 94 050 93 620 93 204 92 803 92 417 92 044 91 685 91 339 91 004 90 681 90 370 90 071 89 783 89 506 89 241 88 987 88 744 88 512 88 291 88 076 87 866 87 663 87 465 87 272 87 086 86 904 86 729 86 559 86 394 86 235 86 081 85 935 85 795 85 663 85 538 85 420 85 309 85 205 85 108 85 018 84 935 84 859
Life Table synthetic (fictitious) cohort exposed to the risks of the period at their various ages. Whichever way the life-table survivor column has been produced, Eqn. (4) can always be used to calculate mean numbers of time-units lived under the risk for which the life-table was constructed. If the time unit is a year and n is chosen so large that no one lives more than n years (symbolized by letting n l _ and %(_) l 0), eo : _ is called the life expectancy of the ! life table and is normally denoted eo . The subscript 0 ! for a newborn indicates that the calculation is made individual. Similar computations can be made at any age x and one gets expected remaining lifetimes of the form eox: l n %(xjt) dt\%(x), with approximation ! to Eqn. (4). formulas nsimilar
3. Multiple-decrement tables When suitable data are available, the risk of death (or force of mortality) may be partitioned according to mutually exclusive causes of death. Let µk(t) be death intensity or force of mortality for cause k at age t, and let the total force of mortality be µ(t) l k µk(t). Then the probability that an individual who is alive at age x will die from cause k before age xj1 is qx(k) l
&!" p µ (xjt) dt
(5)
t x k
where t px is the probability of surviving to age xjt for an individual who is alive at age x. For given risk intensities oµk(:)q, q(k) can be computed by numerical x integration in Eqn. (5). A corresponding column of q(k) x values may then be added to the life table for each k, and other cause-specific life table functions may also be computed. Note that q(k) is influenced by risk x intensities other than µk(:), for 1
p l exp t x
&
5
t
4
3
2
k µj(xjs) ds j ! 6 7 8
where the exponent depends on all µj(:), including those for j k. Several further life-table functions can be defined by formal reduction or elimination of one or more of the risk intensities in formulas like Eqn. (5). In particular, a single-decrement life table can be computed for each cause k in order to show what the life table would look like in the hypothetical case where this cause were the only one operating in the study population and where it did so with the risk function estimated from the data. The purpose is to see the ‘pure’ effect of the risk in question on a fictitious cohort without interference from other causes. (The decrement probabilities and other features of the single-decrement life table do not depend on any other intensities than the one for which
the table is computed.) This does not mean that one should believe that the total risk intensity can actually be reduced to the risk in focus or that this risk operates independently of other causes. Most single-decrement life-table functions have straightforward interpretations. If, for instance, %k(x) denotes the corresponding survival function, then %k(x)\%k(0) is the probability of surviving to age x in the fictitious table. Suppose, for example, that the study population is a group of unpartnered individuals subject to the competing risks of marital and nonmarital union formation as well as to the risk of death. One can then compute a single-decrement life table based on the risk of transition into a nonmarital union alone and find the probability that an individual would end up in such a union by age 50, say, if marriage were no alternative and death did not operate, even though in reality marriages continue to be formed by group members and unpartnered individuals continue to die. In such a table one may get %k(_) 0 because the event never occurs to some individuals, even if it operates alone. (We will all die but some individuals never enter a nonmarital partnership even if they live to age 100.) Life expectancies eo : computed in the three"& n be interpreted as the mean decrement life table would number of years (say) lived in the unpartnered state between ages 15 and 15jn. In the single-decrement life table for nonmarital-union formation, one possibility is to compute the mean age of entry into a nonmarital union only for those who do enter such a union by age n, that is, the conditional mean
&!% (x) dx\o% (0)k% (n)q n
k
k
k
(6)
Conversely one can compute a cause-deleted life table by eliminating one or more of the cause intensities in the formulas. To ‘remove’ cause k, introduce µ−k(t) l µ(t)kµk(t) and construct a ‘normal’ life table by replacing µ(t) by µ−k(t) everywhere. This would show what the life table would look like if it were possible to eliminate cause k without changing the risk of any other cause. It could be used for example to see how partnering probabilities would appear if the partnering process were undisturbed by mortality. One would then remove the force of mortality from the life-table formulas and operate only with the sum µ−k(t) of the intensities of marital and nonmarital union formation in the computations. Note that this is again a purely hypothetical construct, in fact it is counterfactual for mortality cannot of course be completely removed. Life expectancies can of course be calculated for a cause-deleted life table as for any other life table. If %−k(x) is the table’s survival function and %−k(_) l 0, then a formula for the cause-eliminated life expectancy is e(−k) l _%−k(x) dx\%−k(0). The classical applications ! to cause-of-death ! were removal, and the difference 8835
Life Table e(−k)keo was interpreted as the (fictitious) gain in life ! ! expectancy produced by removal of cause k. For example, Westergaard (1907) found that according to the English life tables of 1881–90 life expectancy for females would increase by 2.7 years if tuberculosis deaths could be eliminated, by a year if cancer deaths could be removed, and by 1.1 years if deaths due to diarrhea and dysentery could be deleted separately. The effects were not additive, and if all three cause-ofdeath groups could be removed, the gain would be 5.6 years and not 4.8. The life expectancy for females in the original table was eo l 47.2 years. ! See also: Age Structure; Life Course: Sociological Aspects
Bibliography Chiang C L 1984 The Life Table and its Applications. Krieger, Malabar, FL Dupa# quier J 1996 L’inention de la table de mortaliteT . Presses universitaires de France, Paris Elandt-Johnson R C, Johnson N L 1980 Surial Models and Data Analysis. Wiley, New York Graunt J 1662 Natural and Political Obserations Made upon the Bills of Mortality. Roycroft, London Hoem J M 1998 The life table. In: Armitage P, Colton T (eds.) Encyclopedia of Biostatistics. Wiley, Chichester, pp. 2235–9 Manton K G, Stallard E 1984 Recent Trends in Mortality Analysis. Academic Press, Orlando Preston S H, Heuveline P, Guillot M 2001 Demography: Measuring and Modeling Populations. Blackwell, Oxford, UK Seal H 1977 Studies in history of probability and statistics, 35: Multiple decrements or competing risks. Biometrika 64(3): 429–39 Smith D, Keyfitz N 1977 Mathematical Demography: Selected Papers. Springer-Verlag, Heidelberg, Germany Westergaard H 1907 The horoscope of the population in the twentieth century. Compte-Rendu de la XI Session de l’Institut international de Statistique aT Copenhague du 26 au 31 aout 1907 Woldemicael G 1999 Infant and child mortality in Eritrea: Levels, trends, and determinants. Ph.D. thesis, Stockholm University
J. M. Hoem Copyright # 2001 Elsevier Science Ltd. All rights reserved.
Lifelong Learning and its Support with New Media: Cultural Concerns
has become an integral and irremovable part of work activities. Learning is a new form of labor, and working often is (and needs to be) a collaborative effort among colleagues and peers. In the emerging information society, an educated person will be someone who is willing and able to consider learning as a lifelong process. More and more knowledge, especially advanced knowledge, is acquired well past the age of formal schooling, and in many situations through educational processes that do not center on traditional schools (Illich 1971). Lifelong learning is more than adult education, which often is restricted to providing people with opportunities to engage in (school-like) learning activities during their adult life. The challenge for lifelong learning is to fundamentally rethink learning, teaching, and education for the information age in an attempt to change mindsets. It involves and engages learners of all ages in acquiring and applying knowledge and skills in the context of authentic, selfdirected problems, and it exploits the possibilities offered by new media. Lifelong learning has emerged as one of the major challenges for the worldwide knowledge society of the future. A variety of recent events supports this claim: (a) 1996 was the ‘European Year of Lifelong Learning,’ (b) UNESCO (United Nations Educational, Scientific and Cultural Organization) has included ‘Lifetime Education’ as one of the key issues in its planning, and (c) the G7\G8 group of countries has named ‘Lifelong Learning’ as a main strategy in the fight against unemployment. Despite this great interest, there are few encompassing efforts to tackle the problem in a coherent way. Lifelong learning is comprehensive; it cannot be investigated in isolation by looking just at one small part of it, such as K-12 education, university education, or worker re-education.
2. Problems in the Information Age 2.1 Lack of Creatiity and Innoation Societies and countries of the future will be successful not ‘because their people work harder, but because they work smarter.’ Creativity and innovation are considered essential capabilities for working smarter in knowledge societies (Drucker 1994); thus, an important challenge is how these capabilities can be learned and practiced. An implicit assumption is made that self-directed and lifelong learning can influence the creativity and innovation potential of individuals, groups, organizations, and countries (Dohmen 1999).
1. Introduction Learning needs to be examined across the lifespan because traditional notions of a divided lifetime— education followed by work—are no longer tenable (Gardner 1991). Professional activity has become so knowledge-intensive and fluid in content that learning
2.2 Coping with Change Most people see schooling as a period of their lives that prepares them for work in a profession or for a change of career. This view has not enabled people to
8836
International Encyclopedia of the Social & Behavioral Sciences
ISBN: 0-08-043076-7