1 Basic Concepts and Measures

1.1 Meaning of "Population"

1.2 The Balancing Equation of Population Change
1.3 The Structure of Demographic Rates
1.4 Period Rates and Person-years
1.5 Principal Period Rates in Demography
1.6 Growth Rates in Demography
1.7 Estimating Period Person-years
1.8 The Concept of a Cohort
1.9 Probabilities of Occurrence of Events

1.1 Meaning of "Popu lation"

To a statisticiao, the term "population" refers to a collection of .items, for example, balis in an
um. Demographers use the term in a similar way to denote Lhe collection of persons alive at
a pecified point intime who meet certain criteria. Thus, they may refer to the "population of
lndia on April l, J 995," or to the "population of Arnerican bl.ack females in Lhe Northeast on
June 1, l900 ." ln both cases the cii ter ia for inclusion in the population need further elaboration:
do we coun t "legal residents" or simply those who can be found within the borders on that
date? What do we mean by "black," or by "Northeast"? Do we refer to rnidnight or noon oa
tbe specified date? It is clear that "the population oflndia on April l, l995" is a shorthand
description of what may be a rather long set of operational choices designed to minimize
bluniness at the boundaries.
But demographers also use the te1m "population" to refer Lo a differenL kind of
collectivity, one that persists through time even thoug h its members are continuously
changing th rough attrition and accession. Thus, "the population of Inclia" may refer to the
aggregate of persons who have ever been ative in the area we define as lnda and possibly
even to those yet to be bom Lhere. The collectivity persis ts even though a virlua lly complete
turnover ofits members occurs at least once each century.
Demographic ana lysis focuses on Lhis enduring collect ivity. IL is particularly addressed
to studyng changes in its size, its growth rates, and its composilion. But while the emphasis
is on understanding aggregate processes, demography is also attentive to the i.mplications of
those processes for individuais. Many of the indexes in common use in demography, such
as life expectancy at birtb and U1e total fertility rate, translate aggregate-level processes
into
2 BASIC CONCEPTS AN D MEASURES

statements about the demographic circumstances faced by an average or randomly-chosea

individual. ln lu rn, a frequent concern ia demo graphy i to trace out the consequences of
cbanges in individual-levei behavior for aggregate processe . Demography is one of the social
science disciplines where micro- and ma cro-levei a nalyses find perhaps tbeir most complete
and satisfactory articulation.

1.2 The Balancing Equation of Population Change

No malter how a population is defined, there are only two ways of enlering it: being bom into
it; or migrating into iL Jf the definition of the population includes a social eleme nt in acldition
to the customary geographic/temporal elemcnts, tben "migration" can include a change in the
socia l label a process often refeITed to as "social rnobility." For example, tbe population of
American higb school graduates can be entered by achieving a h.igh scbool diploma, a fonn
of social migration or mobility. Note in this example that the population carmot be entered al
birth snce the acquisition of the label ofhigh schoo l g raduate requires tbe i nvestment of years
of life. Populations defined by marital status or occupation are other examplcs of populations
tbat cannot normally be entered by birth (except for the default opt ions, unmarriecl and no
occupation) . On the otber hand, populations defined by characteiistics fixed at birth, sucb as
sex, race, or nativily, caimot be entered tluough migration but only tbrougb birtb. So there are
at most two ways of entering a pop ul at ion, birth and in-migration (= im.migration).
Likew ise, there are at most two ways of leavin g a population, deatb and out-
migration (= emjgration). Ali populations can be left through deat b, but only those defined by
charac teiistics not fixed at bi rth can be exited through migration. li one is born in the United
State s, one cannol leave Lhe population of persons bom in the Uniled States by migration, bul
one can obviously leave the popula t io n resident in the United States by migration.
Because there are only four possible ways of entering or Jeaving a popu latio n, we can be
sure that changes in the size of the population mu st be attributable to the magnitude of these
flo ws. ln particular,

N(T) = N(O) + B [O , T] - D[O, T] + J[O , T] - oro, T], (1 .1 )

where
N (T) = nurnber of persons alive in tbe population at time T ,
N(O) = numb er of persons alive in tbe population at time O,
B [O, T ] = number of births in the population between time O and time T,
D[O, T] = number of deatbs in the population betwe en time O and time T,
l[O, T] = number of in-migra ton s between time O and time T,
O[O, T] = number of out-migrations from the population between time O and time T.

The unit of time in this equation, and throughout the book unless otberwise noted, is number
of years. Thus, the time period in which birtbs, deaths, and migrations are occurring is T years
in length. T may be fractional and need not be an integer number.
Kenneth Boulding has caUed this equation the most fundamental in the social sciences. It
is clearly an identity rather than ao approximation ora bypothesized relation. However, when
data are used to estimate the elements of th:is equation, it is no longe r the case that both sidcs
must be equal. Error in measuring any element will cause an imbalance in the equation, unless
two or more errors happen to be exactly offsetting. An imbalance i n the equation is sometimes
 BASIC CONCEPTS AND MEASURES 3

eox 1.1 The Bafancing Equation of Population Change

N(T) = N(O) + B [O, T] - D [, T] + l [O , T] - O[O, T]

Example: Sweden, 1988

Ending Starting Births Deaths In-migralions Out-migraLions

population population betwee n between between betwee n
Jan. l , 1989 Jan. 1 , 1988 Jan. 1 , 1988 Jan. 1, 1988 Jan. 1, 1988 Jan. 1 , 1988
and and and and
Ja n. 1 , 19 89 Jan. 1, 1989 Jan. 1, 1989 Jan. 1 , 1989

N (l 989.0) = N(l988.0) + 8[1988.0, 1989.0J - D[l988.0, 1989.0] + / [ 19 88.0, . 0] - 0[1988.0, 1989.0J

l
1989
8,46 1 ,554 = 8,4 16 , 599 + 11 2,080 - 96,756 + 51 ,092 - 2 1 ,461

Data source: Unitecl Nations, Demographic Yearbook (various years).

referred to as an "error of closure." Box. l .1 demo nstrate s the application of the equation to
datafrom Sweden, which are among the world's most reliable.

1.3 The Structure of Demographic Rates

The balancing equation of population change breaks down the changes in the size of the
populat:ion into four flows. Each flow is the sum of events or transitions occurring to
individuals. Three of the four types of events can be rela ted to an individual present in the
population prior to the event. While death and out-migration can be related to one individuaJ,
birth can be related to two indiv idual parents, assuming that both belong to the population of
interest. Analytical insight can be gained by relating the size of these flows (number of
occurrences) to the size of lhe population produc in g them. This task is normally accomplished
by constructing a demographic "rate."
The term "rate" is used in many lields and its meaning is not consistent. An unemployment
rate, for exarnple, is simply a ratio of the unemployed to the total labor force at a moment
in time. ln dem ography, rates are normally (but not invariably) what are known in stati stical
par.lance as "occmrence/exposure rates." Toe typical form of demographic rates reflects the
fact that the frequency of occurrences can be expected to be higher in a larger population , and
thal the total number of occurrences can also be ex.pected to be higher the longer the members
of the population are exposed to the "risk" of the occurrence. The amount of ex.posure in
the denominator of an occuJTence/exposure rate combines these two features - the number of
persoos in the population and the length of the time frame ..11 which ex.posure is counted. The
mos.t conventional occurrence/exposure rate in demography takes tbe form of:

Number
- - of- Occurrences
- - - - - of- Exposure
Rate= - Person.-years - - - - - - - - -
to th.e Risk of Occurrence

Demographic rates thus contain in tbe numerntor a count of the number of events occur
rin.g within some defined time period, and in the denominator an estimate of the number of
"perso n-yea rs" lived in the population during that time period. The number of person-years
 BASIC CONCEPTS AND MEASURES

functions in partas an i.ndicator of the popula tion's amount of exposure to the tisk of the
event, hence the term occurrence/exposure rate. When person-yerus are used in the
denominator, a rate is referred to as an "a nnu alize d" rate.
Unlike occurrences, the number of person-years lived is rarely directly observed or counted.
Nevertheless, the concept is central in demography. To deal with the concept in a population
that is continuously changing its membership, it is usefuJ to represent individual exposures
as "life-li.nes." A life-line extends from an individual's bi1ih (A) to the point wbere he or she
experiences some terminal event ( B ), usually death . Occurrences of interesl , 0;, can be added
to the life-line, as illustrated below:

A B
1-_- _. ._ _ _ _ _ _ _ _ _ ......._ ......_ , Time

ln arder to better coonect events and exposure to the risk of experiencing the event, a life-line
is sometimes restiicted: if we are interested in the risk of giving birtb, for instance , we may
restrict analysis of life-lines to a certain age range. ln our exposition, event A and B are sim ply
bi1ih and death respectively, but the concept can readily be extended to other types of bouoding
eve nts.
For a group of individuais, however the group might be defined, the concept of the
occunence/exposure rate can be illust:rated by a set of life-lines for each member of tbe
group G:

1----------------------------------+Time

wbere 0 j are the event occurrences in group G and A; and B; represent the birtb and death of
individual i in tbe group. The rate for the group defined over tbeir eotire life times is

G
Rate = L aN;
=-c'-'E _
LiEGT;

where N; is tbe total numb ero fo ccurrences ia the lifetime of individual i, T; is the lengtb of time
between A; and B;, and
rue a member of group G.
L;E G is an instruction to take the surn across all individuals (i) who
 BASIC CONCEPTS AND MEASURES 5
1

1 1

1.4 Period Rates and Person-years

A period rate for a population is constructed by limiting the count of ocCLmences and exposure
Limes to those pertaining to members of the population dming a specified period of time:

Nu mb er o f O ccu rre nce s b etw ee n T ime O and T

Rate [O, T] =- - - - - - - - - - - - - - - - - -

Person-years Livecl in lhe Population between Time O and T

If a person lives one year between time O and time T , be or she has cont1ibuted one person
year to the denominator of the period rate. If a person li ves 24 hours between O and T , he
or she has conlributed 11365th of a person-year. The contributions from all individuais who
were aJive in the population at any time between O and T are simply added togetb.er in order
to produce the denomi.nator for our rates .
The idea is easily grasped by referring again to life-lines. If we are interested in pe1iod
O to T , all life-lines can be truncated lo the "window" O to T, since we will not count any
occurrences out sid e that interval. Figure 1.1 shows the l ife-l i ne s of 7 individuais in a small
hypotbeticaJ population duri.ng the period from 12:00 A.M., January 1, 1981 to 12:00 A.M. on
January 1, 1982.
Person 1, for exarnple, is a member of the pop ul at io,n for the entire year, whereas person
6 is born on April 1 and dies on October 1 , thereby contributing only 6 months or one-half
of a person-year to the sum of perso n-years. Adding exposure across i nd ividu aJs would be a
convenient way to estimate person-year s lived in count ry that had a population register which
recorded exact dates of birth, dea th, and migration for each individual.
An alternative method of compu ti ng period person-years is to ignore individual histor ies,
such as those provided by a population register, and simply record the number of persons alive

.. 1 t- - - - ---i-- - - - ---1-- - - - --+-------------------1

2 34

1 Person 3 f- Person-years
number 4 1 contributed by
5 each person
6 >-
7>-
Total= 4

a. Lile-lines for seven individuais who live in a population ai any time

., between Jan. 1, 1981 and Dec. 31, 1981

l' 7 -
6 -
Number oi 5 -
persons 4t-----<
alive 3-
2- 4* 6* 5* 4* Total = 4
1-
o - - - ---'-- - - - - - - - - - -
- -
Jan. 1,1981 April 1, 1981 July 1, 1981 Oct.1, 1981 Jan.1, 1982
1 Time

b. Lile-lines converted into numbers oi persons alive at each moment

Figure 1.1 Demonstration of the equivalence of the two method s for recording perso n-years
 6 BASIC CONCEPTS AND MEASURES

at various points in time during tbe year. Jn our example, there were 4 persons aLive from
to
January 1, 1981 to April 1, 1981, so that this quarter-year contributed 4() =1 person year. of
The next quarter contributed 6() = 1 . 5 person-years, and so on to a total of 4.75 person-years rer
contributed during ali of 1981. This value is of comse the sarne number derived by following
personal histories, as demonstrated in figure J . 1 .
ln tbis alterna.tive approach, what we have don e is to estimate the area un der tbe N(t) curve 1 !..
between January 1, 1981 and January 1, 1982 . N(t) is defined as the number of persons alive
at time t. An areais found by taking the he.ight of a figure times its width. Jn our case, N(t) is
wi
thc'
the height and the proportion of the yem that coITesponds to our measurement of N(t) is the e q'
width. Since heigbt represents persons and width represents fractions of a yem, it is natural to of
measure lhe product in units of person -years.
ln our example the number of person-years was: TI

py [ 1981 .00, 1982.00] = 4(.25) + 6(.25) + 5(.25) + 4(.25) = 4.75

Th.is sum can be written in conventional notation as:
TI
4
PY[ l 981.00, 1982.00] = L Ni t:,.i
i=I

where Ni is the number of persons a]jve in the ith quarter and t:,.i is the fraction of a year T
represented by that quarter. Had we measured the size of the population each day instead of
each quaiter , the sum would be represented as: 1

1
PY [198 1.00 , 1982.00] = N (Jan. 1 , 1981 ) -
365 T
l
+ N ( Jan. 2, 1981) -
365
+
1
+N(Dec. 31, 198 1 ) -
365
365
=LN;b.j
i=I

If we were able to measure the height , N(t), in tiny intervals of time dt, whe-e dt
represents the width of the interval, the area under the curve could be represented more
accurately as:

1982.00

PY[198 l. OO, 19 82.00] = / N(t) dt

19 81.00

Here an integral sign has replaced the sununation sign and for the fraction of a year represented
by the time inte rval, dt has replaced !:,.; .
We have seen that areas under a curve can be represented in two ways , us in g either
algebraic or calculus notation. ln demog raphy, aJgebraic notation satisfie s a practical need
that a.rises when measurement occurs in discrete intervals. But calculus is often preferred for
its compact notation and for its far more extensive body of theorems having direct applicabil i ty
 BASIC CQNCEPTS AND MEASURES 7
to popula tio n processes. We will use algebra and calculus interchangeably i n this volume. One
of the most frequent uses of calculus will occur in the issue we have already encountered,
representng the area under a curve.

1.5 Principal Period Rates in Demography

We can now apply the concept of period rate to demograpbic events of interest, in particular
tbe four components of the balancing equa ti o n of population cbange. When the elements of
equation (1. 1 ), the balancing equation of population growth, are each divided by the number
of person -years lived between O and T, we define four rates:

The Crude Birth Rate between times O and T:

Number of births in the population between times O and T
CBR[ ,O T] = N
umb er of person -years J1' m. Ih e at1.011b etween and T
ved popul t.unesO

The Crude Death Rate between times O and T :

Number of deaths in the population between times O and T
CDR [O, T] =- - - - - - - - - - - - - - - - - - - -
Number of person-years lived in the population between times O and T

The Crude Rate of In-migratoo between times O and T:

Number of in-migrations into the population between times O and T
CRIM [O, T] = . . . .
Number ofperson-years ltved m the populat10n between hmes O and T

The Crude Rate of Ou t- mig ration between times O and T:

Number of out-migrations from the population between times O and T
CROM[O, T] =- - - - - - - - - - - - - - - - - - -
Number of person -years lived in the population between times O and T

We could label the crude birth rate as we have defined it as the "true" crude birth rate, since
it incJudes tbe actual births and actual person-years in the numerator and denominator, respec
tively. Throughout the book, the term " rates" wilJ refer to the true or actual rates prevailing in
a popu latio n. These should be distinguished from the "recorded" or "estimated" rates that are
produced when data are used to estimate the value of the true rate.
A person is normaliy counted as having migrated during the period O to T if he or she has
changed his or her p1incipal place of residence during the period in a way that crosses the
administrative boundaries defining "the population " of a region.
As is especia1ly clear from our definion of the crude rate of in-migrat ion , the connection
between exposure and event is not alway s very precise in demography. No member of a
population is l iterally exposed to the 1isk of in-migrating into that same popul atio n; those
aL risk are all outside of the population. Like any definitions, these contain an element of
arbitrminess, and we could have chosen to put another element in the denominator. What the
crude rate of in -migralion expresses is the rate at which the population is growing as a result
of in -migration. The other rates also indicate the rate at which the population is changing as
a result of births, deaths , or out-migration. Using person-years as the denominator for a1l the
majorrates in demog raphy provides a firm basis for developing and i_ntegratng many different
functions and formulas involving population growth. This advantage should become evident
in tbe course of this volume.
 It is import ant to keep in mind the distinction between the reference period to wbich a rate
pertains (i.e., the period for which the values me calculated) and the un.it in which exposure time is
measmed. As noted, tbe conventional practice i to count expos ure in the form of person
years lived, thus crealing "annua li.zed " rates. They express tbe number of evcnts occurring per N<T
year of exposur e . But a period rate need not refor to a single yearof the popu lation's expe rience .
For examp le, we can readily define a crude death rate for 1990 - 1. Here the number of events in

PY
the numerator would include ali deaths for calendar years 1990 and 1991 , and the denominator

C
would include ali person-years lived i..n 1990 as weU as those lived in 1991. Since both the
numerator and denominator are, in size, approxirnately double what they would be if they
referred to only a single calendar year, defining tbe rale over a 2-year period does not affect
the
scale of tbe rate . It is still an ann ualized rate, expressing the number of events per person - Perso

year .
Likewise, we could define a crude death rate for May 1992 , in which both numerator and N(l9
denominator would be approximately one-twelfth of their value for ali of 1992. The scale of pY
the rate, and its annualized nacure, is preserved.
CGR
Although a period rate in demography apparently can accommodale any length of reference
period, il is important Lo recognize that it must bave some reference period. Tbe phrase, " the 8,461
crude birth rate of the Uniled States ," has no rneaning and lhere is no way to calculate its
va l ue. We must know in what period to count births for the nurn.erator and person-years for
the denominator.
CG/l

0.00
1.6 Growth Rates in Demography
Dow
1.6.1 Crude growth rate
Let us rearrange the balancing equation of population change (1 .1), by subtracting N (O) from
both sides and lhen dividin g bo!h sides by the total of person-years lived betweeo O and T,
PY [O, TJ: Th
dis
N(T) - N(O) B[O, n D[O, Tl no,
T] 0[0, T]
PY[O, T] = PY[O, T] - PY[O, T] + PY[O, PY[O, T]
T]
CGR [O, T] = CBR ro, T] - CDR [O, T] + CRTM [O, T] - CROM [O, (1.2) 1.
T]
As
CRNI[O, TJ + CRNM[O, T]

we
.
Here we have defined the crude growth rate belween O and T, CGR[O, T], as the change i.n the
size of population divided by person-years lived between O and T. lf N(T) exceeds N(O), then ra
gr
the growth rate will be positive; if N(O) exceeds N(T), it will be negative. Clear ly, the crude ap
growth rate as we have defined it is simply equal lo the crude birth rate mi.nus the crude death
rate plus the crude rale of in-rn.igration minus tbe crude rate of out-migration.
The difference between lhe crude birth rale and the crude death rate is usually termed the
crude rate ofnatural increase (CRNI); also, the difterence between the crude rale of in-migratio n
and the crude rate of out-rn.igration is usually Lermed the crude rate of net m igration (CRNM).
So the crude growtll rate will equal the crude rate of natural increase plus the crude rate of
net mig ratio n. Box 1.2 illustrales the calculation of crude demographic rates, again using the
Swedish data in box 1.1 and es tima ti.ng the person-years lived in 1988 by the population size
on July 1, 1988 . Table 1 .1 presents the estimated value of demographic rates for major regions

in
of th e worJd.
 BASIC CONCEPTS ANO MEASURES 9
luch a rate
osure time
of person
Box 1.2 Principal Period Rates in Demography
uningper
N(T) - N(O)= 8 [0, T] _ DfO. T] + /[O, T] _ OLO, Tl
perience. f
pY[O, T] PY[O, T] PY [O, T] PY[ O, TJ PY[O, T]
events in
nominator CGR[O, 7'] = CBR [O , TJ- CDR IO, TI+ CRIM [O , TJ - CROM LO, TJ

both the = CRNl [O , T] + CRNM[O. TJ

be if they
Example: Sweden, 1988
affect the
rson-year. Person-years lived in Swedcn between .January l, 1988 and January l , 1989 = 8,438,477 (mid-year population)
!rator
N( l989.0)- N(l988.0)= 8 [19 88.0, 19 89.01 0 119 88.0, 1989.01 /[ 1988.0, 1989.0 J 0 [1 988.0, 19 89.0]
and PY[ l 988.0, 1989.0) P Y[ l9 88.0, 1989.0]- PY[l988.0. 1989.0] + PY[ 1988.0, 1989.0] - PY[l988.0, 19 89.0]
escale of
CGRL l9 88.0, 1989.0J = CBR [ 1988.0, 1989.0) - CD/1 [1988.0, 1989.01 + CJUM [1988 .0, 1989.0] - CROM I 19 88.0. 1989.0)

reference
rase, "the 8,461,554 - 8,416,599 =
1 12,080 96 ,756 5 1 , 0 92 2 1 ,461
8,438,477
- - - - + -
culate its 8.438,477 8,438,477 8,438,477 8,438,477

years for -
0.00533 = 0.01328 - O.OI 147 + 0.00605 0.00254

CGR[ l 988.0, 1989.0] = CRNI [ 1988.0, 1 9 89.01 + CRNM[ l 9 88.0, 1989.01

0.00533 = 0.00182 f 0.00351

+

Data somte: United Nations, Demographic Yearbook (var io us years).

;co) from
p and T, Tbe crude growth rate is only one of severa! types of growth rate encount ered in
demography. The term "growth rate" is used to refer to other measures as wel l, and it
is important to distinguisb the various forms.

(1.2) 1.6.2 lnstantaneous growth rate

As any rate, the crude growth rate can be computed for any period of time. Wbat happens when
we compute the growth rate during a very short period of time, between time t and t + !::,.t, as
!::,.t approaches O? Denote the population change, N(t +!::,.1) - N(t), as !::,.N (t) and lhe growlli
e in the rate as r(t). Since the person-years lived over the period [t , t + !::,.t] is now N(t)!::,.1, the crude
0), then growlh rate for the period is r(t) = !::,.N(t)/N(t)!::,.t. But the linrit of !::,.N(t)/M when M
e crude approaches O is simply the derivative ofthe N(t) function, designated dN(t)ldt. Therefore:
le death
dN(t)
ed the t::,.N(t) dln[N(t)]
gration lim
r(t) = ) dt =----
N(t (1.3)
t.t->O natural log arithm. Tbe time interval is very sbort, dt years, so that r(t)
dt pertains to l11e tiny interval of time between t and t + dt. Because it is measured intime uns
where "ln" ofyears, r(t) continues to be an annualized rate. It is referred as "the growth rate at time t" or
refers to tbe "the inslantaneous growth. rate at time t ." It is, of course, also the crude growth rate in the tiny
interval oflime from t to t+ dt.
Table 1.1: Population size and components of change in major areas of the world, 1995- 2000

Majorarea Population size Births Df!.aths Net Crude Crud Crud Crude rate Crude rate
(thousands) (thousands) (thousands) intemationa growth e e ofnatural ofnet
l migran.ts rate birth death in.crease migration.
( thousan.ds) (percenrage) rate rate ( per 1000) (per 1000
(per 1000) (per 1000 ) )
1995 2000 1995- 2000 1995-2000 1995-2000 1995-2000 1995- 2000 1995-2000 1995- 2000 1995- 2000

World 5,666,360 6,055,049 649,050 260,360 o 1.33 22.1 8.9 13.2 o.o
Africa 696,963 784,445 140,575 51,655 - 1,435 2.37 38.0 13.9 24.l - 0.4
Asia 3,436,281 3,682,550 389,765 137,460 - 6,035 1.38 21.9 7.7 14 .2 - 0.3
Europe 727,912 728,887 37,465 41,240 4,750 0.03 10.3 11.3 - 1.0 1.3
Latin America and 479,954 519,143 57,770 16,225 -2,355 1.57 23.1 6.5 16.6 - 0.9
the Caribbean
Nortbern America 296,762 309,631 20,860 12,640 4,650 0.85 13.8 8.3 5.5 3.l
Oceania 28,488 30,393 2,635 1,135 405 1.30 17.9 7.7 10.2 2.8

Source: United Nations. 1999.

C/)
.o., o.o \
o
5 =-.
i ,. ,.g> - 1
;r _
..
 BASIC CONCEPTS ANO MEASURES 11
Tbe concept of the instantaneous growth rate enables us to develop a new expression for
populaon chaoge over a longer time interval. Integrating formula (1.3) between exact time s
o and T (also measured n years) , gives:
f T
r(t)dt = f T
dlnN(t)
- - - dt = lnN(t)]
' dt o
o T o
So:

J T
r(t) dt =ln (1.4)

(N(T))
N(O
o )

Taking exponentials on both sides we have:

e, Ol
T r(r)dl _
-
N(T)
--
N(Q)

or

N(T=) N(O)eftr (t )di (1.5)

Formula (1.5) is extremely important in demo graphy. Jt appears in many guises in many
differenl applica tions . It expresse s the change in population size during a paiiicular discrete
time pe1iod (io this case from O to T) as a simple function of the set of instantaneous growth
rates that prevailed during that period. Note tbat the proportionate growth in population over
the period, N(T)IN(O), is a simple function of the sum of growth rates. The order in which
those growth rates are applied is immate ria l ; ali that matters is their sum.
Viewing r(t) as a co ntinuously vai-ying function raises questions about the commonly
encountered te rm, "exponential growth." Any growth that occurs, including zero growth or
negative growth, must obey equation (1.5). An exponential appears in that formula beca use we
have de.fmed our measme of growth - the growth rate - in proportionate tenns. ln thi.s sense
the term "exponential growth" is a reduodancy ; all growth is exponential by our measure of
growth as the proportionate rate of change in population size. When people use the terrn "expo
nential growth" they are often (but not invai"iably) refeJTing to an N(t) sequence produced by
a constanJ positive growth rate within some time interval. Such a seque1Jce is probably more
predsely characterized by the term Malthus chose for it, "geometric growth," or by "constant
growth rate." If the instantaneous growth rate is in fact constant between time O and tme T at
some value r*, tben equation (1.5) simplifies to:

N(T) = N( O) re * T
(1.6)

This formula foUows from the fact that:

f
T

r* dt 1
= r * = r* T - r* O = r*. T
o
12 BASIC CONCEPTS ANO MEASURES

Rean anging equation (1 .6 ) and tak:ing natu ral logar ilhms gives: 1

ln (N(T))
N(O)
1 1.6.5
Weha
r* = - - - -
T (1.7) O and
inclu
Equa tio n ( l.7) shows that, if the inslantaneous growth rate is constant during the interva l O basic
to T, one can solve for its value by observin g the population size at the beginning and end of is con
the interval. sarne
unles
1.6.3 Mean annualized growth rate fuucti
Fr
lf we divide both sides of equatio n (1.4) by T, the length of the time interval over which
growtb is occuning, we have:

t T l n [N ( T ) J
1 0 N(O)
r(t) T
dt T
The left-hand side of thls equation is si mply the mean val ue of the in sta ntaneo us growth rate
A
over the peiiod O to T , which we will designate as r [O , T]. ll is the area under the r(t) func tion
between O and T, divided by the Je ngth of the time interval. Thus: Oand
' rates
does
l n [N ( T) J
Lhat
i
r[ O , T] = N(O) (1.8) to a
T figur
Note that tbe right- hand side of equation (1. 8 ) is identi cal to that of (1 .7 ); if tbe growth rate is s
consta nt between O and T , equation (1.8) provides a way of estimating its vaJue. But (1.8) is dist
clearly a more general expression since it requires no assumption of constancy. Perforrning the
s imple operation given by the right-hand side of equation (1 .8) provides the "mea n annualized
growth rate belween O and T ."

1.6.4 Ooubling time

If popuJation size double s belween time O and time T, then N (T)IN(O) = 2 and:

ln[ N (T) / N(O)] = ln[2] = .693

A population thus doubles in size beyond some i nitial date wheneve r the sum of its annualized
growth rates beyond that date equal s 0 .693 . If the growth rate is constant al r* , the population
will double whenever the producl of;, * and T , tbe Je ngt h of time (in yea rs) over which it is
app li e d, is 0.693.
So with coastaot growth rate r*,
.693
Doubling time
= r*
Under a cons ta nt a nnual growtb rate of 0.03, the population will double in .693/ .03 =
23. 1 years. With a co ns tant growth rate of 0.01, it will double in .693/ .01 = 69.3 years. S1ce Fi
-e 693 = 1 / e 693 = 0 .5, a population w ill be reduced to half of its initial size wheneve r the
sum of ammal grow U1rates equa]s - .693. ass
 BASIC CONCEPTS ANO MEASURES 13

1.6.5Comparis on of crude growth rate and mean annualized growth rate

We ha ve now developed two formulas for period growth rate s over the discrete in terva l between
0 and T: the crude growth rate anel the mean annualized growth rate. This section, which is in
(1.7)
clnded for complete ness and can be skipped by many readers , compares the two rates. The basic
lesson is that the two growth rates will be the sarne when tbe instantaneous growth rate is
be interval O
g and end of constant during the period O to T. Otherwise , the two rates will not , in general, have the sarne
value . However, d iffere nce s between them can usually be ignored for practical pur poses unless tbe
period of measurement i s very long (say, longer than 1O years) anel the growth rate function , r (t )
, is very inegular.
from (1.2), the crude grow th rate between O and T can be written as:
h ich growtb
B [O , T] - D [O , T] + J [O , T] - O[O, T ]
CGR[O , T] = T
.fo N(1) c/1
N(T) - N(O)
T (1.9)
.fo N(t)dt
growth rate
f function
As is clear in (1.8), r [O, T ] does not depend on the order iu which growth rates occur between
O and T. The 11U1nerator of CGR[O , T] in (1.9) is also independent of the order in which growth
rates occur. But the denominator of CGR[O, T] in '1.9), person -ye rus lived between O and T ,
does depend on the order iu which growth rates occur. A d i s tribution of positive growth rates
tha t is heavily skewed toward the beginning of the period will raise person-ye rus lived relative
(1.8) to a distribution that is skewed toward the end of the perio d. This tendency is illustrated in
figure 1.2. 1
owth rate is
So i t is clear tha t, in general, there can be no equality between CGR and r. An "ea rly"
ut (1.8) is distribution of growth rates will lowe r CGR relative to r,
and a "la te" dis tli bution wiJI
'orming the
\ann ualized
,
...
,l

.... ,
.' /

'I
i i II
:' I
Early growth ./ /
I
.. 1

unualized Population Constant ,... /

op ulat i o size N(t) growth_ .....- /
I
. I
n hich i t ,, /
/
is ........ .,// Late growth
,. .,,

o T
Time t
T
The sum oi growth rate s, r(t)dt, is the sarne in the three cases, since N(O) and N(T) are lhe sarne.

r3/ .03 = f o
P erson-years lived - the area under the N(/) curve - are different, however.
ars.
Since Figure 1:2 Po pulation grow th sequences between times O and T under three different as
nleve r t he us m ptton s about the time sequence of growth rates
14 BASIC CONCEPTS AND MEASURES

raise CGR relative to r. There is, bowev er, one ci.rcumstance in whicb CGR will equal r. This
occurs when the growth rates are constant between O and T. Suppose tbat r(t) = r* for O ::: 1.7
t::: T. Then:
The

J J
T T T
part
N(t) dt = N(O)/ ' dt = N(O) e,,*. dt pers
o o o
l . T N (O) e,*. 1 f
= N (O) - e' 1 ] = o
r* o
r* Mo
N(T) - N(O) (1.10)
r*

Substituting expression ( 1 .10 ) for person-years lived betw een O and T into equation (1.9) gives:

. N(T) - N(O)
CGR[O, T]= [ N (T) , : N (O) J = r*

ln the case of a constant growth rate , wt also have:

al

J
T est
r [O, T] = r* dt = r* wi
o yc
gr
So in the case of constant growth rates - and , except for rare ci_rcum stances, only in tbis case - jf
the crude growth rate will equal r. Differences between the two will normally be ttivial in size th
un less tbe growth rate sequence is extremely enatic and the time period (O to T) very long, th
say a decade or more. po
If one wants to ens ure that thecrude growth ratecalculated by (1.9) is in fact equal to the

I
mean of annual ized growth rates, then a simpie rule for computing person-years is i nd icated:
compute pers on-years lived during the period as though the growth rate were constant
throughout. Under th.is ci rcum stance, the denominator for calculating aU crude rates would
be:

N(T)- N(O)= [N(T)- N(O)] T

j. N(t) dt = r[O, T] ln (N(T))

N(O) '
if r 1- o

o T N(O),
if r =o
Although we defined the "mean annualized growth rate" as the average of period rates,
in equation (1.8) it does not have person-years i n tbe denominator, which was said to be
a typical feature of a demographic rate. ln this format, it shares tbe characteristic of many it
e
rates in conunon usage, such as a mean rate of speed or mean rate of infl atio n. But unde r the
simplifying assumption that the "mean annualized growth rate" is constant during the interval
of measuremen t, its value is in fact iden tical to that of the crude growth rate, which does y
explicitly contain person-years in the denominator.
 BASIC CONCEPTS AND MEASURES 15

1.7 Estimating Period Person-years

The above argurnent suggests that, if oneknew nothing about the path of N (t ), or r (t ) , dw-
ing a particular yeai:, one sh uld assume constancy of the growth rate during the period and
estimate person-yeais lived dunng the year as:

PY[O, l=] N(l) - N(O)

N(l)- N(O)
r[O, 1) ln [N(J)J
N(O)

(1.10)
More generaJly, when tbe period is not necessaiily one year long,
1n (1.9) gives:
(1.11)
P Y[O T]= [N(T)- N (O )]
, [N( T )]
ln --

T
N(O)

Using equation (1.11) to estimate person-years has the advantage of forcing consistency
between tbe crude growtb rate for the pe1iod and the mean annualized growtb rate for tbat
period, and it would be exactly coITect if tbe growth rate were constant during the pe1iod.
But it does require observations on populatin size at the beginning and end of tbe period. It
is often the case (e.g., in the United States) tbat population size estimates are only available
at mid-year. It wiJJ usually be perfectly acceptable to use the mid-y eai population size as an
estimate of person-yeais lived du1ing the year. The mid-yeai approximation to person-years
will be exactly correct if the N(t) sequence is linear between the beginning and end of tbe
year, as demonstrated in figure 1.3. Even if the N(t) sequence is a product of a constant
growth rate, the errar in using tbe mid-year approximation will be very small . For example,
if r = 0.03 (rapid by hist01ical standards), the ratio of true person-yeais lived in a year to the
Jn this case - mid-year popLLlation will be 1.00004. The mid-year population will always underestimate
irivial in the true number of person-years lived if the population is changing ata constant rate, whether
size positive or negative.
) very long, More caution is necessary in using mid-period approximations to estimate person-y eais
when the interval of time for which an estimate is sought extends far beyond a year. For
l to themean example, if we estimate the person-years lived during a 10-year peiiod in a population growing
ed: compute at 3 percenl a year by taking the mid-period population times IO (i.e., mid-height times width),
ho u t. then the ratio of true person-years l ived to our estimated person-years will be 1.0038. Tliis
Under errnr of about four-tenths of 1 percent is too large to ignore for most purposes. Note that if we
had used the aiithmetic mean of beginning and end-peri od populations (times 10) as our
estimate of person-years lived in this example, we would have overestimated true person-years
by thc factor l.0075. So this procedure provides an even poorer estimate of person-years than
does the m.id-period population in a populatioo having a constant positive growth rate.
If rnid-yeai population estimates U-e available for each year during a 10-year pe1iod, a
sensible way to estimate person-years lived dming the period would be simply to add up the
1O estirnates. If observations aie available at the beginning, middle, and end of the period,
riod rates, tben it is possible to ascertain whether growth is more nearly Jineai or exponential and to use
said to be the corresponding approximation for each half-period.
ic of many Although it is convenient and fairly accurate to estimate person-years lived dming a p,u-tic
t under tbe u lar year as the population size in tlle middle of the year, it is important to remember that tbe
the interval resulting demographic rate should not be expressed as a number of occurrences divided by a
hich does
16 BASIC CONCEPTS AND MEASURES

When N (t) follows a linear growth

pattern, the estimate of person-years
Population lived using the mid-period population
size N(t) times period lenglh will be accurate rate o
because lhe overestimale for the firsl
half-period is exactly offsel by lhe N(O)
undereslimale for lhe second half-
period, i.e., lhe two triangles have N(5)
o T/2 T equal areas.
N(l
Time t

1.

When N (f) follows an exponential

growlh pattern, lhe lwo shaded surfaces
Populalion have different areas and the mid-year
size N(t) approximation, N (T/2) T, will under-
ealimate person-years lived during 2.
lhe period.
N(t) a

o T/2 T
Time t

Figure 1.3 Approximation of person-years lived by mid period population times period length

number of peop]e. The unit in which exposure-time is measured (usually, person -years) must
not disappear, or confusion is inevita ble. We are nsing the mid-year population as an estim ate
of person -yea rs Lived during the perio d, and not as a substitute for person-years. The risk of
confusion is greatest wben an annualized rate is being estimated for a petiod that is not one
year in Jength. Box. 1.3 illustrates the computation of growth rates and person-ye ars lived
during a 10-yea..r period in a hypothetical population with a constant annualized growth rate
of0 .03.

1.8 The Concept of a Cohort 3.

Almost as important to demography as the concept of a population is tbe concept of a cohort.
A cohort is the aggregate of all u nits that experience a particular demographic event during
a specific time interval. As in the case of a population, a cohort always has some specific
geographic referent , whether it is explicit or implicit. A cobort usually consists ofpeople, bu!
it may a]so consist of entities (e.g., marriages) formed by a demographic event. The cohort
is usually identified verbally both by the event itself and by the time period in which it is
experienced. Some examples of cohorts are:
"US birth cohm1 of 1942," which refers to all persons bom as US citizens in calendar year
1942;
 BASIC CONCEPTS AND MEASURES 17

Box 1.3 11/ ustration of Ca/cu/ation of Cirowth Rates and Person-years

Suppose that a population had 100,000 persons at time O and that it grew ata conslant annualized growtl1
rate of 0.03. Then:

N(O) = 100, 000

N (5) = 100,000 - e5 03 = 116,! 83
N(IO) = 100,000 e 10- .D3 = 134,986

1, Calc ula ting the mean annualized growth rate between t = O and 1 = 10:

l n ( -N- (l ln(-
13 4 ,98 6)
- -
r[O 10=] O)) - J 00,000 = 0.0300
N (O)
, . 10 10
2. Estimating perso n- y, e uslived between 1 = O and 1 = 1 O:

a) Assuming a constant growth rate:

1

PY[O T]= N(T) - N(O)= N(lO) - N(O)= 13 4,986 - 100,000=

' r [O , T] i' [O , 10] 0.03 ' '

b) Assuming growtb is Linear and using the mid-peiiod approximation:

PY[ O, T]= N(T/2) T

PY[O, 10] = N(5) 10 = 116,183 10 = 1,161,830

e) Assuming growth is linear and using tl1e mean of initial and final popu lation sizes:

PY [O , T] = [ N(O) J T
N(T )

PY [O , 10=] [ N(O) N(IOJ) 10

-_ [ 1000,00
2+13 4, 986 ] . 1 O_ - 1,174 9, 3O

3. Calculating crude growth rates based upon vmious estimates of person-years Lived:
3 4 ,986
a) CGR [O, T] = -- - = 0 .0300
1,166,200
b) CGR [O, T ] = 3 4 986 = 0.030 1
1 ,161,830
e) CGR [O, T] = 34 986 = 0.0298 1

1,174,930 1
18 BASIC CONCEPTS AND MEASURES

" French marriage cohort of 1990," which refers to aJl marriages contracted in France d Luign
the calendar year 1990;
"French fernale man-iage cobort of 1990," wbich refers to ali women who rnarried in France
in 1990;
"Austrian immigrant coho1t of L995," which refers to ali immigrants into Austria in L995.
The most frequently encountered type of cohort is a birth cohort. Persons who are born
during the same petiod are destined to pass through life together, in the sense that they will
reach their xth birthday dming a period exactly x years beyond that which defined thei.r cobort
membership. For the US birth cohort of 1942, all would reach their 10th birthday (assuming
that they survived) i n 1952, the- 15th birthday in 1957, and so on. The time period that
ci.rcumscribes the cohort need not be one year ill le ngth; it is common to deaJ with such entities
as the US birth cobort of 1918- 22, for exarnp le .
To calculate a rate for a cohort, we sirnply rest1ict the counting of occurrences and person
yeais of exposure to people who were born during tbe period that defines mernbersbip in the
cohort. The lines below show tJ1e counting schema for a birth cohort defined by birth i.i1 the
period ao to a 1 :

A, B,

A, B,

0,

1--- - - - - - - - --t- - - - - - - - - - - - - - - - - - - - - -+ Time

Allbough those life-lines refer to a bi1th cobort , the concept can clearly be extended to other
types of cohorts.

1.9 Probabilities of Occurrence of Events

We can define an additional concept for a cohort that is -npossible for a popula tion : the
concept of a probability. Tbe term is used u1 demography in a rnanner similar to its usage in
statistics.
It refers to the chance that some event will occttr, rather than to the rate at wlch it occurs.
T hus , for example, we may compute the probability that a marriage would end in a divorce for a
given birth cohort by counting, over all members of the cohor t, the nurnber of ma1Tiages and
tbe number of divorces over the cohort's lifetime:

D Number of Divorces
q . = N umb er i,FM
arn ages
ln doing so, we have used a "relative frequency" approach to estimating the probability
of divorce. We have said, in effect, Lhat om best guess about the true underlying probability
 BASIC CONCEPTS AND MEASURES 19
of divorce in the cohort is the observed frequency of divorce. Tbe situation is analogo us to
drawing balis oul of a vety large um . If we draw a sample of 10 balis and 2 of them are red,
theo the reJative frequency of red balls in that drawing is 0.2. This relative frequency is also
the maximum likelihood estimator of tbe t:rue proportion of balis in the entire urn that are red,
assuro.ing that d1e outcome of drawings is independent. That is, a proportion in tbe entire um
of 0.2 is more likely tban any otber proportion to have given rise to the observed sarnp le of
[O baJls of which 2 are red. Many introductory statistics texts contain a clear discussion of
mamum likeliliood estimalion.
The structure of a probability in demography is thus quite different from the structure of
a rate:

Number of Occurrences
Rate -----------
= Number of Person-years Lived

P /V'Ja
l b . .tltty = --------------
Number of
Number of Preceding Events or Triais

The denominator ofthe probability indicates tbat itis not possible to define a probability unless
Lhereis some even.t or trial (equ ivalent to tbe act of drawing balis out of an urn). Since each
occurrence in tbe numerator (e.g., divorce) must be preceded by an event in the denominator
(rnai.Tiage), tbe number of occu1Tences cannot exceed the number of preceding eveots. Thus
tbe probability cannot exceed one and , since we are only dealing with positive quan tit ies ,
probabilities cannot be nega tive.
Populations do not have probabiJities except insofar as they pertain to cohorts that are
incJuded in tbe population. Although we could count the number of marriages in a populalion
during some calendar year and the number of divorces during that year, lhe two numbers
combined do not give a sensible estimate of the probability of divorce becau se they don'tapply
to the sarne cohort. We are, in effect , counting events (or triais) in one um and occmTences in
another. Ifwe happened to choose a year in a smal l population where no one manied but there
was a divorce , our popul ation 's probability of divorce qD would be 1/0 = oo, an obviously
absurd outcome . Only when we count the events pertaining to the cohort at risk of the evenl
can we properly define a probability.
The coocepts of cohorts and of probabilities that certain eveots will occur to cohorts can
be applied to a vast number of situations extending weU beyond demography 's customary
range. .They are central to aJI ana lysis of longitudinal data in the social and health sciences.
Perhaps their major utility derives from the fact that they translate aggregate -leve l measures
uno implications for individua is. They help "locate" d1e individual in an od1erwise amorphous
and u11differentiated population .
Despite its conceptual simplicity, analysis of data on actual cohorts suffers from severa}
major practical lintations. Fi.rst, computing cohort rates and probabilities requires complete
informatioo 011 eacb individual until he or she has died (or at least has ceased to be " at risk"
of the event of interest). We may Iose trnck of some individuai s, for instance, when they move
out of the area of the stu dy. Out-migration is part of a more general problem called " l o ss to
follow-up ." We deal with one way of coping with this problem in chapter 4. A more serious
practical problem is that, by the time the cohort's expetience is completely observed, much
of lhe experience may be ancient. ln order to provide more timely information, demographers
rcly primarily 011 data for recent periods. The measmes that are constmcted from period rates

include life expectancy, expected years to be lived in the single state, total fertility rate, net
reproduction rate and gross reproduction rate. They also include probabilities of dying, giving birth,
migrating, and so on. In constructing these and other measures, demographers re]y on the concept
of a cohort, but adapt that concept to deal with data pertai1ng to a pe1iod. The principal adaptation is
the introduction of "hypothetical cohorts," a concept that wiU be encountered frequently in the
remainder of this volume.