Coaie and Trussell.

6 The general theory on which these of children do not live with their mothers or cannot be
methods are based îs essentially the same, but they properly iinked to them because of poor information,
arrive at somewhat different multipliers because the data the parity-ratio approach is veiy likely to be better. A
bases used in each case are difTerent. Since the Sullivan detailed description of the Preston-Palloni method is not
variant has no obvious advantages over that proposed inciuded here, in part because in most cases where mor­
by Trussell, whereas the latter is based on a wider range tality needs to be estimated indirectly, age distributions
of cases, the Trussell procedure is described here. It are at best only moderately reliable, and in part because
must be mentioned, however, that the multipliers the data required for its application are not as widely
presented are a more recent and more satisfactory ver- available as the proportions of children dead. However,
sion of those originally proposed by Trussell in 1975. the user who has access to the former data for cases
It is important to take note that this method of estima- where biases due to fertility change may be a problem is
tion is based on the assumption that fertility and child- encouraged to consider the application of this method.
hood mortality have remained constant in the recent Probably a more widespread problem is posed by de­
past. If, for example, fertility has been changing, the clining mortality. The procedures outlined above all
ratios of average parities obtained from a cross-sectional assume that a constant pattern and level of mortality
survey will not replicate accurately the experience of have prevailed in the recent past of the population under
any cohort of women and will not provide a good index study. In most countries, however, mortality has been
of the distribution in time of the births to the women of declining.
each age group. Feeney* 9 10 was the first to examine the effects of chang­
The problems caused by declining fertility can be ing mortality on the performance of the child-mortality
avoided when data for true cohorts are available (from estimation procedure. Using infant mortality as an
censuses or surveys taken five or 10 years apart). In this index of mortality level in a one-parameter logit life-
case, an estimation method specifically designed for table system, he calcuiated the proportions of children
cohorts experiencing fertility change should be used.7 dead that would be observed if infant mortality were
Preston and Palloni8 propose an alternative approach changing linearly through time. On the basis of these
to estimate the time location of births, which circum- simulated cases, he showed that for plausible annual
vents all the problems associated with changing fertility. rates of change in infant mortality, the ţ(l) values
This approach is closely reiated to the “own-children” estimated from data on children ever bom and surviving
procedure for estimating fertility from an age distribu­ for different age groups of mother could be matched
tion (see chapter VIII, section C). If it is possible to link, with the 9(1) values prevalent during a set of years
within households, the records of mothers and their sur- before the survey; and that this set of years was, for all
viving children, it becomes possible to tabulate surviving practicai purposes, invariant with respect to the rate of
children according both to their own age and to that of mortality change. Using this empirical finding, Feeney
their mothers. Given an age pattern of mortality, say, developed an estimation procedure to establish the set of
from one of the Coale-Demeny regional model life years to which infant mortality rates estimated from data
tables, the combination of the proportion of children on children ever bom and children surviving refer. This
dead and the age distribution of the surviving children procedure was developed from data generated by using
of women from some particular age group uniquely a one-parameter logit life-table system derived from the
determines a level of mortality. The age distribution of general standard (see chapter I, subsection B.4) and the
surviving children is used to define the age distribution Brass fertility polynomial. 0 The use of q( 1), infant mor­
of children ever bom without recourse to fertility tality, as an indicator of mortality level and as the
models. In an actual application, the choice between the estimated parameter makes the underlying age pattern
age distribution of surviving children and the ratio of of mortality important to the results, since similar
consecutive parities to estimate the real distribution of overall levels of mortality (life tables with the same
births over time, depends upon data availability and expectation of life at birth, for example) can be associ­
upon a rough assessment of the likelihood each ated with markedly different infant mortality rates. As a
approach has of yieiding the best possible estimate. In result, the Feeney method is likely to yield biased q( 1)
cases where age-reporting is good and most children live estimates when the mortality pattern in early childhood
with their mothers, the approach suggested by Preston of the population under study does not resemble that
and Palloni may be the better, particularly if fertility is embodied by the general standard. For this reason,
changing. In cases where age-reporting or completeness Feeney’s original method is not described in detail.
of enumeration is poor, or where a sizeable proportion It is fairly straightforward to apply Feeney’s approach
to data generated with other mortality models. Coaie
* Ansley J. Coaie and T. James Trussell. “Model fertility schedules:
variations in the age structure of childbearing in human populations".
Population Index, voi. 40. No. 2 (April 1974), pp. 185-258. 9 Griffith Feeney, “Estimating infant mortality rates from child sur-
' This method is presented in section E and it uses the estimation vivorship data by age of mother”. Asian and Pacific Census Netvsletter.
equations and coefficients given in tables 70-71. voi. 3, No. 2 (November 1976), pp. 12-16: and Griffith Feeney. "Es­
* Samuel H. Preston and Alberto Palloni, “Fine-tuning Brass-type timating infant mortality trends from child survivorship data". Popula­
mortality estimates with data on ages of surviving children”, Popula­ tion Sludies, voi. XXXIV. No. I (March 1980). pp. 109-128.
lion Buuetin of the United Nations, No. 10-1977 (United Nations publi- 10 W. Brass. Methods for Estimating Fertility and Mortalityfronţ Lim­
cation, Sales No. [Link].6), pp. 72-87. ited and Defective Dala.

74