Skip to Content

Introduction to child mortality analysis


The level of child mortality in a country has long been treated as an index of general development. Over the last two decades, the international community has established periodic targets for the reduction of child mortality. The World Summit for Children in 1990 set the goal of reducing 5q0 by one-third, or to below 70 per thousand, between 1990 and 2000. The 1994 International Conference on Population and Development reiterated this goal, but set the additional longer-term target of reducing 5q0 to below 45 per thousand by 2015. The Millennium Summit in 2000 then adopted a set of eight development goals, of which the fourth included a target of reducing 5q0 by two-thirds between 1990 and 2015. The report of the Commission on Information and Accountability for Women’s and Children’s Health, established by the Secretary-General of the United Nations, has recently reaffirmed the importance of frequent reporting on 5q0. It is thus clear that the measurement of infant and child mortality has a very high priority both at the national and at the international level. This section discusses broad options for such measurement, with details of the various methods discussed in other sections.

Data for estimating under-5 mortality

Data on child mortality from sources other than civil registration systems (and often the data in civil registration systems as well) come almost exclusively from mothers. The data are collected either about all (generally live-born) children or about a specified child or group of children. They are collected with varying amounts of detail, bearing in mind the difficulty of recalling details of births long in the past. Lack of detail in data collection may subsequently be compensated for at the analysis stage by making assumptions, primarily about past age patterns of child mortality and fertility, but with increased uncertainty.

The fact that data are collected from mothers is important, because it raises the issue of selection bias. If all the children (alive or dead) in a target population over an extended period had mothers who were still alive and also still in the population, then each child has an equal probability of being reported and there will be no bias. However, in practice mothers are themselves subject to mortality and migration risks and, as a result, at a subsequent date it will no longer be the case that all children in a target population and only those children in the target population will be reported. The largest threat in terms of bias is in populations substantially affected by HIV/AIDS. In such populations, in the absence of extensive use of antiretrovirals, there will be a strong correlation between the mortality risks of mothers and of their children, such that child mortality as reported by surviving women will underestimate true child mortality. This issue is discussed in more detail in the section on effects of HIV on methods of child mortality estimation. The longer the time lag between the birth of a child and the data collection, the greater the potential for such biases.

We distinguish below between data collected through a household interview (including data collection exercises that aim for universal coverage of a population, such as a census, and those that collect data from a representative sample of a population, such as Demographic and Health Surveys) on the one hand and data collected at a health facility, such as a birthing centre, on the other.

Data collected by household interviews

Surveys collecting data through household interviews can range from universal (a population census) to small and focused samples. The amount and complexity of data that can be collected in an interview varies by survey size: in general, large surveys permit only short questionnaires and limited interviewer training, whereas small surveys can (subject to financial constraints) use much longer questionnaires and interviewers can be trained more carefully to collect much more detailed information. More detailed information can support analysis of a wider range of indicators, whereas simple data for a larger sample or for the whole population permit much greater disaggregation by geography or socio-economic factors. The two sources may thus be seen as complementary rather than competing. In this section, we describe types of data from the most complex to the simplest. We then proceed to describe the process of data evaluation and analysis.

The full birth history

A full birth history (FBH) collects information from surveyed mothers about each live-born child to whom the woman has ever given birth. In most settings, information is collected from all women of reproductive age (normally 15 to 49). However, in situations where it is considered inappropriate to ask never-married women about childbearing, questions may be limited to ever-married women. The minimum information collected about each child is name, date of birth (often collected as month and year), whether the child is still alive, and if the child has died, age at death. In Demographic and Health Surveys (DHS) age at death is collected in days if the child dies before 28 days, in months if the child dies between 28 days and 24 months, and in years thereafter. From an analyst’s point of view the use of years as units is less than optimal since deaths cannot be located with adequate precision in calendar time.

In the DHS, data are collected by order of the birth from the earliest to the latest, with probes between each to make sure no child is unreported. Additional information is often collected on whether the child was one of a multiple birth, the child’s sex, whether the child is living with the mother, and, if so, the line number of the child on the household listing. The data collected thus allow the location of each birth and, if the child has died, its death in time, permitting the calculation of conventional occurrence-exposure rates and use of standard life table analytic methods to estimate mortality risks for age ranges of childhood.

In the DHS, the full birth history is prefaced by a detailed summary birth history (SBH), whereby each woman is asked about the number of sons and daughters living at home, the number of sons and daughters living elsewhere, and the number of sons and daughters that have died. The FBH is then collected, and the FBH and SBH are field-edited to ensure consistency. Because this edit is done in the field, DHS provides no evidence on the frequency, magnitude or direction of discrepancies between the FBH and the initial SBH.

Some surveys have collected full pregnancy histories, adding all pregnancies that did not result in a live birth. This has not been the norm in the DHS, although the model questionnaire does include the collection of data about pregnancy losses that occurred in the five years or so before the survey. It has been argued that a pregnancy history will collect better data about live births than a birth history, by increasing the likelihood that very early neonatal deaths will get reported, but there are no experimental data to bear this out.

As will be clear, the FBH is a complex data collection instrument that requires extensive fieldworker training and careful field supervision. It is only used when the woman herself can be interviewed; proxy reports are not allowed. It is not appropriate for very large-scale data collection exercises such as a population census.

The truncated birth history

Some surveys have used truncated birth histories (TBH) to reduce fieldwork costs and possible interviewee fatigue. A TBH collects the same information as a FBH about each birth, but limits the number of births for which information is collected either by time period (for example, to births in the preceding 5 years) or by number (for example, for the most recent births up to a limit of three). Experience suggests that TBHs are a risky form of data collection for both theoretical and practical reasons. A time limitation drastically reduces information about mortality risks of somewhat older children (for example with a five-year cut-off, only children born between 4 and 5 years ago will provide information about mortality between the ages of 4 and 5, and will provide no information about mortality above age 5). It also runs a practical risk that children that have died may be intentionally shifted out of the time window. A number limitation raises concerns about selection bias (women with more than 3 recent births will only report on the most recent 3) and also selective reporting (potentially favouring reporting on surviving children rather than children who have died).

The summary birth history

In a summary birth history (SBH), the only information collected is aggregate numbers of children ever born and children still alive (or dead). The amount of detail varies, from just two questions (number of children ever born and number of children still alive) to the detailed SBH asking about boys and girls separately, and enquiring about surviving children living at home versus those living elsewhere, as described above for the DHS FBH sequence of questions. It is believed, without any empirical basis as far as we know, that the more detailed form of the questions improves data quality. Unlike the FBH, the SBH is often used with proxy reports. In a census, for example, a single household informant will usually provide the information. The informant will sometimes be the woman herself, but by no means always. In some cases a male head of household will provide information for all the women living there.

The SBH does not provide a basis for direct calculation of standard mortality indicators. Events (deaths) are not located explicitly in time, and there is no basis for calculating precise exposure times. Thus conventional occurrence-exposure rates cannot be calculated from the data. Instead, the data are analyzed using an indirect estimation of child mortality approach, whereby the age (or sometimes the duration of marriage or the time since first birth) of a group of women is taken as an indicator of the average exposure time of their children to the risk of dying, and model age patterns of fertility and mortality are used to convert the proportions dead of children ever borne by women in a group into a standard life table function.

Survival of a recent birth

The SBH does not in practice provide a very recent estimate of child mortality, for reasons explained in the section on indirect estimation of child mortality. Censuses and large-scale surveys have attempted to obtain a more recent estimate by asking about the survival or otherwise of a recent child. (Information on a recent birth is also used as a basis for fertility estimation.) There are essentially two forms of the questions in common use. One asks whether a woman had a live birth in a specified time period (such as the 12 months before the survey) and whether that child is still alive. The other asks for the date (month and year) of the woman’s most recent live birth and whether the child is still alive. Either form of question can be used to calculate the proportion alive of (a large majority of) children born in the 12 months before the survey. If births are distributed evenly across the 12 months, the proportion alive estimates the life table parameter

1( 1 L 0 l 0 )

. The second form of question can also be used to calculate survival to later ages of childhood, as proposed by the Brass-Blacker procedure below.

In practice, as described in the section on assessment of recent fertility data, the number of births reported in the last 12 months is often implausible (usually too low), raising issues about selectivity or reporting bias possibly related to a child’s survival.  Conversion of the parameter

1( 1 L 0 l 0 )

into the generally-used index 5q0 is also dependent on the choice of a model life table system.

Blacker and Brass (2005) suggest using data from the second format (date of most recent birth) to estimate child mortality from survival of most recent births in the two years before a census or survey.  The authors show that for typical age patterns of child mortality, the proportion dead among such births can in the absence of selection bias be converted into an estimate of the infant mortality rate 1q0 simply by multiplying by a factor of 1.09.  However, this adjustment factor is based on the assumption that the most recent births in the last 24 months are not selected from all births in the last 24 months (i.e. including all births of women with more than one birth in the period) on the basis of survival.  This assumption turns out to be surprisingly far off the mark. Analysis of DHS full birth history data shows that for women with more than one birth in the period, the earlier birth(s) are much less likely to have survived than the most recent.  A major reason for this is that birth intervals after an early child death are systematically shorter than those after a child that survives infancy, with the consequence that births in the last 24 months that do not survive are more likely to be followed by a subsequent (potentially surviving) birth. When this happens, the child that died does not get reported as a most recent birth.  In a sample of African DHSs, the proportion dead of most recent births in the last 24 months is biased downwards relative to the proportion dead among all births in the 24 months by over 20 percent on average. Even in Bangladesh, a population with long birth intervals where the effect would be expected to be smaller, the bias still reaches almost 16 percent.

In view of concerns about both data quality in the case of survival of most recent births in the 12 months before a survey and selection bias when data for 24 months are used, the analysis of information on survival of most recent births is not recommended.

Data collected at health facilities

A major expense of household surveys is the cost of getting an interviewer to the (correct) household. Much of this expense can be eliminated by taking advantage of respondents coming to the interviewer, such as on visits to health facilities. Health facilities are also likely to record births and deaths that occur at the facility as part of a routine health management information system. The problem with using such data for child mortality estimation is selection bias in that we can never be sure that the women who visit a facility are representative of all mothers. To improve coverage, experiments are being conducted to find out whether health extension workers or other health staff working in communities can collect adequate data on births and deaths. Such an approach is akin to a sample registration system.

Two approaches to collecting data from women at the health facility have been proposed, and one, the previous birth technique (PBT), has been extensively developed (Brass and Macrae 1984). In essence, what this method does is ask women coming to a facility to give birth whether their previous birth, if they had one, is still alive. Given typical birth interval distributions, the proportion of previous births that have died can be interpreted as a probability of dying by an exact age of childhood. The data are extremely easy and inexpensive to collect, and can provide an indication of child mortality trends even if the level may be affected by selection bias. Such trends, however, will not be accurate if the selection bias changes materially over time.

The second approach starts with the assumption that women who visit a health facility are a biased sample of all mothers, but does not assume a priori the direction of the bias. Instead of assuming that women with healthier than average children or women with less healthy children will predominate, it aims to estimate the selection probabilities so as to be able to adjust for bias. For example, women visiting a health facility could be asked their age, children ever born and children still alive, plus a number of additional questions about their socio-economic characteristics. The children ever born and children still alive could then be modelled onto the socio-economic structure of the whole population, available for example from a population census. To our knowledge, this approach has never been tested.

Further reading and references

In addition to the two papers cited above, the interested reader is directed to the work of Hill (1984, 1991), and Manual X (UN Population Division 1983) for further historical reading on indirect estimation; to Rajaratnam, Tran, Lopez et al. (2010) for recent developments in modelling under-5 mortality, as well as Preston (1985) and Rutstein and Rojas (2003) for a description of direct estimation of child mortality from survey data.

Blacker J and W Brass. 2005. "The estimation of infant mortality from proportions dying among births in the past 24 months", Southern African Journal of Demography 10(1-2):25-42.

Brass W and S Macrae. 1984. "Childhood mortality estimated from reports on previous births given by mothers at the time of a maternity: I. Preceding-births technique", Asian and Pacific Census Forum 11(2):5-8.

Hill K. 1984. "An evaluation of indirect methods for estimating mortality," in Vallin, J, JH Pollard and L Heligman (eds). Methodologies for the Collection and Analysis of Mortality Data. Liège, Belgium: Ordina Editions, pp. 145-176.

Hill K. 1991. "Approaches to the measurement of childhood mortality: A comparative review", Population Index 57(3):368-382. doi:

Preston SH. 1985. "Mortality in childhood: Lessons from the WFS," in Cleland, J and J Hobcraft (eds). Reproductive Change in Developing Countries. Oxford: Oxford University Press, pp. 253-272.

Rajaratnam JK, LN Tran, AD Lopez and CJL Murray. 2010. "Measuring under-five mortality: Validation of new low-cost methods", PLoS Medicine 7(4):e1000253. doi:

Rutstein S and G Rojas. 2003. Guide to DHS Statistics. Calverton, MD: ORC Macro.

UN Population Division. 1983. Manual X: Indirect Techniques for Demographic Estimation. New York: United Nations, Department of Economic and Social Affairs, ST/ESA/SER.A/81.