unregistered. Figure 3 shows that among infants who medical staff recorded as live-born and who died at 0-6 days, registration was reasonably effective because it was the hospital’s responsibility. But among older infants, where parents were responsible for registration at civil offices, compliance was poor. It is not clear from this exercise how many of these infants’ births were also unregistered. Nonetheless, if non-registration of infant deaths occurs in some other countries in the region on a similar scale as in Armenia, then this clearly constitutes one important source of undercounting in official infant mortality rates. Figure 3: Infant deaths in Armenia registered at medical facilities but unregistered at Civil Acts Registration Bureau, 2000 (absolute number from survey of 519 deaths) 7-28 days 29-365 days all infant deaths regis tered unregis tered Source: Ministry of Health and UNICEF (2002) Infant mortality in Armenia: A review of procedures and registration, classification and related medical services, Ministry of Health Armenia and Republic of Armenia: Yerevan, Table 15. Note: Medical facilities issue medical death certificates in cases where they are involved in treatment of the infant. If the infant is less than a week old, medical facilities are responsible for registering deaths with the Civil Acts Registration Bureau (CARB), which compiles official vital statistics for the National Statistical Service. If the infant is more than 6 days old at time of death, parents are responsible for registering the death directly with the CARB, without which the infant’s death will be excluded from the official count. The total number of infant deaths reported by the National Statistical Service in 2000 was 540 (Source: MONEE Project Database). 20 5. ESTIMATING INFANT MORTALITY FROM SURVEY DATA While the preceding three sections outline issues that strongly point to undercounting of official infant mortality rates in several countries, alternative estimates, derived from survey data, cannot simply be assumed to be ‘correct’. This section seeks to add to existing literature on the survey based estimation of infant mortality in two important ways: first, to point to some specific problems in the estimation of infant mortality from survey data in former communist countries during the transition; and second (and more generally), to question some assumptions made about the indirect estimation of infant mortality from limited maternity histories. Calculating infant mortality from survey data As Table 1 shows, survey based estimates of infant mortality are based on three main survey types – the DHS, the RHS and the MICS. In addition, one LSMS conducted in Tajikistan in 1999 included a module that allows the calculation of infant mortality rates. All surveys were based on random stratified samples, and all aimed to be representative of either national populations, or of women of reproductive age (15-44 or 15-49).12 The way infant mortality rates are computed depends on what type of information is collected in a survey. The DHS, RHS and the Tajikistan LSMS ask respondents to report all their pregnancy outcomes in some detail. For all live births, date of birth, survival status and (for not surviving children) age at death were obtained. The mortality rates calculated from these surveys are based on the deaths of children over a specific time period, usually five years preceding the survey.13 In MICS, on the other hand, women are asked to report the total number of births they have had in their life and total number of children who were born alive but later died. Because no information is collected regarding dates of birth (except for the very first birth) or age at death, mortality rates can be estimated only indirectly, using a method developed by Brass (1964). This method (discussed in more detail below) produces estimates of mortality for several reference years, which are usually averaged into a final estimate, represented by a mid-point year. Non-sampling and sampling error Asking women about their children who have died is a difficult matter, not least because the death of children is invariably painful for a mother to talk about. There is a danger that some mothers may withhold some details, or 12 For information sources on the surveys, see Appendix I. 13 Data for the most recent period will include some children who have not been exposed to the risk of dying for a full year. To compensate for this, a method of synthetic cohort probabilities is used to compute probability of death for this group. A description of this procedure can be found in Sullivan et al (1994). 21 ‘overlook’ some deaths, or even terminate the interview. Responses such as these will affect the quality of the survey. Even where mothers answer all questions, they may interpret them in different ways. While respondents to all surveys are asked to report pregnancy outcomes according to the WHO definition of live birth, it is not always clear how respondents themselves interpret both questions and events. In the case of a death that occurred shortly after birth and where medical staff assisted at the birth (as is the case with most births in CEE/CIS countries), there are three possible ways how the mother might report this event to the interviewer: * she will report what she recalls herself * she will repeat what she has been told by medical personnel * she will report