example the reclassification of infant deaths as stillbirths, also appears to have increased over the 1990s. In some countries, deaths of older infants may be recorded as deaths of children aged over one year. And it is clear that in a few countries, the registration of births and infant deaths is less than complete: unregistered births and deaths are not included in official statistics. It must be added that most of this hard evidence pertains to CIS countries. Similar investigations have generally not been carried out in the countries of South Eastern Europe. Can surveys be used to provide reliable alternative estimates of infant mortality in the CEE/CIS region? The answer is complex. For the most part, they appear accurate enough to support the contention that official data greatly understate the true infant mortality rate. But some factors that are perhaps specific to CEE/CIS countries may increase uncertainty. With some estimates of infant mortality from survey data, the degree of uncertainty is particularly wide. The paper is organized as follows. Section 1 discusses how infant mortality is typically measured, and presents estimates from different sources for countries in the region. Section 2 looks at the definition of ‘live birth’, which is particularly important in the CEE/CIS context. Section 3 examines misreporting of infant deaths, and Section 4 considers the problem of nonregistration of births and infant deaths. In Section 5, attention is focused on uncertainty associated with survey based estimates. Section 6 concludes. 1. MEASURING INFANT MORTALITY The infant mortality rate represents the probability that a child will die before reaching age of 1 year in a specific period of time (usually a year) and in a specific geographical area (usually a country). There are number of ways of calculating infant mortality rates. For example, the method used by the 3 Russian statistical office, Goskomstat (and generally recognised as acceptable), m = where N and M are, respectively, the number of live births in the year for which the rate is computed and the number of infant deaths that follow these births in the same year; N -1 is the number of live births in the previous year, and M -1 is the number of deaths of infants in the current year, who were born in the previous year (Goskomstat, 2002).2 Table 1 shows recent infant mortality rates for all 27 countries in the CEE/CIS region. The official rates for 2001, provided by national statistical offices and available for all countries except Tajikistan, are calculated from civil registration data on births and deaths. The rates range from 4 per 1,000 live births in Czech Republic and Slovenia, to 44 per 1,000 live births in Kyrgyzstan. Survey estimates, calculated from household surveys where women of childbearing age are asked about their reproductive histories, are available for 11 countries in the region. These surveys, Demographic Health Surveys (DHS), Multiple Indicator Cluster Surveys (MICS) Reproductive Health Surveys (RHS) and Living Standards Measurement Surveys (LSMS), have been carried out since the 1970s in many developing countries where there are usually few official data available for monitoring well-being and for planning public services such as health care. Most of the surveys used in this analysis were carried out in the late 1990s, and estimates of infant mortality refer to a particular point in time or to a 5 year period prior to the survey year. For four countries (Azerbaijan, Kazakhstan, Tajikistan and Uzbekistan) there are two independent survey estimates. In the case of two surveys, however (MICS 2000 for Albania and Uzbekistan), no period or point estimate is reported, diminishing somewhat the value of estimates from these surveys. Table 1 indicates that in nine out of the ten countries with a period-specific survey estimate, infant mortality rates from the survey data are considerably higher than the official rates for the corresponding years. This is also true where two surveys have been carried out in one country. For Romania, the survey estimate is a third greater. In Azerbaijan, the two survey based estimates of 79 and 74 infant deaths per 1,000 live births are over four times as big as the corresponding official rates. Ukraine is the only country where the difference between the official rate and the survey estimate is small. Yet in this case, the survey estimate is likely to greatly understate the true picture, for reasons discussed at different points throughout this paper.3 2 Rates calculated using formulas such as these are simplified, and do not, for example, take account of migration. However, the impact of such factors is generally believed to be small. For a more detailed discussion of the calculation of infant mortality rates, see Shyrock et al (1976). 3 It is worth noting that surveys carried out in some countries in the region around the late 1990s, including Bosnia-Herzegovina (MICS 2000), Moldova (RHS 1997), Russia (RHS 1996), and 4 Table 1: Official infant mortality rates and survey based estimates (deaths per 1,000 live births) Official rate (2001) Survey estimate Survey name and year Years of survey estimate Average official rate for corresponding years Absolute difference (survey estimate less official rate) Tajikistan 89 MICS 2000 (1993) 47 42 Tajikistan - 79 LSMS 2000 (1996-2000) - Kyrgyzstan 44 61 DHS 1997 (1993-1997) 29 32 Turkmenistan 20 74 DHS 2000