and Vallin address specific issues and provide corrections for data deficiencies. The data from Central Asia, especially for males, are assessed to be of poorer quality than those from the northern states. Andersen and Silver base their assessment on the anomalies in the age patterns of mortality in the data from Central Asia. These anomalies include higher mortality rates in urban than rural areas and an apparent ''crossover" in mortality at older ages, favoring rural areas, which they ascribe to age exaggeration among the rural population. They also base their assessment of data quality in the Central Asian states on comparisons with age patterns of mortality in Russia and Latvia, where the quality of mortality data appears better, and with patterns of age misreporting in data among the same ethnic groups in Xinjiang, China. Their general conclusion is that despite possible recent improvements in the quality of mortality data from Central Asia, the health situation there is likely to be worse than appears from official statistics. Past errors in vital statistics were not entirely due to the willful misreporting and coverups common during communist regimes, as has sometimes been implied; and many important causes of misreporting persist and limit analysts' ability to make sense of mortality trends and differences.
Of particular concern are two problems that relate to the reliability and interpretation of the infant mortality rate. The first relates to the Soviet definitions of live birth and infant death, which differ from the World Health Organization (WHO) standard definitions, with the result that births and deaths in the first months of life are underestimated. Since mortality at these ages represents a substantial proportion of infant deaths, particularly at lower mortality levels, the impact of this definitional issue can be substantial. Both Murray and Bobadilla (for 1989) and Kingkade and Arriaga (for 1990) calculate and present adjusted infant mortality rates for the NIS, based on different correction procedures (Table 1-5). From these analyses, it is clear that the range of correction factors is very wide. The actual factor applied to any one state depends largely on the assumptions underlying the methodology used by the authors. However, the impact of the adjustments on the level of life expectancy at birth is fairly minimal, resulting generally in less than a year of difference in estimated levels.
The second problem related to infant mortality is changes in registration coverage over time, which result in apparently increasing infant mortality. This problem is suspected to be most severe in the Central Asian states. Yet although there is some evidence that part of the trends in child mortality in Central Asia could be explained by changes in the completeness of registration, there are no good estimates of the magnitude of underregistration in the 1990s.
Incomplete registration of adult deaths and errors in the declaration of age at death are suspected to affect estimates of mortality at older ages, but in general have less impact on overall life expectancy than errors in infant mortality. Anderson and Silver identify problems with misstatement of age at the time of death in Central Asia, leading to implausibly low levels of mortality among the elderly in Tajikistan. Murray and Bobadilla attempt to measure the extent of underregistr-