NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

National Research Council (US) Committee on Population; Bobadilla JL, Costello CA, Mitchell F, editors. Premature Death in the New Independent States. Washington (DC): National Academies Press (US); 1997.

Cover of Premature Death in the New Independent States

Premature Death in the New Independent States.

Show details

4Issues of Data Quality in Assessing Mortality Trends and Levels in the New Independent States

Barbara A. Anderson and Brian D. Silver


This chapter addresses issues of data quality that affect the interpretation of reported mortality levels and trends in the New Independent States (INS). It presents an overview of data quality issues for readers who are not necessarily specialists in demography or familiar with the quality and types of data that are available from this part of the world. We examine data from selected regions and dates, while drawing the reader's attention to broader issues and the existing literature on the quality of data from the former Soviet Union. Our focus is on the traditionally Moslem NIS countries, including the Central Asian states of Kyrgyz, Tajikistan, Turkmenistan, and Uzbekistan, plus Kazakstan and Azerbaijan, which are linked both historically and culturally to Central Asia; these are cases in which real levels and trends in mortality, both past and present, are obscured by data error. Russia and Latvia are cases in which the reported adult mortality patterns and evidence of increasing mortality can be believed, and they are therefore used as a frame of reference for the reliability of the Central Asian data; these cases are fairly typical of the European part of the NIS. To aid in the analysis, we also draw on some detailed data from Xinjiang (in China), where one finds major ethnic groups that are culturally similar to Turkic groups in the Central Asian states. The purpose of the analysis is to identify ways of improving data collection in the NIS, especially Central Asia, so that policies and interventions related to health and mortality can be more effectively developed and targeted.

It may be noted that although mortality rates are normally the highest among infants and the elderly, these are the ages for which error due to age misstatement and underreporting of deaths is most likely to occur. In this chapter, we first discuss problems with Soviet data on infant mortality and the elderly as a general caution to researchers who are not familiar with data from the region. In the data analysis, however, we focus on an age range for which we can have more confidence in the data. For much of the analysis, we examine data for ages 10-79; for some of the analysis, though, we focus on the age range 20-59.

Our approach to studying demographic trends in the former Soviet Union and the NIS is to start with official statistics, but to view them with a critical eye. Scholars have devoted less attention to the evaluation and adjustment of demographic statistics in this region than to the evaluation and adjustment of economic statistics.1 We do not subscribe to the view that all of the data from the Soviet Union were fabricated or intentionally altered to make the state or political leaders look good or to mask negative trends in popular welfare. A frequent concomitant of such a point of view is a readiness to accept official data from the region only when they reveal negative trends or facts.

Nor do we subscribe to the view that the data are ''in the ballpark" and reliable enough for designing appropriate health and welfare interventions. While we agree that the available data provide a fairly clear picture of the main problems in public health and welfare for some regions and purposes, issues of data quality are too substantial to ignore. Acceptance of reported mortality data at face value would lead to errors in evaluating the impact of intervention strategies, because changes in data quality can obscure changes in real demographic behavior or outcomes. Moreover, some of the mortality rates, including cause-specific rates, have been extremely volatile in response to short-term factors and may now be at or near their peaks. Consequently, there is considerable risk of confusing the effects of policy interventions with "regression effects."2

We assess the plausibility of the reported figures by looking for internal consistency and by comparing them with levels and patterns in reported statistics from other countries. On occasion, formal tests for the consistency of age and mortality data have been applied to data from the Soviet Union and the NIS. Because of the lack of needed data, however, the formal application of consistency checks is not yet feasible for most regions of the former Soviet Union and for most types of mortality data. Furthermore, some methods for estimating error require untenable assumptions about the data. For example, methods of estimating the underregistration of deaths using vital registration and intercensal survival rates work reasonably well only if there is no appreciable age exaggeration in the census or death registration, a precondition that does not exist in data from Central Asia. Hence, a naive application of so-called formal checks for completeness of registration would give a false impression (most likely an underestimate) of the extent of underreporting of mortality in this region.

We have devoted a great deal of effort to examining the demographic information system in this part of the world and what biases it might impart. Often we have had to use indirect methods or to compare patterns from the region with those in other countries because the lack of detailed data or access prohibits direct checks on data accuracy. However, some data problems are easy to detect. For example, there were more persons reported alive at ages 11-15 in the 1970 census than were reported at ages 0-4 in the 1959 census. Although immigration of young children between 1959 and 1970 could have led to this result, in principle the only plausible explanation is that there was an undercount of young children in the 1959 census. A similar pattern occurs in later censuses.3 It is also not possible that the proportion of children who were physically or mentally handicapped was more than 10 times greater in the relatively developed Baltic republics than in the relatively undeveloped Central Asian republics (Anderson et al., 1987). Similarly, there is an obvious deficiency in the reported data showing that the month in which the lowest number of infant deaths occurred in the Soviet Union was December, while the month in which the highest number of infant deaths occurred was January (Anderson and Silver, 1988), and this pattern persists into the post-Soviet period for many regions.

These and other patterns of error in reported demographic data require careful analysis before one can best assess what was actually true, as opposed to what was reported to be true. The existence of error in the data does not mean that the data were deliberately "faked" and ought to be dismissed out of hand. In many cases, the error probably occurred for other reasons. Moreover, the data did not suddenly get better just because the Soviet Union broke up in 1991 and was replaced by multiple new governments, each with varying capabilities and commitments to the reform and improvement of demographic statistics. Nor did a large treasure trove of previously unpublished but validated data suddenly become available (Anderson et al., 1994).

Users of the data need to be aware of how the data were and are generated and to what extent the data in the hands of the government (whether published or not) reflect the true situation among the population. For example, because of differential access to and utilization of services, a great deal of data based on program usage may be an inaccurate reflection of the actual level of program need, both overall and by category of the population (region, urban-rural residence, sex, and other characteristics). A clear instance of this is the published information about disability (Anderson et al., 1987). The same issue must be considered in the analysis of a wide variety of data on morbidity, as well as some data on mortality. For example, the relatively high incidence of and mortality from cervical cancer in Estonia as compared with Finland appears to be due mainly to more effective mass screening in the latter (Aareleid et al., 1993).

The next section identifies various problems with Soviet and post-Soviet mortality data and describes our approach to analyzing the data. The following section presents mortality data for Russia and Latvia, areas where those data quality problems are less severe; thus these data can be viewed as relatively reliable, providing a frame of reference for the reliability of the data for the

Central Asian states. Next is a section examining how the identified data quality problems apply to the Central Asian data, thereby limiting their utility in policy and intervention terms. The final section presents conclusions and recommendations for improving the collection of mortality data in the NIS.

Some Problems with Soviet and Post-Soviet Mortality Data

Detailed data on mortality among the Soviet population were published sparsely before 1975 and almost completely suppressed between 1975 and 1986. The four relatively bountiful years in the publication of population and health statistics during the glasnost period have been followed since the demise of the Soviet Union in 1991 by a decrease in the amount of published data. Recently, however, life tables for 1992 for some of the new states have appeared, and life tables by ethnic group4 for 1990 and for the rural and urban populations of republics in 1990 have been published. Data are now plentiful enough to allow detailed examination of reported mortality conditions by age, sex, country, and rural-urban residence so that earlier conclusions about the plausibility or implausibility of the reported data can be examined more concretely. The following subsections describe various specific problems with Soviet and post-Soviet mortality data; the final subsection explains our approach to data analysis.

Lack of Microdata

One persistent problem with demographic data in the former Soviet Union is that, with few exceptions, only aggregate data have been published or are available in archives. This allows the detection of some data problems, but microdata would be much more useful in detailed analyses of the sources of the problems and in the construction of recommendations for data improvements. The lack of microdata stems partly from a view of such data as the property of government statistical agencies and partly from the lack of any tradition of public availability of data for independent analysis (Anderson et al., 1994). International agencies, such as the United Nations Economic Commission for Europe, have met with only partial success in convincing countries of the former Soviet Union to release census microdata. Many event registries in the NIS, in particular those for cancer, are not up to world standards (Rahu, 1992).

Data Comparability and the Demise of the Soviet Union

The dissolution of the Soviet Union created some problems for the analysis of demographic change in general. We have addressed these problems at some length elsewhere (Anderson et al., 1994). First, some of the NIS countries have begun to use new definitions and data collection procedures for population and health statistics. For example, in 1991 the three Baltic states shifted from the Soviet definitions of live birth and infant death to a standard that is close to the one recommended by the World Health Organization (WHO). This shift increases the reported infant mortality rates for the Baltic states by about 23 percent over what they would have been using the Soviet definitions.5 Russia began to shift to the WHO definitions in 19936 (see Kingkade and Arriaga in this volume).

A second potential problem is that one role of the State Committee on Statistics (Goskomstat) of the Soviet Union was to audit and attempt to improve the quality and consistency of procedures for vital registration and population enumeration throughout the country. Now that the Soviet Union is gone, the quality of population and health data in many of the successor states could deteriorate unless these states are able to develop a strong program of internal auditing and management of the collection of data, or perhaps obtain advice and expertise from abroad.

A third problem is that as the successor states undergo multiple crises, including civil violence and economic hardship, they are not likely to give high priority to the collection and evaluation of population statistics. In general, the most common kinds of error in mortality data tend to lead to underregistration of deaths, to exaggeration of age at death, or to exaggeration of the ages of the enumerated population—errors that in turn are likely to lead to apparent reductions in mortality. Although the rising mortality in the successor states might suggest that underreporting and underregistration are not very important, in fact there is evidence of substantial error in the Central Asian states, Kazakstan, and Azerbaijan. This means that infant mortality in the past was far higher than was implied by the reported data, in some cases by a factor of three or four. 7 Hence, it is difficult to know what baseline to use for interpreting trends in infant mortality in these regions. Use of the reported infant mortality rate would be very misleading; adjusted or corrected infant mortality rates cannot yet be applied consistently for all the countries because of a lack of detailed data.

Construction of Life Tables

As the new states have to deal with the collection, reworking, and analysis of population data, not only are there problems related to maintaining and improving the data collection system, but there are also questions about the consistency over time of the methods used to create summary statistics, including life tables.

The accuracy of life tables depends on the accuracy and completeness of two kinds of information: the enumeration of the population by age and sex, and the number of deaths by age and sex. It also depends on how some technical issues in life-table construction are handled. There have been only a few publications concerning the accuracy of Soviet life tables. Information about the construction of the 1958-1959 life table was published in the 1959 Soviet census summary volume (USSR TsSU, 1962-1963:254-279). Andreev et al. (1975) describe the methods used to construct the 1968-1971 life table and provide some comparisons with the methods used to construct the 1958-1959 life table. Kingkade (1985, 1987, 1989) presents a useful discussion of many aspects of Soviet life tables.

There have been some publications about age distributions and under-enumeration by age in Soviet censuses (Anderson and Silver, 1985a; Blum and Chesnais, 1986; Kingkade, 1985). We know that when constructing life tables, the Soviet authorities did not always use the reported number of people by age, for either the younger or older ages (USSR, TsSU, 1962-1963). The way life tables are closed at the older ages is a technical issue, but it can make a substantial difference in estimates of expectation of life at birth (Anderson and Silver, 1989a; Arriaga, 1984; Vaupel, 1986). For the Soviet Union as a whole, there were also changes over time in life-table calculation in response to problems with the data. In constructing life tables, Goskomstat used a Gompertz-Makeham function to estimate mortality rates above certain ages, in lieu of using the reported age-specific mortality data. A Gompertz-Makeham formula is commonly applied to smooth mortality rates at very old ages. If mortality is understated because of age exaggeration in either the census or death registration, this procedure increases estimated mortality above the age at which it is first applied. Kingkade (1987) has calculated that Goskomstat applied a Gompertz-Makeham function to reported data at ages 90 and above in the 1958-1959 life table, at ages 70 and above in the 1968-1971 life table, and at ages 63 and above in the 1984-1985 life table.

That a Gompertz-Makeham function was applied at a younger age in each succeeding life table suggests that Soviet statisticians became increasingly aware of problems in reported mortality data for the older ages. (See also Kingkade and Arriaga in this volume.) One consequence of applying the Gompertz-Makeham function at progressively lower ages in successive life tables, however, was to lower the estimated expectation of remaining life (ex) at all ages (Anderson and Silver, 1989a).

Hence, as researchers and policymakers study trends and levels of mortality in the post-Soviet period, they need to be aware that overall measures of mortality, such as expectation of life at birth and expectation of remaining life at all ages, may be substantially affected by the methods used in the construction of life tables. If new life tables do not apply adjustments as rigorous as those applied in previous life tables for regions in which the reported ex values were implausibly high, the country's population may appear to be experiencing mortality improvements when in fact it is experiencing primarily a change in the methods used for calculating life tables.

What methods are used to construct life tables in the NIS? Most of the NIS countries do not have specialists with sufficient training to construct life tables. Some that do have such specialists have adopted different methods from those used by the Soviet (later Russian) statistical agencies, so that there can be problems of comparability across time and regions (Katus, 1994b). Some researchers who have access to official raw data on births and deaths construct their own life tables rather than relying on official ones (Shkolnikov, 1994; Shkolnikov et al., 1994). International agencies that receive data from the NIS countries usually do not evaluate those data beyond checking for basic internal consistency. In short, there is little or no standardization in approach at the present time. If the standard or the approach changes, or if it differs across regions, then comparisons over time or by region will be affected.

In publications such as the United Nations Demographic Yearbook, data from a given country are designated as accurate or as estimates based on the statement of the country that contributed the data, rather than any assessment conducted by United Nations staff. Users sometimes think that because the data are not designated as estimates or of questionable quality, they have been judged accurate as the result of some kind of data quality assessment.

A critical question for any consumer of official statistics from the NIS, especially for the less-developed regions, is how the statisticians have addressed or taken into account known problems in previous data.

Age Heaping and Age Exaggeration

Two basic problems with age data affect mortality estimates: age heaping and age exaggeration. Both of these problems are common for populations in less-developed countries, and there is evidence that they create problems with data from the former Soviet Union, especially Central Asia. Garson (1986, 1991) and Bennett and Garson (1983) have shown the implausibility of both the high number of reported centenarians in Soviet censuses and the low reported mortality rates among the elderly.

A common form of age heaping occurs when too many people claim to have an age that ends in a zero, a 5, or an even number, or too many claim to have been born in a year that ends in a zero, a 5, or an even number.8 Although age heaping causes some problems in itself, it can be taken as an indicator of other problems with age data (Ewbank, 1981). Extensive age heaping has been documented in many parts of the world, including Latin America (Nuñez, 1984; Kamps, 1976). It has also been documented for the Central Asian republics by Soviet demographers (Sachuk and Minaeva, 1976) and for Russia in the 1959 census, as well as in death registration for 1958 (Urlanis, 1976). We have found evidence of severe age heaping in the 1990 Census of China for Uighurs and Kazaks, traditionally Moslem peoples who speak a Turkic language and are closely related to Moslem nationalities in former Soviet Central Asia (Anderson and Silver, 1994c). When responding to the 1990 Census of China, 14 percent of male Uighurs in Xinjiang Uighur Autonomous Province claimed to have been born in a year that ended in a zero.

Another problem is age exaggeration, whereby people claim to be older than they actually are, or the age at death of persons who have died is reported as older than was actually the case. There is clear evidence of age exaggeration in Xinjiang (Coale and Li, 1991). Coale and Li note that in 1982, although the population of Xinjiang comprised only 1.3 percent of the population of China, 47 percent of all males in China reported to be aged 95-99 were from Xinjiang. Our more recent research shows that the problems with the age data from Xinjiang are due to the data from Uighurs and Kazaks in that province (Anderson and Silver, 1994c).9 Note that such patterns of age exaggeration may make it inappropriate to use standard techniques for estimating census undercounting using intercensal survival techniques.

Mortality Crossovers

Problems with mortality and age data are sometimes indicated by the presence of mortality crossovers. In this situation, population A has lower age-specific mortality rates than population B below a certain age, but population B has lower age-specific mortality rates above that age. Such a crossover has been observed for black and white males in the United States and is often observed in developing countries, with the urban population having lower mortality rates below a certain age and the rural population having lower mortality rates above that age.

One point of view argues that such crossovers often reflect real differences among groups, with selectivity removing the more frail members of a population at young ages. The survivors, then, are very vigorous and experience low mortality rates for the remainder of their lives (Manton and Stallard, 1984; Manton et al., 1979; Nam et al., 1978; Vaupel et al., 1979).

Another point of view argues that the crossover from higher to lower age-specific death rates is a result of underestimation of death rates at the older ages in the population that has crossed over into lower reported mortality (Myers, 1978; Rosenwaike and Logue, 1983; Rosenwaike and Preston, 1984; Coale and Kisker, 1986; Dechter and Preston, 1991). An increasing body of research has documented situations in which a mortality crossover or surprisingly low reported mortality rates at older ages could not possibly represent the actual risks of dying (Condran et al., 1991; Dechter and Preston, 1991).

It has been suggested that urban-rural mortality crossovers indicate deficiencies in mortality data from the Soviet Union (Anderson and Silver, 1989a; Dmitrieva and Andreev, 1987). Increases over time in the age at which rural-urban mortality rates cross over has also been interpreted as indicating improvements in data quality over time (Anderson and Silver, 1994a). In the Soviet Union as a whole, there was a rural-urban crossover for males at ages 20-24 in 1938, at ages 45-49 in 1959, and at ages 55-59 in 1986. Even in 1989, there was a rural-urban crossover at ages 35-39 for males and at ages 70-74 for females in Kyrgyz, at ages 25-29 for males and ages 65-69 for females in Tajikistan, at ages 15-19 for males and ages 75-79 for females in Turkmenistan, and at ages 30-34 for males and ages 50-54 for females in Azerbaijan. The sex differences in these cases suggest a process by which males are given preference in access to medical care, a phenomenon found in some other Moslem societies (Anderson and Silver, 1994a).10

Even if a mortality crossover is the result of error in the data, this error can stem from various sources, including (1) omission of deaths of older people, (2) overstatement of the ages of the population alive at a given time, and (3) overstatement of the age at death of older people. Further research is needed before we can attribute the error to these or other sources. Later, however, we shall provide additional evidence on the issue.

Problems with Infant Mortality Data

Although this chapter is concerned mainly with adult mortality, it is relevant to discuss briefly some problems with Soviet infant mortality data. When births and infant deaths are incompletely recorded, it is likely that both the birth and the death will not be recorded if an infant dies shortly after birth. The result is a higher proportion of infant deaths than of births being omitted from official statistics. However, if births and infant deaths are counted more completely over time, the reported infant mortality rate will increase even if the actual infant mortality rate has not changed.

The strange rise and fall of infant mortality rates in the Soviet Union during the 1970s shows strong evidence of the effects of both increasingly complete reporting of births and infant deaths and some deliberate falsification of data in the locales to mask the true infant mortality rates (Anderson and Silver, 1986b, 1994b; Ksenofontova, 1994). Also, the error in the reported rates occurred predominantly in rural areas and in the more rural republics of the former Soviet Union—Central Asia, Kazakstan, and Moldova.11

The reported rural infant mortality rates were lower than urban rates in the early 1950s and became consistently higher than the urban rates only after 1967. In fact, the sharp rise in reported infant mortality in the Soviet Union as a whole between 1971 and 1976 was accompanied by a sharp increase in the ratio of rural-to-urban infant mortality rates. It is likely that the main factors involved in the lower reported rural than urban infant mortality are underreporting of rural births and infant deaths and misattribution of infant deaths as deaths that occurred in the second year of life, in particular the thirteenth month (Anderson and Silver, 1994b; Blum and Pressat, 1987; Ksenofontova, 1990). However, it is also possible that rural infant deaths were being misattributed to the urban population.12 Even in the 1980s, both Goskomstat and the Ministry of Health of the Soviet Union were dissatisfied with the quality of registration of infant deaths and took steps to improve it (USSR Ministerstvo, 1984).

Expectation of Life at Birth

Measures of expectation of remaining life at any age, including at birth, are summary measures of mortality above that age. Contradictory trends at different ages can cancel each other out. Moreover, recent experience in the former Soviet Union shows that these "averages" can change rather quickly in either direction. Finally, such measures are especially susceptible to changes in mortality rates at the older ages (Anderson and Silver, 1989a; Vaupel, 1986). For all of these reasons, it is a good idea when studying mortality to disaggregate the mortality experience by age and to be wary of summary measures that may be especially susceptible to error in the data, despite the temptation to rely on the expectation of life at birth as a handy overall indicator.

Approach to Data Analysis

As noted in the introduction, given the substantial problems with infant mortality data and with mortality data for advanced ages (see also Anderson and Silver, 1986b, 1989a, 1994b), this chapter concentrates on ages at which the data are generally relatively reliable. In parts of the analysis we examine data for ages 10-79; in other parts, we concentrate on ages 20-59. While neither of these age ranges is consistent with the formal definition of "working ages" in the Soviet Union (ages 16-59 for men and 16-54 for women), they are useful for purposes of the present analysis.

The first post-World War II life tables for the Soviet Union were produced for 1958-1959. For both males and females, published values of expectation of life at birth increased from 1959 through 1964 (for an overview of trends, see Anderson and Silver, 1990b; see also the chapters in this volume by Shkolnikov et al., Vassin and Costello, and Murray and Bobadilla in this volume). Expectation of life at birth fell from 1964 through 1979 and then increased through 1990. Recent information has shown that expectation of life at birth has fallen since 1990 in many of the NIS countries. Turning points around 1964, 1980, and 1991 appear for many different regions of the former Soviet Union. All of these inflection points are much sharper for males than for females. Their source is still not clear, especially concerning the 1964 and 1980 reversals. Neither of these turning points appears to be related to any obvious changes in health care expenditures, environmental or public health crises, or other policy changes. However, the link between the reported sharp decline in mortality in the mid-1980s and the anti-alcohol campaign is well documented (Shkolnikov and Vassin, 1994; Shkolnikov et al., 1994; see also the chapters by Treml and by Shkolnikov and Nemtsov in this volume).13

We concentrate on data for 1978-1979 and 1990. For these dates we have life tables by age and sex for rural and urban populations for every republic of the Soviet Union. In 1978-1979, reported expectation of life at birth was about at its low point since 1959, and in 1990 expectation of life at birth had substantially recovered from its earlier decline. We also look at data for Russia for 1992.

In the next section we discuss recent mortality trends in Russia and in Latvia. The situation in Russia has been an object of great concern. Latvia is also interesting because of the high level of economic development and the high quality of data. Most of the reported mortality levels and trends in Russia and Latvia probably reflect the actual mortality situation. We have not had data to use in making comparisons of regions below the level of the whole republic.14 However, other scholars have done this for provinces within Russia (Shkolnikov and Vassin, 1994; Velkoff, 1992; Velkoff and Miller, 1995). Recent data for the Baltic states, Russia, Ukraine, and Belarus are generally trustworthy, especially at the working ages. Data for other regions of the NIS, especially for Central Asia, Kazakstan, and Azerbaijan, are more problematic.15

Our discussion of Russia and Latvia is followed by an examination of the mortality situation in the four Central Asian states (Kyrgyz, Tajikistan, Turkmenistan, and Uzbekistan), plus Azerbaijan and Kazakstan. The health problems and high mortality in these areas deserve special attention, but we show that there are also serious problems with the mortality data from these areas that make the assessment of real trends in mortality highly problematic. Although recent mortality data are more accurate than those from earlier periods, we think that in many areas, even recent data portray a mortality situation substantially better than that which has actually occurred. We show the implausibility of the data through internal comparisons; comparisons with patterns in Russia and Latvia; and comparisons with the situation elsewhere in the world, especially in Sweden and among Uighurs, a traditionally Sunni Moslem, Turkic ethnic group in Xinjiang in northwest China.

Mortality Trends in Russia and Latvia

Mortality patterns in Russia have, of course, been the subject of great interest. Yet the study of Russian mortality has been hindered until recently by the lack of detailed published data. Although life tables were published for many other republics of the Soviet Union, life tables for Russia for the post-World War II period were not published until 1988 (for the years 1970-1971 and later). Hence, as the divergence between mortality trends in the Soviet Union as a whole and those in other developed countries became especially evident in the early 1970s (Vallin and Chesnais, 1974), it remained virtually impossible for scholars to identify the regional (republic) components of the Soviet trends, including Russia's contribution.

However, after examining age-specific death rates and expectation of life at birth for the Soviet Union as a whole and for individual republics, Dutton (1979) speculated correctly that poor survival of men in the Soviet-era Russian Federation was responsible for a large portion of the high mortality of men and for increases in their age-specific mortality in the Soviet Union as a whole. Some years later, using inferential methods, we estimated life tables for Russia for 1958-1959 and 1969-1970 (Anderson and Silver, 1986a). The latter turned out to be very similar to the table ultimately reported for Russia for 1970-1971.

Detailed scholarly study of mortality in Russia as a whole and for its separate regions is still at an early stage, but recent work has been of high quality (see the chapters by Vassin and Costello and by Murray and Bobadilla, in this volume; Shkolnikov and Vassin, 1994; and Shkolnikov et al., 1994). Moreover, the quality of data for Russia has improved such that observed trends for the last two decades or so can be taken as fairly reliable, especially for the working ages.16 For this reason, the substantial increases in mortality among Russian men after the mid-1960s, as well as increases in mortality in Russia in 1992 and 1993, can be interpreted as real—not as an artifact of changes in data quality.

The Baltic states, especially Latvia and Estonia, have long had high-quality demographic data. We examine Latvia as an example of the situation in the Baltic states. The course of mortality in Latvia and the possible effects of the Soviet regime on the course of mortality have been the subject of recent scholarly examination (Krumin , 1993, 1994).

Among all populations, age-specific death rates are high in infancy and childhood, rise slowly from about age 10 through later adulthood, and then rise more rapidly after about age 50. Coale and Demeny (1983) summarize the typical pattern of change in mortality at different ages as the overall mortality conditions in a population change. Among all populations, the age-specific mortality rate at ages 20-24 is lower than at ages 70-74. However, the age-specific mortality rate in a given population can be relatively high or low for that age range in comparison with other populations.

To judge whether an age-specific mortality rate at one age is relatively high or low as compared with the age-specific mortality rate at another age, we need a standard for comparison. We use Coale-Demeny West model life tables for this purpose. With every age-specific mortality rate from a population of interest, we associate the expectation of life at birth from the Coale-Demeny West model life table that has the same age-specific mortality rate. If the pattern of mortality by age were the same in the population of interest as in the Coale-Demeny West tables, the expectation of life at birth associated with every age would be the same; a plot of the fitted or implied expectation of life at birth across ages would be a horizontal line. However, if mortality were low at one age in comparison with another, the age with relatively low mortality would be associated with a relatively high implied or fitted expectation of life at birth; in this case a plot of the implied expectation of life at birth would not be horizontal. 17

For the purposes of this discussion, the choice of which Coale-Demeny family to use as a standard or whether to use another standard, such as the U.N. General Pattern (United Nations, 1982a), makes little difference to our substantive conclusions. We do not use conformity with the standard as an absolute test of data quality. We use it instead to provide a more readily interpretable metric for comparing mortality at different age levels and for different populations.18

Figure 4-1a shows the expectation of life at birth, e(0), associated with age-specific mortality rates for females in Russia in 1978-1979, 1990, and 1992, while Figure 4-1b shows the values for females in Latvia in 1978-1979 and 1990. In Russia, mortality rates for females declined between the late 1970s and 1990; this mortality decline was lost between 1990 and 1992. For females in Latvia, mortality was quite low at all ages even in 1978-1979; between the late 1970s and 1990, mortality declined at some ages and increased at others.

Figure 4-1a. Implied e(0) for Russia in 1978-1979, 1990, and 1992, females.

Figure 4-1a

Implied e(0) for Russia in 1978-1979, 1990, and 1992, females.

Figure 4-1b. Implied e(0) for Latvia in 1978-1979, 1990, and 1992, females.

Figure 4-1b

Implied e(0) for Latvia in 1978-1979, 1990, and 1992, females.

Figure 4-2a shows the expectation of life at birth for males in Russia associated with age-specific mortality rates, while Figure 4-2b shows the values for males in Latvia. The implied expectation of life is sharply lower for Russian men at the older working ages as compared with what would be expected if their mortality were consistent with the Coale-Demeny ''level" of that found among younger men. A similar, although less extreme, pattern by age is seen for men in Latvia.

Figure 4-2a. Implied e(0) for Russia in 1978-1979, 1990, and 1992, males.

Figure 4-2a

Implied e(0) for Russia in 1978-1979, 1990, and 1992, males.

Figure 4-2b. Implied e(0) for Latvia in 1978-1979, 1990, and 1992, males.

Figure 4-2b

Implied e(0) for Latvia in 1978-1979, 1990, and 1992, males.

This pattern for males in Russia and Latvia is probably due partly to deaths related to smoking and alcohol consumption. After a period of rising mortality among men from the mid-1960s through 1980, mortality fell until 1990. Shkolnikov and Vassin's (1994) examination of mortality change in Russia by month makes it clear that the fall and rise in male mortality in the mid-1980s was substantially a result of the effects of the anti-alcohol campaign. However, not only were the gains among Russian males from the late 1970s through 1990 lost between 1990 and 1992, but real mortality among older working-age Russian men in 1992 was higher than in the late 1970s.

Figures 4-3a and b show the implied expectation of life at birth from age-specific mortality rates for residents of Russia and Latvia on the one hand, and ethnic Russians and ethnic Latvians in the Soviet Union as a whole on the other hand. The values for Russians and for Russia are virtually identical. However, the mortality levels in Latvia are somewhat lower than those in Russia. This is because almost half the population of Latvia comprises people—primarily Russians—who are not ethnic Latvians. Ethnic Russians in Latvia have higher mortality rates than ethnic Latvians (Krumin, 1994). As a result, mortality rates for all residents of Latvia are higher than those for ethnic Latvians.

Figure 4-3a. Implied e(0) for Russians and Latvians in 1988-1989, and RSFSR and Latvian SSR in 1990, males.

Figure 4-3a

Implied e(0) for Russians and Latvians in 1988-1989, and RSFSR and Latvian SSR in 1990, males.

Figure 4-3b. Implied e(0) for Russians and Latvians in 1988-1989, and RSFSR and Latvian SSR in 1990, females.

Figure 4-3b

Implied e(0) for Russians and Latvians in 1988-1989, and RSFSR and Latvian SSR in 1990, females.

As discussed earlier, age-specific mortality rates are low at the adult ages among all populations. It is important to bear in mind that because mortality rates at some ages are typically very low, those rates even if doubled would cause only a few days' reduction in the average length of life for the population.

Table 4-1 shows the percentage of people in Russia and Latvia alive at age 20 who would be expected to die before reaching age 60 given the age-specific mortality rates in effect. For both males and females in Russia and in Latvia, there was a decline in that percentage between 1978-1979 and 1990. Between 1990 and 1992, the percentage of 20 year olds who would die before age 60 increased for both males and females in Russia. In 1992, the age-specific mortality rates imply that 36 percent of men reaching age 20 would die before reaching age 60. By world standards, the survival rate of men in Russia from ages 20 to 60 is extremely low (Anderson and Silver, 1994a). The level of mortality between ages 20 and 60 for Russian men in 1992 is consistent with an expectation of life at birth of 52 years. Moreover, a recent report by the Russian Federation Ministry of Health (Russia Minzdrav, 1994) indicates that mortality rates in Russia rose considerably between 1992 and 1993.

TABLE 4-1. Percentage of 20-Year-Olds Expected to Die by Age 60, for Russia in 1978-1979, 1990, and 1992, and Latvia in 1978-1979 and 1990.


Percentage of 20-Year-Olds Expected to Die by Age 60, for Russia in 1978-1979, 1990, and 1992, and Latvia in 1978-1979 and 1990.

This very high level of mortality among Russian men at working ages has substantial policy implications. In a period of social disruption, high levels of male mortality mean that high levels of widowhood exacerbate the effects of high divorce rates in breaking up families. The increase in female-headed households resulting from high adult mortality contributes to high levels of poverty. Households headed by women have long been a major segment of the poor in Russia.

Age-Specific Mortality Rates in the Traditionally Moslem NIS Countries

Many problems with mortality data from less-developed countries are found in the data for the traditionally Moslem NIS countries. As discussed earlier, all of these problems result in reported mortality rates lower than the actual rates. As the quality of the data improves, the mortality rates increase, even if the actual mortality situation has not changed.

Figure 4-4 shows the implied levels of expectation of life at birth for males in Sweden in 1989, for Uighur males in Xinjiang in 1990, and for males in Latvia in 1990. For Sweden, there is a comparatively horizontal line. The results for

Figure 4-4. Implied e(0) for Sweden (1989), Uighurs in Xinjiang (1990), and Latvia (1990), males.

Figure 4-4

Implied e(0) for Sweden (1989), Uighurs in Xinjiang (1990), and Latvia (1990), males. NOTE: Fitted levels are based on Coale-Demeny (1983) male West model.

Latvia show a decline with age in the implied level of expectation of life at birth, both because of high mortality at the older working ages and because more recent cohorts were born into a generally more favorable mortality situation than earlier cohorts. The data for Uighurs show a higher implied expectation of life at birth at older than at younger ages; indeed, among Uighur men aged 75-79. the implied expectation of life is as high as that for Swedish men.

A higher implied expectation of life at older ages does not by itself indicate that the data are poor, since a variety of mortality conditions and causes of death could produce such a shape. (See also Shkolnikov et al. and Kingkade and Arriaga, in this volume.) However, the very low mortality rates that are implied at older ages for Uighurs as compared with Swedes are clearly implausible given the known public health conditions in Xinjiang as compared with Sweden. Thus those low rates suggest poor data quality (see Coale and Li, 1991; Anderson and Silver, 1994c).

Further indications of data quality problems result from examining the implied expectation of life at birth based on age-specific mortality rates for urban and rural populations of Uzbekistan, Azerbaijan, Russia, and Latvia for 1978-1979 and 1990. Except for Latvia, the implied expectation of life at birth is higher at older ages among rural than among urban populations. This is more evident for 1978-1979 than for 1990. We interpret this rural-urban crossover as another indication of problems with data quality. Although this crossover occurs in Russia, the age at which it occurs is much later in Russia than in Azerbaijan and Uzbekistan, and the distance between rural and urban populations is much smaller for Russia than for the other two.

In Latvia, the rural population comprises predominantly ethnic Latvians (72 percent in 1989), while the urban population contains a high proportion of ethnic Russians (41 percent). In the traditionally Moslem NIS countries, Russians and members of other European groups are concentrated in urban areas. The rapid increase in the implied expectation of life at birth with increasing age for rural males in Azerbaijan and Uzbekistan is not plausible. If detailed mortality data were available by urban-rural residence and ethnic group within the former republics of the Soviet Union, the actual sources of these strange patterns would be clear. We think the figures for Uzbekistan and Azerbaijan would be similar, even if data only for the indigenous ethnic group were shown.

We have much less faith in rural than in urban data both because rural deaths (especially infant deaths) appear to be much less well enumerated than urban and because we find indirect evidence of many rural deaths being attributed to urban populations (Anderson and Silver, 1994b). Usually, mortality conditions are better in urban than in rural locales (United Nations, 1980:34; 1982b:88,106, 136, 164). Worse urban than rural mortality and crossovers in mortality rates between urban and rural areas provide evidence to support the conclusion that the actual mortality rates in rural areas have been much higher than the reported rates.19

We think rural mortality rates at older ages are underestimated for a combination of reasons: exaggeration of age in the census (or base population estimate), exaggeration of reported ages at death, and underregistration of deaths. Our research in Xinjiang in Chinese Central Asia, however, suggests that underregistration of deaths may not be the main culprit. Uighurs outnumber Han Chinese in Xinjiang. In addition, in the data from China we used, deaths were reported in the census rather than in the vital registration system. However, even when life tables for Uighurs in Xinjiang are constructed on the basis of census data alone (using the count of persons by age in the population and the reported deaths of persons by age in the 6 months preceding the census), patterns of implausible mortality rates at older ages similar to those in the former Soviet Central Asian republics appear in the Uighur population of Xinjiang (Anderson and Silver, 1994c).

Figures 4-5a through d show data for Latvia, Russia, and Uzbekistan for 1990. If these data are to be believed, males in rural Uzbekistan had much better mortality conditions than males in rural Russia and Latvia; the comparison is similar, but less extreme, for urban males. The mortality levels for females in all three former republics are similar. We do not think it possible that the actual mortality rates of males were lower in rural Uzbekistan than in rural Latvia in 1990.

Figure 4-5a. Implied e(0) for urban population of Latvia, Russia, and Uzbekistan in 1990, males.

Figure 4-5a

Implied e(0) for urban population of Latvia, Russia, and Uzbekistan in 1990, males.

Figures 4-6a and b show the implied expectation of life at birth for males and females in the six traditionally Moslem republics. Among those republics, Kazakstan and Azerbaijan have a relatively high level of socioeconomic development, and Tajikistan and Uzbekistan a relatively low level. One would suppose that the implied expectation of life at birth would be higher in the more-developed republics. The results for females have a certain amount of plausibility: the more-developed republics generally have higher implied expectation of life at birth. For males, however, the implied expectation of life at birth tends to be higher the less developed the republic.

Figure 4-6a. Implied e(0) for total population of six Moslem republics in 1990, males.

Figure 4-6a

Implied e(0) for total population of six Moslem republics in 1990, males.

Figure 4-6b. Implied e(0) for total population of six Moslem republics in 1990, females.

Figure 4-6b

Implied e(0) for total population of six Moslem republics in 1990, females.

Figure 4-5b. Implied e(0) for rural population of Latvia, Russia, and Uzbekistan in 1990, males.

Figure 4-5b

Implied e(0) for rural population of Latvia, Russia, and Uzbekistan in 1990, males.

Figure 4-5c. Implied e(0) for urban population of Latvia, Russia, and Uzbekistan in 1990.

Figure 4-5c

Implied e(0) for urban population of Latvia, Russia, and Uzbekistan in 1990. females.

Figure 4-5d. Implied e(0) for rural population of Latvia, Russia, and Uzbekistan in 1990, females.

Figure 4-5d

Implied e(0) for rural population of Latvia, Russia, and Uzbekistan in 1990, females.

Some of the differences in mortality among the urban populations of these republics reflect differences in the ethnic composition of the urban population, but except for Kazakstan, the rural population of each republic comprises predominantly indigenous populations. Men of European background may consume more alcohol than men from indigenous ethnic groups in Central Asia; thus the indigenous men may have lower mortality from causes directly related to alcohol consumption. However, the magnitude of implied life expectancy for older men from some of the Central Asian republics is so high and so inconsistent with age-specific mortality rates at younger ages as to be out of the range of relationships of mortality at different ages in any well-recorded populations. The data indicate that rural males in Tajikistan have an extremely high implied expectation of life at birth: over 70 years for all age groups from 45-49 through 75-79.

Table 4-2 shows the percentage of people alive at age 20 who would be expected to die before they reached age 60 given the age-specific mortality rates in the given year. The data are shown for the six traditionally Moslem republics; for total, urban, and rural populations; and for 1978-1979 and 1990.

TABLE 4-2. Percentage of 20-Year-Olds Expected to Die by Age 60, for Six Traditionally Moslem NIS countries, 1978-1979 and 1990.


Percentage of 20-Year-Olds Expected to Die by Age 60, for Six Traditionally Moslem NIS countries, 1978-1979 and 1990.

We have argued that the reported mortality rates for men in the Moslem republics are not plausible. At both dates shown in Table 4-2, the working-age mortality of men is reported as greater in urban than in rural areas in every Central Asian republic, Kazakstan, and Azerbaijan. However, the gap lessens over time. In every traditionally Moslem republic, the percentage of men in urban areas dying at working ages declines, while in all of these republics except Kyrgyz and Uzbekistan, the percentage dying in rural areas at working ages increases. This is exactly what should happen if urban data were much more accurate than rural and if the quality of rural data improved over time.

The data for women appear to be much more reasonable than those for men. In every case, except for Azerbaijan and Kazakstan in 1978-1979, the percentage expected to die between ages 20 and 60 is higher in rural than in urban areas, and the values for rural and urban Kazakstan are almost identical. In addition, the more-developed republics tend to have a lower estimated proportion dying between ages 20 and 60.

The health and mortality of the working-age population are a matter of great policy concern in the NIS. Policy planning that accepted as accurate the values shown in Table 4-2 would be in serious error. It is not plausible that men in Tajikistan, the least-developed republic, have the lowest working-age mortality anywhere in the NIS.

Conclusions and Recommendations

The serious health and mortality situation in the NIS deserves policy and scientific attention. In none of the states has mortality reduction kept up with the reductions seen in most of the developed world during the last 20 years.

It is beyond the scope of this discussion to provide a comprehensive overview of the data needs in each of the NIS countries. Instead, we have focused on interpreting reported levels and trends in mortality in the region. Consistent with this focus, we make some recommendations here for data collection and improvements in data quality that would strengthen the ability of policymakers to monitor and interpret mortality trends in the region.

As we have indicated, the overall high mortality rates in the NIS, both in the aggregate and by cause, are grounds for concern and action. But because of poor data quality, interventions to improve health conditions that would also improve data quality would be likely to produce equivocal results. For example, efforts to reduce infant mortality rates in Central Asia could also produce more complete reporting of infant deaths. Interventions that were actually lowering infant mortality could lead to apparent increases in infant mortality (or to a slower decrease in infant mortality than was actually occurring)—and perhaps to premature abandonment of policies and programs that were actually working. The same thing could happen in Russia, in which generally lower data quality in some of the predominantly non-Russian regions (e.g., Chechnya, Ingushetia, Daghestan, Balkaria) could disguise actual improvements in infant mortality if these improvements were accompanied by more complete registration of births and infant deaths.

Improvements in the accuracy of reporting of ages of the base population and of the deceased, or more complete registration of deaths, would also be likely to occur if there were a concerted effort to reduce adult mortality. The reported adult mortality rates would probably show a smaller reduction (and possibly even an increase), even while actual adult mortality rates were declining.

Thus, addressing the issue of data quality and building in standards for evaluating program success are essential if one hopes to obtain a realistic picture of program efficacy.

Addressing Data Quality in Russia and the European NIS Countries

Following improvements in mortality among Russian adults between 1980 and 1990, the situation deteriorated at least until 1993. Close attention to changes in mortality in Russia in the post-Soviet era is warranted. The deterioration has been more serious for men than for women. More data that would allow examination of recent mortality trends in Russia and elsewhere in the European NIS countries by age, ethnic group, and cause of death would be very worthwhile.

The work of Russian and French scholars in this area is interesting and important (see Meslé et al., 1992; Shkolnikov et al., 1994).

Though applying varying methods, the chapters in this volume provide a convergent picture of the cause-structure of adult mortality and of recent trends by age in the NIS. But the volatility and rapid changes in some of these rates suggest the need for care in designing intervention strategies. How much of the rapid increase in mortality in Russia between 1992 and 1993, for example, is actually attributable to a deterioration in health programs, medical services, and public sanitation, and how much to the general economic crisis, inflation, unemployment, deteriorating diet, and declining social support for the elderly or lone individuals? This is not intended as an argument against intervention. It is intended as an argument for caution in identifying the effects that could be expected to result from primarily medical interventions when broader social and economic institutional factors may account for a substantial portion of the change in health outcomes. Moreover, cost-benefit analyses of the likely payoff from alternative forms of intervention and alternative delivery systems are needed.

As a general methodological note, we would argue for greater attention to the sociology, geography, and politics of health problems and policy. How should one balance claims of ''efficiency" (or maximum return for the intervention dollar) against claims of "equity" or "fairness," which may entail ensuring attention to various interests and constituencies, including women, children, ethnic minorities, regions, and NIS countries? A narrow focus on the goal of maximizing the "increase in life expectancy" or minimizing the "reduction in length of working life" could lead to a policy devoting the greatest attention to adult Slavic men, whose injurious smoking and drinking habits have somehow justified this attention. What other goals are also worthy of attention, and what are the costs and benefits of pursuing these alternatives?

Addressing Data Quality in the Traditionally Moslem NIS Countries

Former Soviet Central Asia, Azerbaijan, and Kazakstan are regions in which high mortality rates ought to be of concern. High rates of infant mortality should obviously be special targets of policy initiatives. Although, relatively speaking, the mortality rates among adults do not appear to be as serious a problem as infant mortality rates, we advise caution before reaching such a conclusion.

The poor quality of mortality data for the region has masked probable high mortality rates at older ages. Future improvements in data quality are likely to make it difficult to assess the effects of initiatives to improve public health and medical care because, as noted above, improved quality of data is likely to raise the apparent mortality rates, at least for a while.

Levels and Trends

A major problem in interpreting data from Central Asia stems from the difficulties involved in discerning levels and trends. The level of mortality in Central Asia is high, even if some of the published statistics do not show this. However, it is virtually impossible to describe a trend in mortality in that region with any confidence since mortality levels were certainly grossly underestimated in the past.

If one needed to make a best guess for a life table to assign to a Central Asian population, picking one consistent with the reported age-specific mortality rates of women, such as women in their 30s, would probably be the best strategy. However, this would give only a rough approximation of mortality at other ages and would usually still result in the conclusion that mortality conditions for men were better than was actually the case. We know that any real factors that influence mortality, such as smoking, alcohol consumption, and hypertension, very likely have different effects on males and females, so the use of female mortality rates as a standard is risky.

Because of the serious problems with reported age-specific mortality rates, especially for men, it seems unlikely that cause-of-death or morbidity data for Central Asia can tell us very much about trends. If our explanation for the urban-rural crossover for men is accurate, it is also likely that the selectivity of men obtaining health care in urban areas is cause-specific, which will therefore influence cause-of-death data by rural-urban residence. Whether men go to urban places for health care will relate to the complaint and thus to the cause of death if they die.

Serious attention must be paid to the registration and data collection system in order to track trends. Given our findings regarding implausibly low reported mortality in Xinjiang (China), where death reports did not come from the registration system, this is not just a question of fixing the registration system. Error in the mortality data is also strongly affected by people's knowledge and reporting of their ages. A complex approach to improving the accuracy of reporting of ages is needed; we have discussed some possible steps with Chinese statistical authorities. It would not be easy to obtain substantial improvement, but a passive approach in which one simply waits until the entire population has completed secondary education is not very compelling. And an approach that essentially ignores the problem and its effects on mortality data should also be unacceptable.

In examining levels and trends in mortality in Central Asia, compositional effects must also be taken into account. The urban parts of Central Asia are heavily populated by Russians and members of other European groups. Because of interregional and international migration, mortality rates in urban areas are subject to change as a result of changing population composition, especially as many Europeans leave Central Asia, a process that has been going on for decades (Anderson and Silver, 1989c, 1990a). Changing population composition also affects reported fertility rates, since the indigenous population has long had much higher fertility than Russians and other Europeans in the region. Obtaining information on mortality and fertility rates by ethnic group would be very helpful in controlling for the effects of changing population composition on mortality and fertility rates.

Need for Microdata

One way to address mortality data problems in the region is through the collection and dissemination of microdata. Using microdata, demographic patterns by ethnic group, as well as by education and other important social characteristics, can be examined. The analysis of microdata in ethnically diverse regions has been helpful in China (Anderson and Silver, 1994c, 1995), and officials in China's statistics office have shown interest in this line of research for Xinjiang and other provinces, such as Guangxi and Yunnan. Release of the microdata from the 1989 Soviet census for scientific and policy analysis would be a great help in locating more precisely the sources of problems with data from the traditionally Moslem NIS countries. It would be valuable to examine this kind of microdata before a new census is conducted so that ways to minimize problems in the next census can be devised.

Need for New Data

To obtain more reliable data on mortality, the new states in Central Asia should consider including the Brass child mortality questions in health surveys and perhaps on the next census (Brass, 1975). Brass's methods require that questions be asked about the number of surviving children and the number of children ever born. Sometimes, questions about the ages of surviving children are also asked (Preston and Palloni, 1977).

In addition, surveys that ask questions about health behaviors, such as alcohol consumption, smoking, use of prenatal care, and health checkups for children, along with socioeconomic and demographic information, would aid in discerning risk factors for mortality among various populations. Demographic and health surveys of all types are needed in this part of the world. The surveys should also attempt to collect detailed birth and pregnancy history data, as well as mortality history (in households), to help in providing correctives to official registration data. Although the Chinese model of asking mortality questions in censuses has some limitations (see Anderson and Silver, 1994c), it may be useful when combined with registration data, and it could provide especially valuable information about the social, family, and household conditions related to infant and adult mortality.

Improving the Vital Registration System

Many of the NIS countries have discussed, and some have already undertaken, revisions of their systems of vital registration. We urge attention to the design and management of these systems, including assessment of the needs and possibilities for technological improvements that might improve data quality and the utility of and access to data for policy planners and researchers. Improvement in vital registration data and census data collection requires technical expertise and a substantial commitment of state resources.

For many of the NIS countries, issues of data collection and population registration are highly politicized. Should people be classified by citizenship and by ethnic group membership? Should the internal passport system be abolished or perhaps changed in form? Should a population registry be implemented, and if so, what information should be gathered, and who should have access to it? Which types of marital unions should be registered or recognized'?

Despite the politicization of some issues, effective planning of social policy requires accurate and up-to-date information on population composition and dynamics. Improvement in registration systems requires careful study of the situation in each state. Known problems that are characteristic of the vital statistics in certain states (e.g., age heaping, age exaggeration, misreporting of date of death) need special study and attention.

Planning for Censuses, and Training and Developing the Capabilities of Local Specialists

Elsewhere (Anderson et al., 1994) we have discussed some of the major tasks and opportunities in the development of population statistics in the NIS. All of the new states are likely to begin planning a population census within the next few years. Of the 15 states, only Russia has conducted a microcensus since the 1989 census of the Soviet Union.

Preparation for the census will require substantial technical assistance in most of the NIS countries. This is so not only because of the cost of the census and the competition for state funds, but also because of the lack of trained and experienced personnel in many of the NIS countries. Furthermore, a wide variety of technical issues must be addressed concerning the design of the census questionnaire, the choice of the unit of enumeration, definitions and operational rules, management of field operations, data entry, and data analysis.

We would also add a related task: preservation and archiving of the original microdata from previous and future censuses. We emphasize this point because of the sad state of the data from recent censuses of China. For the 1982 census of China, tapes containing original microdata are in bad shape. Many of the tapes cannot be read, and there appears to be no plan in place to rescue the data while it might still be possible to do so. We urge attention to the condition of data tapes from previous Soviet censuses, as well as the establishment of a policy for preservation and distribution of the data from Moscow to the locales.

Some help in the development of new systems of population statistics has already been provided by international and multilateral organizations. The extent and focus of this assistance should be studied, with an eye toward training and developing the capabilities of local scientific, technical, and administrative workers in population, health, and medical statistics. Although there is a shortage of trained personnel, there are well-qualified demographers in many of the NIS countries who also know the local situation extremely well. Technical assistance would likely be misguided and perhaps ignored if current local experts did not play a major role in planning for data collection and analysis, and if no attention were given to the training and upgrading of skills of local experts.


We thank Victoria A. Velkoff, Vladimir M. Shkolnikov, and the State Statistical Bureau of the People's Republic of China for providing some of the data used in this analysis. Research on this paper was supported in part by NICHD Grant No. P30 HD-10003.


  • Aareleid, T., E. Pukkala, H. Thomson, and M. Hakama 1993. Cervical cancer incidence and mortality trends in Finland and Estonia: A screened vs. an unscreened population. European Journal of Cancer 29A(5):745-749. [PubMed: 8471334]
  • Anderson, B.A. 1986. Family, marriage, and fertility in Russian and Soviet censuses. Pp. 131-154 in R.S. Clem, editor. , ed., Research Guide to Russian and Soviet Censuses . Ithaca, NY: Cornell University Press.
  • Anderson, B.A., and B.D. Silver 1985. a Estimating census undercount from school-enrollment data: An application to the Soviet censuses of 1959 and 1970. Demography 22 (May):289-308. [PubMed: 3996694]
  • 1985. b "Permanent" and "present" populations in Soviet statistics. Soviet Studies 37 (July):386-402.
  • 1986. a Sex differentials in mortality in the Soviet Union: Regional differences in length of working life in comparative perspective. Population Studies 40 (July): 191-214.
  • 1986. b Infant mortality in the Soviet Union: Regional differences and measurement issues. Population and Development Review 12 (December):705-738.
  • 1988. The effects of the registration system on the seasonality of births: The case of the Soviet Union. Population Studies 42 (July):303-320.
  • 1989. a The changing shape of Soviet mortality, 1958-1985: An evaluation of old and new evidence. Population Studies 43 (July):243-265. [PubMed: 11621871]
  • 1989. b Patterns of cohort mortality in the Soviet population. Population and Development Review 15 (September):471-501.
  • 1989. c Demographic sources of the changing ethnic composition of the Soviet Union. Population and Development Review 15 (December):609-656.
  • 1990. a Growth and diversity of the population of the Soviet Union. Pp. 155-177 in Samuel H. Preston, editor. , ed., World Population: Approaching the Year 2000, in The Annals of the American Academy of Political and Social Science 510 (July).
  • 1990. b Trends in mortality of the Soviet population. Soviet Economy 6 (July-September): 191-251. [PubMed: 12343067]
  • 1994. a A comparison of Soviet mortality in the working ages: 1959-1988. Pp. 295-338 in W. Lutz, editor; , S. Scherbov, editor; , and A. Volkov, editor. , eds., Demographic Trends and Patterns in the Soviet Union before 1991 . London: Routledge.
  • 1994. b The geodemography of infant mortality in the Soviet Union, 1950-1990. Research Reports. University of Michigan Population Studies Center 94-316 (August).
  • 1994. c Ethnicity and mortality in northern China. Pp. 752-772 in 1990 Population Census of ChinaProceedings of International Seminar, Beijing: China Statistical Publishing House.
  • 1995. Fertility and sex ratios at birth in China: The effects of parity and sex composition of previous children, based on ethnic comparisons in Xinjiang. Population Studies 49:211-226.
  • Anderson, B.A., K. Katus, and B.D. Silver 1994. Developments and prospects for population statistics in countries of the former Soviet Union. Population Index 60 (Spring):4-20. [PubMed: 12287682]
  • Anderson, B.A., B.D. Silver, and V.A. Velkoff 1987. Education of the handicapped in the USSR: Exploration of the statistical picture. Soviet Studies 39 (July):468-488.
  • Andreev, E., A. Kardash, K. Shaburov, and G. Pavlov 1975. Algoritm rascheta pokazatelei tablits smertnosti i srednei prodolzhitel'nosti predstoiashchei zhizni [Algorithm for tabulating life table indicators and the average length of remaining life]. Vestnik statistiki 3(March):28-35.
  • Arriaga, E. 1984. Measuring and explaining changes in life expectancies. Demography 21 (February):83-96. Baranov, A.A., V.Iu. Al'bitskiy, and Iu.M. Komarov; [PubMed: 6714492]
  • 1990. Tendentsii mladencheskoi smertnosti v SSSR v 70-80-e gody [Trends in infant mortality in the USSR in the 70s and 80s]. Sovetskoe zdravookhranenie 3:3-7. [PubMed: 2142812]
  • Bennett, N.G., and L.K. Garson 1983. The centenarian question and old-age mortality in the Soviet Union, 1959-1970. Demography 20 (November):587-606.
  • Blum, A., and J. Chesnais 1986. La pyramide des âges de l'Union Soviétique au récensement de 1979. Population 41 (November-December): 1043-1058.
  • Blum, A., and A. Monnier 1989 Recent mortality trends in the U.S.S.R.: New evidence. Population Studies 43 (July):243-266.
  • Blum, A., and R. Pressat 1987. Une nouvelle table de mortalité pour l'URSS (1984-1985). Population 42 (November-December):843-862.
  • Brass, W. 1975. Methods for Estimating Fertility and Mortality from Limited and Defective Data . Chapel Hill, NC: Carolina Population Center, Laboratories for Population Statistics.
  • Campbell, D.T., and H.L. Ross 1968. The Connecticut crackdown on speeding: Time series data in quasi-experimental analysis. Law and Society Review 3 (August):33-53.
  • Coale, A.J., P. Demeny, and B. Vaughan 1983. Regional Model Life Tables and Stable Populations , Second edition. New York: Academic Press.
  • Coale, A.J., and G. Guo 1989. Revised regional model life tables at very low levels of mortality. Population Index 55 (Winter):613-643. [PubMed: 12342771]
  • 1990. Note. Population Index 56 (Spring):26-41.
  • Coale, A.J., and E. Kisker 1986. Mortality crossovers: Reality or bad data? Population Studies 40 (November):389-401.
  • Coale, A.J., and S. Li 1991. The effect of age misreporting in China on the calculation of mortality rates at very high ages. Demography (May):293-301. [PubMed: 2070900]
  • Condran, G.A., C. Himes, and S.H. Preston 1991. Old-age mortality patterns in low-mortality countries: An evaluation of population and death data at advanced ages, 1950 to the present. Population Bulletin of the United Nations 30:23-60.
  • Dechter, A.R., and S.H. Preston 1991. Age-misreporting and its effects on adult mortality estimates in Latin America. Population Bulletin of the United Nations 31/32:1-16. [PubMed: 12343668]
  • Dmitrieva, R., and Ie. Andreev 1987. O srednei prodolzhitel'nosti zhizni naseleniia SSSR [On the average life expectancy of the population of the USSR]. Vestnik statistiki 12 (December):31-39. [PubMed: 12178705]
  • Dutton, J., Jr. 1979. Changes in Soviet mortality patterns, 1959-77. Population and Development Review (June):267-291.
  • Ewbank, D. 1981. Age Misreporting and Age-Selective Underenumeration: Sources, Patterns, and Consequences for Demographic Analysis . Committee on Population and Demography. Report No. 4. Washington, D.C.: National Research Council.
  • Feshbach, M. 1960. The Soviet Statistical System: Labor Force Recordkeeping and Reporting . U.S. Census Bureau. Series P-90, No. 12. Washington, D.C.: Government Printing Office.
  • 1962. The Soviet Statistical System: Labor Force Recordkeeping and Reporting Since 1957. U.S. Census Bureau. Series P-90, No. 17. Washington, D.C.: Government Printing Office.
  • 1972. Soviet industrial labor and productivity statistics. Pp. 195-228 in V.G. Treml, editor; and J.P. Hardt, editor. , eds., Soviet Economic Statistics . Durham, NC: Duke University Press.
  • Garson, L.K. 1986. The Centenarian Question: Old Age Mortality in the Soviet Union 1897-1970. Ph.D. dissertation. Princeton University. [PubMed: 11622921]
  • 1991. The centenarian question: Old-age mortality in the Soviet Union, 1897-1970. Population Studies 45 (July):265-278. [PubMed: 11622921]
  • Kamps, J.E. 1976. La declaración de la edad en los censos de población de la América Latina . Serie C. No. 1004. Costa Rica: CELADE.
  • Katus, K., Compiler 1994. a Life Tables: Counties, 1986-1991 . Population Statistics of Estonia. RU Seeria C, No. 4. Tallinn: Estonian Interuniversity Population Research Centre.
  • 1994. b Infant Mortality: Counties, 1965-1993 . Population Statistics of Estoni. RU Seeria C, No. 5. Tallinn: Estonian Interuniversity Population Research Centre.
  • Kingkade, W.W. 1985. Evaluation of Selected Soviet Population Statistics . U.S. Bureau of the Census, Center for International Research. CIR Staff Paper No. 9. Washington, D.C.
  • 1987. Changes in the Treatment of Old-Age Mortality in Soviet Official Life Tables. Manuscript. Center for International Research, U.S. Bureau of the Census. Washington, D.C.
  • 1989. Content, organization, and methodology in recent Soviet population censuses. Population and Development Review 15 (March): 123-138.
  • Krumin, J. 1993. [Length of life-trends and problems of increase]. Riga: University of Latvia.
  • 1994. Changing mortality patterns in Latvia, Lithuania, and Estonia. Pp. 403-419 in W. Lutz, editor; , S. Scherbov, editor; , and A. Volkov, editor. , eds., Demographic Trends and Patterns in the Soviet Union Before 1991 . London: Routledge.
  • Ksenofontova, N.Iu. 1990 Nekotorye tendentsii mladencheskoi smertnosti v poslednee desiatiletie [Some trends in infant mortality in the last decade]. Pp. 116-134 in A.G. Volkov, editor. , ed., Demograficheskie protsessy v SSSR [Demographic Processes in the USSR]. Moscow: Nauka.
  • 1994. Trends in infant mortality in the USSR. Pp. 359-378 in W. Lutz, editor; , S. Scherbov, editor; , and A. Volkov, editor. , eds., Demographic Trends and Patterns in the Soviet Union Before 1991 . London: Routledge.
  • Manton, K.G., and E. Stallard 1984. Recent Trends in Mortality Analyses . New York: Academic Press. Manton, K.G., S.S. Poss, and S. Wing;
  • 1979. The black-white mortality crossover: Investigation from the perspective of the components of aging. The Gerontologist 19 (June):291-300. [PubMed: 551024]
  • Meslé, F., V. Shkolnikov, and J. Vallin 1992. Mortality by cause in the USSR in 1970-1987: The reconstruction of time series. European Journal of Population 8:281-308. [PubMed: 12285970]
  • Myers, R.J. 1978. An investigation of the age of an alleged centenarian. Demography 15 (May):235-236. [PubMed: 658564]
  • Nam, C.B., N.L. Weatherby, and K.A. Ockay 1978. Causes of death which contribute to the mortality crossover effect. Social Biology 25 (Winter):306-314. [PubMed: 757307]
  • Nuñez. L. 1984. Una aproximación al efecto de la mala declaración de la edad en la información demográfica recabada en México . Mexico City: Dirección General del Registro Nacional de Población y Identificación Personal.
  • Preston, S.H., and A. Palloni 1977. Fine-tuning Brass-type mortality estimates with data on ages of surviving children. Population Bulletin of the United Nations (10):72-87. [PubMed: 12310592]
  • Rahu, M. 1992. Cancer epidemiology in the former Soviet Union. Epidemiology 3 (September):464-470. [PubMed: 1391142]
  • Rosenwaike, I., and B. Logue 1983. Accuracy of death certificate ages for the extreme aged. Demography 20 (November):569-585.
  • Rosenwaike, I., and S.H. Preston 1984. Age overstatement and Puerto Rican mortality. Human Biology 56:503-525. [PubMed: 6489994]
  • Russia, Ministerstvo zdravookhranenie 1994. K zdrorovoi Rossii [Toward a Healthy Russia]. Moscow: Ministerstvo zdravookhraneniia RF.
  • Sachuk, N.N., and V.P. Minaeva 1976. Regional'nye osobennosti urovnia dolgoletiia naseleniia SSSR [Regional peculiarities of the level of long-living of the population of the USSR]. Gerontologiia i geriatriia 1976:Ezhegodnik (Kiev):87-93.
  • Shenfield, S.D. 1992. The struggle for control over statistics: The role of the central statistical administration within the inclusive statistical system of the USSR. Pp. 89-120 in J.R. Millar, editor. , ed., Cracks in the Monolith: Party Politics in the Brezhnev Era . Armonk, NY: M.E. Sharpe.
  • Shkolnikov, V. 1994. Life Expectancy and Causes of Death in Russia: The Overview of Trends in 1970-1992. Paper presented at the Seminar on the Geodemography of the Former Soviet Union. August. Radford University, Radford, Virginia.
  • Shkolnikov, V., F. Meslé, and J. Vallin 1994. Prodolzhitel'nost' zhizni i smertnost' naseleniia Rossii: Novoe nastupleniia neschastnikh sluchaev, travm i nasil'stvennoi smerti . Moscow: Tsentr demografii i ekologii cheloveka.
  • Shkolnikov, V.M., and S.A. Vassin 1994. Spatial differences in life expectancy in European Russia in the 1980s. Pp. 379-402 in W. Lutz, editor; , S. Scherbov, editor; , and A. Volkov, editor. , eds., Demographic Trends and Patterns in the Soviet Union Before 1991 . London: Routledge.
  • Sinelnikov, A.B. 1988. Dinamika urovn'ia smertnosti v SSSR [Dynamics of the level of mortality in the USSR]. In L.L. Rybakovskiy, editor. , ed., Naselenie SSSR za 70 let [Population of the USSR During 70 Years]. Moscow: Nauka.
  • Treml, editor; , Vladimir G., editor; , and John P. Hardt, editor. , eds. 1972. Soviet Economic Statistics . Durham, N.C.: Duke University Press.
  • United Nations 1980. Patterns of Urban and Rural Population Growth . New York: United Nations.
  • 1982. a Model Life Tables for Developing Countries . New York: United Nations.
  • 1982. b Levels and Trends of Mortality Since 1950 . New York: United Nations.
  • Urlanis, B.Ts. 1976. Narodonaselenie: Issledovaniia, publitsistika[Human Population: Research and Dissemination]. Moscow: Statistika.
  • USSR, Ministerstvo zdravookhraneniia SSSR 1984. O dalneishim sovershenstovovanii vedeniia meditsinskoi documentatsii, udosoveriaioushchei sluchai rozhdenii i smertei [On the Further Improvement of the Procedures for Medical Documentation Certifying Cases of Births and Deaths]. Prikaz No. 1300, 19 noiabria 1984. Moscow: Ministerstvo zdravookhraneniia SSSR.
  • USSR, TsSU Itogi vsesoiuznoi perepisi naseleniia 1959 goda: Svodnyi tom [Results of the All-Union. 1962.
  • 1963. Census of Population of 1959: Summary Volume]. Moscow: Gosstatizdat.
  • Vallin, J., and J. Chesnais 1974. Evolution récente de la mortalité en Europe, dans les pays Anglo-Saxons et en Union soviétique, 1960-1970. Population 29 (July-October):861-898.
  • Vaupel, J.W. 1986. How change in age-specific mortality affects life expectancy. Population Studies 40 (February): 147-158. [PubMed: 11611920]
  • Vaupel, J.W., K.G. Manton, and E. Stallard 1979. The impact of heterogeneity in individual frailty on the dynamics of mortality. Demography 16 (August):439-454. [PubMed: 510638]
  • Velkoff, V.A. 1990. Trends in infant mortality in the Soviet Union, 1984-1988. Presented at the International Conference on Mortality and Health Care Systems. Varna, Bulgaria.
  • 1992. Trends and Differentials in Infant Mortality in the Soviet Union for the Years 1970-1988 . Unpublished Ph.D. dissertation. Princeton University.
  • Velkoff, V.A. and J.E. Miller 1995. Trends and differentials in infant mortality in the former Soviet Union: How much is due to misreporting? Population Studies 49 (July):241-258. [PubMed: 11639795]


1. On the characteristics and problems of economic statistics in the Soviet Union, see Feshbach (1960, 1962, 1972), Shenfield (1992), and Treml and Hardt (1972).

2. For a discussion of the issues involved, see the classic study by Campbell and Ross (1968).

3. A greater than 100 percent apparent intercensal survival of young children persisted in the 1970 census (Anderson and Silver, 1985a; Kingkade, 1985). An analysis of the 1979 and 1989 censuses reveals a similar, though perhaps less serious, pattern of undercounting of young children.

4. The Russian word "natsional'nost" is translated into English as "ethnic group," since the English word "nationality" has connotations of ''citizenship."

5. Using detailed new statistics medical registration of births from Estonia, Katus (1994a) has calculated the differential in the infant mortality rate due to the shift to be 16.6 percent.

6. The shift to the WHO definitions of live birth and infant death was a good idea, but it creates problems of data comparability. Many other Soviet definitions were not standard. The Soviet definition of a family, for example, was different from that used anywhere else in the world (Anderson, 1986).

7. For a discussion of problems and opportunities in population statistics in the NIS, see Anderson and Silver (1994b). In that paper, we speculated that in the Central Asian states, reported infant mortality rates would fall, because of an increase in the proportion of infant deaths not being recorded. We also speculated that this would be interpreted as an indication of the positive consequences of throwing off Soviet control. In Uzbekistan and Tajikistan, both the reported decline in infant mortality and this rosy interpretation have in fact occurred (personal communication from Vladimir Shkolnikov).

8. Other patterns can also occur. Among Han Chinese, there is some evidence of heaping on a 12-year cycle corresponding to the animal years of the lunar calendar. See Anderson and Silver (1994c).

9. That this is not, strictly speaking, a characteristic of Moslem populations, but depends on other cultural characteristics, is illustrated by the case of the Hui (so-called Moslem Chinese), who also reside in large numbers in Xinjiang, but do not show any sign of the age heaping observed for the Uighurs and Kazaks. It appears likely that the Hui use the Chinese lunar calendar to reckon their ages.

10 It has been speculated that the higher mortality rates in urban than in rural areas could be real because of worse public health conditions, environmental hazards, and epidemics of communicable diseases in cities—akin, perhaps, to the experience in the United Kingdom in the eighteenth and nineteenth centuries. The mortality risks in the Soviet Union and its successor states in the latter half of the twentieth century have on the whole been significantly higher in rural than in urban areas.

11. We have estimated that the Soviet definition of a live birth and an infant death led to a reported infant mortality rate 22 to 25 percent lower than that which would have resulted from using the WHO-recommended definitions (Anderson and Silver, 1986b). But in Central Asia, the definitional difference is only a small fraction of the error. Baranov et al. (1990) estimate that in 1970, while the reported infant mortality rate for Central Asia was 36 infant deaths per 1,000 live births, the actual rate was 128 using Soviet definitions and 161 using the WHO definitions of live birth and infant death. Although Ksenofontova (1994) questions the methods used by Baranov et al. (1990), her own estimates are not much lower than those resulting from the latter methods for this period. For further discussion, see Anderson and Silver (1990b, 1994b). For an examination of cause of death for infant mortality in Central Asia as a way of detecting data error, see Velkoff (1990, 1992) and Velkoff and Miller (1995).

12. This is due in part to inconsistent application of rules for attributing deaths to the permanent place of residence of the deceased rather than the place of occurrence of the event. See Anderson and Silver (1985b, 1994a).

13. For further discussion of trends in mortality by age and region, see Anderson and Silver (1989b, 1990b), Blum and Monnier(1989), Dutton (1979), and Sinelnikov (1988).

14. For Estonia, infant mortality rates and life tables by county for the Soviet period and the early 1990s have just been published (Katus, 1994a, 1994b).

15. We have studied seasonal patterns of registered births in the republics of the former Soviet Union as an indicator of the overall quality of the vital registration system (Anderson and Silver, 1988). The rank ordering of the republics in the plausibility of the seasonal pattern of births corresponds closely to our evaluation of the quality of mortality data by republic.

16. This conclusion is based in part on an analysis of mortality data for Russian provinces we undertook in collaboration with Vladimir Shkolnikov and Sergei Vassin.

17. The U.N. program COMPAR, part of MORTPAK, was used to calculate the implied levels of expectation of life at birth. When the implied expectation of life at birth was greater than 80 years, it is plotted here as 82. There has been work on model life tables at very low levels of mortality (Coale and Guo, 1989, 1990). It is not plausible that in the traditionally Moslem republics of the former Soviet Union, actual mortality would be consistent with an expectation of life at birth of more than 80 years.

18. For discussion of the selection of a standard as a common metric and for an assessment of the plausibility of the reported "shape" of mortality curves in different regions of the former Soviet Union, see Anderson and Silver (1989a).

19. This was also argued by Dmitrieva and Andreev (1987).

Copyright 1997 by the National Academy of Sciences. All rights reserved.
Bookshelf ID: NBK233384


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (5.7M)

Related information

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...