Theoretically, China has advantages in measuring and evaluating demographic changes, especially the levels and trends of fertility, given a collection of sources of data in population statistics. However, as shown in previous chapters, not all data meet the purpose of estimating fertility with reasonable assurance. In addition, the misunderstanding of data sources and confusion in data use have greatly misled the judgment in most analyses on the quality of data. Therefore, it is very important to evaluate the roles of different data sources in fertility estimation and to estimate the extent of underreporting in each source.
As is the practice of most demographers, this thesis also considers the two registration-based statistics, the hukou statistics and the family planning statistics, not to be suitable for fertility estimation. The omission of both married women and their births on a large scale in the hukou statistics precludes it from being used in fertility analysis. Similarly, the family planning statistics are unsuitable, because they suffer from both the concealment of births by individual couples and the falsification of birth records by local program workers. However, this does not mean that these two sets of statistics are quite useless. Regarding the out-of-plan births appearing in official statistics, it is almost certain that it would no longer be necessary for them to be concealed. In this case, the number of registered births in these statistics constitutes the minimum to evaluate the estimated number of births from other sources. For example, if any survey estimates or census enumerates fewer births or younger children than the registrations, one can safely reject the completeness of that source. A good example is that the 2000 census enumerated fewer children aged one year than were registered in the family planning statistics in 1999, as mentioned in Chapter 3 (see also columns 1-3 in Table 6.6).
There are also two important lessons from the analysis of these two sources. First, in the hukou statistics, many couples who had out-of-plan births were eager to justify their births through their being registered either as adopted children or as in- migrants, while many couples were not enthusiastic about registering their ‘in-plan’ births in a timely fashion due to the diminished importance of hukou in daily life.
Second, the low quality of family planning statistics primarily resulted from the manipulation and falsification of birth records by local program workers. The SFPC village surveys and many careful local studies demonstrated that face-to-face interviews carried out by trained enumerators could obtain more truths directly from respondents, once they successfully avoided interference from local cadres. Moreover, a large number of births were missed by the omission of young mothers rather than by the outright underreporting of births.
The situation is different in surveys and censuses, because these investigations were carefully designed and their data were gathered by trained enumerators through face- to-face interviews. In this case, it was expected that most underreporting of births came directly from respondents, although local cadres still played an important part. One of the disadvantages for nationwide surveys and censuses is that the operations have to rely heavily on local co-operation. For the sample survey, the smaller the survey, the more likely it is to employ trained enumerators to conduct household interviews. The larger the enumeration, the more heavily it relies on co-operation by local cadres, which is especially striking in the population census, mobilising 5-6 million enumerators across the country. Therefore, there are differences in the extent of underreporting of births, so the upward adjustment needed is dependent on whether one chooses census or surveys for analysis.
Empirical evidence suggests that fertility rates estimated from censuses and surveys are quite similar, although those estimated from censuses were slightly lower. The total fertility rates directly calculated from censuses were 2.62 in 1981, 2.25 in 1989 and 1.22 in 2000. The first two were quite close to the results from other sources, such as 2.70 in 1981 reported in the 1982 one-per-thousand survey, and 2.24 reported in 1989 from the 1992 survey. The census-recorded total fertility for 2000 of 1.22 was noticeably lower than the 1.45 observed in the SFPC 2001 survey, and 1.38 reported in the SSB 2001 annual survey. However, the fertility estimated using the own-children method, which can avoid the effects of migration of young mothers as mentioned in Chapter 4, was 1.38 in the census, still close to the survey-estimated figure (Retherford et al. 2004).
All the three most recent censuses undercounted newborns and younger children, but the problem in the 2000 census is more serious, mainly because of the huge number of migrants and the shortcomings in census procedures. Nevertheless, the 10 percent census long form is still usable in fertility analysis, because its samples had already been chosen based on households before the census enumeration. More importantly, it was less affected by the campaign to put back ‘newfound’ people, very likely including many double-counts, into the short form in the second-round census enumeration (Zhang and Cui 2003). Compared to annual surveys, the 1, 000-times- larger coverage of the census long form suggests its employment of enumerators on a large scale and hence more biases. By the same token, the 1995 one percent sample census may suffer from more underreporting of births than an annual survey, but is better than the census long form.
Both the 1982 and 1990 censuses have excellent age-sex reporting except in younger age groups, but this is not the case in the 2000 census, even in the long form. Therefore, when it is necessary to use the age structure of the whole population or of women of reproductive age, it is better to use the previous two censuses. Three life tables derived from the 1982, 1990 and 2000 censuses, which reported a decline in mortality level (Banister and Hill 2004). Since all three censuses suffered some extent of underreporting of deaths, it is safer to use the middle one, the 1989-90 census-based life table with a moderate mortality level.
There are many advantages in using the SSB annual surveys in reconstruction of fertility levels: (1) the only available time series of data; (2) the negligible sampling errors due to large sample size with nationwide representation; (3) its independence from the family planning program; and (4) the detected extent of underreporting of births in each survey from a more intensive post-enumeration survey. As examined in Chapter 5, the SSB official estimates adjusted fertility too high. From the information available, the fertility and births estimated from the unadjusted survey data are still more complete than all other sources.
The SFPC quinquennial surveys provide valuable sources to examine the problems in the ‘current’ status surveys. Because the three most recent surveys included only a single group of women aged 15-49, and suffered from underreporting of births close
to the survey time, the fertility series estimated from these surveys were lower than those from the annual surveys, and their smaller sample sizes imply greater sampling variations in detailed subgroup analyses. All these considerations make these retrospective surveys inferior to annual surveys in fertility analyses. The 1992 survey is without dispute not suitable, but both the 1997 and 2001 surveys are of comparatively good quality, proved by internal consistency, and consistencies between surveys. The 2001 survey has more complete reporting of births covering the whole decade, and is used in analysis in following chapters.
Consequently, for the purpose of reconstructing the path to below-replacement fertility in China in the 1990s, the primary data used in this research come from three sources: (1) the unadjusted data of the SSB annual surveys from 1991 to 1994 and 1996 to 1999; (2) data from the 1995 sample census and the long form of the 2000 census; and (3) the 2001 SFPC retrospective survey. In addition, the 1982 and 1990 census data and the 1990 census-based life tables (1989-90) are used when it is necessary to use the age structure of women of reproductive age.