25

number of births is reported. However, this is usually close to the unweighted total. c Tajikistan: fertility rate is average of official rates for 1998-2000 Sources: Authors’ calculations; Macro International (1995), pages 14-15; see also Appendix 1. 17For discussion of the estimation of change in infant mortality over time see UNICEF (2003), pages 49-51. 24 Model error (the Brass method) Uncertainty due to sampling and non-sampling error is an inherent component of all survey based estimates. However, in the case of the indirect estimation of mortality, using techniques such as the Brass method, new sources of error are introduced, mainly as a result of some modelling assumptions that need to be made. The limited range of questions asked of MICS respondents does reduce the chances of non-response or inaccurate response discussed above. But it does also mean that this indirect estimation technique must be used to estimate infant mortality with these data. Table 1 shows that in the CEE/CIS region, MICS data have been used for this purpose in the case of Albania, Azerbaijan, Tajikistan and Uzbekistan. The Brass method allows mortality rates to be estimated from aggregate information on the number of children born to women in 5-year age groups (15-19, 20-24 etc), and the proportion of children in each group who die. What the Brass method does is convert the observed proportion of children dead among those born to women in a certain age group into an estimate of probability of dying by some exact age, using a set of multipliers which depend on this summary information from the surveys, and data derived from model life tables. There are several versions of the Brass method. The Trussel version (1975), discussed here, is based on model life tables (developed by Coale and Demeny, 1983) which are aggregated into four groups, North, South, West and East, each describing mortality patterns in different regions of Europe and some other countries in 19th and 20th centuries. Thus the Trussel version produces four different sets of mortality estimates for each set of birth and mortality data. Table 10 demonstrates an application of the Trussel version of the Brass method to Azerbaijan MICS data. Number of children born, average number of children per woman in each age cohort, and proportion who have died are calculated from the survey data. 

1     2     3     4     5      6     7     8      9      10      11      12     13      14      15      16     17      18       19       20     21     22     23     24      25