In many less developed countries, measuring child mortality is difficult due to non-functioning or seriously incomplete vital registration systems. Under such conditions, the preceding birth technique has gained popularity in recent years because of its simplicity and minimal data requirements. The technique makes innovative use of data collected at the time of a woman's delivery and requires only that she state whether her preceding birth, if any, is still alive. The proportion dead of preceding births, $\Pi$, is taken to approximate the probability of dying between birth and age 24 months--q(24).
Although trials of the preceding birth technique in many countries have generally been supportive of the method's performance, a disproportionate number of these studies have simply compared $\Pi$ with the q(24) estimated from life tables based on the same sample of preceding births. The practical value of the technique, however, rests with whether it allows us to make inferences about the mortality conditions faced by the overall population of births.
The purpose of this study is to evaluate systematically whether $\Pi$ is consistent not only with life table estimate of q(24) based on the preceding births, but also with that based on all births in a recent five-year period. To do so, we use data from 24 World Fertility Surveys. Since the preceding birth technique has been criticized for its reliance on births delivered in clinics, the consistency of $\Pi$ derived from births associated with doctor-assisted deliveries with the life table estimate of q(24) based on all births in a recent five-year period is also explored. In addition, comparisons between $\Pi$ and the life table estimate are undertaken with births classified by maternal age, residence, maternal education, birth order and gender. Unlike previous studies, techniques of statistical inference are employed to determine whether observed percent differences are statistically significant both in individual comparisons and in grouped comparisons.
Our results demonstrate that overall, differences between $\Pi$ and q(24) are statistically significant when q(24) is derived from life tables based on either the next-to-last births or all births in the five-year period. Across the 24 countries, differences associated with the former comparisons average about 5 percent, while differences associated with the latter comparisons average about 8 percent, albeit both with considerable variance. The frequency with which $\Pi$ is seen to exceed q(24) indicates a definite upward bias of the PBT estimate. In addition, cross-sectional comparisons across countries suggest that $\Pi$ may not adequately distinguish mortality differences between populations and instead may exaggerate differences between high and low mortality populations.
