• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site

Working paper

New methods for estimating detailed fertility schedules from abridged data

MPIDR Working Paper . WP. Max Planck Institute for Demographic Research, 2018. No. WP 2018-001.
Grigoriev P., Anatoli I. M., Vasily P. G., Jdanov D., Shkolnikov V.
Background and Aim Occasionally, there is a need to split aggregated fertility data into a fine grid of ages. For this purpose, several disaggregation methods have been developed. Yet these methods have some limitations. We seek to identify a method that satisfies the following criteria: 1) shape – the estimated fertility curves should be plausible and smooth; 2) fit – the predicted values should closely trace the observed values; 3) non-negativity – only positive values should be returned; 4) balance – the estimated five-year age group totals should match the input data; and in case of birth order data 5) parity – the balance by parity has to be maintained. To our knowledge, none of the existing methods fully meets the first four criteria. Moreover, no attempt has been made to extend the restrictions to criterion (5). To address the disadvantages of the existing methods, we introduce two alternative approaches for splitting abridged fertility data: namely, the quadratic optimization (QO) method and the neural network (NN) method. Data and Methods We mainly rely on high-quality fertility data from the Human Fertility Database (HFD), Additionally, we use a large and heterogeneous dataset from the Human Fertility Collection (HFC). The performance of the proposed methods is evaluated both visually (by examining of the obtained fertility schedules), and statistically using several metrics of fit. The QO and NN methods are tested against the current HFD splitting protocol (HFD method) and the calibrated spline (CS) method. Results The results of thorough testing suggest that both methods perform well. The main advantage – and a distinguishing feature – of the QO approach is that it meets all of the requirements listed above. However, it does not provide a fit as good as that of the NN and CS methods. In addition, when it is applied to birth order data, it can sometimes produce implausible shapes for parity 1. To account for such cases, we have developed individual solutions, which can easily be adapted to account for other cases that might occur. While the NN method does not satisfy the balance and parity criteria, it returns better results in terms of fit than the other methods. Conclusions The QO method satisfies the needs of large databases such as the HFD and the HFC. While this method has very strict requirements, it returns plausible fertility estimates regardless of the nature of the input data. The NN method appears to be a suitable alternative for use in individual cases in which the priority is given to the fit criterion.