Extrapolation of IUD retention rates: an evaluation of some models
Rao, K.A.; Sholapurkar, M.V.
Artha Vijnana Journal of the Gokhale Institute of Politics and Economics Poona 21(1): 1-12
1979
ISSN/ISBN: 0004-3559 PMID: 12265164 Document Number: 496324
Contraceptive retention rates play a vital role in estimating the number of births averted. Generally, the data based on follow-up studies are of a short duration of 1 or 2 years, necessitating the use of the continuation curves for extrapolating the retention rates. The most widely used curve for extrapolating the IUD retention rates is negative exponential of Mauldin et al. Modified versions of the above curve and other alternative curves have also been proposed. An attempt is made in this discussion to evaluate precited curves so as to find a suitable method of extrapolation for Indian conditions, i.e., to determine a method which would give better results for a short span of 5-6 years and for which estimates of the parameters could be obtained over desk calculators. The evaluation is carried out in 2 stages. At the 1st instance, assuming that the retention rates are available only for a period of 24 months. Extrapolation is done for over a period of 60 months and the results thus obtained are compared with actual life table values. Robustness of these curves are tested by fitting them using only a segment of the data, i.e., using data from 13-24 months. Estimated values of the retention rates for the retrospective and prospective periods are then computed and compared with actual life table values. The curve which has got the minimal sum of squares of the deviations is considered as the best. The curves that are considered for evaluation are negative exponential curve, mixture of 2 exponentials, modified negative exponential, and modified type 3 exponential. The data for the analysis was taken from a retrospective survey conducted by the Population Center, Bangalore. Modified type 3 exponential gives a better fit, compared to single and the modified exponential functions. There is some evidence to show that the total curve of cumulative retention rates is a mixture of 2 or more curves. The analysis reveals that in cases where single exponential must be resorted to, it is advisable to use the later part of the available data, as the extrapolation based on it minimizes the underestimation in the later months.