Predicting Fertility in Sub-Saharan Africa Based on Patterns of Contraceptive Use

Publication date: 2015

Predicting Fertility in Sub-Saharan Africa Based on Patterns of Contraceptive UseHEALTH POL ICY P R O J E C T BACKGROUND ■ In sub-Saharan Africa (SSA), expected national fertility levels and country-level observations demonstrate repeated mismatches in magnitude and/or direction. ■ Bongaarts’ Proximate Determinants (PD) model (1978)—the most robust and commonly used method for examining fertility changes—does not account for all the variation in observed fertility, often producing sizeable residuals. Bongaarts’ Proximate Determinants Framework: Total Fertility Rate (TFR) = Cm * Cc * Ca * Ci * TF Cm index of marriage, or union Cc index of contraception Ca index of induced abortion Ci index of lactational infecundability TF total fecundity CONTACT US Health Policy Project 1331 Pennsylvania Ave NW, Suite 600 Washington, DC 20004 www.healthpolicyproject.com email: policyinfo@futuresgroup.com Tel: +1.202.775.9680 Fax: +1.202.775.9684 The Health Policy Project is a five-year cooperative agreement funded by the U.S. Agency for International Development under Agreement No. AID- OAA-A-10-00067, beginning September 30, 2010. HPP is implemented by Futures Group, in collaboration with Plan International USA, Avenir Health (formerly Futures Institute), Partners in Population and Development, Africa Regional Office (PPD ARO), Population Reference Bureau (PRB), RTI International, and the White Ribbon Alliance for Safe Motherhood (WRA). The information provided in this document is not official U.S. Government information and does not necessarily represent the views or positions of the U.S. Agency for International Development. PRESENTED BY Ellen Smith Bernice Kuang Kaja Jurczynska Health Policy Project, Futures Group Population Association of America 2015 Annual Meeting April 30–May 2, 2015 San Diego, CA 1 -5 Eastern Africa Middle Africa Southern Africa Western Africa 0 -1 -2 -3 -4 Ethiopia 2011 Kenya 2008–09 M adagascar 2008–09 M alaw i 2010 M ozam bique 2011 Rw anda 2010 Tanzania 2010 U ganda 2011 Zam bia 2007 Zim babw e 2010–11 C am eroon 2011 C had 2004 C ongo 2011–12 G abon 2012 Lesotho 2009 N am ibia 2006–07 C ôte d’Ivoire 2011–12 Benin 2011–12 Burkina Faso 2010 G hana 2008 G uinea 2012 M ali 2006 N iger 2012 N igeria 2013 Senegal 2012–13 Chart A. TFR Residuals in SSA, Bongaarts Original Predicted-Observed HEALTH POL ICY P R O J E C T Adjustment A Eliminating CPR-Postpartum Insusceptibility (PPI) Overlap Contraceptive users who were also considered postpartum insusceptible were removed from the CPR estimate. This adjustment eliminates the problem of double counting PPI contraceptive users as protected against the risk of pregnancy, since these women are also counted as protected in the Ci index. Adjustment B Accounting for the CPR-TFR Timing Mismatch We interpolated CPR for 27 months prior to each survey date. Twenty-seven months was used because it is the midpoint of when the births that contribute to the TFR were conceived. This adjustment aligns the timing of the CPR and TFR metrics. Adjustment C Customizing Total Fecundity (TF) to Each Country TF was customized for each country, rather than assuming a constant 15.3 (Bongaarts) or 21 (Stover). We estimated TF for each survey by algebraically solving the PD equation for fecundity [TF = TFR / (Cm * Cc * Ci * Ca)]. Survey-specific TFs were averaged for each country. Judging Predictive Accuracy TFR Confidence Intervals All previous analyses used arbitrary cutoffs, such as +/- 0.5, when judging the accuracy of TFR predictions. For a more evidence-based approach, we used a 95% confidence interval to judge our TFR predictions. The confidence intervals were based on reported sampling errors. Chart C. Average Country-Specific Total Fecundity, by Variation 30 25 20 15 10 5 0 Eastern Africa Middle Africa Southern A frica Western Africa Ethiopia Kenya M adagascar M alaw i M ozam bique Rw anda Tanzania U ganda Zam bia Zim babw e G abon C had C am eroon C ongo Lesotho M ali Benin Burkina Faso C ôte d’Ivoire G hana G uinea N iger N igeria Senegal Bongaarts Original Sexually Active Variation JS Variation TF = 21 TF = 5.3 INDIVIDUAL ADJUSTMENTS Chart B. Average CPR by Country Before and After PPI Exclusion by Variation Bongaarts Original: CPR Total Bongaarts Original: CPR excl. PPI Sexually Active Variation: CPR Total Sexually Active Variation: CPR excl. PPI JS Variation: CPR Total JS Variation: CPR excl. PPI 60% 50% 40% 30% 20% 10% 0% Eastern Africa Middle Africa Southern Africa Western Africa Ethiopia Kenya M adagascar M alaw i M ozam bique Rw anda Tanzania U ganda Zam bia Zim babw e C am eroon C had C ongo G abon Lesotho N am ibia C ôte d’Ivoire Benin Burkina Faso G hana G uinea M ali N iger N igeria Senegal DISCUSSION: TFR LEVEL ■ Low accuracy of predicted TFR without adjustments. ■ When adjustments are implemented simultaneously, predictive accuracy reaches at least 50% across variations, a marked improvement. ■ Though the CPR-PPI overlap and CPR interpolation adjustments are commonly suggested in the literature, their application does not render a notable improvement in predictive accuracy. ■ The customized TF adjustment produced the largest individual and combined improvement on predicted TFR. It is unlikely that the range of TF values is due to true variation in biological fertility. Instead, the range of TF values points to the importance of country-level factors exogenous to the PD framework. CONCLUSION ■ Following all adjustments to both TFR level and TFR intersurvey change, the overall accuracy rates are still not suitable for the degree of certainty that policy and program planners need. ■ The key to improving accuracy may lie in better understanding country-specific patterns (e.g. variation in behaviors and sociological patterns), as evidenced by the range of calculated TF values. ■ New or revised methods for measuring determinants are also required. ■ Until more research is done, it will remain difficult to predict short-term fertility change to a pragmatic level of certainty. It is therefore important to adjust expectations about the certainty of fertility predictions—and the impact of determinants—among demographic researchers, donors, policymakers, and program planners in the field. DISCUSSION: TFR INTERSURVEY CHANGE ■ Approximately half of all unadjusted TFR change predictions fell within the relevant 95% confidence interval. ■ Unlike TFR level, the adjustments, when combined, do not lead to substantial improvements in predictive accuracy. ■ When applied individually, the adjustments to any PD variation do not produce major improvements in the accuracy of predicted TFR intersurvey change. Bongaarts Original Sexually Active Variation John Stover Variation CPR-PPI Overlap 14% 23% 18% Interpolated CPR 9% 26% 12% Country-Specific TF 51% 58% 51% Table B. TFR Predictive Accuracy After Each Adjustment TFR Level Results Bongaarts Original Sexually Active Variation John Stover Variation Before Adjustments Accuracy 12% 17% 14% Average Residual (absolute) 1.27 1.02 1.01 After Adjustments Accuracy 55% 63% 45% Average Residual (absolute) .39 .30 .32 Table A. TFR Predictive Accuracy and Residuals Bongaarts Original Sexually Active Variation John Stover Variation Before Adjustments 50% 50% 47.5% After Adjustments 50% 50% 57.5% Table C. TFR Intersurvey Change Predictive Accuracy TFR Intersurvey Change Results RESULTS Research Question To what extent can the accuracy of predicting fertility in SSA using the proximate determinants framework be improved by implementing revisions, with emphasis on the contraception index? METHODOLOGY Identical adjustments were applied to three variations of the proximate determinants model. Revisions to the contraceptive index are prioritized because contraception is the most commonly recognized and intuitive fertility inhibitor with a rights-based policy lever: 1. Bongaarts’ Original PD Model, with a focus on married/in-union women 2. Sexually Active Variation, identical to Bongaarts’ original with the exception of customization for sexually active women 3. John Stover Variation, which features Stover’s revisions (1998) to the original indices Predicted TFRs—65 for TFR level and 40 for changes in TFR between surveys—were computed before and after these adjustments, for all three variations. ■ Thus there is an unfulfilled demand for better explaining, understanding, and communicating how fertility changes. Accurately predicting fertility is critical for understanding how populations may be expected to change, and for managing expectations about the possible impacts of TFR-affecting policy levers

View the publication

You are currently offline. Some pages or content may fail to load.