Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Estimating the current mean age of mothers at the birth of their first child from household surveys

Estimating the current mean age of mothers at the birth of their first child from household surveys Background: Estimates of the period mean age at first birth are readily available for countries with accurate vital statistics (i.e., in much of the developed world). In contrast, in most developing countries vital statistics are lacking or incomplete and estimates of the period mean age at first birth are therefore often unavailable. The Demographic and Health Surveys (DHS) program provides a large set of demographic and health statistics for many developing countries, but not the mean age at childbearing or the mean age at first birth. Methods: We propose two different methods for the estimation of the period mean age at first birth from information collected in DHS surveys. The first method is the same as the one used in populations with accurate vital statistics and is based on a weighted average of single year of age first birth rates. The second is the singulate mean age at first birth. Results: A comparison of the two estimates obtained from the latest surveys in 62 countries shows excellent agreement in countries in which there is no evidence of a rise in childlessness. But, as expected on theoretical grounds, there is less agreement in populations that have experienced an increase in the proportion childless. Conclusions: Based on these results, we recommend the first method. The measure is relatively straightforward to calculate and, since it refers to recent births, is presumably more accurately reported than indicators based on events that occurred in the more distant past. This measure makes it possible for the first time to assess recent trends in the onset of childbearing in developing countries with multiple DHS surveys and to compare recent period estimates of the mean age at first birth among countries. Background to delay childbearing and increase access to family Becoming a parent for the first time is one of life’s most planning [6, 7]. Thus, the age at which women have a important and influential events. It signals the onset of first birth is an important indicator of the success of the responsibility for insuring the well-being and success these efforts. Finally, delayed childbearing slows popu- of one’s offspring and of the next generation. For lation growth through increasing the length between women, the age at which they have a first birth can have generations and decreasing population momentum [8]. implications for schooling, labor force participation, and Estimates of both cohort and period mean ages at first overall family size [1]. Early childbearing is also associ- birth are available for countries with reliable vital statistics. ated with elevated risks to the health of the mother and For example, EUROSTAT [9] and the Human Fertility her child [2]. As a consequence, there is a large literature Database [10] provide historical estimates for many coun- on the individual, social, and cultural determinants and tries in Europe and other high-income countries for single consequences of this event and its timing in the life cycle years from the 1980s to around 2010 and for a substantial [3–5]. In addition, a renewed interest in the wellbeing of number of birth cohorts. In contrast, in most developing adolescent girls has led to investmentsinprogramsintended countries vital statistics are lacking or incomplete and esti- mates of period and cohort mean ages at first birth are therefore often unavailable. The Demographic and Health * Correspondence: [email protected] Surveys (DHS) program – under which nationally Population Council, 1 Dag Hammarskjold Plaza, New York, NY 10017, USA © 2015 Bongaarts and Blanc. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Bongaarts and Blanc Population Health Metrics (2015) 13:25 Page 2 of 6 representative household surveys are conducted in devel- An alternative approach to estimating the period mean oping countries – provides many valuable statistics on age at first birth is to rely on a method that is widely demographic and health processes, but does not report on used to estimate the mean age at first marriage, called the period age at childbearing or age at first birth (mean the “singulate mean age at marriage” [11, 12]. The appli- or median). Instead, the standard reports provide the co- cation of this approach to estimate the mean age at first hort median age at first birth as calculated from a birth birth was first mentioned by Casterline and Trussell [13] history reported retrospectively by women of reproductive and subsequently implemented by Afzal and Kiani [14] age. In principle, the DHS could also report the mean age and Booth [15]. The equation is as follows at childbearing for cohorts of women but such means max paðÞ ; t −paðÞ ; t a would be biased downward because of incomplete child- max max M ðÞ t ¼ ð2Þ bearing experience of all but the oldest women. 1−paðÞ ; t max For many analytic purposes estimates of period mea- sures are of greatest interest because, in contrast to co- M*(t) = Average age at first birth at time t hort medians, they allow assessments of recent trends in p(a,t) = Proportion of women that has not yet given the timing of the onset of childbearing for specific refer- birth at age a and time t ence periods. The objective of this research note is to p = The proportion of women that has never had a max propose two different methods for the estimation of birth at a max mean age at first birth from information collected in This mean age at first birth is defined as the mean age DHS and similar surveys. Both measures are unaffected at which women would bear their first child if they went by changes in the population age structure, thus allow- through the reproductive years experiencing the age- ing undistorted comparisons of the timing of the onset specific proportions childless observed at time t. of childbearing between populations and over time In Additional file 1 we demonstrate that the two within populations. Estimates are calculated for the most means are equal (i.e., M (t)= M* (t)) under the condition recent surveys in 62 countries. that the shape of the function p(a,t) by age is invariant with respect to time. This implies that p(a,t) can shift to Methods high or lower ages over time (with corresponding The equation for estimating the period mean age at first changes in first birth rates and in the mean age) but with birth used widely in countries with vital statistics [10] is no change in shape and with constant p . max Estimates of M(t) and M*(t) were obtained with equa- max tions (1) and (2) for the most recent DHS surveys in 62 ðÞ a þ 0:5 baðÞ ; t MtðÞ ¼ ð1Þ developing countries for which data files are available max baðÞ ; t for public use (and with sample sizes of married women above 3000). The number of respondents in each survey where varies but typically is between 5000 and 10,000 women M(t) = Average age at first birth at time t of reproductive age. For many countries several surveys b(a,t) = the age-specific birth rate for birth order one are available, so time series of M(t) and M*(t) can also at (single) age a and time t. be calculated. Further details about the surveys are avail- a = the highest age at which first births are observed able on the DHS website [16]. max This period mean age at first birth is defined as the Estimates of b(a,t) are obtained from birth histories with mean age at which women would bear their first child if a simple variant of the standard DHS method for calculat- they went through the reproductive years having the first ing age-specific birth rates by age for the three years be- birth rates observed in a particular period. fore the survey [17]. To estimate b(a,t) two changes are Numbers of births recorded in vital statistics are typic- made in this method: (1) birth rates are calculated by sin- ally large and birth rates are available by single age and gleyearrather than byfive year age intervalsand (2)the single year. As a result, annual estimates of M(t) can be numerators of the birth rates exclude births of order two estimated. and higher. Estimates of p(a,t) are also calculated with a In contrast, in applications of this equation to DHS variant of the standard DHS method estimating the pro- surveys samples of births in a single year are relatively portion nulliparous by single year of age rather than five small. To obtain more robust estimates of the mean age year age intervals. at first birth for a survey, we calculate b(,a,t) by single Finally, it should be noted that values of p(a,t) are sub- year of age for a period of three years before each sur- ject to substantial sampling errors at ages above 40, be- vey. In addition, we exclude surveys with sample sizes of cause the proportions childless at these ages are usually currently married women below 3000 to minimize sam- less than five percent and the number of respondents is pling errors. smaller than at lower ages. To minimize the effects of Bongaarts and Blanc Population Health Metrics (2015) 13:25 Page 3 of 6 these errors on estimates of the mean age at firth birth, and the solid markers are spread around the diagonal in the value of a is set at 40 years and p is estimated Fig. 1 with a standard error of 0.32 years. The second max max as the average of single age values of p(a,t) between ages cluster of countries with open circles includes several 35 and 45. surveys in which M*(t) is usually substantially higher than M(t). This finding is likely attributable to an up- Results ward bias in M*(t) when values of p are rising (the max Figure 1 plots the estimates of M(t) on the horizontal rare cases in this cluster with M(t) higher than M*(t) are axis and the value of M*(t) on the vertical axis. Each probably attributable to measurement or reporting er- marker represents the most recent survey in each of the rors). Our working assumption therefore is that M(t) is 62 countries. The results are presented in two clusters: an unbiased estimator of the mean age at first birth even the solid markers represent surveys in which p is less in surveys in the second cluster. In addition, all except max than 5 % and the open circles represent surveys with one of the countries in the second cluster have a mean p >5 %. This distinction is made to separate observa- age at birth of 22 or higher. This result is not unex- max tions in which the conditions are met for M(t) to be pected as there tends to be a positive correlation be- equal to M*(t) from observations in which they are not. tween age at first birth and the proportion of women As noted in Additional file 1, a key condition for the who remain childless. equality of M(t) and M*(t) is that p is constant. Un- A full analysis of levels and trends in all 62 countries is max fortunately, it is not easy to determine the rate of change beyond the scope of this methodological study, but a few in p , because some countries have only one survey findings can be noted. Estimates of M(t) vary widely max and, even in countries with multiple surveys, the rate of among countries from a low of 19.1 in Niger (2006) to a change in p is erratic due to small sample sizes. In- high of 24.7 in the Maldives (2009). The unweighted aver- max stead, we assume that countries with p less than 5 % ages of M(t) for countries in each of four regions are pre- max have seen little change in p over time, thus approxi- sented in Table 1. The low value for sub-Saharan Africa is max mating the condition that p is constant. In surveys unsurprising since this continent has not progressed as far max where p is higher than 5 % there has likely been through the fertility transition as the other regions. North max change over time because early in the fertility transition Africa/West Asia and South Asia have the highest aver- p is typically a very small number. ages and Latin America has intermediate values. max It is therefore expected that M(t) is closer to M*(t) for Figure 2 presents trends in M(t) for selected countries surveys in the cluster with p <5 %. As is evident from in the developing and developed world. Estimates for max Fig. 1, this is indeed the case. For these surveys the aver- Egypt, Nigeria, India, Kenya, and Bangladesh show very age value of M(t) and M*(t) are respectively 21.0 and modest increases from the 1990s to near 2010. The 21.2, a difference of only 0.2 year (which is not statisti- mean ages at first birth for the Japan, Czech Republic, cally significant). However, the agreement is not perfect UK, and US are mostly substantially higher and have been rising at a more rapid pace than in the five devel- oping countries included in the figure. As noted, DHS published reports provide estimates of the retrospectively reported cohort median age at first birth. These medians are estimated from birth histories y = 1.0092x R² = 0.9352 obtained from respondents of reproductive age. The age at first birth is calculated by subtracting the woman’s date of birth from the date of birth of her first child. Medians for the cohorts aged 25–29 at the time of the survey and above are available for nearly all DHS surveys because the medians are reached before age 25 22 Pmax<0.05 (i.e., at least half of women have had a birth before age 25). 21 Pmax>0.05 Table 1 Average and standard deviation of country estimates of M(t) by region Average of M(t) Standard deviation N Sub-Saharan Africa 20.9 1.1 33 Latin America 21.7 0.9 11 18 19 20 21 22 23 24 25 26 Mean age M (years) South Asia 22.7 1.6 10 Fig. 1 Period mean age at first birth (M* vs M) North Africa/West Asia 23.3 1.1 8 Mean age M* (years) Bongaarts and Blanc Population Health Metrics (2015) 13:25 Page 4 of 6 21.0 Mean (period) 20.5 Japan UK Czech. Rep 20.0 US Egypt Nigeria 19.5 Medians (cohort) India Kenya Bangladesh 19.0 40-44 35-39 30-34 25-29 (age at time of survey) 18.5 18.0 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 Fig. 2 Period mean age at first birth for selected developing and Fig. 3 Period mean and cohort median age at first birth, based on developed countries five surveys in Kenya For a small number of surveys medians are available for refers = 2010.5–12.5 = 1998.0). This approach allows the the cohorts aged 20–24 when the median is below age 20. comparison of cohort medians reported in different sur- These cohort medians have the advantage of being veys and of cohort and period means [18, 19]. With fully available for all DHS surveys but there are also draw- accurate reporting of the timing of first births the lines of backs: 1) the median refers to past experience of cohorts medians plotted in Fig. 3 would exactly overlap (assuming and is therefore not as current as is preferable for many no selectivity of migration and mortality). For example, analytic purposes; 2) the retrospective reporting of the women aged 35–39 should report a median that is the date of the first birth may suffer from recall errors that same as the median reported by women aged 25–29 in a are likely to increase as the time since the event rises; survey conducted ten years earlier. The fact that the lines and 3) the cohort median as calculated by DHS is not do not match indicates misreporting. In particular, it independent of the quantum of first births and can seems that the older cohorts have moved the time of change over time even if the mean is constant. The first the first birth closer to the survey date so that their re- two of these disadvantages also apply to cohort mean ported medians are higher for most years than the me- ages at first birth, a measure we do not discuss because dians reported by younger cohorts for the same years. it is very rarely used as it can only be estimated accur- This pattern is consistent with earlier analyses of data ately for women who have completed their childbearing. quality undertaken by Blanc and Rutenburg [18] and To illustrate, Fig. 3 presents the estimates of the me- Gage [20]. dians obtained from women aged 25–29, 30–34, 35–39 Figure 3 also plots the time series of the period mean and 40–44 from six surveys in Kenya. Time series of age at birth, M(t) as a solid line based on five surveys be- medians are plotted as the thin lines, with one line for tween 1989 and 2008/9. (The points in this line are plot- each of the age groups of women. Each data point is ted 1.5 years before the survey date to account for the plotted in the year in which a given cohort reaches its fact that the mean is based on births in a three year median age. For example, if women aged 30 to 34 re- period before the survey.) The period mean shows a rise ported a median age at first birth of 20 years in a survey between the 1989 and 1998 surveys but remains flat from 1998 to 2008/9. conducted during 2010 then this data point is plotted at 1998.0 years. This assumes that women aged 30 to 34 The period means and cohort medians are not directly are on average 32.5 years old and with a median age at comparable because they are different metrics of differ- ent distributions, but by plotting the data in comparable first birth of 20 years, their first birth occurred 12.5 years before the survey (i.e., age at survey – median age at first years (as discussed above) some tentative conclusions birth = 32.5–20 = 12.5). The reference date to which the can be reached. In particular, the medians reported by women aged 25–29 are lower than the means. This pat- median age at first birth applies is therefore 12.5 years be- fore the survey date (i.e., reference date of survey – time tern is as expected because the distribution of first births before the survey to which the median age at first birth is skewed to higher ages. Comparisons of period means Average age (years) Average age (years) Bongaarts and Blanc Population Health Metrics (2015) 13:25 Page 5 of 6 and cohort medians in other countries yield broadly birth history as well as reporting of their own birth dates similar results (data not shown). by women. It should be emphasized that survey data and any mea- We assessed two methods to estimate the period mean sures derived from them are subject to various reporting age at first birth. The first method is the same as the and non-reporting errors including omission of births, one used in populations with accurate vital statistics, displacement of births in time, and variations in sample and the second is the singulate mean age at first birth. A selection and implementation [18, 21–24]. In particular, comparison of the two estimates obtained from 62 DHS misreporting of the date of recent births has implications surveys shows excellent agreement in countries in which for assessing levels and trends in fertility. As shown by there is no evidence of an increase in childlessness. But, as Schoumaker [24], in a number of countries with DHS expected on theoretical grounds, there is less agreement surveys such errors are non-trivial and lead to underesti- in populations that have experienced a rise in the propor- mation of total fertility rates (TFR). Given that the calcula- tion childless. We therefore prefer the first method. The tion of M(t) is based on recent births, the known biases in measure is readily calculated as a straightforward variant the reporting of distant first births by older women are of the standard procedure used by DHS to estimate period likely to be minimized. Interestingly, our estimate of M(t) fertility rates and its reference period (the three years prior remains unaffected if errors are proportionally the same at to the survey) is the same as the published total fertility all ages. The reason is that age-specific birth rates b(,a,t) rates. In addition, it refers to recent births and is, there- appear in the numerator and the denominator of Equation fore, presumably more accurately reported than indicators 1. An error of say 10 % in all b(,a,t) values would lead to based on events that occurred in the distant past. Since an error of 10 % in the TFR, but there would be no error this new measure makes it possible for the first time to as- in M(t). In reality errors in b(,a,t) are likely to vary some- sess recent trends in the onset of childbearing in develop- what by age and that would lead to a bias in M(t). Further- ing countries with multiple DHS surveys and to compare more, errors in birth histories would not affect M*(t), recent period estimates of the mean age at first birth unless women misreport their childlessness status at the among countries, we suggest that it be considered for in- time of the survey. clusion in published DHS reports. In addition to reporting errors in the birth history, the mean age at first birth estimates could be biased by Endnotes women’s misreports of their own date of birth, especially DHS surveys do not provide estimates of birth rates if the misreporting is linked to fertility. If, for example, a or proportions ever having a birth for women under age woman who has begun childbearing early overstates her 15. However, in the average survey 2.3 % of 15 year olds age due to negative social norms around early childbear- have ever given birth and very small proportions of all ing or if an interviewer estimates her age based on her births therefore occur below age 15. These are estimated childbearing status (in places where knowledge of birth as follows: The proportion ever having a birth at age 14 dates is uncommon), then the mean age at birth would is assumed to be one third of the proportion at age 15. be overestimated. The completeness and accuracy of The proportion ever having a birth at age 13 is assumed birth date reporting, of both women and their children, to be one third the proportion at age 14, etc. Age- is likely to have improved over time, a factor that should specific birth rates under age 15 are calculated directly be kept in mind when assessing trends. from these proportions. Furthermore, the two means are not exactly compar- able because the first method estimates the mean for the Conclusion three years before the survey and the second method esti- The timing of the onset of parenthood is a key indicator mates the mean at the time of the survey. As a result, the used in studies of the determinants and consequences of timing of the means is about 18 months apart. This im- early childbearing as well as an indicator of the success of plies that when childbearing is being postponed, the various programmatic interventions. Annual estimates of first mean is slightly lower than the second. For ex- the period mean age at first birth from vital statistics are ample if the mean is rising at a rate of 1 year per dec- widely available in most developed countries. In contrast, ade (i.e., 0.1 per year) then the two means will differ by vital statistics of high quality are lacking in the large ma- 0.15 years. jority of developing countries and sample surveys such as the DHS are the primary source of demographic and Additional file health indicators. The published indicators from these Additional file 1: Estimating the mean age at first birth. Includes data include the retrospectively reported median but not two equations that provide alternative estimates of the age-standardized the period mean age at first birth. Both medians and mean age at first birth [25]. (DOCX 17 kb) means are dependent on the quality of reporting in the Bongaarts and Blanc Population Health Metrics (2015) 13:25 Page 6 of 6 Abbreviations 19. Feeney G. The population census as a time machine, Demography – DHS: Demographic and health surveys; TFR: Total fertility rates. Statistics – Information Technology Letter, Number 4, 15 January 2014, available online (last accessed 27 February 2015): http://demographer.com/ dsitl/04-population-census-as-time-machine. Competing interests 20. Gage A. An assessment of the quality of data on age at first union, first The authors declare that they have no competing interests. birth, and first sexual intercourse for Phase II of the Demographic and Health Surveys Program. Occasional Papers No. 4. Calverton, Maryland: Authors’ contributions Macro International Inc; 1995. JB carried out the statistical analysis and drafted the manuscript. AB helped 21. Arnold F. Assessment of the quality of birth history data in the to interpret the results and draft the manuscript. Both authors read, reviewed demographic and health surveys. In: An Assessment of DHS-I Data Quality. and approved the final manuscript. Methodological Reports no. 1. Maryland: Institute for Resource Development/Macro Systems, Inc; 1990. p. 83–111. 22. Hertrich V, Lardoux S. Estimating age at first union in Africa. Are census and Acknowledgements survey data comparable? Population–E. 2014;69(3):357–89. This research was supported by a grant from the William and Flora Hewlett 23. Pullum T. An assessment of the age and date reporting in the DHS Foundation to the Population Council. The authors are grateful to Katharine surveys,1985-2003, DHS Methodological Reports 5. Calverton: Maryland McCarthy for data analysis assistance. Macro International Inc; 2006. 24. Schoumaker B. Quality and Consistency of DHS Fertility Estimates, 1990 to Received: 14 January 2015 Accepted: 8 September 2015 2012. Methodological Reports No. 12. Rockville, Maryland: ICF International, DHS; 2014. 25. Bongaarts J, Feeney G. Estimating mean lifetime. Proc Natl Acad Sci. References 2003;100(23):13127–33. 1. National Research Council and Institute of Medicine. Growing up Global. The Changing Transitions to Adulthood in Developing Countries. Panel on Transition to Adulthood in Developing Countries. In: Cynthia L, editor. Committee on Population and Board on Children, Youth, and Families. Division of Behavioral and Social Sciences and Education. Washington, DC: The National Academies Press; 2005. 2. Blanc AK, Winfrey W, Ross J. New findings for maternal mortality age patterns: aggregated results for 38 countries. PLoS One. 2013;8(4):e59864. 3. Nations U. Adolescent Fertility since the International Conference on Population and Development (ICPD) in Cairo. New York: United Nations Population Division; 2013. 4. Dixon-Mueller R. How young is “too young? Comparative perspectives on adolescent sexual, marital, and reproductive transitions. Stud Fam Plan. 2008;39(4):247–62. 5. Gupta N, Mahy M. Adolescent childbearing in sub-Saharan Africa: Can increased schooling alone raise ages at first birth? Demogr Res. 2003;8(4):93–106. 6. DFID. A new strategic vision for girls and women: stopping poverty before it starts. London: UK Department for International Development; 2011. 7. Ki-Moon B. Global Strategy for Women’s and Children’s Health. New York: United Nations; 2010. 8. Bongaarts J. Population policy options in the developing world. Science. 1994;263(5148):771–6. 9. EUROSTAT. http://appsso.eurostat.ec.europa.eu/nui/submitViewTableAction .do;jsessionid6=rYMZ9wi2xbfmambPuCVcQOk3W6yPPH26neBsnUxmIVgQU UW3K9wQ!1673912419. Accessed March 15 2015. 10. Human Fertility Database. Max Planck Institute for Demographic Research (Germany) and Vienna Institute of Demography (Austria). 2014. http://www.humanfertility.org. Accessed 12 January 2015. 11. Hajnal J. Age at marriage and proportions marrying. Popul Stud. 1953;7(2):111–36. 12. United Nations. World Marriage Data 2012. In: United Nations, Department of Economic and Social Affairs, Population Division. 2013. http://www.un.org/en/development/desa/population/publications/dataset/ marriage/wmd2012/MainFrame.html. Accessed 12 January 2015. 13. Casterline JB, Trussell J. Age at first birth. WFS Comparative Studies no. 15. Submit your next manuscript to BioMed Central Voorburg, Netherlands: ISI; 1980. 14. Afzal M, Kiani MF. Mean ages at parities: an indirect estimation. Pak Dev Rev. and take full advantage of: 1995;34(4 Pt. II):545–61. 15. Booth H. Trends in mean age at first birth and first birth intervals in the • Convenient online submission Pacific Islands. Genus. 2001;LVII(3–4):165–90. • Thorough peer review 16. ICF International. The Demographic and Health Surveys. http://dhsprogram.com/. Accessed 12 January 2015. • No space constraints or color figure charges 17. Rutstein S, Rojas G. Guide to DHS statistics. Demographic and Health • Immediate publication on acceptance Surveys. Calverton, MD: ORC Macro; 2006. • Inclusion in PubMed, CAS, Scopus and Google Scholar 18. Blanc A, Rutenberg N. Assessment of the quality of data on age at first sexual intercourse, age at first marriage and age at first birth in the • Research which is freely available for redistribution Demographic and Health Surveys. In: An Assessment of DHS-I Data Quality. DHS Methodological Reports. Columbia, MD: Institute of Resource Submit your manuscript at Development/Macro System; 1990. p. 41–79. www.biomedcentral.com/submit http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Population Health Metrics Springer Journals

Estimating the current mean age of mothers at the birth of their first child from household surveys

Bongaarts, John; Blanc, Ann
Population Health Metrics , Volume 13 (1) – Sep 14, 2015
Download PDF
6 pages

Loading next page...
 
/lp/springer-journals/estimating-the-current-mean-age-of-mothers-at-the-birth-of-their-first-i1rjPVkuS0

References (44)

Publisher
Springer Journals
Copyright
Copyright © 2015 by Bongaarts and Blanc.
Subject
Medicine & Public Health; Public Health; Epidemiology
eISSN
1478-7954
DOI
10.1186/s12963-015-0058-9
pmid
26379478
Publisher site
See Article on Publisher Site

Abstract

Background: Estimates of the period mean age at first birth are readily available for countries with accurate vital statistics (i.e., in much of the developed world). In contrast, in most developing countries vital statistics are lacking or incomplete and estimates of the period mean age at first birth are therefore often unavailable. The Demographic and Health Surveys (DHS) program provides a large set of demographic and health statistics for many developing countries, but not the mean age at childbearing or the mean age at first birth. Methods: We propose two different methods for the estimation of the period mean age at first birth from information collected in DHS surveys. The first method is the same as the one used in populations with accurate vital statistics and is based on a weighted average of single year of age first birth rates. The second is the singulate mean age at first birth. Results: A comparison of the two estimates obtained from the latest surveys in 62 countries shows excellent agreement in countries in which there is no evidence of a rise in childlessness. But, as expected on theoretical grounds, there is less agreement in populations that have experienced an increase in the proportion childless. Conclusions: Based on these results, we recommend the first method. The measure is relatively straightforward to calculate and, since it refers to recent births, is presumably more accurately reported than indicators based on events that occurred in the more distant past. This measure makes it possible for the first time to assess recent trends in the onset of childbearing in developing countries with multiple DHS surveys and to compare recent period estimates of the mean age at first birth among countries. Background to delay childbearing and increase access to family Becoming a parent for the first time is one of life’s most planning [6, 7]. Thus, the age at which women have a important and influential events. It signals the onset of first birth is an important indicator of the success of the responsibility for insuring the well-being and success these efforts. Finally, delayed childbearing slows popu- of one’s offspring and of the next generation. For lation growth through increasing the length between women, the age at which they have a first birth can have generations and decreasing population momentum [8]. implications for schooling, labor force participation, and Estimates of both cohort and period mean ages at first overall family size [1]. Early childbearing is also associ- birth are available for countries with reliable vital statistics. ated with elevated risks to the health of the mother and For example, EUROSTAT [9] and the Human Fertility her child [2]. As a consequence, there is a large literature Database [10] provide historical estimates for many coun- on the individual, social, and cultural determinants and tries in Europe and other high-income countries for single consequences of this event and its timing in the life cycle years from the 1980s to around 2010 and for a substantial [3–5]. In addition, a renewed interest in the wellbeing of number of birth cohorts. In contrast, in most developing adolescent girls has led to investmentsinprogramsintended countries vital statistics are lacking or incomplete and esti- mates of period and cohort mean ages at first birth are therefore often unavailable. The Demographic and Health * Correspondence: [email protected] Surveys (DHS) program – under which nationally Population Council, 1 Dag Hammarskjold Plaza, New York, NY 10017, USA © 2015 Bongaarts and Blanc. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Bongaarts and Blanc Population Health Metrics (2015) 13:25 Page 2 of 6 representative household surveys are conducted in devel- An alternative approach to estimating the period mean oping countries – provides many valuable statistics on age at first birth is to rely on a method that is widely demographic and health processes, but does not report on used to estimate the mean age at first marriage, called the period age at childbearing or age at first birth (mean the “singulate mean age at marriage” [11, 12]. The appli- or median). Instead, the standard reports provide the co- cation of this approach to estimate the mean age at first hort median age at first birth as calculated from a birth birth was first mentioned by Casterline and Trussell [13] history reported retrospectively by women of reproductive and subsequently implemented by Afzal and Kiani [14] age. In principle, the DHS could also report the mean age and Booth [15]. The equation is as follows at childbearing for cohorts of women but such means max paðÞ ; t −paðÞ ; t a would be biased downward because of incomplete child- max max M ðÞ t ¼ ð2Þ bearing experience of all but the oldest women. 1−paðÞ ; t max For many analytic purposes estimates of period mea- sures are of greatest interest because, in contrast to co- M*(t) = Average age at first birth at time t hort medians, they allow assessments of recent trends in p(a,t) = Proportion of women that has not yet given the timing of the onset of childbearing for specific refer- birth at age a and time t ence periods. The objective of this research note is to p = The proportion of women that has never had a max propose two different methods for the estimation of birth at a max mean age at first birth from information collected in This mean age at first birth is defined as the mean age DHS and similar surveys. Both measures are unaffected at which women would bear their first child if they went by changes in the population age structure, thus allow- through the reproductive years experiencing the age- ing undistorted comparisons of the timing of the onset specific proportions childless observed at time t. of childbearing between populations and over time In Additional file 1 we demonstrate that the two within populations. Estimates are calculated for the most means are equal (i.e., M (t)= M* (t)) under the condition recent surveys in 62 countries. that the shape of the function p(a,t) by age is invariant with respect to time. This implies that p(a,t) can shift to Methods high or lower ages over time (with corresponding The equation for estimating the period mean age at first changes in first birth rates and in the mean age) but with birth used widely in countries with vital statistics [10] is no change in shape and with constant p . max Estimates of M(t) and M*(t) were obtained with equa- max tions (1) and (2) for the most recent DHS surveys in 62 ðÞ a þ 0:5 baðÞ ; t MtðÞ ¼ ð1Þ developing countries for which data files are available max baðÞ ; t for public use (and with sample sizes of married women above 3000). The number of respondents in each survey where varies but typically is between 5000 and 10,000 women M(t) = Average age at first birth at time t of reproductive age. For many countries several surveys b(a,t) = the age-specific birth rate for birth order one are available, so time series of M(t) and M*(t) can also at (single) age a and time t. be calculated. Further details about the surveys are avail- a = the highest age at which first births are observed able on the DHS website [16]. max This period mean age at first birth is defined as the Estimates of b(a,t) are obtained from birth histories with mean age at which women would bear their first child if a simple variant of the standard DHS method for calculat- they went through the reproductive years having the first ing age-specific birth rates by age for the three years be- birth rates observed in a particular period. fore the survey [17]. To estimate b(a,t) two changes are Numbers of births recorded in vital statistics are typic- made in this method: (1) birth rates are calculated by sin- ally large and birth rates are available by single age and gleyearrather than byfive year age intervalsand (2)the single year. As a result, annual estimates of M(t) can be numerators of the birth rates exclude births of order two estimated. and higher. Estimates of p(a,t) are also calculated with a In contrast, in applications of this equation to DHS variant of the standard DHS method estimating the pro- surveys samples of births in a single year are relatively portion nulliparous by single year of age rather than five small. To obtain more robust estimates of the mean age year age intervals. at first birth for a survey, we calculate b(,a,t) by single Finally, it should be noted that values of p(a,t) are sub- year of age for a period of three years before each sur- ject to substantial sampling errors at ages above 40, be- vey. In addition, we exclude surveys with sample sizes of cause the proportions childless at these ages are usually currently married women below 3000 to minimize sam- less than five percent and the number of respondents is pling errors. smaller than at lower ages. To minimize the effects of Bongaarts and Blanc Population Health Metrics (2015) 13:25 Page 3 of 6 these errors on estimates of the mean age at firth birth, and the solid markers are spread around the diagonal in the value of a is set at 40 years and p is estimated Fig. 1 with a standard error of 0.32 years. The second max max as the average of single age values of p(a,t) between ages cluster of countries with open circles includes several 35 and 45. surveys in which M*(t) is usually substantially higher than M(t). This finding is likely attributable to an up- Results ward bias in M*(t) when values of p are rising (the max Figure 1 plots the estimates of M(t) on the horizontal rare cases in this cluster with M(t) higher than M*(t) are axis and the value of M*(t) on the vertical axis. Each probably attributable to measurement or reporting er- marker represents the most recent survey in each of the rors). Our working assumption therefore is that M(t) is 62 countries. The results are presented in two clusters: an unbiased estimator of the mean age at first birth even the solid markers represent surveys in which p is less in surveys in the second cluster. In addition, all except max than 5 % and the open circles represent surveys with one of the countries in the second cluster have a mean p >5 %. This distinction is made to separate observa- age at birth of 22 or higher. This result is not unex- max tions in which the conditions are met for M(t) to be pected as there tends to be a positive correlation be- equal to M*(t) from observations in which they are not. tween age at first birth and the proportion of women As noted in Additional file 1, a key condition for the who remain childless. equality of M(t) and M*(t) is that p is constant. Un- A full analysis of levels and trends in all 62 countries is max fortunately, it is not easy to determine the rate of change beyond the scope of this methodological study, but a few in p , because some countries have only one survey findings can be noted. Estimates of M(t) vary widely max and, even in countries with multiple surveys, the rate of among countries from a low of 19.1 in Niger (2006) to a change in p is erratic due to small sample sizes. In- high of 24.7 in the Maldives (2009). The unweighted aver- max stead, we assume that countries with p less than 5 % ages of M(t) for countries in each of four regions are pre- max have seen little change in p over time, thus approxi- sented in Table 1. The low value for sub-Saharan Africa is max mating the condition that p is constant. In surveys unsurprising since this continent has not progressed as far max where p is higher than 5 % there has likely been through the fertility transition as the other regions. North max change over time because early in the fertility transition Africa/West Asia and South Asia have the highest aver- p is typically a very small number. ages and Latin America has intermediate values. max It is therefore expected that M(t) is closer to M*(t) for Figure 2 presents trends in M(t) for selected countries surveys in the cluster with p <5 %. As is evident from in the developing and developed world. Estimates for max Fig. 1, this is indeed the case. For these surveys the aver- Egypt, Nigeria, India, Kenya, and Bangladesh show very age value of M(t) and M*(t) are respectively 21.0 and modest increases from the 1990s to near 2010. The 21.2, a difference of only 0.2 year (which is not statisti- mean ages at first birth for the Japan, Czech Republic, cally significant). However, the agreement is not perfect UK, and US are mostly substantially higher and have been rising at a more rapid pace than in the five devel- oping countries included in the figure. As noted, DHS published reports provide estimates of the retrospectively reported cohort median age at first birth. These medians are estimated from birth histories y = 1.0092x R² = 0.9352 obtained from respondents of reproductive age. The age at first birth is calculated by subtracting the woman’s date of birth from the date of birth of her first child. Medians for the cohorts aged 25–29 at the time of the survey and above are available for nearly all DHS surveys because the medians are reached before age 25 22 Pmax<0.05 (i.e., at least half of women have had a birth before age 25). 21 Pmax>0.05 Table 1 Average and standard deviation of country estimates of M(t) by region Average of M(t) Standard deviation N Sub-Saharan Africa 20.9 1.1 33 Latin America 21.7 0.9 11 18 19 20 21 22 23 24 25 26 Mean age M (years) South Asia 22.7 1.6 10 Fig. 1 Period mean age at first birth (M* vs M) North Africa/West Asia 23.3 1.1 8 Mean age M* (years) Bongaarts and Blanc Population Health Metrics (2015) 13:25 Page 4 of 6 21.0 Mean (period) 20.5 Japan UK Czech. Rep 20.0 US Egypt Nigeria 19.5 Medians (cohort) India Kenya Bangladesh 19.0 40-44 35-39 30-34 25-29 (age at time of survey) 18.5 18.0 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 Fig. 2 Period mean age at first birth for selected developing and Fig. 3 Period mean and cohort median age at first birth, based on developed countries five surveys in Kenya For a small number of surveys medians are available for refers = 2010.5–12.5 = 1998.0). This approach allows the the cohorts aged 20–24 when the median is below age 20. comparison of cohort medians reported in different sur- These cohort medians have the advantage of being veys and of cohort and period means [18, 19]. With fully available for all DHS surveys but there are also draw- accurate reporting of the timing of first births the lines of backs: 1) the median refers to past experience of cohorts medians plotted in Fig. 3 would exactly overlap (assuming and is therefore not as current as is preferable for many no selectivity of migration and mortality). For example, analytic purposes; 2) the retrospective reporting of the women aged 35–39 should report a median that is the date of the first birth may suffer from recall errors that same as the median reported by women aged 25–29 in a are likely to increase as the time since the event rises; survey conducted ten years earlier. The fact that the lines and 3) the cohort median as calculated by DHS is not do not match indicates misreporting. In particular, it independent of the quantum of first births and can seems that the older cohorts have moved the time of change over time even if the mean is constant. The first the first birth closer to the survey date so that their re- two of these disadvantages also apply to cohort mean ported medians are higher for most years than the me- ages at first birth, a measure we do not discuss because dians reported by younger cohorts for the same years. it is very rarely used as it can only be estimated accur- This pattern is consistent with earlier analyses of data ately for women who have completed their childbearing. quality undertaken by Blanc and Rutenburg [18] and To illustrate, Fig. 3 presents the estimates of the me- Gage [20]. dians obtained from women aged 25–29, 30–34, 35–39 Figure 3 also plots the time series of the period mean and 40–44 from six surveys in Kenya. Time series of age at birth, M(t) as a solid line based on five surveys be- medians are plotted as the thin lines, with one line for tween 1989 and 2008/9. (The points in this line are plot- each of the age groups of women. Each data point is ted 1.5 years before the survey date to account for the plotted in the year in which a given cohort reaches its fact that the mean is based on births in a three year median age. For example, if women aged 30 to 34 re- period before the survey.) The period mean shows a rise ported a median age at first birth of 20 years in a survey between the 1989 and 1998 surveys but remains flat from 1998 to 2008/9. conducted during 2010 then this data point is plotted at 1998.0 years. This assumes that women aged 30 to 34 The period means and cohort medians are not directly are on average 32.5 years old and with a median age at comparable because they are different metrics of differ- ent distributions, but by plotting the data in comparable first birth of 20 years, their first birth occurred 12.5 years before the survey (i.e., age at survey – median age at first years (as discussed above) some tentative conclusions birth = 32.5–20 = 12.5). The reference date to which the can be reached. In particular, the medians reported by women aged 25–29 are lower than the means. This pat- median age at first birth applies is therefore 12.5 years be- fore the survey date (i.e., reference date of survey – time tern is as expected because the distribution of first births before the survey to which the median age at first birth is skewed to higher ages. Comparisons of period means Average age (years) Average age (years) Bongaarts and Blanc Population Health Metrics (2015) 13:25 Page 5 of 6 and cohort medians in other countries yield broadly birth history as well as reporting of their own birth dates similar results (data not shown). by women. It should be emphasized that survey data and any mea- We assessed two methods to estimate the period mean sures derived from them are subject to various reporting age at first birth. The first method is the same as the and non-reporting errors including omission of births, one used in populations with accurate vital statistics, displacement of births in time, and variations in sample and the second is the singulate mean age at first birth. A selection and implementation [18, 21–24]. In particular, comparison of the two estimates obtained from 62 DHS misreporting of the date of recent births has implications surveys shows excellent agreement in countries in which for assessing levels and trends in fertility. As shown by there is no evidence of an increase in childlessness. But, as Schoumaker [24], in a number of countries with DHS expected on theoretical grounds, there is less agreement surveys such errors are non-trivial and lead to underesti- in populations that have experienced a rise in the propor- mation of total fertility rates (TFR). Given that the calcula- tion childless. We therefore prefer the first method. The tion of M(t) is based on recent births, the known biases in measure is readily calculated as a straightforward variant the reporting of distant first births by older women are of the standard procedure used by DHS to estimate period likely to be minimized. Interestingly, our estimate of M(t) fertility rates and its reference period (the three years prior remains unaffected if errors are proportionally the same at to the survey) is the same as the published total fertility all ages. The reason is that age-specific birth rates b(,a,t) rates. In addition, it refers to recent births and is, there- appear in the numerator and the denominator of Equation fore, presumably more accurately reported than indicators 1. An error of say 10 % in all b(,a,t) values would lead to based on events that occurred in the distant past. Since an error of 10 % in the TFR, but there would be no error this new measure makes it possible for the first time to as- in M(t). In reality errors in b(,a,t) are likely to vary some- sess recent trends in the onset of childbearing in develop- what by age and that would lead to a bias in M(t). Further- ing countries with multiple DHS surveys and to compare more, errors in birth histories would not affect M*(t), recent period estimates of the mean age at first birth unless women misreport their childlessness status at the among countries, we suggest that it be considered for in- time of the survey. clusion in published DHS reports. In addition to reporting errors in the birth history, the mean age at first birth estimates could be biased by Endnotes women’s misreports of their own date of birth, especially DHS surveys do not provide estimates of birth rates if the misreporting is linked to fertility. If, for example, a or proportions ever having a birth for women under age woman who has begun childbearing early overstates her 15. However, in the average survey 2.3 % of 15 year olds age due to negative social norms around early childbear- have ever given birth and very small proportions of all ing or if an interviewer estimates her age based on her births therefore occur below age 15. These are estimated childbearing status (in places where knowledge of birth as follows: The proportion ever having a birth at age 14 dates is uncommon), then the mean age at birth would is assumed to be one third of the proportion at age 15. be overestimated. The completeness and accuracy of The proportion ever having a birth at age 13 is assumed birth date reporting, of both women and their children, to be one third the proportion at age 14, etc. Age- is likely to have improved over time, a factor that should specific birth rates under age 15 are calculated directly be kept in mind when assessing trends. from these proportions. Furthermore, the two means are not exactly compar- able because the first method estimates the mean for the Conclusion three years before the survey and the second method esti- The timing of the onset of parenthood is a key indicator mates the mean at the time of the survey. As a result, the used in studies of the determinants and consequences of timing of the means is about 18 months apart. This im- early childbearing as well as an indicator of the success of plies that when childbearing is being postponed, the various programmatic interventions. Annual estimates of first mean is slightly lower than the second. For ex- the period mean age at first birth from vital statistics are ample if the mean is rising at a rate of 1 year per dec- widely available in most developed countries. In contrast, ade (i.e., 0.1 per year) then the two means will differ by vital statistics of high quality are lacking in the large ma- 0.15 years. jority of developing countries and sample surveys such as the DHS are the primary source of demographic and Additional file health indicators. The published indicators from these Additional file 1: Estimating the mean age at first birth. Includes data include the retrospectively reported median but not two equations that provide alternative estimates of the age-standardized the period mean age at first birth. Both medians and mean age at first birth [25]. (DOCX 17 kb) means are dependent on the quality of reporting in the Bongaarts and Blanc Population Health Metrics (2015) 13:25 Page 6 of 6 Abbreviations 19. Feeney G. The population census as a time machine, Demography – DHS: Demographic and health surveys; TFR: Total fertility rates. Statistics – Information Technology Letter, Number 4, 15 January 2014, available online (last accessed 27 February 2015): http://demographer.com/ dsitl/04-population-census-as-time-machine. Competing interests 20. Gage A. An assessment of the quality of data on age at first union, first The authors declare that they have no competing interests. birth, and first sexual intercourse for Phase II of the Demographic and Health Surveys Program. Occasional Papers No. 4. Calverton, Maryland: Authors’ contributions Macro International Inc; 1995. JB carried out the statistical analysis and drafted the manuscript. AB helped 21. Arnold F. Assessment of the quality of birth history data in the to interpret the results and draft the manuscript. Both authors read, reviewed demographic and health surveys. In: An Assessment of DHS-I Data Quality. and approved the final manuscript. Methodological Reports no. 1. Maryland: Institute for Resource Development/Macro Systems, Inc; 1990. p. 83–111. 22. Hertrich V, Lardoux S. Estimating age at first union in Africa. Are census and Acknowledgements survey data comparable? Population–E. 2014;69(3):357–89. This research was supported by a grant from the William and Flora Hewlett 23. Pullum T. An assessment of the age and date reporting in the DHS Foundation to the Population Council. The authors are grateful to Katharine surveys,1985-2003, DHS Methodological Reports 5. Calverton: Maryland McCarthy for data analysis assistance. Macro International Inc; 2006. 24. Schoumaker B. Quality and Consistency of DHS Fertility Estimates, 1990 to Received: 14 January 2015 Accepted: 8 September 2015 2012. Methodological Reports No. 12. Rockville, Maryland: ICF International, DHS; 2014. 25. Bongaarts J, Feeney G. Estimating mean lifetime. Proc Natl Acad Sci. References 2003;100(23):13127–33. 1. National Research Council and Institute of Medicine. Growing up Global. The Changing Transitions to Adulthood in Developing Countries. Panel on Transition to Adulthood in Developing Countries. In: Cynthia L, editor. Committee on Population and Board on Children, Youth, and Families. Division of Behavioral and Social Sciences and Education. Washington, DC: The National Academies Press; 2005. 2. Blanc AK, Winfrey W, Ross J. New findings for maternal mortality age patterns: aggregated results for 38 countries. PLoS One. 2013;8(4):e59864. 3. Nations U. Adolescent Fertility since the International Conference on Population and Development (ICPD) in Cairo. New York: United Nations Population Division; 2013. 4. Dixon-Mueller R. How young is “too young? Comparative perspectives on adolescent sexual, marital, and reproductive transitions. Stud Fam Plan. 2008;39(4):247–62. 5. Gupta N, Mahy M. Adolescent childbearing in sub-Saharan Africa: Can increased schooling alone raise ages at first birth? Demogr Res. 2003;8(4):93–106. 6. DFID. A new strategic vision for girls and women: stopping poverty before it starts. London: UK Department for International Development; 2011. 7. Ki-Moon B. Global Strategy for Women’s and Children’s Health. New York: United Nations; 2010. 8. Bongaarts J. Population policy options in the developing world. Science. 1994;263(5148):771–6. 9. EUROSTAT. http://appsso.eurostat.ec.europa.eu/nui/submitViewTableAction .do;jsessionid6=rYMZ9wi2xbfmambPuCVcQOk3W6yPPH26neBsnUxmIVgQU UW3K9wQ!1673912419. Accessed March 15 2015. 10. Human Fertility Database. Max Planck Institute for Demographic Research (Germany) and Vienna Institute of Demography (Austria). 2014. http://www.humanfertility.org. Accessed 12 January 2015. 11. Hajnal J. Age at marriage and proportions marrying. Popul Stud. 1953;7(2):111–36. 12. United Nations. World Marriage Data 2012. In: United Nations, Department of Economic and Social Affairs, Population Division. 2013. http://www.un.org/en/development/desa/population/publications/dataset/ marriage/wmd2012/MainFrame.html. Accessed 12 January 2015. 13. Casterline JB, Trussell J. Age at first birth. WFS Comparative Studies no. 15. Submit your next manuscript to BioMed Central Voorburg, Netherlands: ISI; 1980. 14. Afzal M, Kiani MF. Mean ages at parities: an indirect estimation. Pak Dev Rev. and take full advantage of: 1995;34(4 Pt. II):545–61. 15. Booth H. Trends in mean age at first birth and first birth intervals in the • Convenient online submission Pacific Islands. Genus. 2001;LVII(3–4):165–90. • Thorough peer review 16. ICF International. The Demographic and Health Surveys. http://dhsprogram.com/. Accessed 12 January 2015. • No space constraints or color figure charges 17. Rutstein S, Rojas G. Guide to DHS statistics. Demographic and Health • Immediate publication on acceptance Surveys. Calverton, MD: ORC Macro; 2006. • Inclusion in PubMed, CAS, Scopus and Google Scholar 18. Blanc A, Rutenberg N. Assessment of the quality of data on age at first sexual intercourse, age at first marriage and age at first birth in the • Research which is freely available for redistribution Demographic and Health Surveys. In: An Assessment of DHS-I Data Quality. DHS Methodological Reports. Columbia, MD: Institute of Resource Submit your manuscript at Development/Macro System; 1990. p. 41–79. www.biomedcentral.com/submit

Journal

Population Health Metrics – Springer Journals

Published: Sep 14, 2015

There are no references for this article.