Is Swaziland census data suitable for fertility measurement?
© The Author(s) 2016
Received: 11 September 2015
Accepted: 7 January 2016
Published: 14 July 2016
This study seeks to assess fertility data in Swaziland based on the 1976, 1986, 1997 and 2007 censuses. By utilising single-year age-sex published raw data, demographic evaluation tools—sex ratio, age ratio, Whipple’s index, and modified Whipple’s index—were used to assess age misreporting as several fertility measures rely on the quality of age data. In addition, using published descriptive census tables for women in the reproductive lifespan, 15–49, data on children ever born or parity (P) derived were evaluated for incorrect reporting of parities using the el-Badry technique. Further, the relational Gompertz model was applied to adjust data on reported aggregated births in the last year or current fertility (F) relying on its intrinsic P/F ratios feature as an adjustment and diagnostic tool for consistency checks on fertility and parity distributions. The evidence of some age reporting distortions or age misreporting is not too severe, but moderate, and therefore fertility estimates in the four censuses of Swaziland cannot be said are invalidated. The data on parity and current fertility conforms to expected or typical patterns of fertility distributions of African populations. The study concludes that the census data were of reasonable quality for fertility estimation.
The efforts in evaluating and monitoring the past 2015 Millennium Development Goals (MDGs) has been so crucial for most African countries towards achieving set targets. The post-MDGs era now entails the realisation of the need of reliable data so crucial for developmental planning.
The article examines the importance of evaluating fertility data prior to estimation as utilising poor quality data leads to deriving implausible estimates of fertility. This in turn affects the decision-making process in development and policy planning which results in designing inappropriate programme interventions. The paper focuses on evaluating age and sex data and fertility data on children ever born (parity) and recent births reported for past years in censuses. Recent births reflect current fertility whilst parity data provide lifetime fertility measures.
Whilst the position on quality of demographic data in censuses also surveys has been contested and investigated using various demographic and statistical tools for some Southern African countries such as South Africa (Moultrie and Timæus 2002; Phillips 1999), Zimbabwe and Botswana (Blanc and Rutstein 1994; Thomas and Muvandi 1994), such scientific enquiry is missing for Swaziland in spite of regular census enumerations done. Thus, fertility data remains under-examined for Swaziland. Further, the in-depth analysis of Swaziland census data has been limited due to paucity of data. Previous actual raw data sets such as for 1976 and 1986 are not in the reach or easily accessible for public use. The articles seek to examine the suitability of Swaziland 1976, 1986, 1997 and 2007 census data for deriving plausible fertility estimates by bringing to the fore data quality issues when deriving fertility estimates. Fertility was about 7 children per woman in the 1960s (Cohen 1993) and was 4 at the 2006/7 Demographic and Health Survey (DHS) and 2007 census.
As a preliminary step, census data are evaluated on accuracy and quality of age reporting errors using classic demographic tools such as age ratios, sex ratios and Myers’, Whipple’s and joint score indices on pinpointing irregularities on age-sex composition (Brass 1996; Shryock et al. 1976; Spoorenberg 2007). Also employed are the diagnostic tools of the P/F ratios inherent in the relational Gompertz model and the el-Badry correction method to evaluate fertility data. The multiple census data sources allows for consistency checks on age-sex distributions and comparisons of fertility data (Joyner et al. 2012; Moultrie et al. 2013; United Nations 2004) amongst the four censuses of Swaziland. Therefore, the purpose is to demonstrate, but not necessarily quantify, the possible errors that may arise from estimating fertility using Swaziland census data. This study focuses on consistency checks using average parity, age and sex distributions and age pattern of fertility as basis for evaluating fertility data derived from the four censuses.
In contrary to less flawed demographic data for developed countries, for African countries in the past four decades, the data has been deemed as either defective or inadequate with respect to its poor quality (Cleland 1996; Cohen 1993, 1998; Brass 1996; Moultrie et al. 2013; United Nations 1983; Potter 1977; Hill 1990). Due to an incomplete vital registration system (VRS) in most developing countries, the census, therefore, remains as the reliable and main source of fertility as well as mortality measurement (United Nations 2004). Indeed a census is not an exceptional stand-alone or alternative data source to a VRS or survey, but it is indispensable and complementary for demographic measurement.
In deriving fertility estimates, the potential errors in census data are numerous or inescapable. The types, causes and effects of errors on estimates of fertility are highlighted in demographic literature (National Research Council 2004; Schoumaker 2014; Vergauwen et al. 2015; Potter 1977; Brass 1996; Joyner et al. 2012; Cleland 1996; Spoorenberg 2014; Phillips 1999; United Nations 1983, 2004; Hill 1990; Moultrie et al. 2013). This literature highlights common quality issues on age-sex and fertility data which relate to omission, duplication, misreporting, mistiming or misplacement of birth events or age, selection bias and non-responsiveness or response irregularities due to memory lapse or recall problems and/or negligence as well as other motives in reporting/recording. A distorted age-sex structure through bias and errors in age-sex data has implications on fertility estimates and their projections (Spoorenberg 2014). The United Nations (1983) notes that the effect of age misreporting on fertility is complex, but the overall effect tends to be small.
In overview, it is difficult to uncover, distinguish or correct all errors in data and hence the concern is on reducing and knowing their bias or overall effect (National Research Council 2004; Schoumaker 2014). Errors (of different types) which may have occurred on a simultaneous basis could lead to a similar effect on fertility (Schoumaker 2014) or an offset (National Research Council 2004; Shryock et al. 1976). Further, possible effects of errors or bias, in magnitude or direction, which Brass (1996) cautions their difficulty in proving, can be compensated by adjustments using appropriate methods of applications and assumptions in fertility estimation (Hill 1990; Moultrie et al. 2013). For example, information on parities (P) and current fertility (F) has led to utilisation of a resourceful demographic tool—lifetime/current fertility (P/F) ratio. This ratio is vital to evaluation of fertility data and estimation of fertility using many indirect methods derived on this principle such as the preferable relational Gompertz model (Moultrie et al. 2013). Thus, the United Nations (2004) and Moultrie et al. (2013) posit that though data may portray errors on some aspects, it may be possible to come up with robust levels and trends of fertility.
Data and methods
The data for this study are based on published raw and aggregated census tabulations in the 1976, 1986, 1997 and 2007 censuses undertaken by the Central Statistics Office (CSO) of Swaziland. Published raw data on population distribution by age and sex were used. Tabulated aggregated demographic data collected from census reports were fertility variables on recent births and children ever born. The question on children ever born was asked from women over some minimum age, 12 or 15 years in the censuses.
Consistent with modern census questioning on fertility, the four censuses in Swaziland are similar in content. Although raw data for 1976 and 1986 censuses are not accessible, published aggregate data on age, sex and fertility are credible for appraisal of census data quality.
With the available raw data, in the 2007 census, 20 % of the sample has missing “unknown” or “non-stated” parities needed for assessing and correcting parity distributions, whereas the full data set for 1997 census has adequate fertility variables. To maintain consistency on data use, single-year and aggregated data were derived from descriptive tables of published census reports for the total population. The methods used to evaluate age-sex and fertility data in the four censuses are explained in the next sections.
Quality of age-sex data
Classic demographic ratios and indices such as sex ratios, age ratios, Whipple’s index and Myers’ blended index are the several methods used for the appraisal of age-sex data for errors in age misreporting and digit preference or avoidance (Shryock et al. 1976; Moultrie et al. 2013; Spoorenberg 2007). The discussion on methods which follows draws extensively from these authors.
Both internal consistency (meeting typical patterns on age and or sex distribution) and external consistency (comparison of data across censuses from the same country) are vital in data assessment (Moultrie et al. 2013). Therefore, for both age and sex ratios, an age range from birth to older ages was assessed with the expectation that an oldest 49-year-old woman in the 1976 census would be around 80 years in the 2007 census.
The 23–62 age range for Whipple’s index was altered to 23–52 to represent women in the reproductive lifespan for each census data under the same linearity assumptions. The digits 0 and 5 in the numerator are distributed evenly or linearly over 5-year ages in the denominator, with the exception of ages lower than 23 and older ages greater than 62 (Shryock et al. 1976; Spoorenberg 2007) or 52 in the case of this study. A woman aged 49 probably would be able to respond to an age of 50 in a census due to a 0 digit preference. According to Spoorenberg (2007), Whipple’s index is that the extent of digit preference is meant for terminal digits 0 and 5. Age heaping at any age other than 0 or 5 may occur which can be detected using Myers’ blended index. The index can be applied to an age range of 10–89 for all terminal digits, 0–9 (Shryock et al. 1976).
Since a shorter age range for women aged 15–49 is considered in this context, a Whipple-type version of Myers’ index would be considered since it uses the same principles of Myers’ index. Spoorenberg (2007), following the work proposed in 1992 by Noumbissi of modifying Whipple’s index to detect age heaping at age 0 and 5 separately, extended the index to each terminal digit, i, for 0 to 9 and hence the term “digit-specific modified Whipple’s index” (W i ). Based on Spoorenberg (2007) algebraic expressions of W i , an age range of 21–52 was considered representing women in the reproductive lifespan for each census.
Quality of fertility data
Recent births and children ever born data for women aged 15–49 provide the basis for establishing a number of standard fertility measures and evaluating fertility data in censuses or surveys. Experience from African data on recent births suggests that though they are frequently underreported the emerging age pattern of fertility is usually fairly accurate (United Nations 2004).
Symptoms of omission on parity data according to the United Nations (1983) are average parities that fail to increase rapidly enough as age increases. In some specific cases, average parities for women aged 40–44 and 45–49 may actually fall below those for women aged 35–39 even when there is no reason to suppose that fertility has been increasing. Literature suggests that the shortcomings in the completeness of reporting of parities may be achieved by using the P/F ratio method of the relational Gompertz model for adjusting upwards the underreporting for births, after correcting for defective parities using the el-Badry technique (United Nations 1983, 2004; Moultrie et al. 2013).
Generally, with children ever born data, parity of childless women is often incorrectly recorded as if they had an “unknown” or a “not stated” parity. This scenario may increase average parities, especially for younger women and reduce the proportion of childless women. This becomes an issue of concern when proportion of parity in a “not stated” category is over 2 % for all women in the reproductive lifespan. In such cases, the el-Badry correction is employed to detect and correct distorted parities (Moultrie et al. 2013; Schoumaker 2014; United Nations 1983, 2004).
As prior mentioned, the relational Gompertz model is regarded as a robust and the most improved P/F method for preparing fertility estimates (Moultrie et al. 2013). The method provides an effective way of evaluating the extent of age and birth misreporting errors in census data as well as correcting or adjusting the fertility schedule accounting for the errors occurred. The spreadsheets “FE_Relational Gompertz” and “FE_elBadry_0” (Moultrie et al. 2013) were used to adjust age-specific fertility rates and parities.
A typical pattern for age-specific fertility rates on recent births from census data is characterised by a right-skewed concave shape showing lowest fertility rates for older ages nearing end of childbearing, lower fertility at the beginning of the reproductive lifespan for the youngest age group 15–19. In between, a peak of childbearing is observed especially prior to the age of 30 (Moultrie et al. 2013). Basing on a typical fertility curve, a number of demographic fertility models such as the Coale-Trussell and Gompertz models have been derived to fit fertility data (United Nations 1983).
Using the spreadsheet by Moultrie et al. (2013) for the relational Gompertz model, the inputs of reported age-specific fertility rates and suitably corrected average parities, using the el-Badry method, are used to generate a corrected fertility schedule which conforms to expected alpha (α) and beta (β) parameter ranges. The relational Gompertz model applies in the range − 0.3 < α < 0.3 and 0.8 < β < 1.25. The α and β measure age location and spread of fertility schedule, respectively.
The fittings of α and β were done in such as way that accurate reporting is observed when a corresponding set of F value and P value points strikingly linearly coincides on the same line. This also implies constant fertility when P/F ratio equals 1 (or is very close to unity in real data). Possible irregularities in data occur if the P/F value diverges from constant fertility. Alternatively, this implies a fertility trend—decline or increase in fertility (Brass 1996; Moultrie et al. 2013).
The consistency checks on fertility data with regard to age-sex distribution, average parities and age-specific fertility rate plausibility or expected fertility patterns are employed to assess the quality of fertility data in census data.
Whipple’s index (W) for women aged 23–52, and total modified Whipple’s index (W tot) for women aged 21–52, Swaziland, 1976–2007 PHC
With regard to sex ratio, for most populations, it is expected that male mortality is little in excess of that for females as shown in Fig. 3 for Swaziland. The sex ratios at births are slightly below 100 in all the censuses indicating the possibility of underreporting of births or ages. The estimated sex ratios at birth were at 95.7, 93.3, 97.5 and 98.0 for the 1976, 1986, 1997 and 2007 censuses, respectively. The sex ratios in the middle ages, especially for ages 20–35, are overly below 100 reflecting either or a combination of excess male mortality due to HIV/AIDS, high emigration of males for labour or lower sex ratio at birth.
Figure 4 tends to exhibit an irregular pattern of age ratios in all the censuses for women aged 30–49. This suggests misclassification of ages or probable undercounting for women in the middle to later childbearing years.
The results indicate a similar trend of peaked childbearing for women aged 20–29 in the 1976, 1986 and 1997 censuses, which is characteristic of most African countries. The 2007 census, however, illustrates a somewhat different shape of flattened peaked fertility distribution for ages 20–39. This pattern shows a similarity of fertility behaviour amongst the represented women, of which, the explanation needs further investigation. Possibly, underreporting of births may have occurred for young women aged 20–29, who normally expected to be at the peak of their childbearing. Otherwise, older women appear to have increased or just delayed childbearing.
Discussion and conclusion
The study aimed at appraisal of age-sex and fertility data as components of census data used for deriving reasonable fertility estimates. Multiple census data are employed to ascertain the level of consistency in estimates or patterns. In developing countries, age and sex data are much relied on in fertility measurement using indirect techniques of estimation. It is vital that the age-sex structure be reasonably complete and accurate to ensure quality on the resulting estimates of basic demographic indices.
According to Moultrie et al. (2013), the basis of such enquiry looks at dimensions on data of age-sex structure, current fertility and lifetime fertility (or average parity) patterns as employed in this paper. Accordingly, consistency checks on the expected or plausible patterns with regard to the three stated dimensions were done using published data from the 1976, 1986, 1997 and 2007 Swaziland censuses. As a caution or a limitation, reported or published data utilised could have been edited or manipulated, but the extent of such manipulation cannot be verified when raw data are unavailable (e.g. 1976 and 1986 censuses).
The population distribution, Myers’ index, Whipple’s index, age ratios and sex ratios were instrumental in checking for consistency or anomalies in the multiple census data. The assessment of the age and sex distribution for all the four censuses using the above-mentioned demographic tools yielded a number of observations.
The quality of age reporting in the 1976 and 1986 censuses is moderate or reasonable. In three decades, from 1976 to 2007, the quality of age data in Swaziland improved from rough to highly accurate using Whipple’s indices criteria. Both Whipple’s index and total modified Whipple’s index showed consistency in detecting age preference. The preference for ages ending with digit 0 and 5 including 8 for Swaziland decreased drastically over time. This marked decrease was seen also for digits 1, 3, and 7, which had the highest avoidance. No significant differentials in the age reporting between censuses were evident in the census data.
Overall, a consistent irregular pattern of age ratios and sex ratios in singles years of ages in all four censuses reflects age misreporting often due to understating or overstating of ages. Its effect on fertility is uncertain (United Nations 1983).
The jagged pattern of age ratios suggests age exaggeration or displacement. Affirming this is the possible underreporting of children as reflected in the age structure, which begins with an “L”, shaped pattern or depression for single-year ages. The age distribution in single years of population also confirms age misreporting through age heaping or digit preference. The lesser extent of digit preference for the recent 2007 and 1997 censuses compared to the earlier 1986 and 1976 censuses reflects an improvement in the quality of age reporting in the three decades. This may be linked to possibly many factors such as improvement in the education of the populace or data collection procedures in the context of Swaziland.
Swaziland appears to have lower sex ratios at birth (above 90 but less than 100) in all censuses suggesting an undercount of male births if not excess male mortality. This is uncommon in most world population but typical for some African countries (Garenne 2004; Moultrie et al. 2013).
The lower sex ratios at birth for Swaziland, much lower than 103, the average for Africans (Garenne 2004), are characteristic of Eastern and Southern African countries. Thus, according to Garenne (2004), a low value within the range of 0.87–1.03, as found in the censuses, shows good data.
Recent births and parity data collected from censuses often are limited or inaccurate requiring the use of corrective demographic techniques (United Nations 1983). Reported data on the average parities appears to be increasing with women’s age as expected in all censuses, although underreporting of births for older women is much poorer in 1976 in particular probably due to lower levels of education attained by women then. Similarly, the age pattern of fertility appears to be plausible showing a concave shape as expected with the exception of the 2007 census where a flat topped fertility pattern is observed.
However, applying the relational Gompertz technique, a modified P/F ratio method using the reported seemingly less fault current fertility and parity data shows fertility is underestimated in Swaziland. Thus, adjustment demographic methods when applied correctly are imperative in evaluating and adjusting demographic data as supported by Hill (1990). For older women aged 45–49 in all the censuses, the reported age-specific fertility rates were seemingly higher than those adjusted, which reflects faulty age reporting for older women.
Estimates of fertility may be flawed by certain degree of errors in reporting of ages and births. The analysis of results indicates some irregularities in age structure in all the four censuses of Swaziland. The irregularities in the age distribution reflected in the sex ratios may reflect net migration for young adults.
The extent of age reporting errors in census data is moderate as evaluated as shown in the several methods applied for the four census data and therefore estimates of fertility from the census data can be derived of reasonable quality. Hence, methods of fertility measurement using the reported parity-fertility data and age-sex population structure can be dependable when best methods and robust assumption on generating fertility estimates are made. However, their usefulness becomes questionable when age-sex structure and fertility data utilised are defective and incomplete, of which the study found otherwise.
Actual raw data were not available for the study, except for the 1997 census which was correctly verified with the published census data. The limitation on the published data is that further manipulation and analysis of data was limited. The available raw data requires that data documentation and archiving practice standards should be implemented addressing the processing of data such as on editing.
The authors acknowledge the Central Statistics Office (CSO) of Swaziland support for providing the census data and reports used in this study. The authors are grateful to the Department of Higher Education and Training grant of the Faculty of Humanities, University of the Witwatersrand, for the financial support to organise the writing retreat at which this manuscript was prepared. The support of the DST-NRF Centre of Excellence in Human Development towards the research is hereby acknowledged. Opinions expressed and conclusions arrived at are those of the authors and not necessarily to be attributed to either the Centre of Excellence in Human Development or CSO.
Open AccessThis 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.
- Blanc, A. K., & Rutstein, S. O. (1994). The demographic transition in Southern Africa: yet another look at the evidence from Botswana and Zimbabwe. Demography, 31(2), 209–215.View ArticleGoogle Scholar
- Brass, W. (1996). Demographic data analysis in less developed countries: 1946-1996. Population Studies, 50(3), 451–467.View ArticleGoogle Scholar
- Cleland, J. (1996). Demographic data collection in less developed countries. Population Studies, 50(3), 433–450.View ArticleGoogle Scholar
- Cohen, B. (1993). Fertility levels, differentials, and trends. In K. A. Foote, K. H. Hill, & L. G. Martin (Eds.), Demographic change in sub-Saharan Africa (pp. 8–67). Washington DC: National Academy Press.Google Scholar
- Cohen, B. (1998). The emerging fertility transition in sub-Saharan Africa. World Development, 26(8), 1431–1461.View ArticleGoogle Scholar
- Garenne, M. (2004). Sex ratios at birth in populations of Eastern and Southern Africa. Southern African Journal of Demography, 9(1), 91–96.Google Scholar
- Hill, K. (1990). Demographic estimation from deficient or defective data. In B. Fetter (Ed.), Demography from scanty evidence: central Africa in the colonial era (pp. 133–142). Boulder: L. Rienner Publishers.Google Scholar
- Joyner, K., Peters, H. E., Hynes, K., Sikora, A., Taber, J. R., & Rendall, M. S. (2012). The quality of male fertility data in major U.S. surveys. Demography, 49(1), 101–124.View ArticleGoogle Scholar
- Moultrie, T. A., & Timæus, I. M. (2002). Trends in South African fertility between 1970 and 1998: an analysis of the 1996 Census and the 1998 Demographic and Health Survey. http://www.mrc.ac.za/bod/trends.pdf. Accessed 21 May 2015.Google Scholar
- Moultrie, T. A., Dorrington, R. E., Hill, A., Hill, K., Timæus, I., & Zaba, B. (2013). Tools for demographic estimation. Paris: International Union for the Scientific Study of Population.Google Scholar
- National Research Council. (2004). The 2000 Census: Counting Under Adversity. Panel to Review the 2000 Census (Committee on National Statistics, Division of Behavioral and Social Sciences and Education). Washington: The National Academies Press.Google Scholar
- Phillips, H. (1999). Demographic survey data in South Africa: an evaluation of methodology and quality. Southern African Journal of Demography, 7(1), 1–10.Google Scholar
- Potter, J. E. (1977). Problems in using birth-history analysis to estimate trends in fertility. Population Studies, 31(2), 335–364.View ArticleGoogle Scholar
- Schoumaker, B. (2014). Quality and consistency of DHS fertility estimates, 1990 to 2012 (DHS Methodological Reports No. 12). Rockville: ICF International.Google Scholar
- Shryock, H. S., Siegel, J. S., et al. (1976). The methods and materials of demography. (Condensed ed., studies in population). New York: Academic Press, Inc.Google Scholar
- Spoorenberg, T. (2007). Quality of age reporting: extension and application of the modified Whipple’s Index. Population (English edition), 4(62), 729–741.View ArticleGoogle Scholar
- Spoorenberg, T. (2014). Reverse survival method of fertility estimation: an evaluation. Demographic Research, 31(9), 217–246.View ArticleGoogle Scholar
- Thomas, D., & Muvandi, I. (1994). The demographic transition in Southern Africa: reviewing the evidence from Botswana and Zimbabwe. Demography, 31(2), 217–227.View ArticleGoogle Scholar
- United Nations. (1983). Manual X: indirect techniques for demographic estimation. New York: United Nations.Google Scholar
- United Nations. (2004). Handbook on the collection of fertility and mortality data. http://unstats.un.org/unsd/demographic/standmeth/handbooks/Handbook_Fertility_Mortality.pdf. Accessed 21 May 2015.Google Scholar
- Vergauwen, J., Wood, J., De Wachter, D., & Neels, K. (2015). Quality of demographic data in GGS wave 1. Demographic Research, 32(24), 723–774.View ArticleGoogle Scholar