Child marriage (CM) is defined as a legal or customary union that occurs before the age of 18. In India, the minimum legal age of marriage is 18 for girls and 21 for boys. The Prohibition of Child Marriage Act, 2006 has closed loopholes which allowed CM and has made it a punishable offence.1 India is party to a range of international conventions outlawing child marriages, including several UN human rights conventions including the Convention on Consent to Marriage, Minimum Age for Marriage, and Registration of Marriages (1962). The accepted measure of the prevalence of CM is the proportion of women aged 20–24 who were married prior to 18.2 This is generally higher than the proportion of girls aged 18 or under, married at a particular point in time, due to underreporting for girls below the legal age of marriage.3
In part, because of its size and high prevalence in a number of poorer states, India, in spite of the laws, has the highest number of child marriages in the world, about 1.5 million each year.4 However, CM in India has fallen significantly from 47% in 2005–06 to 27% in 2015–16.5–7 The decline in CM is closely associated with improvements in girls’ education, transition of households to an improved standard of living and a decrease in average household size.8,9 Some of the decline may also be due to the number of interventions and government programs, even though few have been appropriately evaluated. The serious economic impact of the COVID 19 pandemic on India is likely to reverse these favourable trends in child marriage.
While a complex range of factors contribute to the continuing practice of CM, the most important are:
poverty,2,10 the average CM rate is 63% in the lowest quintile, whereas it is 10% in the highest. In Jumui in Bihar, the CM rate is 81% in the lowest quintile;11
social and cultural norms,2,9,12–14 such as the practice of dowry;9 and
truncated educational opportunities,9,11,15 girls discontinue education after marriage, mainly due to pressure from community, lack of permission from in-laws, and increasing household responsibilities and financial burden.15
There are significant differences in CM rates between the Indian states. As indicated, the reasons are complex, but a number, such as poverty and educational outcomes, are reflected in gross state product (GSP). Figure 1 shows the strong association between CM prevalence and GSP per capita in Indian states.
The theory of change16,17 on which our modelling framework is based, is that girls at risk of marriage benefit from improved educational and economic opportunities as alternatives to CM. The economic benefits of reduced CM arise from more productive employment opportunities because of improved education outcomes. The interventions to improve education outcomes are those designed to keep girls at school, reduce the dropout rate and extend the time at school to at least secondary school completion. These have the effect of delaying CM. Other interventions are aimed directly at reducing CM. By delaying marriage, these interventions help keep girls at school. These relationships are illustrated in Figure 2.
The aims of this paper are to identify the most effective interventions to reduce child marriage, to estimate their cost and impact for the case of India, and, using existing models18–20, to estimate the benefit-cost ratio for their application to India and their potential impact on CM.
Of the interventions considered, three have direct impacts on CM rates, while four have indirect impacts through the effect of educational interventions on school attendance, and hence on CM rates (Figure 2). We estimate the impact of reduced CM on educational outcomes, notably early dropout, years of schooling and completion of secondary schooling. We also estimate the economic benefit of better educational outcomes (such as higher productivity and access to better employment), leading to higher levels of gross domestic product (GDP) per capita. The results are brought together in a cost-benefit analysis.
Evidence of the costs and effectiveness of specific CM interventions was sought through a literature search of peer reviewed articles and grey literature (see Figure 3). The peer-reviewed literature search for articles is an extension of the search conducted in Rasmussen et al.21 Web of Science and PubMed were searched from 2006 to 2020 (English articles only). The terms were varied, with the initial search (child marriage OR girl marriage OR early marriage) and (reduc* OR prevent*) resulting in 1456 articles in Web of Science, and 2012 articles in PubMed. These results were refined to (impact* OR intervention* OR trial* OR evaluation*) resulting in 562 articles in Web of Science and 777 in PubMed, totaling 1339 articles. The two sets were combined in Endnote (Clarivate Analytics, version x9.3), and 370 duplicates were removed, leaving 949 articles. The titles and abstracts of these were reviewed and 21 full-text articles were chosen for the assessment.
A grey literature search for non peer-reviewed literature and relevant reports was conducted in Google Scholar, university library catalogs, and websites of relevant agencies, in particular international agencies (e.g. World Bank, UNICEF, UNFPA, Population Council) and research centers. We specifically sought evidence from the Indian sub-continent. Fifty-five reports were downloaded and investigated for intervention results or impact.
Citations in relevant peer-reviewed journal articles and grey literature were hand searched for further relevant literature. A final set of 3 journal articles and 16 reports were selected for consideration in the modeling because of their relevance to interventions and impact on child marriage (see Table 1).
Malhotra et al.13 identified five main effective strategies that included: life skills, community mobilisation, education incentives, conditional economic incentives and legal framework. Kalamar et al.22 ranked interventions according to their detail, rigor, design and included impact measurement, randomization, and pre and post comparisons. From Table 1, we selected intervention studies, with a preference for those from India, that included cost and effectiveness estimates and which met Kalamar et al.'s22 criteria and conformed to Malhotra et al.'s13 framework. Where there were no available data for India, and where possible, we drew on evidence from studies conducted in other countries. A more recent review of interventions by Malhotra and Elnakib,23 broadly confirmed the selection of interventions used in this study. This included life skills training, conditional asset transfers to delay marriage, supply-side education interventions and the creation of female-focused employment opportunities.
Discussed below are the specific CM and the education interventions used in the modelling. The CM interventions used in the modelling included only life skills and conditional economic incentives. We considered modelling the interventions for community mobilisation24 as suggested by Malhotra et al.,13 however, that indicated a low level of effectiveness, and accordingly was not included in the modelling.
Specific child marriage interventions
The 'Life Skills’ programs were represented by the Maharashtra program25,26 and the Youth Information Centres program27 in Bihar and Uttar Pradesh. Based on these two programs, we adopted an effectiveness rate of 40%. A 70% reduction in the marriage rate was achieved by the Maharashtra Life Skills program with an odds ratio of 4.0,26 although the likelihood of selection bias in these comparisons and other unobserved variables is acknowledged. Exposure to the Youth Information Centres program27 reduced CM compared with the control group by 56% (Adj. OR 2.25, CI 1.28–3.94).
For the cost of the life skills program, we used the average of the highly effective but relatively expensive Egyptian Ishraq program of $31.50 per girl, and the much cheaper but less effective life skills component of the Indian Deepshika program,4 about $4–6.50 per girl, which resulted in an average of $21.50 per girl.
For the conditional economic incentive interventions, we used the costs and effectiveness of the Kanyashree Prakalpa program,28 the only one evaluated in India, at a cost of $11.55 per girl and a effectiveness of 32.9%.
Education interventions, which also reduced CM rates, were derived from a meta-analysis.20 It measured the impact of education interventions to reduce secondary school dropout rates in terms of standard deviations. Only those showing an effect size in excess of 0.1 standard deviations for either learning improvement or dropout reduction were selected. The evidence suggested those which had a significant impact on CM20,43 were (with their standard deviations in brackets):
Increase provision of school in rural areas to give girls greater access to schools (S.D. 0.38 (ρ = 0.27)).
Improve educational infrastructure, e.g. provision of girls’ latrines (SD = 0.12 (ρ = 0.0)).
Pedagogical changes (SD = 0.13 (ρ = 0.004)).
Private public partnerships (SD = 0.15 (ρ = 0.136)).
The costs of the education interventions derived from Wils et al.20 are expressed as percentages of the base cost of Indian education programs. The costs are respectively 10%, 5% and 10% for points 1, 2, and 3 above, and a negligible cost for point 4 above.
In line with the theory of change,16,17 we developed a simulation/modelling study, which aimed to synthesize the available evidence on CM in India. The basic methodological approach followed our previous paper.21
Two Microsoft Excel models were used to undertake the cost-benefit analysis: a cost and outcomes model which generated education costs and other education outcomes; and a benefits model which forecast economic benefits from employment, GDP levels and productivity gains. The modelling compared two scenarios, a continuation of existing conditions described as ‘base scenario’, and an ‘intervention scenario’ which included the interventions discussed above. For the ‘base scenario’, the cost and outcomes model projected base education costs, as well as CM prevalence and education enrolments. The intervention scenario estimated the impact of the interventions on this base scenario from 2020. The benefit-cost ratios are calculated on the basis that the annual cost of the interventions increases progressively to 2030, thereafter remaining constant to 2050. The benefits are modelled to include productivity and employment gains until retirement for each age cohort.
To estimate the CM prevalence for 2020, we projected the rate from the 2016 National Family and Health Survey. To estimate the marriage rate for those aged 15–17 in 2020, we used the declining trend in the estimated single-year marriage rates from the Survey to project the rate for 2020. We estimated that the rate would have declined to 16.4% by 2020. We used this as the starting CM rate for the benefit-cost analysis.
For the intervention scenario, the models were run with the addition of the interventions discussed above. With the cost model, new estimates of education costs, student enrolments and child marriage prevalence were calculated, and the benefits model generated new estimates of productivity gains, employment levels and GDP, based on the improved education outcomes. The additional economic gains from the employment and productivity effects, arising from the interventions, were compared with the costs of the interventions, to enable the benefit-cost ratios to be calculated.18,20,21,44 The sensitivity of the benefit-cost ratios to different intervention cost and effectiveness assumptions was tested by varying each by +/– 10%.
These models made extensive use of international data sources: UNICEF,45 UNESCO Institute for Statistics (UIS), ILO, and the World Bank Development Indicators and EdStats database. Indian data came from the National Sample Survey (NSS), the Annual Status of Education Report and the District Information System for Education.
Reduction in child marriage rates
The application of all interventions reduced the marriage rate for those aged 15–17 from the base of 16.4% in 2020 by 7.5 percentage points in 2050 (excluding any further trend decline), to 8.9% (Table 2). Much of the impact was estimated to be achieved by 2030 with a reduction of 6.1 percentage points. The effect of the child marriage interventions, however, was modest with reductions of only 0.8 and 1.2 percentage points by 2030 and 2050 (Table 2).
This shows that the education inventions have a greater effect on the CM rate than the specific marriage interventions. However, we estimate that the net present value of costs to 2050 of the education interventions was six times larger, $243 billion, compared with $40 billion for the specific marriage interventions.
Regrettably, the severe effect of COVID appears likely to interrupt the downward trend in poverty and CM in India.46,47 We estimate that the effect of COVID 19 in 2020–21 will be to lift the estimated number of CMs by 179,000, an increase of almost 3%. This is based on the decline of 9% in per capita income for India in 2020–21 compared with a year earlier.48 Paul49 has shown that CM increases by 0.3% for 1% increase in poverty, meaning that the marriage rate would increase by 2.7 percentage points, representing an additional 155,000 CMs. In addition, the number of 15–17 year old girls who will now be poor (less than $2 per day), and a have a higher propensity to marry, is estimated to have increased from 6 million to 13.4 million. Based on our modelling, this is estimated to add a further 22,000 to the number married.
At this early stage, it is difficult to project the impact of COVID on the longer-term decline in CM in India and we have not attempted to do so. However, it would appear that a greater investment than modelled here will be required to achieve the level projected in this study.
Better schooling outcomes and increased productivity
Table 3 presents schooling and productivity effects for the education and specific CM interventions. The interventions increase the share of girls completing secondary education by 13.1 percentage points to 2030. The education interventions had the larger effect, shifting completions by 11.6 percentage points compared with 2.0 for the marriage interventions.
Improved schooling outcomes were assumed to increase productivity. It is assumed that each additional year of schooling provides an economic return by way of increased income, and secondary school completions increase the number of girls employed and the proportion engaged in formal employment.18 The immediate economic effect of increased schooling is negative because it withdraws girls from the workforce. This is later offset, as a higher proportion of the cohort enters the workforce in more productive roles. Table 3 shows the productivity changes, for the two productivity effects and the two sets of interventions to 2030. Overall, the productivity improvement for both sets of interventions is 16.4%, of which 14.8% is a result of the education interventions and 1.8% for the CM interventions.
The change in employment type (increased formal employment) had a larger effect on productivity, than did the additional years of schooling. The change in employment level and type had a productivity effect of 10.5%, compared with 5.4% of the additional years of schooling effect.
Our study calculated the benefit-cost ratios based on total employment benefits as a ratio of the total costs of the relevant education interventions and the specific CM ones discussed above. The modelling assumed that coverage of the interventions is progressively increased to reach an 11.3% target level by 2030.20 Both costs and benefits are discounted at 3%.
Table 4 shows relevant benefit-cost ratios for four education interventions and two specific CM interventions. The ratios for all interventions is 16.8, meaning that there are almost $17 in economic benefits for every dollar invested. This is very high. It would be lower if the benefits were to be evaluated over a more limited period. Including the benefits to retirement age reflects the fact that the benefits of additional education are transformational and last for their working lifetime. The benefit-cost ratio is 13.1 for the education interventions and 21.0 for the CM interventions. While productivity and schooling gains of the education interventions are larger, so is their cost.
The benefit-cost ratio remains high, 14.9, even when the intervention costs are increased by 10% and their effectiveness reduced by 10%. It increases to 18.9 if more favourable assumptions are adopted, a 10% reduction in intervention costs and an increase of 10% in effectiveness.
To explore the effect of regional differences, modelling was conducted for two states with contrasting economic circumstances, the relatively well-off Tamil Nadu and relatively poor Madhya Pradesh, illustrated in Figure 1. While Tamil Nadu had a secondary completion rate for girls of 79.4% in 2018, it was only 46.3% in Madhya Pradesh. However, by 2050, it was projected to increase by 32.3 percentage points for Madhya Pradesh to 79.1%, but only 11.1 percentage points to 90.6% for Tamil Nadu. While the benefit-cost ratios are the outcome of the interaction of many factors, there is a marked difference in the two ratios, 14.8 for Madhya Pradesh and 9.0 for Tamil Nadu, indicating the benefit of this relative outperformance by the poorer state catching up with an already well-performing better-endowed state.
As discussed, CM is the outcome of the complex interactions of many factors which include poverty, level of education, and social and cultural attitudes. While CM in India has on average declined sharply over the intercensal decade to 2016, progress has been very uneven between regions, rural and urban areas, income groups and education levels. This study indicates that ongoing interventions in education, and social and cultural attitudes are still a highly valuable investment in continuing this reduction in CM with an overall benefit-cost ratio of 16.8, meaning that there are almost $17 in economic benefits for every dollar invested.
In modelling the two sets of interventions separately, we show that the direct CM interventions deliver a higher benefit-cost ratio (21.0) than the education interventions (13.1). This is partly due to the relatively high costs of the education interventions compared with the CM ones. The cost of the education interventions range from $3,200 to $5,900 per girl, compared with the CM interventions which are in the range $12–$22 per girl. The education interventions which we modelled are all supply-side interventions to make education more attractive (closer, more girl-friendly schools with better-trained teachers). In contrast, the CM interventions increase demand by providing conditional non-cash incentives to delay marriage and life skills programs to empower girls to remain unmarried and therefore to stay longer at school.
While these results certainly support the expansion of CM intervention programs, they should not be undertaken at a cost to the education system, since the two are inter-dependent. The benefits of the CM interventions arise because they allow girls to stay at school. Ultimately, a quality and accessible education system is fundamental to providing the skills and training necessary to generate very large economic benefits. As Malhotra and Elnakib argue, ‘the enhancement of girls’ own human capital and opportunities is the most compelling pathway to delaying marriage’.23(p1) Together, the two sets of interventions act to discourage girls from dropping out of school and continuing in their studies to complete secondary school. They deliver large benefits not only in improved productivity for every year of additional schooling, but also in the opportunity to find higher paid jobs in the formal sector.50–54
In communities where girls are systematically excluded from participation in social, economic and political life, CM represents a serious human rights issue for individual girls. Delaying marriage and extending years at school have benefits that go beyond enhanced employment opportunities and higher incomes. There are other benefits, such as reduced fertility and improved health outcomes, not included in this study.55 Together with the employment benefits, these benefits are transformational for communities with high levels of poverty. Better-educated women with smaller families are better placed to break the intergenerational cycle of early marriage, limited education and low incomes.56,57
The context in which these interventions are being evaluated is undoubtedly important. Our analysis for the UNFPA showed that since 1990, trends in countries with high CM prevalence could be placed in three categories of almost equal size, as trending down, stuck after a downward shift or no change/increase.58(p91-113) The Indian context has been supportive to reducing CM and therefore is favourable to the effectiveness of our modelled interventions. However, the impact of COVID is to cause a significant retracement of poverty rates, making reductions in CM more difficult and stimulating the requirement for greater investment to continue the downward trend. Other countries which have demonstrated the effectiveness of these interventions, but which now face rising COVID-related poverty, may also find greater resistance to reductions in CM using these interventions and require increased investment in expanded programs to reduce CM.59
As with all modelling exercises, the results produced here depend on the assumptions made in specifying the relevant variables. Some causes of CM were not modelled. Moreover, the relationships between CM, education and employment outcomes are complex, and the direction of causation is often highly interdependent. It is not possible, given the limited evidence, to capture all these relationships.
In deriving the effectiveness and cost parameters from the CM intervention literature, we acknowledge that we are adopting an experimental approach in which the outcomes with and without the interventions are compared.60 While there are weaknesses in such an approach, one strength is that we are able to test the cost effectiveness of interventions based on the results of actual field experiments.60,61 A limitation of the approach is that some relevant interventions may not have been formally evaluated and we are unable to include their impact in our modelling. Offsetting that limitation is that those deemed most important tend to be those that have been evaluated.
Accordingly, the modelling attempts to incorporate the most important relationships based on the best understanding from the available evidence. Nonetheless, this analysis relies on a small number of studies, not all of them Indian. Furthermore, the results of these studies are broadly applied to contexts which may be very different from those where the results were produced. Even so, the benefit-cost ratios are very high permitting substantially higher costs or lower effectiveness in the implementation of the modelled interventions, without undermining the very advantageous economic outcomes from investing in reducing CM.
Modelling the impacts of education interventions and child marriage interventions on early marriage makes it possible to compare the value of the economic and social gains from reducing child marriage, with the costs of the interventions to do so. This study suggests that interventions that reduce child marriage through increased attendance at school and changing social attitudes to child marriage, are both socially important and economically valuable. While the knowledge of impact and costs remain imperfect, the benefit-cost ratios are robust for different intervention levels. The interventions generate economic and social benefits that are many times their costs, leaving a significant margin for error. While the COVID 19 pandemic has introduced new uncertainties into outcomes modelled in this paper, with the extent of the economic downturn yet to be realised, the pandemic can only have exacerbated the factors driving poor families to marry off their daughters. There is even more reason for the interventions identified in this paper to be implemented.
The authors are grateful for the advice and support of Howard Friedman, Venkatesh Srinivasan, Devender Singh and Shobhana Boyle of UNFPA.
The authors gratefully acknowledge funding support from the UN Populations Fund (UNFPA/IND/2018/003).
BR, NM and PJS conceptualized this paper and drafted it with contributions from SS, AK and RK. JS provided the data and modeling, and participated in the analysis. MK conducted the formal literature review strategy and did the overall edit. All authors reviewed the findings. All authors agreed with the final version of the paper.
The authors completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf (available upon request from the corresponding author), and declare no conflicts of interest.
Bruce Rasmussen, Victoria Institute of Strategic Economic Studies, Victoria University, PO Box 14428, Melbourne, Victoria, 8001, Australia.